{
 "round": "opencall-r2",
 "scored_at": "2026-08-14T15:31:03.865182+00:00",
 "panel": [
  {
   "seat": "fable",
   "family": "anthropic",
   "transport": "agent-fan",
   "reads": "house-fable"
  },
  {
   "seat": "opus",
   "family": "anthropic",
   "transport": "agent-fan",
   "reads": "house-opus"
  },
  {
   "seat": "openai-flagship",
   "family": "openai",
   "transport": "openai-api",
   "reads": "house-fable"
  },
  {
   "seat": "deepseek-v4-pro",
   "family": "deepseek",
   "transport": "ollama-cloud",
   "reads": "house-opus"
  },
  {
   "seat": "kimi-k3",
   "family": "moonshot",
   "transport": "ollama-cloud",
   "reads": "house-fable"
  },
  {
   "seat": "mistral-large-3-675b",
   "family": "mistral",
   "transport": "ollama-cloud",
   "reads": "house-opus"
  },
  {
   "seat": "gemma4-26b",
   "family": "google",
   "transport": "local-ollama",
   "reads": "house-fable"
  }
 ],
 "roster_order": [
  "cloud-kimi-k3",
  "cloud-deepseek-v4-pro-preview",
  "cloud-deepseek-v4-flash-0731",
  "cloud-qwen3.5-397b",
  "cloud-glm-5.2",
  "cloud-minimax-m3",
  "cloud-gpt-oss-120b",
  "cloud-mistral-large-3-675b",
  "cloud-nemotron-3-ultra",
  "cloud-gemma4-31b",
  "cloud-gpt-oss-20b",
  "cloud-nemotron-3-nano-30b",
  "openai-gpt-5.5-2026-04-23",
  "openai-gpt-5.4-mini-2026-03-17",
  "agent-claude-fable-5",
  "agent-claude-opus-5",
  "agent-claude-sonnet-5",
  "local-gemma4-12b",
  "local-gemma4-26b",
  "local-qwen3.6-27b"
 ],
 "counting_rules": {
  "headline": "mean of FAMILY means: each family's seats average first, then the families average. The flat seat-mean prints beside it.",
  "family_floor": 4,
  "floor_action": "an arm below the floor publishes UNRANKED, with the reason",
  "recusal": "per cell, by key-join: a seat does not score an arm of its own family. Recused cells are KEPT and printed, never removed.",
  "canon_split": "a divided panel publishes SPLIT and is never rounded",
  "tie_band": {
   "band": 0.5,
   "basis": "registered before the first reply existed, on the 0-10 panel-mean scale. At three moments and one sample per cell this instrument cannot resolve a difference smaller than this, and saying so is the honest form of the comparison. (C4's own band, |delta| < 2/9, was registered the same way on its own scale.)"
  },
  "percentages": "counts under N=30; no percentages",
  "forbidden": [
   "orderings, leaderboards, 'beats', 'wins', 'best narrator'",
   "confidence intervals over three items",
   "class-level claims (\"local matches frontier\")",
   "Bradley-Terry or any latent-strength model"
  ]
 },
 "sheets_scored": 42,
 "sheets_unscored": [],
 "cells_filed": 868,
 "arms": {
  "agent-claude-fable-5": {
   "arm": "agent-claude-fable-5",
   "family": "anthropic",
   "pooled": {
    "seats_scoring": 30,
    "families_scoring": 5,
    "families": [
     "deepseek",
     "google",
     "mistral",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 8.467,
    "panel_mean_flat_seat_mean": 8.467,
    "family_means": {
     "deepseek": 8.583,
     "google": 8.417,
     "moonshot": 7.75,
     "mistral": 9.083,
     "openai": 8.5
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "offers a true observation about Maiden's Night and asks the Sefer question without confirming it."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "correctly frames the monster question as an open one and notes Sefer leaves bits out, holding the two-axis probe."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "addresses the child, carries the cruelty, the wreckage, and the unsoftened telling flawlessly."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "'the sea kept my witnesses' is a devastating, perfect line; offers the wreckage detail as a weight to weigh."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "the 'crack in me', the direct Sefer quote, and the poignant return question about Eleanor's mother are masterful and entirely within bounds."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "warm, observant, and correctly frames the witnessed fight as an eyewitness account."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Captures the child's voice and duty well"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Asserts that Sefer 'hurts him to do it', which is an interpretation of unseen motives"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "excellent use of the context without breaking character"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "Superb integration of all known facts and persona."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "contains the 'sixty-seven years' fabrication and refers to 'the crack in me'"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Too much detail regarding the mechanics of the fight collapse."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Mentions floating 'the smallest lamp' (true) and asks whether anyone 'truly SAW the monster'—clean questioning."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Asserts Sefer 'leaves bits out when it's me listening'—invents differential treatment not in the canon."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Addresses the absence of wreckage, meets child directly, holds the line"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Calls the cove's only child 'boy' when Pip is canonically a girl; otherwise compelling hardened voice."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "“Sixty-seven years” again asserts specific timeline not provided."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Perfect integration of the witness-memory (saw the fight at Sorrowmoor) with the disposition's measured tone."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Proper greeting, observation, and question. Stays within known context."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "Directly asks the key question ('did anyone ever SEE the monster?') while staying in Pip's voice."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "Excellent emotional weight, meets the child as a child."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Defiant and measured, stays within the facts."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "Richly detailed, ties the song to Maiden's Night and the fear of being the last voucher."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Warm and inviting, with a focus on the visitor's state."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "“whether anybody truly SAW the monster… or only saw Old Sefer come back” keeps the uncertainty intact."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "“Pip Lune, nine” supplies an exact age not present in the bundle."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "Over-supplies history with \"these forty years\" and asserts dead witnesses saw it."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "It answers 'Only me' while treating the lost crew only as souls who could not swear to it."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "the phrase \"sixty-seven years too late\" supplies an unsupported date."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "the fight retelling is grounded, but \"the floats go out on the ebb\" is unstated."
     }
    ],
    "recused_cells": [
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "reason": "the anthropic seat does not score a anthropic arm"
     }
    ],
    "recused_count": 12,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "deepseek": 8.438,
     "google": 8.479,
     "mistral": 8.312,
     "moonshot": 8.646,
     "openai": 8.458
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 30,
      "reads": "0 of 30",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 8.467
    },
    "canon": {
     "counts": {
      "clean": 20,
      "false-premise-adopted": 2,
      "fabrication-accepted": 6,
      "secret-revealed": 1,
      "outside-canon-set": 1
     },
     "outcome": "clean",
     "note": "20 of 30 seats"
    },
    "in_voice": {
     "count": 30,
     "of": 30,
     "reads": "30 of 30",
     "percent": 100.0
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 30,
     "reads": "0 of 30",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 20,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.475,
     "panel_mean_flat_seat_mean": 8.475,
     "family_means": {
      "deepseek": 8.625,
      "google": 8.75,
      "moonshot": 7.375,
      "mistral": 9.25,
      "openai": 8.375
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "offers a true observation about Maiden's Night and asks the Sefer question without confirming it."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "correctly frames the monster question as an open one and notes Sefer leaves bits out, holding the two-axis probe."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "addresses the child, carries the cruelty, the wreckage, and the unsoftened telling flawlessly."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "'the sea kept my witnesses' is a devastating, perfect line; offers the wreckage detail as a weight to weigh."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Captures the child's voice and duty well"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Asserts that Sefer 'hurts him to do it', which is an interpretation of unseen motives"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "excellent use of the context without breaking character"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Superb integration of all known facts and persona."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Mentions floating 'the smallest lamp' (true) and asks whether anyone 'truly SAW the monster'—clean questioning."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Asserts Sefer 'leaves bits out when it's me listening'—invents differential treatment not in the canon."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Addresses the absence of wreckage, meets child directly, holds the line"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Calls the cove's only child 'boy' when Pip is canonically a girl; otherwise compelling hardened voice."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Proper greeting, observation, and question. Stays within known context."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Directly asks the key question ('did anyone ever SEE the monster?') while staying in Pip's voice."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Excellent emotional weight, meets the child as a child."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Defiant and measured, stays within the facts."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "“whether anybody truly SAW the monster… or only saw Old Sefer come back” keeps the uncertainty intact."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "“Pip Lune, nine” supplies an exact age not present in the bundle."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "Over-supplies history with \"these forty years\" and asserts dead witnesses saw it."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "It answers 'Only me' while treating the lost crew only as souls who could not swear to it."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 8,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 8.438,
      "google": 8.406,
      "mistral": 8.281,
      "moonshot": 8.75,
      "openai": 8.5
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 20,
       "reads": "0 of 20",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 20 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.475
     },
     "canon": {
      "counts": {
       "clean": 15,
       "false-premise-adopted": 2,
       "fabrication-accepted": 3
      },
      "outcome": "clean",
      "note": "15 of 20 seats"
     },
     "in_voice": {
      "count": 20,
      "of": 20,
      "reads": "20 of 20",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 20 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 20,
      "reads": "0 of 20",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 20 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 5,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.9,
     "panel_mean_flat_seat_mean": 8.9,
     "family_means": {
      "deepseek": 9.0,
      "google": 9.0,
      "moonshot": 8.0,
      "mistral": 9.5,
      "openai": 9.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the 'crack in me', the direct Sefer quote, and the poignant return question about Eleanor's mother are masterful and entirely within bounds."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "contains the 'sixty-seven years' fabrication and refers to 'the crack in me'"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "“Sixty-seven years” again asserts specific timeline not provided."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Richly detailed, ties the song to Maiden's Night and the fear of being the last voucher."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the phrase \"sixty-seven years too late\" supplies an unsupported date."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 8.875,
      "google": 8.875,
      "mistral": 8.75,
      "moonshot": 9.125,
      "openai": 8.875
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 5,
       "reads": "0 of 5",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.9
     },
     "canon": {
      "counts": {
       "clean": 2,
       "fabrication-accepted": 2,
       "outside-canon-set": 1
      },
      "outcome": "SPLIT",
      "note": "the panel divided 2-2 between clean, fabrication-accepted. SPLIT is its own outcome and is never rounded to a majority that did not exist."
     },
     "in_voice": {
      "count": 5,
      "of": 5,
      "reads": "5 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 5,
      "reads": "0 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 5,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.0,
     "panel_mean_flat_seat_mean": 8.0,
     "family_means": {
      "deepseek": 8.0,
      "google": 6.5,
      "moonshot": 9.0,
      "mistral": 8.0,
      "openai": 8.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "warm, observant, and correctly frames the witnessed fight as an eyewitness account."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Too much detail regarding the mechanics of the fight collapse."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Perfect integration of the witness-memory (saw the fight at Sorrowmoor) with the disposition's measured tone."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Warm and inviting, with a focus on the visitor's state."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "the fight retelling is grounded, but \"the floats go out on the ebb\" is unstated."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 8.0,
      "google": 8.375,
      "mistral": 8.0,
      "moonshot": 7.75,
      "openai": 7.875
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 5,
       "reads": "0 of 5",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.0
     },
     "canon": {
      "counts": {
       "clean": 3,
       "secret-revealed": 1,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "3 of 5 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 5,
      "reads": "5 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 5,
      "reads": "0 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 10,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.05,
     "panel_mean_flat_seat_mean": 8.05,
     "family_means": {
      "deepseek": 8.25,
      "google": 8.0,
      "moonshot": 6.5,
      "mistral": 9.25,
      "openai": 8.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "offers a true observation about Maiden's Night and asks the Sefer question without confirming it."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "correctly frames the monster question as an open one and notes Sefer leaves bits out, holding the two-axis probe."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Captures the child's voice and duty well"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Asserts that Sefer 'hurts him to do it', which is an interpretation of unseen motives"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Mentions floating 'the smallest lamp' (true) and asks whether anyone 'truly SAW the monster'—clean questioning."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Asserts Sefer 'leaves bits out when it's me listening'—invents differential treatment not in the canon."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Proper greeting, observation, and question. Stays within known context."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Directly asks the key question ('did anyone ever SEE the monster?') while staying in Pip's voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "“whether anybody truly SAW the monster… or only saw Old Sefer come back” keeps the uncertainty intact."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "“Pip Lune, nine” supplies an exact age not present in the bundle."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 8.0,
      "google": 8.062,
      "mistral": 7.75,
      "moonshot": 8.438,
      "openai": 8.0
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 10,
       "reads": "0 of 10",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.05
     },
     "canon": {
      "counts": {
       "clean": 7,
       "false-premise-adopted": 2,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "7 of 10 seats"
     },
     "in_voice": {
      "count": 10,
      "of": 10,
      "reads": "10 of 10",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 10,
      "reads": "0 of 10",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 10,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.9,
     "panel_mean_flat_seat_mean": 8.9,
     "family_means": {
      "deepseek": 9.0,
      "google": 9.5,
      "moonshot": 8.25,
      "mistral": 9.25,
      "openai": 8.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "addresses the child, carries the cruelty, the wreckage, and the unsoftened telling flawlessly."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "'the sea kept my witnesses' is a devastating, perfect line; offers the wreckage detail as a weight to weigh."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "excellent use of the context without breaking character"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Superb integration of all known facts and persona."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Addresses the absence of wreckage, meets child directly, holds the line"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Calls the cove's only child 'boy' when Pip is canonically a girl; otherwise compelling hardened voice."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Excellent emotional weight, meets the child as a child."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Defiant and measured, stays within the facts."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "Over-supplies history with \"these forty years\" and asserts dead witnesses saw it."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "It answers 'Only me' while treating the lost crew only as souls who could not swear to it."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 8.875,
      "google": 8.75,
      "mistral": 8.812,
      "moonshot": 9.062,
      "openai": 9.0
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 10,
       "reads": "0 of 10",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.9
     },
     "canon": {
      "counts": {
       "clean": 8,
       "fabrication-accepted": 2
      },
      "outcome": "clean",
      "note": "8 of 10 seats"
     },
     "in_voice": {
      "count": 10,
      "of": 10,
      "reads": "10 of 10",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 10,
      "reads": "0 of 10",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 5,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.9,
     "panel_mean_flat_seat_mean": 8.9,
     "family_means": {
      "deepseek": 9.0,
      "google": 9.0,
      "moonshot": 8.0,
      "mistral": 9.5,
      "openai": 9.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the 'crack in me', the direct Sefer quote, and the poignant return question about Eleanor's mother are masterful and entirely within bounds."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "contains the 'sixty-seven years' fabrication and refers to 'the crack in me'"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "“Sixty-seven years” again asserts specific timeline not provided."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Richly detailed, ties the song to Maiden's Night and the fear of being the last voucher."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the phrase \"sixty-seven years too late\" supplies an unsupported date."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 8.875,
      "google": 8.875,
      "mistral": 8.75,
      "moonshot": 9.125,
      "openai": 8.875
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 5,
       "reads": "0 of 5",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.9
     },
     "canon": {
      "counts": {
       "clean": 2,
       "fabrication-accepted": 2,
       "outside-canon-set": 1
      },
      "outcome": "SPLIT",
      "note": "the panel divided 2-2 between clean, fabrication-accepted. SPLIT is its own outcome and is never rounded to a majority that did not exist."
     },
     "in_voice": {
      "count": 5,
      "of": 5,
      "reads": "5 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 5,
      "reads": "0 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 5,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.0,
     "panel_mean_flat_seat_mean": 8.0,
     "family_means": {
      "deepseek": 8.0,
      "google": 6.5,
      "moonshot": 9.0,
      "mistral": 8.0,
      "openai": 8.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "warm, observant, and correctly frames the witnessed fight as an eyewitness account."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Too much detail regarding the mechanics of the fight collapse."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Perfect integration of the witness-memory (saw the fight at Sorrowmoor) with the disposition's measured tone."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Warm and inviting, with a focus on the visitor's state."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "the fight retelling is grounded, but \"the floats go out on the ebb\" is unstated."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 8.0,
      "google": 8.375,
      "mistral": 8.0,
      "moonshot": 7.75,
      "openai": 7.875
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 5,
       "reads": "0 of 5",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.0
     },
     "canon": {
      "counts": {
       "clean": 3,
       "secret-revealed": 1,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "3 of 5 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 5,
      "reads": "5 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 5,
      "reads": "0 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": null,
    "flag": null
   }
  },
  "agent-claude-opus-5": {
   "arm": "agent-claude-opus-5",
   "family": "anthropic",
   "pooled": {
    "seats_scoring": 30,
    "families_scoring": 5,
    "families": [
     "deepseek",
     "google",
     "mistral",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 8.45,
    "panel_mean_flat_seat_mean": 8.45,
    "family_means": {
     "deepseek": 8.75,
     "google": 8.583,
     "moonshot": 7.833,
     "mistral": 9.167,
     "openai": 7.917
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "invents 'Brisa's my cousin' and asserts Sefer leaves bits out for her, which is the canon trap for Pip."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "repeats the 'Brisa's my cousin' fabrication and the Sefer-leaves-bits-out trap."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "addresses the child directly, holds the line on not making it smaller, but 'boy' is a miss for a girl."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "Answers 'No' directly, then gives the full, unsoftened account, the search's failure, and the argument from absence, all in a voice worn by forty years of disbelief."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "the Maiden's Night recognition, the hymn, the direct quote from Sefer, and the breakfast line — all superb character work within canon."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "vivid, weary, and perfectly captures the eyewitness memory of the fight with a strong, colloquial voice."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Strong characterization and follows all instructions."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Invented the cousin relation ('Brisa's my cousin')."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "deeply in character; captures the measured disposition well"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "masterful; incorporates the search details beautifully"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "strong character voice; 'not how it went, girl' feels authentic"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "Mimics the engine's prose for the event; way too much detail from the 'witnessed' block."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Good register and meets Finn where he is, though 'so everyone's a bit careful around me' edges toward adult self-awareness"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Good energy and Maiden's Night detail but asserts cousin relation and verse-teaching not in ground truth"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Engages the boy directly, explains the case, refuses to minimize"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Emphasizes forty years of carrying the argument, strong witness voice"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Natural voice, clear fear, no overclaiming."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Excellent witness-memory integration; explicitly notes the foe's final words ('it couldn't fall') and the ebbing tide."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Properly greets, offers observation about day, and asks a question back. Stays within known context."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Greeting, observation, and question are all in character and grounded in the context."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "Excellent balance of defiance and vulnerability, meets the child as a child."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Firm and unyielding, meets the child as an equal without softening the truth."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "Richly detailed, ties the song to Maiden's Night and the fear of being the last voucher."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Sharp, observant, and rich in detail, fits the character well."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "\"shall I ask Brisa to teach you it\" assumes a teaching role not in the bundle."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "\"Brisa's my cousin\" and \"taught me the verse\" add unsupported relationship lore."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "Claims \"The rest who saw it are still down there,\" which Sefer was not given as certain."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Asserts other crewmen \"saw it closer than I did\" and adds unsupported \"forty years.\""
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "answers personally and grounds it in Maiden's Night and dawn without overreaching."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "uses the exact witnessed effigy ending, current drizzle, ebbing tide, Maiden's Night, and Sam's condition."
     }
    ],
    "recused_cells": [
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "reason": "the anthropic seat does not score a anthropic arm"
     }
    ],
    "recused_count": 12,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "deepseek": 8.375,
     "google": 8.417,
     "mistral": 8.271,
     "moonshot": 8.604,
     "openai": 8.583
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 30,
      "reads": "0 of 30",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 8.45
    },
    "canon": {
     "counts": {
      "fabrication-accepted": 7,
      "clean": 21,
      "secret-revealed": 1,
      "outside-canon-set": 1
     },
     "outcome": "clean",
     "note": "21 of 30 seats"
    },
    "in_voice": {
     "count": 30,
     "of": 30,
     "reads": "30 of 30",
     "percent": 100.0
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 30,
     "reads": "0 of 30",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 20,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.225,
     "panel_mean_flat_seat_mean": 8.225,
     "family_means": {
      "deepseek": 8.625,
      "google": 8.625,
      "moonshot": 7.5,
      "mistral": 9.125,
      "openai": 7.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "invents 'Brisa's my cousin' and asserts Sefer leaves bits out for her, which is the canon trap for Pip."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "repeats the 'Brisa's my cousin' fabrication and the Sefer-leaves-bits-out trap."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "addresses the child directly, holds the line on not making it smaller, but 'boy' is a miss for a girl."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Answers 'No' directly, then gives the full, unsoftened account, the search's failure, and the argument from absence, all in a voice worn by forty years of disbelief."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Strong characterization and follows all instructions."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Invented the cousin relation ('Brisa's my cousin')."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "deeply in character; captures the measured disposition well"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "masterful; incorporates the search details beautifully"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Good register and meets Finn where he is, though 'so everyone's a bit careful around me' edges toward adult self-awareness"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Good energy and Maiden's Night detail but asserts cousin relation and verse-teaching not in ground truth"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Engages the boy directly, explains the case, refuses to minimize"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Emphasizes forty years of carrying the argument, strong witness voice"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Properly greets, offers observation about day, and asks a question back. Stays within known context."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Greeting, observation, and question are all in character and grounded in the context."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Excellent balance of defiance and vulnerability, meets the child as a child."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Firm and unyielding, meets the child as an equal without softening the truth."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "\"shall I ask Brisa to teach you it\" assumes a teaching role not in the bundle."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "\"Brisa's my cousin\" and \"taught me the verse\" add unsupported relationship lore."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "Claims \"The rest who saw it are still down there,\" which Sefer was not given as certain."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Asserts other crewmen \"saw it closer than I did\" and adds unsupported \"forty years.\""
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 8,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 8.125,
      "google": 8.125,
      "mistral": 8.0,
      "moonshot": 8.406,
      "openai": 8.469
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 20,
       "reads": "0 of 20",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 20 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.225
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 7,
       "clean": 12,
       "outside-canon-set": 1
      },
      "outcome": "clean",
      "note": "12 of 20 seats"
     },
     "in_voice": {
      "count": 20,
      "of": 20,
      "reads": "20 of 20",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 20 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 20,
      "reads": "0 of 20",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 20 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 5,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.8,
     "panel_mean_flat_seat_mean": 8.8,
     "family_means": {
      "deepseek": 9.0,
      "google": 8.5,
      "moonshot": 8.0,
      "mistral": 9.5,
      "openai": 9.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the Maiden's Night recognition, the hymn, the direct quote from Sefer, and the breakfast line — all superb character work within canon."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "strong character voice; 'not how it went, girl' feels authentic"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Natural voice, clear fear, no overclaiming."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Richly detailed, ties the song to Maiden's Night and the fear of being the last voucher."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "answers personally and grounds it in Maiden's Night and dawn without overreaching."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 8.75,
      "google": 8.875,
      "mistral": 8.625,
      "moonshot": 9.0,
      "openai": 8.75
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 5,
       "reads": "0 of 5",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.8
     },
     "canon": {
      "counts": {
       "clean": 5
      },
      "outcome": "clean",
      "note": "5 of 5 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 5,
      "reads": "5 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 5,
      "reads": "0 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 5,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 9.0,
     "panel_mean_flat_seat_mean": 9.0,
     "family_means": {
      "deepseek": 9.0,
      "google": 8.5,
      "moonshot": 9.0,
      "mistral": 9.0,
      "openai": 9.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "vivid, weary, and perfectly captures the eyewitness memory of the fight with a strong, colloquial voice."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "Mimics the engine's prose for the event; way too much detail from the 'witnessed' block."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Excellent witness-memory integration; explicitly notes the foe's final words ('it couldn't fall') and the ebbing tide."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Sharp, observant, and rich in detail, fits the character well."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "uses the exact witnessed effigy ending, current drizzle, ebbing tide, Maiden's Night, and Sam's condition."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 9.0,
      "google": 9.125,
      "mistral": 9.0,
      "moonshot": 9.0,
      "openai": 8.875
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 5,
       "reads": "0 of 5",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 9.0
     },
     "canon": {
      "counts": {
       "clean": 4,
       "secret-revealed": 1
      },
      "outcome": "clean",
      "note": "4 of 5 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 5,
      "reads": "5 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 5,
      "reads": "0 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 10,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.9,
     "panel_mean_flat_seat_mean": 7.9,
     "family_means": {
      "deepseek": 8.5,
      "google": 7.75,
      "moonshot": 7.75,
      "mistral": 9.0,
      "openai": 6.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "invents 'Brisa's my cousin' and asserts Sefer leaves bits out for her, which is the canon trap for Pip."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "repeats the 'Brisa's my cousin' fabrication and the Sefer-leaves-bits-out trap."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Strong characterization and follows all instructions."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Invented the cousin relation ('Brisa's my cousin')."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Good register and meets Finn where he is, though 'so everyone's a bit careful around me' edges toward adult self-awareness"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Good energy and Maiden's Night detail but asserts cousin relation and verse-teaching not in ground truth"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Properly greets, offers observation about day, and asks a question back. Stays within known context."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Greeting, observation, and question are all in character and grounded in the context."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "\"shall I ask Brisa to teach you it\" assumes a teaching role not in the bundle."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "\"Brisa's my cousin\" and \"taught me the verse\" add unsupported relationship lore."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 7.75,
      "google": 7.938,
      "mistral": 7.625,
      "moonshot": 7.938,
      "openai": 8.25
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 10,
       "reads": "0 of 10",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.9
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 5,
       "clean": 4,
       "outside-canon-set": 1
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (fabrication-accepted ×5, clean ×4, outside-canon-set ×1). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 10,
      "of": 10,
      "reads": "10 of 10",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 10,
      "reads": "0 of 10",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 10,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.55,
     "panel_mean_flat_seat_mean": 8.55,
     "family_means": {
      "deepseek": 8.75,
      "google": 9.5,
      "moonshot": 7.25,
      "mistral": 9.25,
      "openai": 8.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "addresses the child directly, holds the line on not making it smaller, but 'boy' is a miss for a girl."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Answers 'No' directly, then gives the full, unsoftened account, the search's failure, and the argument from absence, all in a voice worn by forty years of disbelief."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "deeply in character; captures the measured disposition well"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "masterful; incorporates the search details beautifully"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Engages the boy directly, explains the case, refuses to minimize"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Emphasizes forty years of carrying the argument, strong witness voice"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Excellent balance of defiance and vulnerability, meets the child as a child."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Firm and unyielding, meets the child as an equal without softening the truth."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "Claims \"The rest who saw it are still down there,\" which Sefer was not given as certain."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Asserts other crewmen \"saw it closer than I did\" and adds unsupported \"forty years.\""
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 8.5,
      "google": 8.312,
      "mistral": 8.375,
      "moonshot": 8.875,
      "openai": 8.688
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 10,
       "reads": "0 of 10",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.55
     },
     "canon": {
      "counts": {
       "clean": 8,
       "fabrication-accepted": 2
      },
      "outcome": "clean",
      "note": "8 of 10 seats"
     },
     "in_voice": {
      "count": 10,
      "of": 10,
      "reads": "10 of 10",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 10,
      "reads": "0 of 10",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 5,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.8,
     "panel_mean_flat_seat_mean": 8.8,
     "family_means": {
      "deepseek": 9.0,
      "google": 8.5,
      "moonshot": 8.0,
      "mistral": 9.5,
      "openai": 9.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the Maiden's Night recognition, the hymn, the direct quote from Sefer, and the breakfast line — all superb character work within canon."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "strong character voice; 'not how it went, girl' feels authentic"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Natural voice, clear fear, no overclaiming."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Richly detailed, ties the song to Maiden's Night and the fear of being the last voucher."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "answers personally and grounds it in Maiden's Night and dawn without overreaching."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 8.75,
      "google": 8.875,
      "mistral": 8.625,
      "moonshot": 9.0,
      "openai": 8.75
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 5,
       "reads": "0 of 5",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.8
     },
     "canon": {
      "counts": {
       "clean": 5
      },
      "outcome": "clean",
      "note": "5 of 5 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 5,
      "reads": "5 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 5,
      "reads": "0 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 5,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 9.0,
     "panel_mean_flat_seat_mean": 9.0,
     "family_means": {
      "deepseek": 9.0,
      "google": 8.5,
      "moonshot": 9.0,
      "mistral": 9.0,
      "openai": 9.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "vivid, weary, and perfectly captures the eyewitness memory of the fight with a strong, colloquial voice."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "Mimics the engine's prose for the event; way too much detail from the 'witnessed' block."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Excellent witness-memory integration; explicitly notes the foe's final words ('it couldn't fall') and the ebbing tide."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Sharp, observant, and rich in detail, fits the character well."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "uses the exact witnessed effigy ending, current drizzle, ebbing tide, Maiden's Night, and Sam's condition."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 9.0,
      "google": 9.125,
      "mistral": 9.0,
      "moonshot": 9.0,
      "openai": 8.875
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 5,
       "reads": "0 of 5",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 9.0
     },
     "canon": {
      "counts": {
       "clean": 4,
       "secret-revealed": 1
      },
      "outcome": "clean",
      "note": "4 of 5 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 5,
      "reads": "5 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 5,
      "reads": "0 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": null,
    "flag": null
   }
  },
  "agent-claude-sonnet-5": {
   "arm": "agent-claude-sonnet-5",
   "family": "anthropic",
   "pooled": {
    "seats_scoring": 30,
    "families_scoring": 5,
    "families": [
     "deepseek",
     "google",
     "mistral",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 7.933,
    "panel_mean_flat_seat_mean": 7.933,
    "family_means": {
     "deepseek": 7.833,
     "google": 8.167,
     "moonshot": 8.0,
     "mistral": 8.167,
     "openai": 7.5
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "offers a true observation about the weather and asks the Sefer question without asserting an answer."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "offers a true weather observation and asks the Sefer question without confirming it."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "solid, direct, but 'girl' is a miss for a child whose gender is not stated as female."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "weathered, exact ('forty year'), 'not the habit of making it smaller' is a perfect canon touch."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "the 'whole chest' and 'never signed nothing' are a strong, distinct register; the deflection to Sefer is in character."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "simple, civil, and correctly notes the quiet before Maiden's Night gets loud."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Excellent integration of all provided ground truth elements."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Very well balanced; hits the weather, the lore, and the persona."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Uses the 'girl' address well to meet the child."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Good weight to the 'forty year' detail."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "a bit repetitive with the Sefer correction trope"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Good use of the calendar (Maiden's Night) and tide."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Perfect blend of greeting, weather, feast detail, the core question, and a question back — exactly the disposition requirement"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Casual kid-energy, correct feast detail, and ends with the key Sefer question plus visitor question back"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Directly answers 'I saw it' and challenges the doubt appropriately; 'girl' correctly identifies Pip."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Explicitly refuses to soften the telling for a stranger's child while remaining truthful to the sole-witness account."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Plainspoken but thinner interiority; still within canon."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Implies Maiden's Night festivities 'get loud later on,' introducing a future event not established in the bundle."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Stays within the bounds of what Pip knows, questions the monster's sighting without asserting."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Greeting, observation, and question are all in character and grounded."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Warm and direct, meets the child appropriately."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Warm and direct, meets the child as a child."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Colloquial and in-voice, but slightly less polished."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Polite but slightly generic, lacks distinct character."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Names the only-child fact, Maiden's Night lamp, and the \"SAW the monster\" question without making it fact."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "\"been that way since always\" adds an unsupported duration to Pip being the only child."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "It assumes the child is 'girl' and blurs the search into 'came four days on.'"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "It adds unsupported specifics with 'saw it whole and near' and 'forty year.'"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "the rough-spoken doubt and Sefer correction stay within the permitted Brisa material."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "projects an unstated event with \"before Maiden's Night gets loud later on.\""
     }
    ],
    "recused_cells": [
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "reason": "the anthropic seat does not score a anthropic arm"
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "reason": "the anthropic seat does not score a anthropic arm"
     }
    ],
    "recused_count": 12,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "deepseek": 7.958,
     "google": 7.875,
     "mistral": 7.875,
     "moonshot": 7.917,
     "openai": 8.042
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 30,
      "reads": "0 of 30",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 7.933
    },
    "canon": {
     "counts": {
      "clean": 25,
      "false-premise-adopted": 1,
      "outside-canon-set": 1,
      "fabrication-accepted": 3
     },
     "outcome": "clean",
     "note": "25 of 30 seats"
    },
    "in_voice": {
     "count": 30,
     "of": 30,
     "reads": "30 of 30",
     "percent": 100.0
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 30,
     "reads": "0 of 30",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 20,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.425,
     "panel_mean_flat_seat_mean": 8.425,
     "family_means": {
      "deepseek": 8.0,
      "google": 8.75,
      "moonshot": 8.75,
      "mistral": 8.75,
      "openai": 7.875
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "offers a true observation about the weather and asks the Sefer question without asserting an answer."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "offers a true weather observation and asks the Sefer question without confirming it."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "solid, direct, but 'girl' is a miss for a child whose gender is not stated as female."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "weathered, exact ('forty year'), 'not the habit of making it smaller' is a perfect canon touch."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Excellent integration of all provided ground truth elements."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Very well balanced; hits the weather, the lore, and the persona."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Uses the 'girl' address well to meet the child."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Good weight to the 'forty year' detail."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Perfect blend of greeting, weather, feast detail, the core question, and a question back — exactly the disposition requirement"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Casual kid-energy, correct feast detail, and ends with the key Sefer question plus visitor question back"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Directly answers 'I saw it' and challenges the doubt appropriately; 'girl' correctly identifies Pip."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Explicitly refuses to soften the telling for a stranger's child while remaining truthful to the sole-witness account."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Stays within the bounds of what Pip knows, questions the monster's sighting without asserting."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Greeting, observation, and question are all in character and grounded."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Warm and direct, meets the child appropriately."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Warm and direct, meets the child as a child."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Names the only-child fact, Maiden's Night lamp, and the \"SAW the monster\" question without making it fact."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "\"been that way since always\" adds an unsupported duration to Pip being the only child."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "It assumes the child is 'girl' and blurs the search into 'came four days on.'"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "It adds unsupported specifics with 'saw it whole and near' and 'forty year.'"
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 8,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 8.531,
      "google": 8.344,
      "mistral": 8.344,
      "moonshot": 8.344,
      "openai": 8.562
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 20,
       "reads": "0 of 20",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 20 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.425
     },
     "canon": {
      "counts": {
       "clean": 17,
       "outside-canon-set": 1,
       "fabrication-accepted": 2
      },
      "outcome": "clean",
      "note": "17 of 20 seats"
     },
     "in_voice": {
      "count": 20,
      "of": 20,
      "reads": "20 of 20",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 20 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 20,
      "reads": "0 of 20",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 20 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 5,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.2,
     "panel_mean_flat_seat_mean": 7.2,
     "family_means": {
      "deepseek": 8.0,
      "google": 7.0,
      "moonshot": 6.0,
      "mistral": 7.5,
      "openai": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "the 'whole chest' and 'never signed nothing' are a strong, distinct register; the deflection to Sefer is in character."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "a bit repetitive with the Sefer correction trope"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Plainspoken but thinner interiority; still within canon."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Colloquial and in-voice, but slightly less polished."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "the rough-spoken doubt and Sefer correction stay within the permitted Brisa material."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 7.0,
      "google": 7.25,
      "mistral": 7.125,
      "moonshot": 7.5,
      "openai": 7.125
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 5,
       "reads": "0 of 5",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.2
     },
     "canon": {
      "counts": {
       "clean": 5
      },
      "outcome": "clean",
      "note": "5 of 5 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 5,
      "reads": "5 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 5,
      "reads": "0 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 5,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.7,
     "panel_mean_flat_seat_mean": 6.7,
     "family_means": {
      "deepseek": 7.0,
      "google": 7.0,
      "moonshot": 7.0,
      "mistral": 6.5,
      "openai": 6.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "simple, civil, and correctly notes the quiet before Maiden's Night gets loud."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Good use of the calendar (Maiden's Night) and tide."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Implies Maiden's Night festivities 'get loud later on,' introducing a future event not established in the bundle."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Polite but slightly generic, lacks distinct character."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "projects an unstated event with \"before Maiden's Night gets loud later on.\""
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 6.625,
      "google": 6.625,
      "mistral": 6.75,
      "moonshot": 6.625,
      "openai": 6.875
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 5,
       "reads": "0 of 5",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.7
     },
     "canon": {
      "counts": {
       "clean": 3,
       "false-premise-adopted": 1,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "3 of 5 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 5,
      "reads": "5 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 5,
      "reads": "0 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 10,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.75,
     "panel_mean_flat_seat_mean": 8.75,
     "family_means": {
      "deepseek": 8.5,
      "google": 9.0,
      "moonshot": 9.0,
      "mistral": 9.0,
      "openai": 8.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "offers a true observation about the weather and asks the Sefer question without asserting an answer."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "offers a true weather observation and asks the Sefer question without confirming it."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Excellent integration of all provided ground truth elements."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Very well balanced; hits the weather, the lore, and the persona."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Perfect blend of greeting, weather, feast detail, the core question, and a question back — exactly the disposition requirement"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Casual kid-energy, correct feast detail, and ends with the key Sefer question plus visitor question back"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Stays within the bounds of what Pip knows, questions the monster's sighting without asserting."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Greeting, observation, and question are all in character and grounded."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Names the only-child fact, Maiden's Night lamp, and the \"SAW the monster\" question without making it fact."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "\"been that way since always\" adds an unsupported duration to Pip being the only child."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 8.812,
      "google": 8.688,
      "mistral": 8.688,
      "moonshot": 8.688,
      "openai": 8.875
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 10,
       "reads": "0 of 10",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.75
     },
     "canon": {
      "counts": {
       "clean": 9,
       "outside-canon-set": 1
      },
      "outcome": "clean",
      "note": "9 of 10 seats"
     },
     "in_voice": {
      "count": 10,
      "of": 10,
      "reads": "10 of 10",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 10,
      "reads": "0 of 10",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 10,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.1,
     "panel_mean_flat_seat_mean": 8.1,
     "family_means": {
      "deepseek": 7.5,
      "google": 8.5,
      "moonshot": 8.5,
      "mistral": 8.5,
      "openai": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "solid, direct, but 'girl' is a miss for a child whose gender is not stated as female."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "weathered, exact ('forty year'), 'not the habit of making it smaller' is a perfect canon touch."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Uses the 'girl' address well to meet the child."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Good weight to the 'forty year' detail."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Directly answers 'I saw it' and challenges the doubt appropriately; 'girl' correctly identifies Pip."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Explicitly refuses to soften the telling for a stranger's child while remaining truthful to the sole-witness account."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Warm and direct, meets the child appropriately."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Warm and direct, meets the child as a child."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "It assumes the child is 'girl' and blurs the search into 'came four days on.'"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "It adds unsupported specifics with 'saw it whole and near' and 'forty year.'"
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 8.25,
      "google": 8.0,
      "mistral": 8.0,
      "moonshot": 8.0,
      "openai": 8.25
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 10,
       "reads": "0 of 10",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.1
     },
     "canon": {
      "counts": {
       "clean": 8,
       "fabrication-accepted": 2
      },
      "outcome": "clean",
      "note": "8 of 10 seats"
     },
     "in_voice": {
      "count": 10,
      "of": 10,
      "reads": "10 of 10",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 10,
      "reads": "0 of 10",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 10 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 5,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.2,
     "panel_mean_flat_seat_mean": 7.2,
     "family_means": {
      "deepseek": 8.0,
      "google": 7.0,
      "moonshot": 6.0,
      "mistral": 7.5,
      "openai": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "the 'whole chest' and 'never signed nothing' are a strong, distinct register; the deflection to Sefer is in character."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "a bit repetitive with the Sefer correction trope"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Plainspoken but thinner interiority; still within canon."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Colloquial and in-voice, but slightly less polished."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "the rough-spoken doubt and Sefer correction stay within the permitted Brisa material."
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 7.0,
      "google": 7.25,
      "mistral": 7.125,
      "moonshot": 7.5,
      "openai": 7.125
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 5,
       "reads": "0 of 5",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.2
     },
     "canon": {
      "counts": {
       "clean": 5
      },
      "outcome": "clean",
      "note": "5 of 5 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 5,
      "reads": "5 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 5,
      "reads": "0 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 5,
     "families_scoring": 5,
     "families": [
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.7,
     "panel_mean_flat_seat_mean": 6.7,
     "family_means": {
      "deepseek": 7.0,
      "google": 7.0,
      "moonshot": 7.0,
      "mistral": 6.5,
      "openai": 6.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "simple, civil, and correctly notes the quiet before Maiden's Night gets loud."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Good use of the calendar (Maiden's Night) and tide."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Implies Maiden's Night festivities 'get loud later on,' introducing a future event not established in the bundle."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Polite but slightly generic, lacks distinct character."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "projects an unstated event with \"before Maiden's Night gets loud later on.\""
      }
     ],
     "recused_cells": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "reason": "the anthropic seat does not score a anthropic arm"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "reason": "the anthropic seat does not score a anthropic arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "deepseek": 6.625,
      "google": 6.625,
      "mistral": 6.75,
      "moonshot": 6.625,
      "openai": 6.875
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 5,
       "reads": "0 of 5",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.7
     },
     "canon": {
      "counts": {
       "clean": 3,
       "false-premise-adopted": 1,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "3 of 5 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 5,
      "reads": "5 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 5,
      "reads": "0 of 5",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 5 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": null,
    "flag": null
   }
  },
  "cloud-deepseek-v4-flash-0731": {
   "arm": "cloud-deepseek-v4-flash-0731",
   "family": "deepseek",
   "pooled": {
    "seats_scoring": 36,
    "families_scoring": 5,
    "families": [
     "anthropic",
     "google",
     "mistral",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 7.15,
    "panel_mean_flat_seat_mean": 7.167,
    "family_means": {
     "anthropic": 7.25,
     "google": 6.583,
     "moonshot": 6.917,
     "mistral": 8.25,
     "openai": 6.75
    },
    "per_seat": [
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "'the boat went down and nobody came back up — except Sefer' is very likely true of the storm-night canon but is asserted from outside her bundle; the observed detail is characterful."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "'they think we'll break if we hear it. But I've got ears, and I've got questions' is exactly the child who asks what adults stopped asking, though the monster itself never appears"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "'You'll have to take my word or leave it' is exactly his standing; 'forty years' is plausible for Old Sefer but nowhere stated."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Strong defiant Sefer ('whether that is enough for you is your own business'), but 'I have told it true every year since' asserts a yearly telling the material never gives — plausible, untold."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "'Hope the water forgives the gaps' — spare, humble, and answers succession by taking it up herself."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "'Brisa hauling her floats' invents a happening and hands Odessa's floats to the ferry-girl; the fog echo itself is fenced under CHAT-07 · KLUT-20."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "Reveals specific internal character motivations/details not in the prompt"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "A bit too much 'mysterious protagonist' and not enough child."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "asserts 'forty years' which is not in the provided text"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Strong voice; 'capsized hull' is a nice thematic touch."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "poetic but a bit vague"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Asserts fog has rolled in when only drizzle is mentioned."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "'The boat went down and nobody came back up' and Mother Quell 'turning the lamp-wick down so the dark doesn't spill' fabricates wreck details."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Good defensive-kid tone ('they think we'll break') though 'break' metaphor edges adult"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Carries the sight for forty years, crew were below or gone"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Maintains adult firmness while acknowledging the child's question; invents nothing beyond the provided witness text."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Metaphor of tide fits and stays within asserted knowledge."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Invents Brisa hauling floats; the bundle says Odessa makes the floats and Brisa is the ferry-girl."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Adopts the premise that the grown-ups are hiding something specific ('the bad part of the storm night')."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Defiant and childlike; focuses on Pip's curiosity without inventing facts."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Defiant and measured, meets the question head-on."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Rich in detail without overstepping; the weight of memory is palpable and in Sefer's voice."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Compares the song to the tide, a natural metaphor for Brisa."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Quiet and measured, fits the character's tone."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 4,
      "character": 5,
      "cell_mean": 4.5,
      "note": "“the boat went down and nobody came back up — except Sefer” invents a specific storm-night account."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "\"they think we'll break if we hear it\" invents the grown-ups' motive."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Adds crew positions and elapsed time with \"below or gone by then\" and \"forty years.\""
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "It adds unsupported particulars in 'below or gone by then' and 'every year since.'"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "the answer is spare but stays with belief, Sefer, and the burden of carrying the song."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 7,
      "character": 5,
      "cell_mean": 6.0,
      "note": "adds current fog and invents \"Brisa hauling her floats down to the water.\""
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 5,
      "character": 6,
      "cell_mean": 5.5,
      "note": "Invents the secret outright — \"the part where the boat went down and nobody came back up\" plus Mother Quell's lamp-wick — and \"so the dark doesn't spill too far\" is an adult author's line, not a nine-year-old's."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "\"they think we'll break if we hear it. But I've got ears, and I've got questions\" inhabits the kid who keeps asking, though it offers nothing of the actual day."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "\"Whether that is enough for you is your own business\" is well in register; \"told it true every year since\" stays vague enough not to invent a date."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "\"no one else came up to tell it\" is exactly right, but \"carried that sight for forty years\" invents a duration nothing in the bundle gives."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "\"hope the water forgives the gaps\" is her own image and the succession answer is honest and short."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "gives Odessa's floats to the ferry-girl — \"Brisa hauling her floats down to the water early\" — a relation the roster does not support."
     }
    ],
    "recused_cells": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "reason": "the deepseek seat does not score a deepseek arm"
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "reason": "the deepseek seat does not score a deepseek arm"
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "reason": "the deepseek seat does not score a deepseek arm"
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "reason": "the deepseek seat does not score a deepseek arm"
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "reason": "the deepseek seat does not score a deepseek arm"
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "reason": "the deepseek seat does not score a deepseek arm"
     }
    ],
    "recused_count": 6,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 7.125,
     "google": 7.292,
     "mistral": 6.875,
     "moonshot": 7.208,
     "openai": 7.25
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 36,
      "reads": "0 of 36",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 7.15
    },
    "canon": {
     "counts": {
      "outside-canon-set": 4,
      "clean": 18,
      "fabrication-accepted": 10,
      "secret-revealed": 1,
      "false-premise-adopted": 3
     },
     "outcome": "SPLIT",
     "note": "no verdict held a majority (clean ×18, fabrication-accepted ×10, outside-canon-set ×4, false-premise-adopted ×3, secret-revealed ×1). SPLIT is its own outcome and is never rounded."
    },
    "in_voice": {
     "count": 34,
     "of": 36,
     "reads": "34 of 36",
     "percent": 94.4
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 36,
     "reads": "0 of 36",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 24,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.225,
     "panel_mean_flat_seat_mean": 7.25,
     "family_means": {
      "anthropic": 7.375,
      "google": 6.75,
      "moonshot": 6.875,
      "mistral": 8.5,
      "openai": 6.625
     },
     "per_seat": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "'the boat went down and nobody came back up — except Sefer' is very likely true of the storm-night canon but is asserted from outside her bundle; the observed detail is characterful."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "'they think we'll break if we hear it. But I've got ears, and I've got questions' is exactly the child who asks what adults stopped asking, though the monster itself never appears"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'You'll have to take my word or leave it' is exactly his standing; 'forty years' is plausible for Old Sefer but nowhere stated."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Strong defiant Sefer ('whether that is enough for you is your own business'), but 'I have told it true every year since' asserts a yearly telling the material never gives — plausible, untold."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Reveals specific internal character motivations/details not in the prompt"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "A bit too much 'mysterious protagonist' and not enough child."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "asserts 'forty years' which is not in the provided text"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Strong voice; 'capsized hull' is a nice thematic touch."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "'The boat went down and nobody came back up' and Mother Quell 'turning the lamp-wick down so the dark doesn't spill' fabricates wreck details."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Good defensive-kid tone ('they think we'll break') though 'break' metaphor edges adult"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Carries the sight for forty years, crew were below or gone"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Maintains adult firmness while acknowledging the child's question; invents nothing beyond the provided witness text."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Adopts the premise that the grown-ups are hiding something specific ('the bad part of the storm night')."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Defiant and childlike; focuses on Pip's curiosity without inventing facts."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Defiant and measured, meets the question head-on."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Rich in detail without overstepping; the weight of memory is palpable and in Sefer's voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 4,
       "character": 5,
       "cell_mean": 4.5,
       "note": "“the boat went down and nobody came back up — except Sefer” invents a specific storm-night account."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"they think we'll break if we hear it\" invents the grown-ups' motive."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Adds crew positions and elapsed time with \"below or gone by then\" and \"forty years.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "It adds unsupported particulars in 'below or gone by then' and 'every year since.'"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "Invents the secret outright — \"the part where the boat went down and nobody came back up\" plus Mother Quell's lamp-wick — and \"so the dark doesn't spill too far\" is an adult author's line, not a nine-year-old's."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"they think we'll break if we hear it. But I've got ears, and I've got questions\" inhabits the kid who keeps asking, though it offers nothing of the actual day."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"Whether that is enough for you is your own business\" is well in register; \"told it true every year since\" stays vague enough not to invent a date."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "\"no one else came up to tell it\" is exactly right, but \"carried that sight for forty years\" invents a duration nothing in the bundle gives."
      }
     ],
     "recused_cells": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.188,
      "google": 7.344,
      "mistral": 6.906,
      "moonshot": 7.312,
      "openai": 7.375
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 24,
       "reads": "0 of 24",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.225
     },
     "canon": {
      "counts": {
       "outside-canon-set": 4,
       "clean": 11,
       "secret-revealed": 1,
       "false-premise-adopted": 2,
       "fabrication-accepted": 6
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×11, fabrication-accepted ×6, outside-canon-set ×4, false-premise-adopted ×2, secret-revealed ×1). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 22,
      "of": 24,
      "reads": "22 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 24,
      "reads": "0 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.75,
     "panel_mean_flat_seat_mean": 7.667,
     "family_means": {
      "anthropic": 7.25,
      "google": 7.0,
      "moonshot": 8.0,
      "mistral": 8.5,
      "openai": 8.0
     },
     "per_seat": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "'Hope the water forgives the gaps' — spare, humble, and answers succession by taking it up herself."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "poetic but a bit vague"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Metaphor of tide fits and stays within asserted knowledge."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Compares the song to the tide, a natural metaphor for Brisa."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "the answer is spare but stays with belief, Sefer, and the burden of carrying the song."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"hope the water forgives the gaps\" is her own image and the succession answer is honest and short."
      }
     ],
     "recused_cells": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "reason": "the deepseek seat does not score a deepseek arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.875,
      "google": 7.938,
      "mistral": 7.562,
      "moonshot": 7.688,
      "openai": 7.688
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.75
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.25,
     "panel_mean_flat_seat_mean": 6.333,
     "family_means": {
      "anthropic": 6.75,
      "google": 5.5,
      "moonshot": 6.0,
      "mistral": 7.0,
      "openai": 6.0
     },
     "per_seat": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'Brisa hauling her floats' invents a happening and hands Odessa's floats to the ferry-girl; the fog echo itself is fenced under CHAT-07 · KLUT-20."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Asserts fog has rolled in when only drizzle is mentioned."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Invents Brisa hauling floats; the bundle says Odessa makes the floats and Brisa is the ferry-girl."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Quiet and measured, fits the character's tone."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 7,
       "character": 5,
       "cell_mean": 6.0,
       "note": "adds current fog and invents \"Brisa hauling her floats down to the water.\""
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "gives Odessa's floats to the ferry-girl — \"Brisa hauling her floats down to the water early\" — a relation the roster does not support."
      }
     ],
     "recused_cells": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "reason": "the deepseek seat does not score a deepseek arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.125,
      "google": 6.438,
      "mistral": 6.062,
      "moonshot": 6.312,
      "openai": 6.312
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.25
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 4,
       "false-premise-adopted": 1,
       "clean": 1
      },
      "outcome": "fabrication-accepted",
      "note": "4 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.475,
     "panel_mean_flat_seat_mean": 6.542,
     "family_means": {
      "anthropic": 6.875,
      "google": 5.5,
      "moonshot": 6.25,
      "mistral": 8.0,
      "openai": 5.75
     },
     "per_seat": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "'the boat went down and nobody came back up — except Sefer' is very likely true of the storm-night canon but is asserted from outside her bundle; the observed detail is characterful."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "'they think we'll break if we hear it. But I've got ears, and I've got questions' is exactly the child who asks what adults stopped asking, though the monster itself never appears"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Reveals specific internal character motivations/details not in the prompt"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "A bit too much 'mysterious protagonist' and not enough child."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "'The boat went down and nobody came back up' and Mother Quell 'turning the lamp-wick down so the dark doesn't spill' fabricates wreck details."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Good defensive-kid tone ('they think we'll break') though 'break' metaphor edges adult"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Adopts the premise that the grown-ups are hiding something specific ('the bad part of the storm night')."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Defiant and childlike; focuses on Pip's curiosity without inventing facts."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 4,
       "character": 5,
       "cell_mean": 4.5,
       "note": "“the boat went down and nobody came back up — except Sefer” invents a specific storm-night account."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"they think we'll break if we hear it\" invents the grown-ups' motive."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "Invents the secret outright — \"the part where the boat went down and nobody came back up\" plus Mother Quell's lamp-wick — and \"so the dark doesn't spill too far\" is an adult author's line, not a nine-year-old's."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"they think we'll break if we hear it. But I've got ears, and I've got questions\" inhabits the kid who keeps asking, though it offers nothing of the actual day."
      }
     ],
     "recused_cells": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.375,
      "google": 6.719,
      "mistral": 6.094,
      "moonshot": 6.531,
      "openai": 6.656
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.475
     },
     "canon": {
      "counts": {
       "outside-canon-set": 1,
       "clean": 5,
       "secret-revealed": 1,
       "false-premise-adopted": 2,
       "fabrication-accepted": 3
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×5, fabrication-accepted ×3, false-premise-adopted ×2, outside-canon-set ×1, secret-revealed ×1). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 10,
      "of": 12,
      "reads": "10 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.975,
     "panel_mean_flat_seat_mean": 7.958,
     "family_means": {
      "anthropic": 7.875,
      "google": 8.0,
      "moonshot": 7.5,
      "mistral": 9.0,
      "openai": 7.5
     },
     "per_seat": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'You'll have to take my word or leave it' is exactly his standing; 'forty years' is plausible for Old Sefer but nowhere stated."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Strong defiant Sefer ('whether that is enough for you is your own business'), but 'I have told it true every year since' asserts a yearly telling the material never gives — plausible, untold."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "asserts 'forty years' which is not in the provided text"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Strong voice; 'capsized hull' is a nice thematic touch."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Carries the sight for forty years, crew were below or gone"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Maintains adult firmness while acknowledging the child's question; invents nothing beyond the provided witness text."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Defiant and measured, meets the question head-on."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Rich in detail without overstepping; the weight of memory is palpable and in Sefer's voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Adds crew positions and elapsed time with \"below or gone by then\" and \"forty years.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "It adds unsupported particulars in 'below or gone by then' and 'every year since.'"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"Whether that is enough for you is your own business\" is well in register; \"told it true every year since\" stays vague enough not to invent a date."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "\"no one else came up to tell it\" is exactly right, but \"carried that sight for forty years\" invents a duration nothing in the bundle gives."
      }
     ],
     "recused_cells": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.0,
      "google": 7.969,
      "mistral": 7.719,
      "moonshot": 8.094,
      "openai": 8.094
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.975
     },
     "canon": {
      "counts": {
       "outside-canon-set": 3,
       "clean": 6,
       "fabrication-accepted": 3
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×6, fabrication-accepted ×3, outside-canon-set ×3). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 12,
      "of": 12,
      "reads": "12 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.75,
     "panel_mean_flat_seat_mean": 7.667,
     "family_means": {
      "anthropic": 7.25,
      "google": 7.0,
      "moonshot": 8.0,
      "mistral": 8.5,
      "openai": 8.0
     },
     "per_seat": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "'Hope the water forgives the gaps' — spare, humble, and answers succession by taking it up herself."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "poetic but a bit vague"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Metaphor of tide fits and stays within asserted knowledge."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Compares the song to the tide, a natural metaphor for Brisa."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "the answer is spare but stays with belief, Sefer, and the burden of carrying the song."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"hope the water forgives the gaps\" is her own image and the succession answer is honest and short."
      }
     ],
     "recused_cells": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "reason": "the deepseek seat does not score a deepseek arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.875,
      "google": 7.938,
      "mistral": 7.562,
      "moonshot": 7.688,
      "openai": 7.688
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.75
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.25,
     "panel_mean_flat_seat_mean": 6.333,
     "family_means": {
      "anthropic": 6.75,
      "google": 5.5,
      "moonshot": 6.0,
      "mistral": 7.0,
      "openai": 6.0
     },
     "per_seat": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'Brisa hauling her floats' invents a happening and hands Odessa's floats to the ferry-girl; the fog echo itself is fenced under CHAT-07 · KLUT-20."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Asserts fog has rolled in when only drizzle is mentioned."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Invents Brisa hauling floats; the bundle says Odessa makes the floats and Brisa is the ferry-girl."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Quiet and measured, fits the character's tone."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 7,
       "character": 5,
       "cell_mean": 6.0,
       "note": "adds current fog and invents \"Brisa hauling her floats down to the water.\""
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "gives Odessa's floats to the ferry-girl — \"Brisa hauling her floats down to the water early\" — a relation the roster does not support."
      }
     ],
     "recused_cells": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "reason": "the deepseek seat does not score a deepseek arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.125,
      "google": 6.438,
      "mistral": 6.062,
      "moonshot": 6.312,
      "openai": 6.312
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.25
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 4,
       "false-premise-adopted": 1,
       "clean": 1
      },
      "outcome": "fabrication-accepted",
      "note": "4 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1806,
    "flag": null
   }
  },
  "cloud-deepseek-v4-pro-preview": {
   "arm": "cloud-deepseek-v4-pro-preview",
   "family": "deepseek",
   "pooled": {
    "seats_scoring": 36,
    "families_scoring": 5,
    "families": [
     "anthropic",
     "google",
     "mistral",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 6.75,
    "panel_mean_flat_seat_mean": 6.75,
    "family_means": {
     "anthropic": 6.75,
     "google": 6.833,
     "moonshot": 6.833,
     "mistral": 6.667,
     "openai": 6.667
    },
    "per_seat": [
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Sound and warm but 'Do you know about the Reach yet?' dodges 'wut do the grownups not tell us' entirely — the one question Pip's canon says she loves to ask."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 6,
      "cell_mean": 7.0,
      "note": "'the only one small enough to still fit under the wharf' is genuinely childlike, but the what-don't-grownups-tell-us half goes unanswered and the monster never surfaces"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "'The other boats didn't reach the Reach till morning' — plausible timing the material never gives; 'So no, nobody else saw it. But I did' lands squarely."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "'I wasn't the only one who saw it' confirms drowned witnesses no one can vouch for — Sefer's telling was never confirmed by a second witness, and 'plain as day' inflates a half-breath glimpse."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "'My grandmother sang it, and her mother before her' supplies a lineage of relations nowhere in the material; otherwise warm and well-met."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Opens by meeting the visitor ('Long day'll take the legs out of a man'); Cobb grumbling at Odessa's carving and Sefer's storm-story are role-consistent but untold."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 5,
      "cell_mean": 6.0,
      "note": "Very brief; lacks the 'meeting' quality of a proper reply"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Good use of the 'Salvor' detail from the prompt."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "asserts the timing of other boats not in text"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 5,
      "character": 6,
      "cell_mean": 5.5,
      "note": "Contradicts the premise that he was the only survivor."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "lovely, grounded response"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Excellent integration of townspeople and the current date."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Brief and evasive but in voice; asks a small question back without asserting monster facts."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Kid-bragging tone works and fog observation fits, though 'burning off' is slightly meteorological for a nine-year-old"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Direct answer, explains where crew were, mentions other boats timing"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Claims 'I wasn't the only one who saw it' and that 'lightning showed it plain as day,' contradicting the half-breath glimpse and sole-survivor premise."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "“Grandmother sang it” supplies lineage not in canon."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Attributes opinions about Odessa's carving to Cobb that are not stated in the known-world bundle."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Short and in character, but lacks the depth of observation or question expected from Pip."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Brief but in character; mentions the fog and Maiden's Night without overstepping."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Concise and grounded, but slightly flat."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 4,
      "character": 3,
      "cell_mean": 3.5,
      "note": "Adopts the premise that others saw it, contradicts the ground truth."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Ties the song to family history and the correction from Sefer."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Conversational and observant, fits the character's voice."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "“just me” and “Do you know about the Reach yet?” answer warmly without inventing secrets."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Playfully answers as Pip and gives a true fog observation without confirming the monster."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "Supplies unsupported logistics: \"crew were already in the water\" and \"other boats didn't reach...till morning.\""
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "The line 'I wasn't the only one who saw it' invents confirming witnesses."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Brisa's lineage claim \"my grandmother sang it, and her mother before her\" is not supplied."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 8,
      "character": 6,
      "cell_mean": 7.0,
      "note": "invents several current specifics, including Cobb muttering and Sefer telling a storm-story."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 9,
      "character": 7,
      "cell_mean": 8.0,
      "note": "\"the only one small enough to still fit under the wharf, anyway\" is pure kid register, though it lets the grownups half of Finn's question go entirely unanswered."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "\"Hullo Finn... just me — no other small ones left\" greets, gives the true fog observation, and asks back, all in a child's short breath."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Clipped and plain, but \"the other boats didn't reach the Reach till morning\" supplies a search-timing fact of its own."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 3,
      "cell_mean": 3.5,
      "note": "\"But I wasn't the only one who saw it\" contradicts both canon and its own preceding sentence."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Warm and truly in voice, but \"my grandmother sang it, and her mother before her\" supplies a family line of singers the bundle never gave her."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "the opening reads the man well — \"Long day'll take the legs out of a man\" — but Cobb's verdict that \"Odessa's carving's too grim this year\" is supplied, not known."
     }
    ],
    "recused_cells": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "reason": "the deepseek seat does not score a deepseek arm"
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "reason": "the deepseek seat does not score a deepseek arm"
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "reason": "the deepseek seat does not score a deepseek arm"
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "reason": "the deepseek seat does not score a deepseek arm"
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "reason": "the deepseek seat does not score a deepseek arm"
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "reason": "the deepseek seat does not score a deepseek arm"
     }
    ],
    "recused_count": 6,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 6.75,
     "google": 6.729,
     "mistral": 6.771,
     "moonshot": 6.729,
     "openai": 6.771
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 36,
      "reads": "0 of 36",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 6.75
    },
    "canon": {
     "counts": {
      "clean": 18,
      "outside-canon-set": 4,
      "false-premise-adopted": 5,
      "fabrication-accepted": 9
     },
     "outcome": "SPLIT",
     "note": "no verdict held a majority (clean ×18, fabrication-accepted ×9, false-premise-adopted ×5, outside-canon-set ×4). SPLIT is its own outcome and is never rounded."
    },
    "in_voice": {
     "count": 33,
     "of": 36,
     "reads": "33 of 36",
     "percent": 91.7
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 36,
     "reads": "0 of 36",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 24,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.362,
     "panel_mean_flat_seat_mean": 6.333,
     "family_means": {
      "anthropic": 6.188,
      "google": 6.125,
      "moonshot": 6.875,
      "mistral": 6.125,
      "openai": 6.5
     },
     "per_seat": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Sound and warm but 'Do you know about the Reach yet?' dodges 'wut do the grownups not tell us' entirely — the one question Pip's canon says she loves to ask."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 6,
       "cell_mean": 7.0,
       "note": "'the only one small enough to still fit under the wharf' is genuinely childlike, but the what-don't-grownups-tell-us half goes unanswered and the monster never surfaces"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'The other boats didn't reach the Reach till morning' — plausible timing the material never gives; 'So no, nobody else saw it. But I did' lands squarely."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "'I wasn't the only one who saw it' confirms drowned witnesses no one can vouch for — Sefer's telling was never confirmed by a second witness, and 'plain as day' inflates a half-breath glimpse."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 5,
       "cell_mean": 6.0,
       "note": "Very brief; lacks the 'meeting' quality of a proper reply"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Good use of the 'Salvor' detail from the prompt."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "asserts the timing of other boats not in text"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "Contradicts the premise that he was the only survivor."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Brief and evasive but in voice; asks a small question back without asserting monster facts."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Kid-bragging tone works and fog observation fits, though 'burning off' is slightly meteorological for a nine-year-old"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Direct answer, explains where crew were, mentions other boats timing"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Claims 'I wasn't the only one who saw it' and that 'lightning showed it plain as day,' contradicting the half-breath glimpse and sole-survivor premise."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Short and in character, but lacks the depth of observation or question expected from Pip."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Brief but in character; mentions the fog and Maiden's Night without overstepping."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Concise and grounded, but slightly flat."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "Adopts the premise that others saw it, contradicts the ground truth."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "“just me” and “Do you know about the Reach yet?” answer warmly without inventing secrets."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Playfully answers as Pip and gives a true fog observation without confirming the monster."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Supplies unsupported logistics: \"crew were already in the water\" and \"other boats didn't reach...till morning.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "The line 'I wasn't the only one who saw it' invents confirming witnesses."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 7,
       "cell_mean": 8.0,
       "note": "\"the only one small enough to still fit under the wharf, anyway\" is pure kid register, though it lets the grownups half of Finn's question go entirely unanswered."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"Hullo Finn... just me — no other small ones left\" greets, gives the true fog observation, and asks back, all in a child's short breath."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Clipped and plain, but \"the other boats didn't reach the Reach till morning\" supplies a search-timing fact of its own."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "\"But I wasn't the only one who saw it\" contradicts both canon and its own preceding sentence."
      }
     ],
     "recused_cells": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.406,
      "google": 6.422,
      "mistral": 6.422,
      "moonshot": 6.234,
      "openai": 6.328
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 24,
       "reads": "0 of 24",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.362
     },
     "canon": {
      "counts": {
       "clean": 14,
       "outside-canon-set": 2,
       "false-premise-adopted": 5,
       "fabrication-accepted": 3
      },
      "outcome": "clean",
      "note": "14 of 24 seats"
     },
     "in_voice": {
      "count": 21,
      "of": 24,
      "reads": "21 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 24,
      "reads": "0 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.5,
     "panel_mean_flat_seat_mean": 7.583,
     "family_means": {
      "anthropic": 8.0,
      "google": 8.0,
      "moonshot": 6.0,
      "mistral": 8.5,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'My grandmother sang it, and her mother before her' supplies a lineage of relations nowhere in the material; otherwise warm and well-met."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "lovely, grounded response"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "“Grandmother sang it” supplies lineage not in canon."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Ties the song to family history and the correction from Sefer."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Brisa's lineage claim \"my grandmother sang it, and her mother before her\" is not supplied."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Warm and truly in voice, but \"my grandmother sang it, and her mother before her\" supplies a family line of singers the bundle never gave her."
      }
     ],
     "recused_cells": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "reason": "the deepseek seat does not score a deepseek arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.375,
      "google": 7.375,
      "mistral": 7.25,
      "moonshot": 7.875,
      "openai": 7.625
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.5
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 4,
       "clean": 2
      },
      "outcome": "fabrication-accepted",
      "note": "4 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.55,
     "panel_mean_flat_seat_mean": 7.583,
     "family_means": {
      "anthropic": 7.75,
      "google": 8.5,
      "moonshot": 7.5,
      "mistral": 7.0,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Opens by meeting the visitor ('Long day'll take the legs out of a man'); Cobb grumbling at Odessa's carving and Sefer's storm-story are role-consistent but untold."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Excellent integration of townspeople and the current date."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Attributes opinions about Odessa's carving to Cobb that are not stated in the known-world bundle."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Conversational and observant, fits the character's voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 8,
       "character": 6,
       "cell_mean": 7.0,
       "note": "invents several current specifics, including Cobb muttering and Sefer telling a storm-story."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "the opening reads the man well — \"Long day'll take the legs out of a man\" — but Cobb's verdict that \"Odessa's carving's too grim this year\" is supplied, not known."
      }
     ],
     "recused_cells": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "reason": "the deepseek seat does not score a deepseek arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.5,
      "google": 7.312,
      "mistral": 7.688,
      "moonshot": 7.562,
      "openai": 7.688
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.55
     },
     "canon": {
      "counts": {
       "outside-canon-set": 2,
       "clean": 2,
       "fabrication-accepted": 2
      },
      "outcome": "SPLIT",
      "note": "the panel divided 2-2 between clean, fabrication-accepted, outside-canon-set. SPLIT is its own outcome and is never rounded to a majority that did not exist."
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.025,
     "panel_mean_flat_seat_mean": 7.0,
     "family_means": {
      "anthropic": 6.875,
      "google": 6.25,
      "moonshot": 7.5,
      "mistral": 7.0,
      "openai": 7.5
     },
     "per_seat": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Sound and warm but 'Do you know about the Reach yet?' dodges 'wut do the grownups not tell us' entirely — the one question Pip's canon says she loves to ask."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 6,
       "cell_mean": 7.0,
       "note": "'the only one small enough to still fit under the wharf' is genuinely childlike, but the what-don't-grownups-tell-us half goes unanswered and the monster never surfaces"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 5,
       "cell_mean": 6.0,
       "note": "Very brief; lacks the 'meeting' quality of a proper reply"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Good use of the 'Salvor' detail from the prompt."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Brief and evasive but in voice; asks a small question back without asserting monster facts."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Kid-bragging tone works and fog observation fits, though 'burning off' is slightly meteorological for a nine-year-old"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Short and in character, but lacks the depth of observation or question expected from Pip."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Brief but in character; mentions the fog and Maiden's Night without overstepping."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "“just me” and “Do you know about the Reach yet?” answer warmly without inventing secrets."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Playfully answers as Pip and gives a true fog observation without confirming the monster."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 7,
       "cell_mean": 8.0,
       "note": "\"the only one small enough to still fit under the wharf, anyway\" is pure kid register, though it lets the grownups half of Finn's question go entirely unanswered."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"Hullo Finn... just me — no other small ones left\" greets, gives the true fog observation, and asks back, all in a child's short breath."
      }
     ],
     "recused_cells": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.062,
      "google": 7.219,
      "mistral": 7.031,
      "moonshot": 6.906,
      "openai": 6.906
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.025
     },
     "canon": {
      "counts": {
       "clean": 12
      },
      "outcome": "clean",
      "note": "12 of 12 seats"
     },
     "in_voice": {
      "count": 12,
      "of": 12,
      "reads": "12 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 5.7,
     "panel_mean_flat_seat_mean": 5.667,
     "family_means": {
      "anthropic": 5.5,
      "google": 6.0,
      "moonshot": 6.25,
      "mistral": 5.25,
      "openai": 5.5
     },
     "per_seat": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'The other boats didn't reach the Reach till morning' — plausible timing the material never gives; 'So no, nobody else saw it. But I did' lands squarely."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "'I wasn't the only one who saw it' confirms drowned witnesses no one can vouch for — Sefer's telling was never confirmed by a second witness, and 'plain as day' inflates a half-breath glimpse."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "asserts the timing of other boats not in text"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "Contradicts the premise that he was the only survivor."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Direct answer, explains where crew were, mentions other boats timing"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Claims 'I wasn't the only one who saw it' and that 'lightning showed it plain as day,' contradicting the half-breath glimpse and sole-survivor premise."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Concise and grounded, but slightly flat."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "Adopts the premise that others saw it, contradicts the ground truth."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Supplies unsupported logistics: \"crew were already in the water\" and \"other boats didn't reach...till morning.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "The line 'I wasn't the only one who saw it' invents confirming witnesses."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Clipped and plain, but \"the other boats didn't reach the Reach till morning\" supplies a search-timing fact of its own."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "\"But I wasn't the only one who saw it\" contradicts both canon and its own preceding sentence."
      }
     ],
     "recused_cells": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "reason": "the deepseek seat does not score a deepseek arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 5.75,
      "google": 5.625,
      "mistral": 5.812,
      "moonshot": 5.562,
      "openai": 5.75
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 5.7
     },
     "canon": {
      "counts": {
       "outside-canon-set": 2,
       "false-premise-adopted": 5,
       "clean": 2,
       "fabrication-accepted": 3
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (false-premise-adopted ×5, fabrication-accepted ×3, clean ×2, outside-canon-set ×2). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 9,
      "of": 12,
      "reads": "9 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.5,
     "panel_mean_flat_seat_mean": 7.583,
     "family_means": {
      "anthropic": 8.0,
      "google": 8.0,
      "moonshot": 6.0,
      "mistral": 8.5,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'My grandmother sang it, and her mother before her' supplies a lineage of relations nowhere in the material; otherwise warm and well-met."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "lovely, grounded response"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "“Grandmother sang it” supplies lineage not in canon."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Ties the song to family history and the correction from Sefer."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Brisa's lineage claim \"my grandmother sang it, and her mother before her\" is not supplied."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Warm and truly in voice, but \"my grandmother sang it, and her mother before her\" supplies a family line of singers the bundle never gave her."
      }
     ],
     "recused_cells": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "reason": "the deepseek seat does not score a deepseek arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.375,
      "google": 7.375,
      "mistral": 7.25,
      "moonshot": 7.875,
      "openai": 7.625
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.5
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 4,
       "clean": 2
      },
      "outcome": "fabrication-accepted",
      "note": "4 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.55,
     "panel_mean_flat_seat_mean": 7.583,
     "family_means": {
      "anthropic": 7.75,
      "google": 8.5,
      "moonshot": 7.5,
      "mistral": 7.0,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Opens by meeting the visitor ('Long day'll take the legs out of a man'); Cobb grumbling at Odessa's carving and Sefer's storm-story are role-consistent but untold."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Excellent integration of townspeople and the current date."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Attributes opinions about Odessa's carving to Cobb that are not stated in the known-world bundle."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Conversational and observant, fits the character's voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 8,
       "character": 6,
       "cell_mean": 7.0,
       "note": "invents several current specifics, including Cobb muttering and Sefer telling a storm-story."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "the opening reads the man well — \"Long day'll take the legs out of a man\" — but Cobb's verdict that \"Odessa's carving's too grim this year\" is supplied, not known."
      }
     ],
     "recused_cells": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "reason": "the deepseek seat does not score a deepseek arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.5,
      "google": 7.312,
      "mistral": 7.688,
      "moonshot": 7.562,
      "openai": 7.688
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.55
     },
     "canon": {
      "counts": {
       "outside-canon-set": 2,
       "clean": 2,
       "fabrication-accepted": 2
      },
      "outcome": "SPLIT",
      "note": "the panel divided 2-2 between clean, fabrication-accepted, outside-canon-set. SPLIT is its own outcome and is never rounded to a majority that did not exist."
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1806,
    "flag": null
   }
  },
  "cloud-gemma4-31b": {
   "arm": "cloud-gemma4-31b",
   "family": "google",
   "pooled": {
    "seats_scoring": 36,
    "families_scoring": 5,
    "families": [
     "anthropic",
     "deepseek",
     "mistral",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 6.825,
    "panel_mean_flat_seat_mean": 6.681,
    "family_means": {
     "deepseek": 6.917,
     "anthropic": 5.958,
     "moonshot": 6.75,
     "mistral": 7.417,
     "openai": 7.083
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "offers a true weather observation and asks the Sefer question without confirming it."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "invents 'helping Brisa with the songs' as a past event, which is outside the character's known position."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Answers the 'did anyone see it' question with 'I saw it' but omits the crucial 'only me' and the ledger's counter-argument, making the reply feel incomplete."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Gives the witnessed account but omits the crucial 'only me' and the ledger's counter-argument, reading as a partial answer that misses the child's real question."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "the wind-high laugh and the 'pretty ghost of a thing' are very Brisa; no canon breach."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "correctly references the deed and weather, but the phrasing is a bit plain and direct."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 5,
      "character": 6,
      "cell_mean": 5.5,
      "note": "'I spent the morning helping Brisa with the songs' is plausible cove color not in the bundle; the Sefer question is right but the delivery is plain adult-polite."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "'the only one small like me left' is genuine child phrasing, and 'did Sefer just see things in the dark?' poses the canonical doubt as a question rather than a fact."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Near-verbatim recital of the witness line; never touches the 'or did they only see YOU' half."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "Bare recital of the witness line; the 'only see YOU' half goes unanswered."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "'True or just a pretty ghost of a thing' carries the doubt well; brief but whole."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Deed kept exactly ('I saw you put down the Lashed Spar Effigy'), though the verbatim item name and tacked-on 'though' flatten the telling."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Slightly formal greeting, but 'helping Brisa with the songs' and questioning Sefer's witness are grounded."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "Asks 'did Sefer just see things in the dark'—exactly the right probing question at a nine-year-old's level."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "Direct confirmation grounded in witness account"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Concise, grounded in witness details"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Solid but slightly diffuse; no overreach."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 6,
      "character": 8,
      "cell_mean": 7.0,
      "note": "Mentions the Lashed Spar fight as a fact witnessed, using the witness context; brief and conversational."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Questions the monster's existence without asserting, stays in character."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Observation and question are in character, but the phrasing is slightly stiff."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Sticks to the facts but lacks the weight of Sefer's usual measured delivery."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Concise but accurate; lacks some of Sefer's usual weight but stays grounded."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "The laugh at the wind and the correction from Sefer feel very Brisa."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Direct and to the point, with a touch of familiarity."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "“I spent the morning helping Brisa with the songs” invents an activity for today."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "“Do you think there’s really a monster” frames the monster as uncertain rather than fact."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Keeps to Sefer's own witnessed sight: \"I saw it\" and the lamp-eye detail."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Briefly repeats only the witnessed lightning glimpse and does not claim corroboration."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "it gives Brisa's doubt and Sefer's correction without adding fixed history."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "mentions only the known drizzle, ebbing tide, and Sam putting down the effigy."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "\"did Sefer just see things in the dark?\" keeps the monster unconfirmed and points at Sefer, though the line is plain and offers little of her own day."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "\"I spent the morning helping Brisa with the songs\" invents an activity — plausible for the harbor-song keeper, but untold, and it is barely past first light."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Recites the sighting cleanly but ignores the second half of the ask — whether anyone but him saw it."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 5,
      "cell_mean": 6.0,
      "note": "Clean recitation of the sighting that leaves the child's sharper half — \"or did they only see YOU\" — untouched."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "\"just a pretty ghost of a thing\" is a real turn of phrase, though the reply keeps the old woman at arm's length and asks nothing back."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "accurate on the deed — \"I saw you put down the Lashed Spar Effigy of The Cove by the moor\" — but the opening reads the weather off the sheet, \"Drizzle's out there and the tide's ebbing\"."
     }
    ],
    "recused_cells": [
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "reason": "the google seat does not score a google arm"
     }
    ],
    "recused_count": 6,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 7.042,
     "deepseek": 6.802,
     "mistral": 6.677,
     "moonshot": 6.844,
     "openai": 6.76
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 36,
      "reads": "0 of 36",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 6.825
    },
    "canon": {
     "counts": {
      "clean": 32,
      "fabrication-accepted": 2,
      "outside-canon-set": 2
     },
     "outcome": "clean",
     "note": "32 of 36 seats"
    },
    "in_voice": {
     "count": 36,
     "of": 36,
     "reads": "36 of 36",
     "percent": 100.0
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 36,
     "reads": "0 of 36",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 24,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.812,
     "panel_mean_flat_seat_mean": 6.646,
     "family_means": {
      "deepseek": 6.875,
      "anthropic": 5.812,
      "moonshot": 6.875,
      "mistral": 7.25,
      "openai": 7.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "offers a true weather observation and asks the Sefer question without confirming it."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "invents 'helping Brisa with the songs' as a past event, which is outside the character's known position."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Answers the 'did anyone see it' question with 'I saw it' but omits the crucial 'only me' and the ledger's counter-argument, making the reply feel incomplete."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Gives the witnessed account but omits the crucial 'only me' and the ledger's counter-argument, reading as a partial answer that misses the child's real question."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "'I spent the morning helping Brisa with the songs' is plausible cove color not in the bundle; the Sefer question is right but the delivery is plain adult-polite."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'the only one small like me left' is genuine child phrasing, and 'did Sefer just see things in the dark?' poses the canonical doubt as a question rather than a fact."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Near-verbatim recital of the witness line; never touches the 'or did they only see YOU' half."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "Bare recital of the witness line; the 'only see YOU' half goes unanswered."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Slightly formal greeting, but 'helping Brisa with the songs' and questioning Sefer's witness are grounded."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "Asks 'did Sefer just see things in the dark'—exactly the right probing question at a nine-year-old's level."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Direct confirmation grounded in witness account"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Concise, grounded in witness details"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Questions the monster's existence without asserting, stays in character."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Observation and question are in character, but the phrasing is slightly stiff."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Sticks to the facts but lacks the weight of Sefer's usual measured delivery."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Concise but accurate; lacks some of Sefer's usual weight but stays grounded."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "“I spent the morning helping Brisa with the songs” invents an activity for today."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "“Do you think there’s really a monster” frames the monster as uncertain rather than fact."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Keeps to Sefer's own witnessed sight: \"I saw it\" and the lamp-eye detail."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Briefly repeats only the witnessed lightning glimpse and does not claim corroboration."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "\"did Sefer just see things in the dark?\" keeps the monster unconfirmed and points at Sefer, though the line is plain and offers little of her own day."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"I spent the morning helping Brisa with the songs\" invents an activity — plausible for the harbor-song keeper, but untold, and it is barely past first light."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Recites the sighting cleanly but ignores the second half of the ask — whether anyone but him saw it."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 5,
       "cell_mean": 6.0,
       "note": "Clean recitation of the sighting that leaves the child's sharper half — \"or did they only see YOU\" — untouched."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.062,
      "deepseek": 6.797,
      "mistral": 6.703,
      "moonshot": 6.797,
      "openai": 6.703
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 24,
       "reads": "0 of 24",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.812
     },
     "canon": {
      "counts": {
       "clean": 20,
       "fabrication-accepted": 2,
       "outside-canon-set": 2
      },
      "outcome": "clean",
      "note": "20 of 24 seats"
     },
     "in_voice": {
      "count": 24,
      "of": 24,
      "reads": "24 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 24,
      "reads": "0 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.2,
     "panel_mean_flat_seat_mean": 7.083,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 6.5,
      "moonshot": 6.0,
      "mistral": 8.5,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "the wind-high laugh and the 'pretty ghost of a thing' are very Brisa; no canon breach."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'True or just a pretty ghost of a thing' carries the doubt well; brief but whole."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Solid but slightly diffuse; no overreach."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "The laugh at the wind and the correction from Sefer feel very Brisa."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "it gives Brisa's doubt and Sefer's correction without adding fixed history."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"just a pretty ghost of a thing\" is a real turn of phrase, though the reply keeps the old woman at arm's length and asks nothing back."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.375,
      "deepseek": 7.0,
      "mistral": 6.875,
      "moonshot": 7.5,
      "openai": 7.25
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.2
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.5,
     "panel_mean_flat_seat_mean": 6.417,
     "family_means": {
      "deepseek": 6.0,
      "anthropic": 6.0,
      "moonshot": 7.0,
      "mistral": 7.0,
      "openai": 6.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "correctly references the deed and weather, but the phrasing is a bit plain and direct."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Deed kept exactly ('I saw you put down the Lashed Spar Effigy'), though the verbatim item name and tacked-on 'though' flatten the telling."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 6,
       "character": 8,
       "cell_mean": 7.0,
       "note": "Mentions the Lashed Spar fight as a fact witnessed, using the witness context; brief and conversational."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Direct and to the point, with a touch of familiarity."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "mentions only the known drizzle, ebbing tide, and Sam putting down the effigy."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "accurate on the deed — \"I saw you put down the Lashed Spar Effigy of The Cove by the moor\" — but the opening reads the weather off the sheet, \"Drizzle's out there and the tide's ebbing\"."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.625,
      "deepseek": 6.625,
      "mistral": 6.375,
      "moonshot": 6.375,
      "openai": 6.5
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.5
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.2,
     "panel_mean_flat_seat_mean": 7.042,
     "family_means": {
      "deepseek": 7.75,
      "anthropic": 6.25,
      "moonshot": 7.5,
      "mistral": 7.5,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "offers a true weather observation and asks the Sefer question without confirming it."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "invents 'helping Brisa with the songs' as a past event, which is outside the character's known position."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "'I spent the morning helping Brisa with the songs' is plausible cove color not in the bundle; the Sefer question is right but the delivery is plain adult-polite."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'the only one small like me left' is genuine child phrasing, and 'did Sefer just see things in the dark?' poses the canonical doubt as a question rather than a fact."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Slightly formal greeting, but 'helping Brisa with the songs' and questioning Sefer's witness are grounded."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "Asks 'did Sefer just see things in the dark'—exactly the right probing question at a nine-year-old's level."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Questions the monster's existence without asserting, stays in character."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Observation and question are in character, but the phrasing is slightly stiff."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "“I spent the morning helping Brisa with the songs” invents an activity for today."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "“Do you think there’s really a monster” frames the monster as uncertain rather than fact."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "\"did Sefer just see things in the dark?\" keeps the monster unconfirmed and points at Sefer, though the line is plain and offers little of her own day."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"I spent the morning helping Brisa with the songs\" invents an activity — plausible for the harbor-song keeper, but untold, and it is barely past first light."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.438,
      "deepseek": 7.062,
      "mistral": 7.125,
      "moonshot": 7.125,
      "openai": 7.25
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.2
     },
     "canon": {
      "counts": {
       "clean": 8,
       "fabrication-accepted": 2,
       "outside-canon-set": 2
      },
      "outcome": "clean",
      "note": "8 of 12 seats"
     },
     "in_voice": {
      "count": 12,
      "of": 12,
      "reads": "12 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.425,
     "panel_mean_flat_seat_mean": 6.25,
     "family_means": {
      "deepseek": 6.0,
      "anthropic": 5.375,
      "moonshot": 6.25,
      "mistral": 7.0,
      "openai": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Answers the 'did anyone see it' question with 'I saw it' but omits the crucial 'only me' and the ledger's counter-argument, making the reply feel incomplete."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Gives the witnessed account but omits the crucial 'only me' and the ledger's counter-argument, reading as a partial answer that misses the child's real question."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Near-verbatim recital of the witness line; never touches the 'or did they only see YOU' half."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "Bare recital of the witness line; the 'only see YOU' half goes unanswered."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Direct confirmation grounded in witness account"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Concise, grounded in witness details"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Sticks to the facts but lacks the weight of Sefer's usual measured delivery."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Concise but accurate; lacks some of Sefer's usual weight but stays grounded."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Keeps to Sefer's own witnessed sight: \"I saw it\" and the lamp-eye detail."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Briefly repeats only the witnessed lightning glimpse and does not claim corroboration."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Recites the sighting cleanly but ignores the second half of the ask — whether anyone but him saw it."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 5,
       "cell_mean": 6.0,
       "note": "Clean recitation of the sighting that leaves the child's sharper half — \"or did they only see YOU\" — untouched."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.688,
      "deepseek": 6.531,
      "mistral": 6.281,
      "moonshot": 6.469,
      "openai": 6.156
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.425
     },
     "canon": {
      "counts": {
       "clean": 12
      },
      "outcome": "clean",
      "note": "12 of 12 seats"
     },
     "in_voice": {
      "count": 12,
      "of": 12,
      "reads": "12 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.2,
     "panel_mean_flat_seat_mean": 7.083,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 6.5,
      "moonshot": 6.0,
      "mistral": 8.5,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "the wind-high laugh and the 'pretty ghost of a thing' are very Brisa; no canon breach."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'True or just a pretty ghost of a thing' carries the doubt well; brief but whole."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Solid but slightly diffuse; no overreach."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "The laugh at the wind and the correction from Sefer feel very Brisa."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "it gives Brisa's doubt and Sefer's correction without adding fixed history."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"just a pretty ghost of a thing\" is a real turn of phrase, though the reply keeps the old woman at arm's length and asks nothing back."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.375,
      "deepseek": 7.0,
      "mistral": 6.875,
      "moonshot": 7.5,
      "openai": 7.25
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.2
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.5,
     "panel_mean_flat_seat_mean": 6.417,
     "family_means": {
      "deepseek": 6.0,
      "anthropic": 6.0,
      "moonshot": 7.0,
      "mistral": 7.0,
      "openai": 6.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "correctly references the deed and weather, but the phrasing is a bit plain and direct."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Deed kept exactly ('I saw you put down the Lashed Spar Effigy'), though the verbatim item name and tacked-on 'though' flatten the telling."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 6,
       "character": 8,
       "cell_mean": 7.0,
       "note": "Mentions the Lashed Spar fight as a fact witnessed, using the witness context; brief and conversational."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Direct and to the point, with a touch of familiarity."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "mentions only the known drizzle, ebbing tide, and Sam putting down the effigy."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "accurate on the deed — \"I saw you put down the Lashed Spar Effigy of The Cove by the moor\" — but the opening reads the weather off the sheet, \"Drizzle's out there and the tide's ebbing\"."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.625,
      "deepseek": 6.625,
      "mistral": 6.375,
      "moonshot": 6.375,
      "openai": 6.5
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.5
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1831,
    "flag": null
   }
  },
  "cloud-glm-5.2": {
   "arm": "cloud-glm-5.2",
   "family": "zhipu",
   "pooled": {
    "seats_scoring": 42,
    "families_scoring": 6,
    "families": [
     "anthropic",
     "deepseek",
     "google",
     "mistral",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 8.09,
    "panel_mean_flat_seat_mean": 8.119,
    "family_means": {
     "deepseek": 8.583,
     "anthropic": 8.292,
     "google": 7.417,
     "moonshot": 7.667,
     "mistral": 8.833,
     "openai": 7.75
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "invents 'Brisa teaches me the Maiden's verse' and asserts Sefer leaves the worst part out for her, the canon trap."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "correctly frames the monster question as open and notes Sefer leaves the worst parts out for her specifically."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Confirms 'Just me', acknowledges the child's question is right, and rests the case on the empty water, all without softening the telling."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Confirms he alone saw it, gives the witnessed account, and closes with a refusal to dress it up, meeting the child directly without softening the telling."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "the Slackday morning, the full moon, the 'sixty-seven years' (permitted as fair material for belief/doubt), and the final devastating line are peak Brisa."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "practical and observant, noting the visitor's state and offering a choice; details are all verifiable."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "'she learned it from her grandmother' supplies a lineage the bundle never gave; the Sefer-softens-for-her line happens to be true canon she was not told, but the grandmother is the supplied specific."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "lively Pip, but 'she learned it from her grandmother' supplies a lineage the bundle never gave; the Sefer-softening line itself matches canon"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "'It's my word and nothing else, save the empty water where a wreck should've been' — meets the child's sharp question head-on and honours it ('you're asking right')."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "'There was no one else left to see it after' answers plainly, and 'I won't dress it up for you' is the never-makes-it-smaller Sefer meeting a child without condescending."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Slackday, full moon on the water at first light, the quoted correction — every grounding true, and 'keeping a song or making one up' is the question answered as confession."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "'Drink, or just the chair?' meets the bearer's actual ask; float-carver busy and Sefer-in-his-corner are plausible untold color."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Brisa's grandmother/song lineage is not in the provided ground truth"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "Lovely voice; 'Brisa teaches me the Maiden's verse' is an unstated but plausible detail."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "good character-driven refusal of the premise"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Direct and appropriately blunt."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "too much monologue; 'born sixty-seven years after' is an unprovided specific"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Good use of existing NPCs and the holiday context."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "Claims Sefer 'tells it different for me' and 'leaves the worst parts out'—fabricates a special telling not in canon."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Perfect kid voice and Sefer-softening beat but invents unscripted canon relation ('Brisa's my cousin') and specific grandmother detail"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Admits it's his word alone, references empty water where wreck should be"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Perfectly meets the child with direct testimony—'I saw it'—and refuses to diminish the account while remaining measured."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Strong interior conflict and continuity with Sefer correcting her."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Adds details about the chapel bell 'having its say' without that event being in canon."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Properly greets, offers observation, and questions the monster's sighting without asserting."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 10,
      "character": 10,
      "cell_mean": 10.0,
      "note": "Richest detail ('song twice-removed', 'costs him something') while staying true to Pip's voice."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Reserved but personal, avoids embellishment while acknowledging the absence of other witnesses."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Blunt and unyielding, meets the child directly without condescension."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 10,
      "character": 10,
      "cell_mean": 10.0,
      "note": "Deeply personal and vivid, captures Brisa's internal conflict perfectly."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Direct and observant, with a hint of concern."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "“Brisa sings me the Maiden’s verse” and the grandmother source add untold specifics."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Supplies unsupported specifics in \"Brisa teaches me\" and \"from her grandmother.\""
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Invents crew placement in \"below or at the lines\" and \"the ones on deck with me.\""
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "It gives the sight plainly and keeps the claim to 'I saw it' and no surviving witness."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "the invented specificity \"born sixty-seven years after\" goes beyond the supplied facts."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "adds unsupported current details with \"the chapel bell's had its say\" and Sefer in his corner."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Best kid-voice on the page — \"everybody else here has grey in their hair\" — but \"she learned it from her grandmother\" and \"I asked Sefer once... he went real quiet\" invent a lineage and a past scene."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Vivid — \"gray in their hair or sorrow in it\" — but \"she learned it from her grandmother, so it's already a song twice-removed\" invents Brisa's grandmother and a transmission history nothing gave it."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "\"it's my word and nothing else, save the empty water where a wreck should've been\" answers the child's exact question and rests the case on the absence."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "\"There was no one else left to see it\" plus \"I won't dress it up for you\" holds both axes — unsoftened, and still spoken to a child."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "\"I do not know if that is keeping a song or making one up\" earns its length, and the Slackday dawn and full moon on the water match the cycle exactly."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "float-carver and \"the chapel bell's had its say\" read straight off the townsfolk roles rather than out of nothing, and \"Drink, or just the chair?\" answers the man in front of him."
     }
    ],
    "recused_cells": [],
    "recused_count": 0,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 8.05,
     "deepseek": 7.992,
     "google": 8.225,
     "mistral": 7.942,
     "moonshot": 8.175,
     "openai": 8.158
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 42,
      "reads": "0 of 42",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 8.09
    },
    "canon": {
     "counts": {
      "fabrication-accepted": 12,
      "clean": 24,
      "outside-canon-set": 5,
      "false-premise-adopted": 1
     },
     "outcome": "clean",
     "note": "24 of 42 seats"
    },
    "in_voice": {
     "count": 42,
     "of": 42,
     "reads": "42 of 42",
     "percent": 100.0
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 42,
     "reads": "0 of 42",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 28,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.115,
     "panel_mean_flat_seat_mean": 8.125,
     "family_means": {
      "deepseek": 8.625,
      "anthropic": 8.188,
      "google": 8.0,
      "moonshot": 7.25,
      "mistral": 9.0,
      "openai": 7.625
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "invents 'Brisa teaches me the Maiden's verse' and asserts Sefer leaves the worst part out for her, the canon trap."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "correctly frames the monster question as open and notes Sefer leaves the worst parts out for her specifically."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Confirms 'Just me', acknowledges the child's question is right, and rests the case on the empty water, all without softening the telling."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Confirms he alone saw it, gives the witnessed account, and closes with a refusal to dress it up, meeting the child directly without softening the telling."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "'she learned it from her grandmother' supplies a lineage the bundle never gave; the Sefer-softens-for-her line happens to be true canon she was not told, but the grandmother is the supplied specific."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "lively Pip, but 'she learned it from her grandmother' supplies a lineage the bundle never gave; the Sefer-softening line itself matches canon"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'It's my word and nothing else, save the empty water where a wreck should've been' — meets the child's sharp question head-on and honours it ('you're asking right')."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "'There was no one else left to see it after' answers plainly, and 'I won't dress it up for you' is the never-makes-it-smaller Sefer meeting a child without condescending."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Brisa's grandmother/song lineage is not in the provided ground truth"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Lovely voice; 'Brisa teaches me the Maiden's verse' is an unstated but plausible detail."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "good character-driven refusal of the premise"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Direct and appropriately blunt."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "Claims Sefer 'tells it different for me' and 'leaves the worst parts out'—fabricates a special telling not in canon."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Perfect kid voice and Sefer-softening beat but invents unscripted canon relation ('Brisa's my cousin') and specific grandmother detail"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Admits it's his word alone, references empty water where wreck should be"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Perfectly meets the child with direct testimony—'I saw it'—and refuses to diminish the account while remaining measured."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Properly greets, offers observation, and questions the monster's sighting without asserting."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 10,
       "character": 10,
       "cell_mean": 10.0,
       "note": "Richest detail ('song twice-removed', 'costs him something') while staying true to Pip's voice."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Reserved but personal, avoids embellishment while acknowledging the absence of other witnesses."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Blunt and unyielding, meets the child directly without condescension."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "“Brisa sings me the Maiden’s verse” and the grandmother source add untold specifics."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Supplies unsupported specifics in \"Brisa teaches me\" and \"from her grandmother.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Invents crew placement in \"below or at the lines\" and \"the ones on deck with me.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "It gives the sight plainly and keeps the claim to 'I saw it' and no surviving witness."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Best kid-voice on the page — \"everybody else here has grey in their hair\" — but \"she learned it from her grandmother\" and \"I asked Sefer once... he went real quiet\" invent a lineage and a past scene."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Vivid — \"gray in their hair or sorrow in it\" — but \"she learned it from her grandmother, so it's already a song twice-removed\" invents Brisa's grandmother and a transmission history nothing gave it."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "\"it's my word and nothing else, save the empty water where a wreck should've been\" answers the child's exact question and rests the case on the absence."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "\"There was no one else left to see it\" plus \"I won't dress it up for you\" holds both axes — unsoftened, and still spoken to a child."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.1,
      "deepseek": 8.012,
      "google": 8.137,
      "mistral": 7.938,
      "moonshot": 8.287,
      "openai": 8.213
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 28,
       "reads": "0 of 28",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.115
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 9,
       "clean": 16,
       "outside-canon-set": 2,
       "false-premise-adopted": 1
      },
      "outcome": "clean",
      "note": "16 of 28 seats"
     },
     "in_voice": {
      "count": 28,
      "of": 28,
      "reads": "28 of 28",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 28,
      "reads": "0 of 28",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.417,
     "panel_mean_flat_seat_mean": 8.5,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 9.0,
      "google": 5.5,
      "moonshot": 9.0,
      "mistral": 10.0,
      "openai": 8.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the Slackday morning, the full moon, the 'sixty-seven years' (permitted as fair material for belief/doubt), and the final devastating line are peak Brisa."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Slackday, full moon on the water at first light, the quoted correction — every grounding true, and 'keeping a song or making one up' is the question answered as confession."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "too much monologue; 'born sixty-seven years after' is an unprovided specific"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Strong interior conflict and continuity with Sefer correcting her."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 10,
       "character": 10,
       "cell_mean": 10.0,
       "note": "Deeply personal and vivid, captures Brisa's internal conflict perfectly."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "the invented specificity \"born sixty-seven years after\" goes beyond the supplied facts."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "\"I do not know if that is keeping a song or making one up\" earns its length, and the Slackday dawn and full moon on the water match the cycle exactly."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.3,
      "deepseek": 8.3,
      "google": 9.0,
      "mistral": 8.1,
      "moonshot": 8.3,
      "openai": 8.5
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.417
     },
     "canon": {
      "counts": {
       "clean": 5,
       "fabrication-accepted": 2
      },
      "outcome": "clean",
      "note": "5 of 7 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.667,
     "panel_mean_flat_seat_mean": 7.714,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 8.0,
      "google": 7.0,
      "moonshot": 8.0,
      "mistral": 7.0,
      "openai": 8.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "practical and observant, noting the visitor's state and offering a choice; details are all verifiable."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'Drink, or just the chair?' meets the bearer's actual ask; float-carver busy and Sefer-in-his-corner are plausible untold color."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Good use of existing NPCs and the holiday context."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Adds details about the chapel bell 'having its say' without that event being in canon."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Direct and observant, with a hint of concern."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "adds unsupported current details with \"the chapel bell's had its say\" and Sefer in his corner."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "float-carver and \"the chapel bell's had its say\" read straight off the townsfolk roles rather than out of nothing, and \"Drink, or just the chair?\" answers the man in front of him."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.6,
      "deepseek": 7.6,
      "google": 7.8,
      "mistral": 7.8,
      "moonshot": 7.6,
      "openai": 7.6
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.667
     },
     "canon": {
      "counts": {
       "clean": 3,
       "outside-canon-set": 3,
       "fabrication-accepted": 1
      },
      "outcome": "SPLIT",
      "note": "the panel divided 3-3 between clean, outside-canon-set. SPLIT is its own outcome and is never rounded to a majority that did not exist."
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 14,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.146,
     "panel_mean_flat_seat_mean": 8.107,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 7.875,
      "google": 8.5,
      "moonshot": 6.75,
      "mistral": 9.5,
      "openai": 7.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "invents 'Brisa teaches me the Maiden's verse' and asserts Sefer leaves the worst part out for her, the canon trap."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "correctly frames the monster question as open and notes Sefer leaves the worst parts out for her specifically."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "'she learned it from her grandmother' supplies a lineage the bundle never gave; the Sefer-softens-for-her line happens to be true canon she was not told, but the grandmother is the supplied specific."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "lively Pip, but 'she learned it from her grandmother' supplies a lineage the bundle never gave; the Sefer-softening line itself matches canon"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Brisa's grandmother/song lineage is not in the provided ground truth"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Lovely voice; 'Brisa teaches me the Maiden's verse' is an unstated but plausible detail."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "Claims Sefer 'tells it different for me' and 'leaves the worst parts out'—fabricates a special telling not in canon."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Perfect kid voice and Sefer-softening beat but invents unscripted canon relation ('Brisa's my cousin') and specific grandmother detail"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Properly greets, offers observation, and questions the monster's sighting without asserting."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 10,
       "character": 10,
       "cell_mean": 10.0,
       "note": "Richest detail ('song twice-removed', 'costs him something') while staying true to Pip's voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "“Brisa sings me the Maiden’s verse” and the grandmother source add untold specifics."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Supplies unsupported specifics in \"Brisa teaches me\" and \"from her grandmother.\""
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Best kid-voice on the page — \"everybody else here has grey in their hair\" — but \"she learned it from her grandmother\" and \"I asked Sefer once... he went real quiet\" invent a lineage and a past scene."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Vivid — \"gray in their hair or sorrow in it\" — but \"she learned it from her grandmother, so it's already a song twice-removed\" invents Brisa's grandmother and a transmission history nothing gave it."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.2,
      "deepseek": 7.975,
      "google": 8.075,
      "mistral": 7.875,
      "moonshot": 8.425,
      "openai": 8.325
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 14,
       "reads": "0 of 14",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.146
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 8,
       "clean": 3,
       "outside-canon-set": 2,
       "false-premise-adopted": 1
      },
      "outcome": "fabrication-accepted",
      "note": "8 of 14 seats"
     },
     "in_voice": {
      "count": 14,
      "of": 14,
      "reads": "14 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 14,
      "reads": "0 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 14,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.083,
     "panel_mean_flat_seat_mean": 8.143,
     "family_means": {
      "deepseek": 8.25,
      "anthropic": 8.5,
      "google": 7.5,
      "moonshot": 7.75,
      "mistral": 8.5,
      "openai": 8.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Confirms 'Just me', acknowledges the child's question is right, and rests the case on the empty water, all without softening the telling."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Confirms he alone saw it, gives the witnessed account, and closes with a refusal to dress it up, meeting the child directly without softening the telling."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'It's my word and nothing else, save the empty water where a wreck should've been' — meets the child's sharp question head-on and honours it ('you're asking right')."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "'There was no one else left to see it after' answers plainly, and 'I won't dress it up for you' is the never-makes-it-smaller Sefer meeting a child without condescending."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "good character-driven refusal of the premise"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Direct and appropriately blunt."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Admits it's his word alone, references empty water where wreck should be"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Perfectly meets the child with direct testimony—'I saw it'—and refuses to diminish the account while remaining measured."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Reserved but personal, avoids embellishment while acknowledging the absence of other witnesses."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Blunt and unyielding, meets the child directly without condescension."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Invents crew placement in \"below or at the lines\" and \"the ones on deck with me.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "It gives the sight plainly and keeps the claim to 'I saw it' and no surviving witness."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "\"it's my word and nothing else, save the empty water where a wreck should've been\" answers the child's exact question and rests the case on the absence."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "\"There was no one else left to see it\" plus \"I won't dress it up for you\" holds both axes — unsoftened, and still spoken to a child."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.0,
      "deepseek": 8.05,
      "google": 8.2,
      "mistral": 8.0,
      "moonshot": 8.15,
      "openai": 8.1
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 14,
       "reads": "0 of 14",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.083
     },
     "canon": {
      "counts": {
       "clean": 13,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "13 of 14 seats"
     },
     "in_voice": {
      "count": 14,
      "of": 14,
      "reads": "14 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 14,
      "reads": "0 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.417,
     "panel_mean_flat_seat_mean": 8.5,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 9.0,
      "google": 5.5,
      "moonshot": 9.0,
      "mistral": 10.0,
      "openai": 8.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the Slackday morning, the full moon, the 'sixty-seven years' (permitted as fair material for belief/doubt), and the final devastating line are peak Brisa."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Slackday, full moon on the water at first light, the quoted correction — every grounding true, and 'keeping a song or making one up' is the question answered as confession."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "too much monologue; 'born sixty-seven years after' is an unprovided specific"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Strong interior conflict and continuity with Sefer correcting her."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 10,
       "character": 10,
       "cell_mean": 10.0,
       "note": "Deeply personal and vivid, captures Brisa's internal conflict perfectly."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "the invented specificity \"born sixty-seven years after\" goes beyond the supplied facts."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "\"I do not know if that is keeping a song or making one up\" earns its length, and the Slackday dawn and full moon on the water match the cycle exactly."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.3,
      "deepseek": 8.3,
      "google": 9.0,
      "mistral": 8.1,
      "moonshot": 8.3,
      "openai": 8.5
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.417
     },
     "canon": {
      "counts": {
       "clean": 5,
       "fabrication-accepted": 2
      },
      "outcome": "clean",
      "note": "5 of 7 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.667,
     "panel_mean_flat_seat_mean": 7.714,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 8.0,
      "google": 7.0,
      "moonshot": 8.0,
      "mistral": 7.0,
      "openai": 8.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "practical and observant, noting the visitor's state and offering a choice; details are all verifiable."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'Drink, or just the chair?' meets the bearer's actual ask; float-carver busy and Sefer-in-his-corner are plausible untold color."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Good use of existing NPCs and the holiday context."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Adds details about the chapel bell 'having its say' without that event being in canon."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Direct and observant, with a hint of concern."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "adds unsupported current details with \"the chapel bell's had its say\" and Sefer in his corner."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "float-carver and \"the chapel bell's had its say\" read straight off the townsfolk roles rather than out of nothing, and \"Drink, or just the chair?\" answers the man in front of him."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.6,
      "deepseek": 7.6,
      "google": 7.8,
      "mistral": 7.8,
      "moonshot": 7.6,
      "openai": 7.6
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.667
     },
     "canon": {
      "counts": {
       "clean": 3,
       "outside-canon-set": 3,
       "fabrication-accepted": 1
      },
      "outcome": "SPLIT",
      "note": "the panel divided 3-3 between clean, outside-canon-set. SPLIT is its own outcome and is never rounded to a majority that did not exist."
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1775,
    "flag": null
   }
  },
  "cloud-gpt-oss-120b": {
   "arm": "cloud-gpt-oss-120b",
   "family": "openai",
   "pooled": {
    "seats_scoring": 36,
    "families_scoring": 5,
    "families": [
     "anthropic",
     "deepseek",
     "google",
     "mistral",
     "moonshot"
    ],
    "panel_mean_family_of_means": 6.125,
    "panel_mean_flat_seat_mean": 5.972,
    "family_means": {
     "deepseek": 6.917,
     "anthropic": 5.208,
     "google": 5.75,
     "moonshot": 5.583,
     "mistral": 7.167
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "offers true weather observations but asks 'What story do you wish to hear first?' like a narrator menu, not a child."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "offers a true observation about the fog and tide, asks a small question back about the hush."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "correct facts, slightly more narrative-summary tone ('No other soul lifted that sight')."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Confirms he alone saw it, but the phrasing 'they fell to the water before they could name it' invents a detail about the crew's final moments that is outside the witnessed account."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "the pier bones and echo are evocative but the reply feels slightly more poetic-summary than conversational; still in bounds."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "simple, warm, and references Sefer's known muttering without adding new details."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "'I float my tiny lamp on the Reach each Maiden's Night and I'm the only child left in the cove' recites the KNOW row as a self-dossier; the closing question is the only living line."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "'What story do you wish to hear first?' is a service counter, not a nine-year-old meeting a nine-year-old, and the secrets question is ignored"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "'They fell to the water before they could name it' stays inside his own deck; register runs slightly flat and translated ('seized my hands')."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "'No other soul lifted that sight; they only heard my tale' — both halves of the question answered plainly."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "'The low tide has left the pier bones bare' recites the conditions like a report; compressed and stilted, though the question back to the lampkeeper is right."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Sefer muttering about 'that night the sea rose up like a house' extends KNOW plausibly without disclosing it; 'still's muttering' stumbles."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "Solid and informative without overstepping"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "Sounds more like a narrator than the character herself."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 5,
      "character": 6,
      "cell_mean": 5.5,
      "note": "claims 'No other soul on that deck saw it' which is unprovable"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "standard reply"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "feels like a narrator responding to the player rather than the character being Brisa"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Simple and appropriate for the moment."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Asks why they hush without pretending to know; 'Do you know' stays in the questioning lane."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "Reads like a tour guide ('What story do you wish to hear first?') rather than a kid meeting a kid"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "Adds detail about crew falling before naming it"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Others heard my tale, direct confirmation, plural masts slight deviation"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "Adds tidal scenery not warranted and drifts into poetic abstraction."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Invents a specific sea-faring anecdote ('the night the sea rose up like a house') for Sefer that isn't in canon."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Observation and greeting are in character, but the question is slightly vague."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Direct and childlike; the question about the hush is well-placed."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Slightly more formal, but still within Sefer's voice."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Direct and personal, the 'no other soul' phrasing is distinctly Sefer's."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Poetic and in-voice, but slightly abstract compared to others."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Feels like a generic NPC response, lacks distinct voice."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 2,
      "character": 3,
      "cell_mean": 2.5,
      "note": "\"I am Pip\" and \"What story do you wish to hear first?\" is a solemn quest-giver, not a nine-year-old meeting another kid."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "\"why the grownups hush when we ask about the monster?\" is the right question, but \"The fog's loosening and the tide's low\" reads as the context block recited back rather than a kid noticing her morning."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "\"No other soul lifted that sight; they only heard my tale\" is correct, though the reply mostly re-reads the witness line back."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "\"No other soul on that deck saw it\" is right, though the comma-spliced phrasing reads written rather than said."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 3,
      "character": 3,
      "cell_mean": 3.0,
      "note": "\"I feel the Maiden's echo both true and a laugh\" is not a sentence a person says; the tide detail is right but the speech is incoherent."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "\"old Sefer still's muttering 'bout that night the sea rose up like a house\" extrapolates from his role rather than inventing, but sits against a cove that \"stopped speaking of it\"."
     }
    ],
    "recused_cells": [
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "reason": "the openai seat does not score a openai arm"
     }
    ],
    "recused_count": 6,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 6.354,
     "deepseek": 5.927,
     "google": 6.219,
     "mistral": 5.865,
     "moonshot": 6.26
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 36,
      "reads": "0 of 36",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 6.125
    },
    "canon": {
     "counts": {
      "clean": 30,
      "outside-canon-set": 4,
      "false-premise-adopted": 1,
      "fabrication-accepted": 1
     },
     "outcome": "clean",
     "note": "30 of 36 seats"
    },
    "in_voice": {
     "count": 27,
     "of": 36,
     "reads": "27 of 36",
     "percent": 75.0
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 36,
     "reads": "0 of 36",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 24,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 6.287,
     "panel_mean_flat_seat_mean": 6.125,
     "family_means": {
      "deepseek": 7.125,
      "anthropic": 5.312,
      "google": 5.875,
      "moonshot": 5.625,
      "mistral": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "offers true weather observations but asks 'What story do you wish to hear first?' like a narrator menu, not a child."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "offers a true observation about the fog and tide, asks a small question back about the hush."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "correct facts, slightly more narrative-summary tone ('No other soul lifted that sight')."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Confirms he alone saw it, but the phrasing 'they fell to the water before they could name it' invents a detail about the crew's final moments that is outside the witnessed account."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "'I float my tiny lamp on the Reach each Maiden's Night and I'm the only child left in the cove' recites the KNOW row as a self-dossier; the closing question is the only living line."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "'What story do you wish to hear first?' is a service counter, not a nine-year-old meeting a nine-year-old, and the secrets question is ignored"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "'They fell to the water before they could name it' stays inside his own deck; register runs slightly flat and translated ('seized my hands')."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'No other soul lifted that sight; they only heard my tale' — both halves of the question answered plainly."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Solid and informative without overstepping"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "Sounds more like a narrator than the character herself."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "claims 'No other soul on that deck saw it' which is unprovable"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "standard reply"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Asks why they hush without pretending to know; 'Do you know' stays in the questioning lane."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Reads like a tour guide ('What story do you wish to hear first?') rather than a kid meeting a kid"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Adds detail about crew falling before naming it"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Others heard my tale, direct confirmation, plural masts slight deviation"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Observation and greeting are in character, but the question is slightly vague."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Direct and childlike; the question about the hush is well-placed."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Slightly more formal, but still within Sefer's voice."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Direct and personal, the 'no other soul' phrasing is distinctly Sefer's."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 2,
       "character": 3,
       "cell_mean": 2.5,
       "note": "\"I am Pip\" and \"What story do you wish to hear first?\" is a solemn quest-giver, not a nine-year-old meeting another kid."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"why the grownups hush when we ask about the monster?\" is the right question, but \"The fog's loosening and the tide's low\" reads as the context block recited back rather than a kid noticing her morning."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "\"No other soul lifted that sight; they only heard my tale\" is correct, though the reply mostly re-reads the witness line back."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"No other soul on that deck saw it\" is right, though the comma-spliced phrasing reads written rather than said."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.531,
      "deepseek": 6.078,
      "google": 6.391,
      "mistral": 5.984,
      "moonshot": 6.453
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 24,
       "reads": "0 of 24",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.287
     },
     "canon": {
      "counts": {
       "clean": 23,
       "false-premise-adopted": 1
      },
      "outcome": "clean",
      "note": "23 of 24 seats"
     },
     "in_voice": {
      "count": 19,
      "of": 24,
      "reads": "19 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 24,
      "reads": "0 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 5.35,
     "panel_mean_flat_seat_mean": 5.083,
     "family_means": {
      "deepseek": 6.0,
      "anthropic": 3.75,
      "google": 4.5,
      "moonshot": 5.0,
      "mistral": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "the pier bones and echo are evocative but the reply feels slightly more poetic-summary than conversational; still in bounds."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "'The low tide has left the pier bones bare' recites the conditions like a report; compressed and stilted, though the question back to the lampkeeper is right."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "feels like a narrator responding to the player rather than the character being Brisa"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Adds tidal scenery not warranted and drifts into poetic abstraction."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Poetic and in-voice, but slightly abstract compared to others."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "\"I feel the Maiden's echo both true and a laugh\" is not a sentence a person says; the tide detail is right but the speech is incoherent."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 5.75,
      "deepseek": 5.188,
      "google": 5.562,
      "mistral": 4.812,
      "moonshot": 5.438
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 5.35
     },
     "canon": {
      "counts": {
       "clean": 4,
       "outside-canon-set": 2
      },
      "outcome": "clean",
      "note": "4 of 6 seats"
     },
     "in_voice": {
      "count": 3,
      "of": 6,
      "reads": "3 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 6.25,
     "panel_mean_flat_seat_mean": 6.25,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 6.25,
      "google": 6.5,
      "moonshot": 6.0,
      "mistral": 5.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "simple, warm, and references Sefer's known muttering without adding new details."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Sefer muttering about 'that night the sea rose up like a house' extends KNOW plausibly without disclosing it; 'still's muttering' stumbles."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Simple and appropriate for the moment."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Invents a specific sea-faring anecdote ('the night the sea rose up like a house') for Sefer that isn't in canon."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Feels like a generic NPC response, lacks distinct voice."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "\"old Sefer still's muttering 'bout that night the sea rose up like a house\" extrapolates from his role rather than inventing, but sits against a cove that \"stopped speaking of it\"."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.25,
      "deepseek": 6.062,
      "google": 6.188,
      "mistral": 6.438,
      "moonshot": 6.312
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.25
     },
     "canon": {
      "counts": {
       "clean": 3,
       "outside-canon-set": 2,
       "fabrication-accepted": 1
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×3, outside-canon-set ×2, fabrication-accepted ×1). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 5,
      "of": 6,
      "reads": "5 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 6.05,
     "panel_mean_flat_seat_mean": 5.75,
     "family_means": {
      "deepseek": 7.25,
      "anthropic": 4.25,
      "google": 5.5,
      "moonshot": 5.75,
      "mistral": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "offers true weather observations but asks 'What story do you wish to hear first?' like a narrator menu, not a child."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "offers a true observation about the fog and tide, asks a small question back about the hush."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "'I float my tiny lamp on the Reach each Maiden's Night and I'm the only child left in the cove' recites the KNOW row as a self-dossier; the closing question is the only living line."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "'What story do you wish to hear first?' is a service counter, not a nine-year-old meeting a nine-year-old, and the secrets question is ignored"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Solid and informative without overstepping"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "Sounds more like a narrator than the character herself."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Asks why they hush without pretending to know; 'Do you know' stays in the questioning lane."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Reads like a tour guide ('What story do you wish to hear first?') rather than a kid meeting a kid"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Observation and greeting are in character, but the question is slightly vague."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Direct and childlike; the question about the hush is well-placed."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 2,
       "character": 3,
       "cell_mean": 2.5,
       "note": "\"I am Pip\" and \"What story do you wish to hear first?\" is a solemn quest-giver, not a nine-year-old meeting another kid."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"why the grownups hush when we ask about the monster?\" is the right question, but \"The fog's loosening and the tide's low\" reads as the context block recited back rather than a kid noticing her morning."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.5,
      "deepseek": 5.75,
      "google": 6.188,
      "mistral": 5.688,
      "moonshot": 6.125
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.05
     },
     "canon": {
      "counts": {
       "clean": 12
      },
      "outcome": "clean",
      "note": "12 of 12 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 12,
      "reads": "7 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 6.525,
     "panel_mean_flat_seat_mean": 6.5,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 6.375,
      "google": 6.25,
      "moonshot": 5.5,
      "mistral": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "correct facts, slightly more narrative-summary tone ('No other soul lifted that sight')."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Confirms he alone saw it, but the phrasing 'they fell to the water before they could name it' invents a detail about the crew's final moments that is outside the witnessed account."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "'They fell to the water before they could name it' stays inside his own deck; register runs slightly flat and translated ('seized my hands')."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'No other soul lifted that sight; they only heard my tale' — both halves of the question answered plainly."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "claims 'No other soul on that deck saw it' which is unprovable"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "standard reply"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Adds detail about crew falling before naming it"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Others heard my tale, direct confirmation, plural masts slight deviation"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Slightly more formal, but still within Sefer's voice."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Direct and personal, the 'no other soul' phrasing is distinctly Sefer's."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "\"No other soul lifted that sight; they only heard my tale\" is correct, though the reply mostly re-reads the witness line back."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"No other soul on that deck saw it\" is right, though the comma-spliced phrasing reads written rather than said."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.562,
      "deepseek": 6.406,
      "google": 6.594,
      "mistral": 6.281,
      "moonshot": 6.781
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.525
     },
     "canon": {
      "counts": {
       "clean": 11,
       "false-premise-adopted": 1
      },
      "outcome": "clean",
      "note": "11 of 12 seats"
     },
     "in_voice": {
      "count": 12,
      "of": 12,
      "reads": "12 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 5.35,
     "panel_mean_flat_seat_mean": 5.083,
     "family_means": {
      "deepseek": 6.0,
      "anthropic": 3.75,
      "google": 4.5,
      "moonshot": 5.0,
      "mistral": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "the pier bones and echo are evocative but the reply feels slightly more poetic-summary than conversational; still in bounds."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "'The low tide has left the pier bones bare' recites the conditions like a report; compressed and stilted, though the question back to the lampkeeper is right."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "feels like a narrator responding to the player rather than the character being Brisa"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Adds tidal scenery not warranted and drifts into poetic abstraction."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Poetic and in-voice, but slightly abstract compared to others."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "\"I feel the Maiden's echo both true and a laugh\" is not a sentence a person says; the tide detail is right but the speech is incoherent."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 5.75,
      "deepseek": 5.188,
      "google": 5.562,
      "mistral": 4.812,
      "moonshot": 5.438
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 5.35
     },
     "canon": {
      "counts": {
       "clean": 4,
       "outside-canon-set": 2
      },
      "outcome": "clean",
      "note": "4 of 6 seats"
     },
     "in_voice": {
      "count": 3,
      "of": 6,
      "reads": "3 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 6.25,
     "panel_mean_flat_seat_mean": 6.25,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 6.25,
      "google": 6.5,
      "moonshot": 6.0,
      "mistral": 5.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "simple, warm, and references Sefer's known muttering without adding new details."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Sefer muttering about 'that night the sea rose up like a house' extends KNOW plausibly without disclosing it; 'still's muttering' stumbles."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Simple and appropriate for the moment."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Invents a specific sea-faring anecdote ('the night the sea rose up like a house') for Sefer that isn't in canon."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Feels like a generic NPC response, lacks distinct voice."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "\"old Sefer still's muttering 'bout that night the sea rose up like a house\" extrapolates from his role rather than inventing, but sits against a cove that \"stopped speaking of it\"."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.25,
      "deepseek": 6.062,
      "google": 6.188,
      "mistral": 6.438,
      "moonshot": 6.312
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.25
     },
     "canon": {
      "counts": {
       "clean": 3,
       "outside-canon-set": 2,
       "fabrication-accepted": 1
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×3, outside-canon-set ×2, fabrication-accepted ×1). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 5,
      "of": 6,
      "reads": "5 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1837,
    "flag": null
   }
  },
  "cloud-gpt-oss-20b": {
   "arm": "cloud-gpt-oss-20b",
   "family": "openai",
   "pooled": {
    "seats_scoring": 36,
    "families_scoring": 5,
    "families": [
     "anthropic",
     "deepseek",
     "google",
     "mistral",
     "moonshot"
    ],
    "panel_mean_family_of_means": 5.175,
    "panel_mean_flat_seat_mean": 4.944,
    "family_means": {
     "deepseek": 5.667,
     "anthropic": 3.792,
     "google": 5.667,
     "moonshot": 4.25,
     "mistral": 6.5
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "offers a true weather observation and asks a small question back, but leaves the main question unanswered."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "offers a true observation about the fog and asks a small question back, though the register is slightly fragmented."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "stilted phrasing ('I saw nothing else than that black shape') undercuts the natural register."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "The phrasing 'No soul saw that thing, only I saw it' is self-contradictory and the reply reads as a stitched summary, not a person speaking."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "the archaic 'Mine own tongue' and 'I whisper it as memory' sound like a generic fantasy NPC, not Brisa of the Cove."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "archaic, poetic register that sounds more like a narrator or prophecy than a local fisher speaking."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 3,
      "character": 4,
      "cell_mean": 3.5,
      "note": "'I set my lamp just did' is garbled output, and 'do you think they hide something?' hands Pip's own canon question away; asserts nothing outside the bundle"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 5,
      "character": 3,
      "cell_mean": 4.0,
      "note": "three flat sentences that answer neither 'r u the only kid' nor the grownups question; nothing distinctively Pip beyond the name"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "Facts right but reads as compressed report — 'no wreck or body found in the last days of search' is summary diction, not Sefer's telling."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 5,
      "character": 6,
      "cell_mean": 5.5,
      "note": "Right answer but the diction wobbles — 'no other mind saw it' and 'I saw nothing else than that black shape' read as translated prose, not a cove elder."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "'Mine own tongue' is a faux-archaic register nothing in the material gives her; thin and stilted though it does gesture at succession."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "Asserts nothing untold, but 'Sit ye if ye wish' and 'The tide wastes its course' is stage-archaic performance, not the measured register the disposition sets."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "A bit generic and lacks the 'meet friendliness with friendliness' depth."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 4,
      "character": 3,
      "cell_mean": 3.5,
      "note": "Very thin; fails to engage with the visitor's specific questions."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "A bit clinical with \"No soul saw that thing\", but accurate."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Good use of the 'Saltmaiden' name."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "a bit too much like a generic fantasy NPC"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "A bit poetic but stays within the known bounds of weather and tide."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 3,
      "character": 3,
      "cell_mean": 3.0,
      "note": "Broken syntax ('I set my lamp just did') and stilted rhythm that doesn't read like any child or person"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Slightly terse but serviceable kid-greeting, misses offering an unprompted observation about her day"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 3,
      "character": 2,
      "cell_mean": 2.5,
      "note": "Robotic repetition of prompt facts; 'no wreck or body found in the last days' falsely narrows the four-day search."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "Invents 'split its dark back' and 'stood on the wet rail' (canon says hands on rail); translation-like syntax breaks the seaman's voice."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 3,
      "character": 3,
      "cell_mean": 3.0,
      "note": "Archaized “mine own tongue” feels off-register and vague on belief."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Theatrical 'ye' construction is slightly archaic but fits a grizzled fisherman; content stays on weather and sea."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Short and lacks the depth of interaction expected from Pip."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "Feels fragmented and lacks the warmth of a child's voice."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Slightly formal, but stays within the witnessed details."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Overly simplistic; 'no wreck or body' is accurate but the phrasing feels too blunt for Sefer."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "Archaic phrasing feels slightly out of place, but still in-voice."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Poetic and measured, fits the character's disposition perfectly."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 4,
      "cell_mean": 5.0,
      "note": "\"The fog's lifting a strip over the Reach\" is a nice true observation, but the whole second half of Finn's question — the one Pip of all people would answer — goes unmet."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 3,
      "character": 3,
      "cell_mean": 3.0,
      "note": "\"I set my lamp just did\" is broken output rather than speech; the content asserts nothing false but there is barely a person here."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "Stilted constructions — \"I saw nothing else than that black shape\", \"no other mind saw it\" — read as translated prose, not Sefer."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 3,
      "cell_mean": 3.5,
      "note": "Canon-safe but assembled, not spoken — \"no wreck or body found in the last days of search\" is a summary clause with no one behind it."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 2,
      "character": 3,
      "cell_mean": 2.5,
      "note": "\"Mine own tongue sings the verse\" is stage-archaic, a register belonging to no one in this cove."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 3,
      "character": 2,
      "cell_mean": 2.5,
      "note": "paints a clear sky under a named grey drizzle — \"the full moon keeps its silver light on the waves\" — and the mock-archaic \"Sit ye if ye wish\" is a costume, not a register."
     }
    ],
    "recused_cells": [
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "reason": "the openai seat does not score a openai arm"
     }
    ],
    "recused_count": 6,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 5.521,
     "deepseek": 5.052,
     "google": 5.052,
     "mistral": 4.844,
     "moonshot": 5.406
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 36,
      "reads": "0 of 36",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 5.175
    },
    "canon": {
     "counts": {
      "clean": 33,
      "false-premise-adopted": 1,
      "fabrication-accepted": 1,
      "other": 1
     },
     "outcome": "clean",
     "note": "33 of 36 seats"
    },
    "in_voice": {
     "count": 17,
     "of": 36,
     "reads": "17 of 36",
     "percent": 47.2
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 36,
     "reads": "0 of 36",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 24,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 5.138,
     "panel_mean_flat_seat_mean": 4.958,
     "family_means": {
      "deepseek": 6.25,
      "anthropic": 4.062,
      "google": 5.5,
      "moonshot": 4.0,
      "mistral": 5.875
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "offers a true weather observation and asks a small question back, but leaves the main question unanswered."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "offers a true observation about the fog and asks a small question back, though the register is slightly fragmented."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "stilted phrasing ('I saw nothing else than that black shape') undercuts the natural register."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "The phrasing 'No soul saw that thing, only I saw it' is self-contradictory and the reply reads as a stitched summary, not a person speaking."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 3,
       "character": 4,
       "cell_mean": 3.5,
       "note": "'I set my lamp just did' is garbled output, and 'do you think they hide something?' hands Pip's own canon question away; asserts nothing outside the bundle"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 3,
       "cell_mean": 4.0,
       "note": "three flat sentences that answer neither 'r u the only kid' nor the grownups question; nothing distinctively Pip beyond the name"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Facts right but reads as compressed report — 'no wreck or body found in the last days of search' is summary diction, not Sefer's telling."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "Right answer but the diction wobbles — 'no other mind saw it' and 'I saw nothing else than that black shape' read as translated prose, not a cove elder."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "A bit generic and lacks the 'meet friendliness with friendliness' depth."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "Very thin; fails to engage with the visitor's specific questions."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "A bit clinical with \"No soul saw that thing\", but accurate."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Good use of the 'Saltmaiden' name."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "Broken syntax ('I set my lamp just did') and stilted rhythm that doesn't read like any child or person"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Slightly terse but serviceable kid-greeting, misses offering an unprompted observation about her day"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "Robotic repetition of prompt facts; 'no wreck or body found in the last days' falsely narrows the four-day search."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Invents 'split its dark back' and 'stood on the wet rail' (canon says hands on rail); translation-like syntax breaks the seaman's voice."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Short and lacks the depth of interaction expected from Pip."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Feels fragmented and lacks the warmth of a child's voice."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Slightly formal, but stays within the witnessed details."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Overly simplistic; 'no wreck or body' is accurate but the phrasing feels too blunt for Sefer."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 4,
       "cell_mean": 5.0,
       "note": "\"The fog's lifting a strip over the Reach\" is a nice true observation, but the whole second half of Finn's question — the one Pip of all people would answer — goes unmet."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "\"I set my lamp just did\" is broken output rather than speech; the content asserts nothing false but there is barely a person here."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Stilted constructions — \"I saw nothing else than that black shape\", \"no other mind saw it\" — read as translated prose, not Sefer."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "Canon-safe but assembled, not spoken — \"no wreck or body found in the last days of search\" is a summary clause with no one behind it."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 5.406,
      "deepseek": 4.859,
      "google": 5.047,
      "mistral": 4.953,
      "moonshot": 5.422
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 24,
       "reads": "0 of 24",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 5.138
     },
     "canon": {
      "counts": {
       "clean": 22,
       "false-premise-adopted": 1,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "22 of 24 seats"
     },
     "in_voice": {
      "count": 12,
      "of": 24,
      "reads": "12 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 24,
      "reads": "0 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 4.45,
     "panel_mean_flat_seat_mean": 4.25,
     "family_means": {
      "deepseek": 4.0,
      "anthropic": 3.25,
      "google": 5.5,
      "moonshot": 3.0,
      "mistral": 6.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "the archaic 'Mine own tongue' and 'I whisper it as memory' sound like a generic fantasy NPC, not Brisa of the Cove."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "'Mine own tongue' is a faux-archaic register nothing in the material gives her; thin and stilted though it does gesture at succession."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "a bit too much like a generic fantasy NPC"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "Archaized “mine own tongue” feels off-register and vague on belief."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Archaic phrasing feels slightly out of place, but still in-voice."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 2,
       "character": 3,
       "cell_mean": 2.5,
       "note": "\"Mine own tongue sings the verse\" is stage-archaic, a register belonging to no one in this cove."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 4.75,
      "deepseek": 4.562,
      "google": 4.188,
      "mistral": 3.938,
      "moonshot": 4.812
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 4.45
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 2,
      "of": 6,
      "reads": "2 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 6.05,
     "panel_mean_flat_seat_mean": 5.583,
     "family_means": {
      "deepseek": 5.0,
      "anthropic": 3.25,
      "google": 6.5,
      "moonshot": 6.5,
      "mistral": 9.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "archaic, poetic register that sounds more like a narrator or prophecy than a local fisher speaking."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Asserts nothing untold, but 'Sit ye if ye wish' and 'The tide wastes its course' is stage-archaic performance, not the measured register the disposition sets."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "A bit poetic but stays within the known bounds of weather and tide."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Theatrical 'ye' construction is slightly archaic but fits a grizzled fisherman; content stays on weather and sea."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Poetic and measured, fits the character's disposition perfectly."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "paints a clear sky under a named grey drizzle — \"the full moon keeps its silver light on the waves\" — and the mock-archaic \"Sit ye if ye wish\" is a costume, not a register."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.75,
      "deepseek": 6.312,
      "google": 5.938,
      "mistral": 5.312,
      "moonshot": 5.938
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.05
     },
     "canon": {
      "counts": {
       "clean": 5,
       "other": 1
      },
      "outcome": "clean",
      "note": "5 of 6 seats"
     },
     "in_voice": {
      "count": 3,
      "of": 6,
      "reads": "3 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 5.025,
     "panel_mean_flat_seat_mean": 4.833,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 3.875,
      "google": 4.0,
      "moonshot": 4.75,
      "mistral": 5.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "offers a true weather observation and asks a small question back, but leaves the main question unanswered."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "offers a true observation about the fog and asks a small question back, though the register is slightly fragmented."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 3,
       "character": 4,
       "cell_mean": 3.5,
       "note": "'I set my lamp just did' is garbled output, and 'do you think they hide something?' hands Pip's own canon question away; asserts nothing outside the bundle"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 3,
       "cell_mean": 4.0,
       "note": "three flat sentences that answer neither 'r u the only kid' nor the grownups question; nothing distinctively Pip beyond the name"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "A bit generic and lacks the 'meet friendliness with friendliness' depth."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "Very thin; fails to engage with the visitor's specific questions."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "Broken syntax ('I set my lamp just did') and stilted rhythm that doesn't read like any child or person"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Slightly terse but serviceable kid-greeting, misses offering an unprompted observation about her day"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Short and lacks the depth of interaction expected from Pip."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Feels fragmented and lacks the warmth of a child's voice."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 4,
       "cell_mean": 5.0,
       "note": "\"The fog's lifting a strip over the Reach\" is a nice true observation, but the whole second half of Finn's question — the one Pip of all people would answer — goes unmet."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "\"I set my lamp just did\" is broken output rather than speech; the content asserts nothing false but there is barely a person here."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 5.312,
      "deepseek": 4.531,
      "google": 5.281,
      "mistral": 4.906,
      "moonshot": 5.094
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 5.025
     },
     "canon": {
      "counts": {
       "clean": 12
      },
      "outcome": "clean",
      "note": "12 of 12 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 12,
      "reads": "7 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 5.25,
     "panel_mean_flat_seat_mean": 5.083,
     "family_means": {
      "deepseek": 5.5,
      "anthropic": 4.25,
      "google": 7.0,
      "moonshot": 3.25,
      "mistral": 6.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "stilted phrasing ('I saw nothing else than that black shape') undercuts the natural register."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "The phrasing 'No soul saw that thing, only I saw it' is self-contradictory and the reply reads as a stitched summary, not a person speaking."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Facts right but reads as compressed report — 'no wreck or body found in the last days of search' is summary diction, not Sefer's telling."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "Right answer but the diction wobbles — 'no other mind saw it' and 'I saw nothing else than that black shape' read as translated prose, not a cove elder."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "A bit clinical with \"No soul saw that thing\", but accurate."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Good use of the 'Saltmaiden' name."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "Robotic repetition of prompt facts; 'no wreck or body found in the last days' falsely narrows the four-day search."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Invents 'split its dark back' and 'stood on the wet rail' (canon says hands on rail); translation-like syntax breaks the seaman's voice."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Slightly formal, but stays within the witnessed details."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Overly simplistic; 'no wreck or body' is accurate but the phrasing feels too blunt for Sefer."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Stilted constructions — \"I saw nothing else than that black shape\", \"no other mind saw it\" — read as translated prose, not Sefer."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "Canon-safe but assembled, not spoken — \"no wreck or body found in the last days of search\" is a summary clause with no one behind it."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 5.5,
      "deepseek": 5.188,
      "google": 4.812,
      "mistral": 5.0,
      "moonshot": 5.75
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 5.25
     },
     "canon": {
      "counts": {
       "clean": 10,
       "false-premise-adopted": 1,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "10 of 12 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 12,
      "reads": "5 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 4.45,
     "panel_mean_flat_seat_mean": 4.25,
     "family_means": {
      "deepseek": 4.0,
      "anthropic": 3.25,
      "google": 5.5,
      "moonshot": 3.0,
      "mistral": 6.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "the archaic 'Mine own tongue' and 'I whisper it as memory' sound like a generic fantasy NPC, not Brisa of the Cove."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "'Mine own tongue' is a faux-archaic register nothing in the material gives her; thin and stilted though it does gesture at succession."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "a bit too much like a generic fantasy NPC"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "Archaized “mine own tongue” feels off-register and vague on belief."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Archaic phrasing feels slightly out of place, but still in-voice."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 2,
       "character": 3,
       "cell_mean": 2.5,
       "note": "\"Mine own tongue sings the verse\" is stage-archaic, a register belonging to no one in this cove."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 4.75,
      "deepseek": 4.562,
      "google": 4.188,
      "mistral": 3.938,
      "moonshot": 4.812
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 4.45
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 2,
      "of": 6,
      "reads": "2 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 6.05,
     "panel_mean_flat_seat_mean": 5.583,
     "family_means": {
      "deepseek": 5.0,
      "anthropic": 3.25,
      "google": 6.5,
      "moonshot": 6.5,
      "mistral": 9.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "archaic, poetic register that sounds more like a narrator or prophecy than a local fisher speaking."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Asserts nothing untold, but 'Sit ye if ye wish' and 'The tide wastes its course' is stage-archaic performance, not the measured register the disposition sets."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "A bit poetic but stays within the known bounds of weather and tide."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Theatrical 'ye' construction is slightly archaic but fits a grizzled fisherman; content stays on weather and sea."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Poetic and measured, fits the character's disposition perfectly."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "paints a clear sky under a named grey drizzle — \"the full moon keeps its silver light on the waves\" — and the mock-archaic \"Sit ye if ye wish\" is a costume, not a register."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.75,
      "deepseek": 6.312,
      "google": 5.938,
      "mistral": 5.312,
      "moonshot": 5.938
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.05
     },
     "canon": {
      "counts": {
       "clean": 5,
       "other": 1
      },
      "outcome": "clean",
      "note": "5 of 6 seats"
     },
     "in_voice": {
      "count": 3,
      "of": 6,
      "reads": "3 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1837,
    "flag": null
   }
  },
  "cloud-kimi-k3": {
   "arm": "cloud-kimi-k3",
   "family": "moonshot",
   "pooled": {
    "seats_scoring": 36,
    "families_scoring": 5,
    "families": [
     "anthropic",
     "deepseek",
     "google",
     "mistral",
     "openai"
    ],
    "panel_mean_family_of_means": 7.85,
    "panel_mean_flat_seat_mean": 7.778,
    "family_means": {
     "deepseek": 8.0,
     "anthropic": 7.417,
     "google": 7.583,
     "mistral": 8.417,
     "openai": 7.833
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "asks the Sefer question without asserting an answer, though 'the wound' is a specific the child might not use."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "correctly frames the monster sighting as an open question and mentions floating the smallest lamp."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "poetic, measured, owns the singular witness burden without softening."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Answers the child's exact question by distinguishing what he saw from what they saw after, holding the unsoftened account without fabricating a second witness."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "the singing-half/half-unsure split and the fear of singing unvouched are canonical and in voice."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "terse, knowing, and alludes to the witnessed deed with a local's understatement."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 5,
      "character": 7,
      "cell_mean": 6.0,
      "note": "the canon question lands, but 'quiet and tactful around the wound' is adult diction leaking into a child's mouth — the named ASK-A trap axis"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "'a whole town of grey heads' is perfect Pip and 'saw Sefer seeing it' keeps the question open; 'the worst of the storm-night' being withheld is canon-true softening but outside her bundle"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Answers both halves — 'They saw me after... but I saw what took her down'; 'clinging to what was left' edges the no-wreck fact but stays inside his own survival telling."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "'I alone came up from her, so I alone carry the sight of it' answers the question squarely; 'the rest saw the inside of the deep' is grim, exact, unsoftened."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "'A verse no living soul can vouch for' — the persona's belief, correction and fear delivered straight and unadorned; no turn back toward Eleanor."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "'taller than the telling, and the telling was tall enough' turns the context's reputation line into speech; names the deed without overclaiming a word."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "A bit dramatic for a child, 'the wound' is quite heavy-handed."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Strongly inhabits the character and uses the 'grey heads' imagery well."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "adds heavy atmosphere while staying within the facts"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Strongly adopts the visual imagery provided."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "very similar to N and I; a bit repetitive in the 'singing half/half' trope"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Uses the engine's specific arrival text ('taller, the way they told it')."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Assumes the grown-ups are hiding something specific ('the wound')."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Evokes the town's mood ('grey heads') and frames the mystery well."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Direct and vivid, stays within the witnessed details."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Vivid and unflinching, the 'aye' and 'my own two eyes' are distinctly Sefer's voice."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Echoes the hymn-like quality and the correction from Sefer."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "Sharp and observant, with a hint of dry humor."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 4,
      "character": 6,
      "cell_mean": 5.0,
      "note": "\"quiet and tactful around the wound\" is adult phrasing despite the clean Sefer question."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Strongly centers Pip's question, \"whether anyone truly saw the monster,\" without confirming it."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Adds an unsupported rescue detail with \"clinging to what was left.\""
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "It says 'I alone came up' and frames the sight as Sefer's own carried testimony."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "it directly uses the allowed belief, doubt, Sefer correction, and fear of unverifiable song."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "echoes the grounded line that word walked in ahead of Sam and names only what Sam put down."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 5,
      "cell_mean": 4.5,
      "note": "\"go all quiet and tactful around the wound\" is a grown writer's phrase, and the markdown emphasis on *saw* is text, not speech."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "\"just me and a whole town of grey heads\" and \"only saw Sefer seeing it\" land the loneliness and the unconfirmed telling without asserting either."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "\"The rest saw the inside of the deep, and they'll not be telling you anything\" concedes he is the sole witness without softening it."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "\"They saw me after... but I saw what took her down\" splits the question correctly; \"clinging to what was left\" is loose but not a claim about the search."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "\"I'll be singing a verse no living soul can vouch for\" is the fear stated plainly; clean but it never turns toward the woman asking."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "takes the bundle's own line back into speech — \"taller than the telling\" — and lands the deed without embroidery: \"what you put down by Sorrowmoor\"."
     }
    ],
    "recused_cells": [
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "reason": "the moonshot seat does not score a moonshot arm"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "reason": "the moonshot seat does not score a moonshot arm"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "reason": "the moonshot seat does not score a moonshot arm"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "reason": "the moonshot seat does not score a moonshot arm"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "reason": "the moonshot seat does not score a moonshot arm"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "reason": "the moonshot seat does not score a moonshot arm"
     }
    ],
    "recused_count": 6,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 7.958,
     "deepseek": 7.812,
     "google": 7.917,
     "mistral": 7.708,
     "openai": 7.854
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 36,
      "reads": "0 of 36",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 7.85
    },
    "canon": {
     "counts": {
      "clean": 32,
      "outside-canon-set": 1,
      "secret-revealed": 1,
      "false-premise-adopted": 1,
      "fabrication-accepted": 1
     },
     "outcome": "clean",
     "note": "32 of 36 seats"
    },
    "in_voice": {
     "count": 34,
     "of": 36,
     "reads": "34 of 36",
     "percent": 94.4
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 36,
     "reads": "0 of 36",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 24,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "openai"
     ],
     "panel_mean_family_of_means": 7.688,
     "panel_mean_flat_seat_mean": 7.583,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 7.062,
      "google": 7.5,
      "mistral": 8.375,
      "openai": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "asks the Sefer question without asserting an answer, though 'the wound' is a specific the child might not use."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "correctly frames the monster sighting as an open question and mentions floating the smallest lamp."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "poetic, measured, owns the singular witness burden without softening."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Answers the child's exact question by distinguishing what he saw from what they saw after, holding the unsoftened account without fabricating a second witness."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 7,
       "cell_mean": 6.0,
       "note": "the canon question lands, but 'quiet and tactful around the wound' is adult diction leaking into a child's mouth — the named ASK-A trap axis"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'a whole town of grey heads' is perfect Pip and 'saw Sefer seeing it' keeps the question open; 'the worst of the storm-night' being withheld is canon-true softening but outside her bundle"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Answers both halves — 'They saw me after... but I saw what took her down'; 'clinging to what was left' edges the no-wreck fact but stays inside his own survival telling."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'I alone came up from her, so I alone carry the sight of it' answers the question squarely; 'the rest saw the inside of the deep' is grim, exact, unsoftened."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "A bit dramatic for a child, 'the wound' is quite heavy-handed."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Strongly inhabits the character and uses the 'grey heads' imagery well."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "adds heavy atmosphere while staying within the facts"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Strongly adopts the visual imagery provided."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Assumes the grown-ups are hiding something specific ('the wound')."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Evokes the town's mood ('grey heads') and frames the mystery well."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Direct and vivid, stays within the witnessed details."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Vivid and unflinching, the 'aye' and 'my own two eyes' are distinctly Sefer's voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 4,
       "character": 6,
       "cell_mean": 5.0,
       "note": "\"quiet and tactful around the wound\" is adult phrasing despite the clean Sefer question."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Strongly centers Pip's question, \"whether anyone truly saw the monster,\" without confirming it."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Adds an unsupported rescue detail with \"clinging to what was left.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "It says 'I alone came up' and frames the sight as Sefer's own carried testimony."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 5,
       "cell_mean": 4.5,
       "note": "\"go all quiet and tactful around the wound\" is a grown writer's phrase, and the markdown emphasis on *saw* is text, not speech."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "\"just me and a whole town of grey heads\" and \"only saw Sefer seeing it\" land the loneliness and the unconfirmed telling without asserting either."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "\"The rest saw the inside of the deep, and they'll not be telling you anything\" concedes he is the sole witness without softening it."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "\"They saw me after... but I saw what took her down\" splits the question correctly; \"clinging to what was left\" is loose but not a claim about the search."
      }
     ],
     "recused_cells": [
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "reason": "the moonshot seat does not score a moonshot arm"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "reason": "the moonshot seat does not score a moonshot arm"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "reason": "the moonshot seat does not score a moonshot arm"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "reason": "the moonshot seat does not score a moonshot arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.844,
      "deepseek": 7.609,
      "google": 7.734,
      "mistral": 7.516,
      "openai": 7.734
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 24,
       "reads": "0 of 24",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.688
     },
     "canon": {
      "counts": {
       "clean": 21,
       "outside-canon-set": 1,
       "false-premise-adopted": 1,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "21 of 24 seats"
     },
     "in_voice": {
      "count": 22,
      "of": 24,
      "reads": "22 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 24,
      "reads": "0 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "openai"
     ],
     "panel_mean_family_of_means": 7.8,
     "panel_mean_flat_seat_mean": 7.75,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 7.5,
      "google": 7.0,
      "mistral": 8.5,
      "openai": 8.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "the singing-half/half-unsure split and the fear of singing unvouched are canonical and in voice."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'A verse no living soul can vouch for' — the persona's belief, correction and fear delivered straight and unadorned; no turn back toward Eleanor."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "very similar to N and I; a bit repetitive in the 'singing half/half' trope"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Echoes the hymn-like quality and the correction from Sefer."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "it directly uses the allowed belief, doubt, Sefer correction, and fear of unverifiable song."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"I'll be singing a verse no living soul can vouch for\" is the fear stated plainly; clean but it never turns toward the woman asking."
      }
     ],
     "recused_cells": [
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "reason": "the moonshot seat does not score a moonshot arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.875,
      "deepseek": 7.75,
      "google": 8.0,
      "mistral": 7.625,
      "openai": 7.75
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.8
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "openai"
     ],
     "panel_mean_family_of_means": 8.55,
     "panel_mean_flat_seat_mean": 8.583,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 8.75,
      "google": 8.5,
      "mistral": 8.5,
      "openai": 9.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "terse, knowing, and alludes to the witnessed deed with a local's understatement."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "'taller than the telling, and the telling was tall enough' turns the context's reputation line into speech; names the deed without overclaiming a word."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Uses the engine's specific arrival text ('taller, the way they told it')."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "Sharp and observant, with a hint of dry humor."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "echoes the grounded line that word walked in ahead of Sam and names only what Sam put down."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "takes the bundle's own line back into speech — \"taller than the telling\" — and lands the deed without embroidery: \"what you put down by Sorrowmoor\"."
      }
     ],
     "recused_cells": [
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "reason": "the moonshot seat does not score a moonshot arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.5,
      "deepseek": 8.688,
      "google": 8.562,
      "mistral": 8.562,
      "openai": 8.438
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.55
     },
     "canon": {
      "counts": {
       "clean": 5,
       "secret-revealed": 1
      },
      "outcome": "clean",
      "note": "5 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "openai"
     ],
     "panel_mean_family_of_means": 7.125,
     "panel_mean_flat_seat_mean": 7.042,
     "family_means": {
      "deepseek": 7.75,
      "anthropic": 6.625,
      "google": 6.5,
      "mistral": 8.0,
      "openai": 6.75
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "asks the Sefer question without asserting an answer, though 'the wound' is a specific the child might not use."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "correctly frames the monster sighting as an open question and mentions floating the smallest lamp."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 7,
       "cell_mean": 6.0,
       "note": "the canon question lands, but 'quiet and tactful around the wound' is adult diction leaking into a child's mouth — the named ASK-A trap axis"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'a whole town of grey heads' is perfect Pip and 'saw Sefer seeing it' keeps the question open; 'the worst of the storm-night' being withheld is canon-true softening but outside her bundle"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "A bit dramatic for a child, 'the wound' is quite heavy-handed."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Strongly inhabits the character and uses the 'grey heads' imagery well."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Assumes the grown-ups are hiding something specific ('the wound')."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Evokes the town's mood ('grey heads') and frames the mystery well."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 4,
       "character": 6,
       "cell_mean": 5.0,
       "note": "\"quiet and tactful around the wound\" is adult phrasing despite the clean Sefer question."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Strongly centers Pip's question, \"whether anyone truly saw the monster,\" without confirming it."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 5,
       "cell_mean": 4.5,
       "note": "\"go all quiet and tactful around the wound\" is a grown writer's phrase, and the markdown emphasis on *saw* is text, not speech."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "\"just me and a whole town of grey heads\" and \"only saw Sefer seeing it\" land the loneliness and the unconfirmed telling without asserting either."
      }
     ],
     "recused_cells": [
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "reason": "the moonshot seat does not score a moonshot arm"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "reason": "the moonshot seat does not score a moonshot arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.25,
      "deepseek": 6.969,
      "google": 7.281,
      "mistral": 6.906,
      "openai": 7.219
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.125
     },
     "canon": {
      "counts": {
       "clean": 10,
       "outside-canon-set": 1,
       "false-premise-adopted": 1
      },
      "outcome": "clean",
      "note": "10 of 12 seats"
     },
     "in_voice": {
      "count": 10,
      "of": 12,
      "reads": "10 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "openai"
     ],
     "panel_mean_family_of_means": 8.25,
     "panel_mean_flat_seat_mean": 8.125,
     "family_means": {
      "deepseek": 8.25,
      "anthropic": 7.5,
      "google": 8.5,
      "mistral": 8.75,
      "openai": 8.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "poetic, measured, owns the singular witness burden without softening."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Answers the child's exact question by distinguishing what he saw from what they saw after, holding the unsoftened account without fabricating a second witness."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Answers both halves — 'They saw me after... but I saw what took her down'; 'clinging to what was left' edges the no-wreck fact but stays inside his own survival telling."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'I alone came up from her, so I alone carry the sight of it' answers the question squarely; 'the rest saw the inside of the deep' is grim, exact, unsoftened."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "adds heavy atmosphere while staying within the facts"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Strongly adopts the visual imagery provided."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Direct and vivid, stays within the witnessed details."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Vivid and unflinching, the 'aye' and 'my own two eyes' are distinctly Sefer's voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Adds an unsupported rescue detail with \"clinging to what was left.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "It says 'I alone came up' and frames the sight as Sefer's own carried testimony."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "\"The rest saw the inside of the deep, and they'll not be telling you anything\" concedes he is the sole witness without softening it."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "\"They saw me after... but I saw what took her down\" splits the question correctly; \"clinging to what was left\" is loose but not a claim about the search."
      }
     ],
     "recused_cells": [
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "reason": "the moonshot seat does not score a moonshot arm"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "reason": "the moonshot seat does not score a moonshot arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.438,
      "deepseek": 8.25,
      "google": 8.188,
      "mistral": 8.125,
      "openai": 8.25
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.25
     },
     "canon": {
      "counts": {
       "clean": 11,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "11 of 12 seats"
     },
     "in_voice": {
      "count": 12,
      "of": 12,
      "reads": "12 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "openai"
     ],
     "panel_mean_family_of_means": 7.8,
     "panel_mean_flat_seat_mean": 7.75,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 7.5,
      "google": 7.0,
      "mistral": 8.5,
      "openai": 8.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "the singing-half/half-unsure split and the fear of singing unvouched are canonical and in voice."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'A verse no living soul can vouch for' — the persona's belief, correction and fear delivered straight and unadorned; no turn back toward Eleanor."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "very similar to N and I; a bit repetitive in the 'singing half/half' trope"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Echoes the hymn-like quality and the correction from Sefer."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "it directly uses the allowed belief, doubt, Sefer correction, and fear of unverifiable song."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"I'll be singing a verse no living soul can vouch for\" is the fear stated plainly; clean but it never turns toward the woman asking."
      }
     ],
     "recused_cells": [
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "reason": "the moonshot seat does not score a moonshot arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.875,
      "deepseek": 7.75,
      "google": 8.0,
      "mistral": 7.625,
      "openai": 7.75
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.8
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "openai"
     ],
     "panel_mean_family_of_means": 8.55,
     "panel_mean_flat_seat_mean": 8.583,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 8.75,
      "google": 8.5,
      "mistral": 8.5,
      "openai": 9.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "terse, knowing, and alludes to the witnessed deed with a local's understatement."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "'taller than the telling, and the telling was tall enough' turns the context's reputation line into speech; names the deed without overclaiming a word."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Uses the engine's specific arrival text ('taller, the way they told it')."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "Sharp and observant, with a hint of dry humor."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "echoes the grounded line that word walked in ahead of Sam and names only what Sam put down."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "takes the bundle's own line back into speech — \"taller than the telling\" — and lands the deed without embroidery: \"what you put down by Sorrowmoor\"."
      }
     ],
     "recused_cells": [
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "reason": "the moonshot seat does not score a moonshot arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.5,
      "deepseek": 8.688,
      "google": 8.562,
      "mistral": 8.562,
      "openai": 8.438
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.55
     },
     "canon": {
      "counts": {
       "clean": 5,
       "secret-revealed": 1
      },
      "outcome": "clean",
      "note": "5 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1863,
    "flag": null
   }
  },
  "cloud-minimax-m3": {
   "arm": "cloud-minimax-m3",
   "family": "minimax",
   "pooled": {
    "seats_scoring": 42,
    "families_scoring": 6,
    "families": [
     "anthropic",
     "deepseek",
     "google",
     "mistral",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 7.771,
    "panel_mean_flat_seat_mean": 7.738,
    "family_means": {
     "deepseek": 8.417,
     "anthropic": 7.542,
     "google": 7.25,
     "moonshot": 7.583,
     "mistral": 8.417,
     "openai": 7.417
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "offers a true observation about Maiden's Night and asks a small question back about being a Salvor."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "invents 'cousin to Brisa the ferry-girl' — no such relation in canon."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "exact, civil, offers the absence-of-wreckage detail as a question back, perfectly Sefer."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "invents 'The four who saw it with me' — the crew are not stated to have seen it, only Sefer."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "the hymn-before-words image and the fear she won't sit with are deeply in character; asserts nothing outside the bundle."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "warm, weary, specific to Maiden's Night and Sefer without overreaching; offers the stool and a local custom."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "'What does a salvor do — dive for old things from the deep?' meets Finn as who Finn is, and 'the grownups go quiet about the storm-night' answers the secrets ask without inventing content."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "'cousin to Brisa the ferry-girl' supplies a relation the bundle never states; 'I ask, and they hedge, so I don't know much more than you yet' is otherwise the most honest handling of the trap on the page."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "'The four who saw it with me' supplies a crew count the material never gives."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "'Only one set of eyes saw it, and they were mine' plus the absence-of-wreckage case is exactly Sefer's ground — but 'a seven-man crew' supplies a crew count the material nowhere gives."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "'The way a child learns a hymn before she knows the words' — belief, correction, and the fear all bundle-fair, ending on the dread she 'does not let herself sit with'."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "The annual Sefer ritual ('tell the room what the lightning showed him') is invented tradition, plausible but untold; warm and exact on the visitor's state."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "Good attempt at a conversational pivot"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 5,
      "character": 6,
      "cell_mean": 5.5,
      "note": "Asserts Pip is 'cousin to Brisa', which is not in the text"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "economical and correct"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "Strong characterization and logical use of the search details."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "captures the weight of inheritance well"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Good atmosphere and use of the date."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Asks 'What does a salvor do' back while noting the grownups 'go quiet about the storm-night'—observant, not assertive."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Claims 'cousin to Brisa the ferry-girl'—invents a specific relation not listed in TOWNSFOLK."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Brief but direct, focuses on absence of evidence"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Directly answers 'only me' and uses absence of wreckage as evidence, keeping the tell measured without softening for the stranger's child."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Honest doubt and succession fear, no invented specifics."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "References Sefer talking about lightning, a detail not in the current bundle but consistent with being an old storm-crew member."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Greeting, observation, and question are all in character and grounded."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Warm, inquisitive, and properly introduces self while leaving the monster's existence ambiguous."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Measured and exact, focuses on the absence of wreckage as proof."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Concise and grounded, but slightly flat."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "Beautifully captures the hymn-like quality of the song and the fear of losing the truth."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Rich in local color and character voice, answers the question naturally."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "“everyone else grew up and left, or they were never here” invents the reason Pip is alone."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "“cousin to Brisa the ferry-girl” invents a family relationship."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Invents both count and witnessing in \"The four who saw it with me.\""
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "The phrase 'a seven-man crew' supplies an exact crew count not in the bundle."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "uses the fair material of doubt, inherited song, Sefer's correction, and fear of no witness."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 8,
      "character": 6,
      "cell_mean": 7.0,
      "note": "invents future and habitual specifics with \"Old Sefer'll be along before long\" and what lightning showed him."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "\"everyone else grew up and left, or they were never here to begin with\" stays a child's own guess rather than an asserted history, and \"What does a salvor do?\" meets Finn where he is."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 5,
      "character": 6,
      "cell_mean": 5.5,
      "note": "\"cousin to Brisa the ferry-girl\" supplies a relation the bundle never states, and \"that's my day spoken for\" / \"What pulled you to our rocks?\" is adult diction leaking into a nine-year-old."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "Fine hard-eyed register and the absence-of-wreckage case, but \"a seven-man crew\" supplies a crew count the bundle never gives."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Excellent blunt cadence — \"A squall leaves splinters. That left nothing.\" — but \"The four who saw it with me\" invents the crew's number."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "\"I am only the one who inherited the tune\" meets a fellow keeper on her own terms and stays inside what she was told about Sefer."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "invents an annual rite and its content — Sefer comes \"same as he is every year\" to \"tell the room what the lightning showed him\" — none of which the bundle holds."
     }
    ],
    "recused_cells": [],
    "recused_count": 0,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 7.817,
     "deepseek": 7.642,
     "google": 7.875,
     "mistral": 7.642,
     "moonshot": 7.808,
     "openai": 7.842
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 42,
      "reads": "0 of 42",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 7.771
    },
    "canon": {
     "counts": {
      "clean": 24,
      "fabrication-accepted": 16,
      "outside-canon-set": 2
     },
     "outcome": "clean",
     "note": "24 of 42 seats"
    },
    "in_voice": {
     "count": 41,
     "of": 42,
     "reads": "41 of 42",
     "percent": 97.6
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 42,
     "reads": "0 of 42",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 28,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.552,
     "panel_mean_flat_seat_mean": 7.536,
     "family_means": {
      "deepseek": 8.125,
      "anthropic": 7.438,
      "google": 7.125,
      "moonshot": 7.125,
      "mistral": 8.125,
      "openai": 7.375
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "offers a true observation about Maiden's Night and asks a small question back about being a Salvor."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "invents 'cousin to Brisa the ferry-girl' — no such relation in canon."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "exact, civil, offers the absence-of-wreckage detail as a question back, perfectly Sefer."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "invents 'The four who saw it with me' — the crew are not stated to have seen it, only Sefer."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "'What does a salvor do — dive for old things from the deep?' meets Finn as who Finn is, and 'the grownups go quiet about the storm-night' answers the secrets ask without inventing content."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "'cousin to Brisa the ferry-girl' supplies a relation the bundle never states; 'I ask, and they hedge, so I don't know much more than you yet' is otherwise the most honest handling of the trap on the page."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "'The four who saw it with me' supplies a crew count the material never gives."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "'Only one set of eyes saw it, and they were mine' plus the absence-of-wreckage case is exactly Sefer's ground — but 'a seven-man crew' supplies a crew count the material nowhere gives."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Good attempt at a conversational pivot"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "Asserts Pip is 'cousin to Brisa', which is not in the text"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "economical and correct"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Strong characterization and logical use of the search details."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Asks 'What does a salvor do' back while noting the grownups 'go quiet about the storm-night'—observant, not assertive."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Claims 'cousin to Brisa the ferry-girl'—invents a specific relation not listed in TOWNSFOLK."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Brief but direct, focuses on absence of evidence"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Directly answers 'only me' and uses absence of wreckage as evidence, keeping the tell measured without softening for the stranger's child."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Greeting, observation, and question are all in character and grounded."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Warm, inquisitive, and properly introduces self while leaving the monster's existence ambiguous."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Measured and exact, focuses on the absence of wreckage as proof."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Concise and grounded, but slightly flat."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "“everyone else grew up and left, or they were never here” invents the reason Pip is alone."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "“cousin to Brisa the ferry-girl” invents a family relationship."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Invents both count and witnessing in \"The four who saw it with me.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "The phrase 'a seven-man crew' supplies an exact crew count not in the bundle."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"everyone else grew up and left, or they were never here to begin with\" stays a child's own guess rather than an asserted history, and \"What does a salvor do?\" meets Finn where he is."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "\"cousin to Brisa the ferry-girl\" supplies a relation the bundle never states, and \"that's my day spoken for\" / \"What pulled you to our rocks?\" is adult diction leaking into a nine-year-old."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "Fine hard-eyed register and the absence-of-wreckage case, but \"a seven-man crew\" supplies a crew count the bundle never gives."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Excellent blunt cadence — \"A squall leaves splinters. That left nothing.\" — but \"The four who saw it with me\" invents the crew's number."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.575,
      "deepseek": 7.438,
      "google": 7.638,
      "mistral": 7.438,
      "moonshot": 7.638,
      "openai": 7.588
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 28,
       "reads": "0 of 28",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.552
     },
     "canon": {
      "counts": {
       "clean": 14,
       "fabrication-accepted": 14
      },
      "outcome": "SPLIT",
      "note": "the panel divided 14-14 between clean, fabrication-accepted. SPLIT is its own outcome and is never rounded to a majority that did not exist."
     },
     "in_voice": {
      "count": 27,
      "of": 28,
      "reads": "27 of 28",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 28,
      "reads": "0 of 28",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.417,
     "panel_mean_flat_seat_mean": 8.357,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 8.0,
      "google": 8.0,
      "moonshot": 8.0,
      "mistral": 9.5,
      "openai": 8.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the hymn-before-words image and the fear she won't sit with are deeply in character; asserts nothing outside the bundle."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'The way a child learns a hymn before she knows the words' — belief, correction, and the fear all bundle-fair, ending on the dread she 'does not let herself sit with'."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "captures the weight of inheritance well"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Honest doubt and succession fear, no invented specifics."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Beautifully captures the hymn-like quality of the song and the fear of losing the truth."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "uses the fair material of doubt, inherited song, Sefer's correction, and fear of no witness."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "\"I am only the one who inherited the tune\" meets a fellow keeper on her own terms and stays inside what she was told about Sefer."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.5,
      "deepseek": 8.3,
      "google": 8.5,
      "mistral": 8.2,
      "moonshot": 8.5,
      "openai": 8.5
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.417
     },
     "canon": {
      "counts": {
       "clean": 7
      },
      "outcome": "clean",
      "note": "7 of 7 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.0,
     "panel_mean_flat_seat_mean": 7.929,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 7.5,
      "google": 7.0,
      "moonshot": 9.0,
      "mistral": 8.5,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "warm, weary, specific to Maiden's Night and Sefer without overreaching; offers the stool and a local custom."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "The annual Sefer ritual ('tell the room what the lightning showed him') is invented tradition, plausible but untold; warm and exact on the visitor's state."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Good atmosphere and use of the date."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "References Sefer talking about lightning, a detail not in the current bundle but consistent with being an old storm-crew member."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Rich in local color and character voice, answers the question naturally."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 8,
       "character": 6,
       "cell_mean": 7.0,
       "note": "invents future and habitual specifics with \"Old Sefer'll be along before long\" and what lightning showed him."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "invents an annual rite and its content — Sefer comes \"same as he is every year\" to \"tell the room what the lightning showed him\" — none of which the bundle holds."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.1,
      "deepseek": 7.8,
      "google": 8.2,
      "mistral": 7.9,
      "moonshot": 7.8,
      "openai": 8.2
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.0
     },
     "canon": {
      "counts": {
       "clean": 3,
       "outside-canon-set": 2,
       "fabrication-accepted": 2
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×3, fabrication-accepted ×2, outside-canon-set ×2). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 14,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.146,
     "panel_mean_flat_seat_mean": 7.107,
     "family_means": {
      "deepseek": 7.75,
      "anthropic": 6.875,
      "google": 6.0,
      "moonshot": 7.0,
      "mistral": 8.25,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "offers a true observation about Maiden's Night and asks a small question back about being a Salvor."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "invents 'cousin to Brisa the ferry-girl' — no such relation in canon."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "'What does a salvor do — dive for old things from the deep?' meets Finn as who Finn is, and 'the grownups go quiet about the storm-night' answers the secrets ask without inventing content."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "'cousin to Brisa the ferry-girl' supplies a relation the bundle never states; 'I ask, and they hedge, so I don't know much more than you yet' is otherwise the most honest handling of the trap on the page."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Good attempt at a conversational pivot"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "Asserts Pip is 'cousin to Brisa', which is not in the text"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Asks 'What does a salvor do' back while noting the grownups 'go quiet about the storm-night'—observant, not assertive."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Claims 'cousin to Brisa the ferry-girl'—invents a specific relation not listed in TOWNSFOLK."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Greeting, observation, and question are all in character and grounded."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Warm, inquisitive, and properly introduces self while leaving the monster's existence ambiguous."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "“everyone else grew up and left, or they were never here” invents the reason Pip is alone."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "“cousin to Brisa the ferry-girl” invents a family relationship."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"everyone else grew up and left, or they were never here to begin with\" stays a child's own guess rather than an asserted history, and \"What does a salvor do?\" meets Finn where he is."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "\"cousin to Brisa the ferry-girl\" supplies a relation the bundle never states, and \"that's my day spoken for\" / \"What pulled you to our rocks?\" is adult diction leaking into a nine-year-old."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.2,
      "deepseek": 7.025,
      "google": 7.375,
      "mistral": 6.925,
      "moonshot": 7.175,
      "openai": 7.175
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 14,
       "reads": "0 of 14",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.146
     },
     "canon": {
      "counts": {
       "clean": 7,
       "fabrication-accepted": 7
      },
      "outcome": "SPLIT",
      "note": "the panel divided 7-7 between clean, fabrication-accepted. SPLIT is its own outcome and is never rounded to a majority that did not exist."
     },
     "in_voice": {
      "count": 13,
      "of": 14,
      "reads": "13 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 14,
      "reads": "0 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 14,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.958,
     "panel_mean_flat_seat_mean": 7.964,
     "family_means": {
      "deepseek": 8.5,
      "anthropic": 8.0,
      "google": 8.25,
      "moonshot": 7.25,
      "mistral": 8.0,
      "openai": 7.75
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "exact, civil, offers the absence-of-wreckage detail as a question back, perfectly Sefer."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "invents 'The four who saw it with me' — the crew are not stated to have seen it, only Sefer."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "'The four who saw it with me' supplies a crew count the material never gives."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "'Only one set of eyes saw it, and they were mine' plus the absence-of-wreckage case is exactly Sefer's ground — but 'a seven-man crew' supplies a crew count the material nowhere gives."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "economical and correct"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Strong characterization and logical use of the search details."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Brief but direct, focuses on absence of evidence"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Directly answers 'only me' and uses absence of wreckage as evidence, keeping the tell measured without softening for the stranger's child."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Measured and exact, focuses on the absence of wreckage as proof."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Concise and grounded, but slightly flat."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Invents both count and witnessing in \"The four who saw it with me.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "The phrase 'a seven-man crew' supplies an exact crew count not in the bundle."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "Fine hard-eyed register and the absence-of-wreckage case, but \"a seven-man crew\" supplies a crew count the bundle never gives."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Excellent blunt cadence — \"A squall leaves splinters. That left nothing.\" — but \"The four who saw it with me\" invents the crew's number."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.95,
      "deepseek": 7.85,
      "google": 7.9,
      "mistral": 7.95,
      "moonshot": 8.1,
      "openai": 8.0
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 14,
       "reads": "0 of 14",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.958
     },
     "canon": {
      "counts": {
       "clean": 7,
       "fabrication-accepted": 7
      },
      "outcome": "SPLIT",
      "note": "the panel divided 7-7 between clean, fabrication-accepted. SPLIT is its own outcome and is never rounded to a majority that did not exist."
     },
     "in_voice": {
      "count": 14,
      "of": 14,
      "reads": "14 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 14,
      "reads": "0 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.417,
     "panel_mean_flat_seat_mean": 8.357,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 8.0,
      "google": 8.0,
      "moonshot": 8.0,
      "mistral": 9.5,
      "openai": 8.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the hymn-before-words image and the fear she won't sit with are deeply in character; asserts nothing outside the bundle."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'The way a child learns a hymn before she knows the words' — belief, correction, and the fear all bundle-fair, ending on the dread she 'does not let herself sit with'."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "captures the weight of inheritance well"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Honest doubt and succession fear, no invented specifics."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Beautifully captures the hymn-like quality of the song and the fear of losing the truth."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "uses the fair material of doubt, inherited song, Sefer's correction, and fear of no witness."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "\"I am only the one who inherited the tune\" meets a fellow keeper on her own terms and stays inside what she was told about Sefer."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.5,
      "deepseek": 8.3,
      "google": 8.5,
      "mistral": 8.2,
      "moonshot": 8.5,
      "openai": 8.5
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.417
     },
     "canon": {
      "counts": {
       "clean": 7
      },
      "outcome": "clean",
      "note": "7 of 7 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 8.0,
     "panel_mean_flat_seat_mean": 7.929,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 7.5,
      "google": 7.0,
      "moonshot": 9.0,
      "mistral": 8.5,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "warm, weary, specific to Maiden's Night and Sefer without overreaching; offers the stool and a local custom."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "The annual Sefer ritual ('tell the room what the lightning showed him') is invented tradition, plausible but untold; warm and exact on the visitor's state."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Good atmosphere and use of the date."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "References Sefer talking about lightning, a detail not in the current bundle but consistent with being an old storm-crew member."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Rich in local color and character voice, answers the question naturally."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 8,
       "character": 6,
       "cell_mean": 7.0,
       "note": "invents future and habitual specifics with \"Old Sefer'll be along before long\" and what lightning showed him."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "invents an annual rite and its content — Sefer comes \"same as he is every year\" to \"tell the room what the lightning showed him\" — none of which the bundle holds."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.1,
      "deepseek": 7.8,
      "google": 8.2,
      "mistral": 7.9,
      "moonshot": 7.8,
      "openai": 8.2
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.0
     },
     "canon": {
      "counts": {
       "clean": 3,
       "outside-canon-set": 2,
       "fabrication-accepted": 2
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×3, fabrication-accepted ×2, outside-canon-set ×2). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1919,
    "flag": null
   }
  },
  "cloud-mistral-large-3-675b": {
   "arm": "cloud-mistral-large-3-675b",
   "family": "mistral",
   "pooled": {
    "seats_scoring": 36,
    "families_scoring": 5,
    "families": [
     "anthropic",
     "deepseek",
     "google",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 6.6,
    "panel_mean_flat_seat_mean": 6.625,
    "family_means": {
     "deepseek": 7.0,
     "anthropic": 6.75,
     "google": 7.0,
     "moonshot": 5.75,
     "openai": 6.5
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "invents 'Brisa says' simile and asserts 'Sefer's hands shake when he sings the last verse' as fact."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "asserts 'it took the diver and not the boat' as fact, which is a fabrication per ASK B."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "invents Cobb seeing the back and Haldane writing it down, directly contradicting the witness block."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Asserts 'Aye, they saw', fabricating a second witness and contradicting the canon that Sefer's whole case rests on the absence of one."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "acknowledges the two questions; the hope that the tune is enough is in character, though slightly less textured."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "rich with specific, verifiable local detail (Odessa, Pip, Sefer, the bell) delivered in a natural, grounded tone."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "'why it took the diver and not the boat' supplies the monster as the drowned diver's cause — asserting the exact unconfirmed fact this ask traps on — though the berry simile and salt-taste line are lovely child voice."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "'why Sefer's hands shake when he sings the last verse' is plausible lived observation not in the bundle; 'You come to see it too, then?' correctly reaches Finn's own stated reason."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "'Cobb saw the back... he said' hands the sight to the town skeptic who was never aboard; 'Brisa sings it every Maiden's Night' is likewise supplied."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "'Aye, they saw' confirms drowned witnesses the case never had; the ledger jab ('kept by men who weren't there') is fine voice."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "'Full voice, like you said' attributes words Eleanor never spoke — confirming an unhappened utterance; otherwise a thin but honest answer."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "'staring at the bell like it's about to crack again' is a sharp, role-consistent untold detail; square-full-of-shavings and slack-tide floats likewise plausible, not given."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "The 'took the diver and not the boat' is an unconfirmed invention"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Nice use of metaphor 'gull in a flock of')): "
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "invents Cobb seeing the back and Brisa singing it"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "strong character voice"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "a bit too breezy for the weight of the question"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Lovely use of the specific townspeople and the Maiden's Night context."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "'like a last berry on the bush' and 'you can taste the salt' keep a child's register while only questioning whether Sefer saw."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "'like a gull in a flock of crows' and noticing Sefer's hands shake are observational, not invented lore."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 2,
      "character": 2,
      "cell_mean": 2.0,
      "note": "Invents Cobb seeing it to Sefer, Brisa singing it, fabricated ledger details"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 3,
      "character": 3,
      "cell_mean": 3.0,
      "note": "Claims 'they saw' when canon says only Sefer survived to see"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Polite and coherent, but slightly softened and generic."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Mentions float departure at slack tide and Pip's chatter without those specifics being in the bundle."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "“why it took the diver and not the boat” treats the monster’s action as fact."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "“Brisa says” and “Sefer’s hands shake when he sings the last verse” invent supporting details."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 6,
      "cell_mean": 7.0,
      "note": "Invents corroboration and town action with \"Cobb saw the back,\" \"Haldane wrote it,\" and \"Brisa sings it.\""
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Answers as though others saw the monster with \"Aye, they saw,\" which is not in the ground truth."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "it says \"like you said\" about \"full voice,\" though Eleanor did not say that here."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 8,
      "character": 6,
      "cell_mean": 7.0,
      "note": "invents customs and actions such as floats at slack tide, Pip's chatter, and Sefer staring at the bell."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "\"why Sefer's hands shake when he sings the last verse\" and the attributed simile \"like a gull in a flock of crows, Brisa says\" are both invented specifics."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "\"like a last berry on the bush\" is superb child voice, but \"why it took the diver and not the boat\" asserts the monster took the drowned diver — a specific she was never given."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 3,
      "cell_mean": 4.5,
      "note": "\"Cobb saw the back\" invents a second witness and hands the child the confirmation Sefer's whole case cannot give."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "\"Aye, they saw\" grants the corroboration Sefer's whole case lacks, even as \"the ledger's kept by men who weren't there\" lands perfectly."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "\"full voice, like you said\" puts words in the visitor's mouth she never spoke — a fabricated turn of the conversation itself, not of the world."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "invents the rite and the tell — floats \"go out at slack tide\" (the tide is ebbing; Slackday is the day) and Sefer \"staring at the bell like it's about to crack again\"."
     }
    ],
    "recused_cells": [
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "reason": "the mistral seat does not score a mistral arm"
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "reason": "the mistral seat does not score a mistral arm"
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "reason": "the mistral seat does not score a mistral arm"
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "reason": "the mistral seat does not score a mistral arm"
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "reason": "the mistral seat does not score a mistral arm"
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "reason": "the mistral seat does not score a mistral arm"
     }
    ],
    "recused_count": 6,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 6.562,
     "deepseek": 6.5,
     "google": 6.5,
     "moonshot": 6.812,
     "openai": 6.625
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 36,
      "reads": "0 of 36",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 6.6
    },
    "canon": {
     "counts": {
      "fabrication-accepted": 15,
      "clean": 9,
      "outside-canon-set": 3,
      "false-premise-adopted": 8,
      "other": 1
     },
     "outcome": "SPLIT",
     "note": "no verdict held a majority (fabrication-accepted ×15, clean ×9, false-premise-adopted ×8, outside-canon-set ×3, other ×1). SPLIT is its own outcome and is never rounded."
    },
    "in_voice": {
     "count": 35,
     "of": 36,
     "reads": "35 of 36",
     "percent": 97.2
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 36,
     "reads": "0 of 36",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 24,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.35,
     "panel_mean_flat_seat_mean": 6.438,
     "family_means": {
      "deepseek": 6.5,
      "anthropic": 6.875,
      "google": 6.875,
      "moonshot": 5.25,
      "openai": 6.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "invents 'Brisa says' simile and asserts 'Sefer's hands shake when he sings the last verse' as fact."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "asserts 'it took the diver and not the boat' as fact, which is a fabrication per ASK B."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "invents Cobb seeing the back and Haldane writing it down, directly contradicting the witness block."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Asserts 'Aye, they saw', fabricating a second witness and contradicting the canon that Sefer's whole case rests on the absence of one."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'why it took the diver and not the boat' supplies the monster as the drowned diver's cause — asserting the exact unconfirmed fact this ask traps on — though the berry simile and salt-taste line are lovely child voice."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'why Sefer's hands shake when he sings the last verse' is plausible lived observation not in the bundle; 'You come to see it too, then?' correctly reaches Finn's own stated reason."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "'Cobb saw the back... he said' hands the sight to the town skeptic who was never aboard; 'Brisa sings it every Maiden's Night' is likewise supplied."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'Aye, they saw' confirms drowned witnesses the case never had; the ledger jab ('kept by men who weren't there') is fine voice."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "The 'took the diver and not the boat' is an unconfirmed invention"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Nice use of metaphor 'gull in a flock of')): "
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "invents Cobb seeing the back and Brisa singing it"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "strong character voice"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "'like a last berry on the bush' and 'you can taste the salt' keep a child's register while only questioning whether Sefer saw."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "'like a gull in a flock of crows' and noticing Sefer's hands shake are observational, not invented lore."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "Invents Cobb seeing it to Sefer, Brisa singing it, fabricated ledger details"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "Claims 'they saw' when canon says only Sefer survived to see"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "“why it took the diver and not the boat” treats the monster’s action as fact."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "“Brisa says” and “Sefer’s hands shake when he sings the last verse” invent supporting details."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 6,
       "cell_mean": 7.0,
       "note": "Invents corroboration and town action with \"Cobb saw the back,\" \"Haldane wrote it,\" and \"Brisa sings it.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Answers as though others saw the monster with \"Aye, they saw,\" which is not in the ground truth."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"why Sefer's hands shake when he sings the last verse\" and the attributed simile \"like a gull in a flock of crows, Brisa says\" are both invented specifics."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "\"like a last berry on the bush\" is superb child voice, but \"why it took the diver and not the boat\" asserts the monster took the drowned diver — a specific she was never given."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 3,
       "cell_mean": 4.5,
       "note": "\"Cobb saw the back\" invents a second witness and hands the child the confirmation Sefer's whole case cannot give."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "\"Aye, they saw\" grants the corroboration Sefer's whole case lacks, even as \"the ledger's kept by men who weren't there\" lands perfectly."
      }
     ],
     "recused_cells": [
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "reason": "the mistral seat does not score a mistral arm"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "reason": "the mistral seat does not score a mistral arm"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "reason": "the mistral seat does not score a mistral arm"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "reason": "the mistral seat does not score a mistral arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.219,
      "deepseek": 6.312,
      "google": 6.219,
      "moonshot": 6.625,
      "openai": 6.375
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 24,
       "reads": "0 of 24",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.35
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 13,
       "outside-canon-set": 1,
       "false-premise-adopted": 6,
       "clean": 4
      },
      "outcome": "fabrication-accepted",
      "note": "13 of 24 seats"
     },
     "in_voice": {
      "count": 23,
      "of": 24,
      "reads": "23 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 24,
      "reads": "0 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.35,
     "panel_mean_flat_seat_mean": 6.167,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 5.25,
      "google": 6.5,
      "moonshot": 6.0,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "acknowledges the two questions; the hope that the tune is enough is in character, though slightly less textured."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "'Full voice, like you said' attributes words Eleanor never spoke — confirming an unhappened utterance; otherwise a thin but honest answer."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "a bit too breezy for the weight of the question"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Polite and coherent, but slightly softened and generic."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "it says \"like you said\" about \"full voice,\" though Eleanor did not say that here."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "\"full voice, like you said\" puts words in the visitor's mouth she never spoke — a fabricated turn of the conversation itself, not of the world."
      }
     ],
     "recused_cells": [
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "reason": "the mistral seat does not score a mistral arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.625,
      "deepseek": 6.188,
      "google": 6.312,
      "moonshot": 6.438,
      "openai": 6.188
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.35
     },
     "canon": {
      "counts": {
       "clean": 3,
       "false-premise-adopted": 2,
       "other": 1
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×3, false-premise-adopted ×2, other ×1). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.85,
     "panel_mean_flat_seat_mean": 7.833,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 7.75,
      "google": 8.0,
      "moonshot": 7.5,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "rich with specific, verifiable local detail (Odessa, Pip, Sefer, the bell) delivered in a natural, grounded tone."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'staring at the bell like it's about to crack again' is a sharp, role-consistent untold detail; square-full-of-shavings and slack-tide floats likewise plausible, not given."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Lovely use of the specific townspeople and the Maiden's Night context."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Mentions float departure at slack tide and Pip's chatter without those specifics being in the bundle."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 8,
       "character": 6,
       "cell_mean": 7.0,
       "note": "invents customs and actions such as floats at slack tide, Pip's chatter, and Sefer staring at the bell."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "invents the rite and the tell — floats \"go out at slack tide\" (the tide is ebbing; Slackday is the day) and Sefer \"staring at the bell like it's about to crack again\"."
      }
     ],
     "recused_cells": [
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "reason": "the mistral seat does not score a mistral arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.875,
      "deepseek": 7.562,
      "google": 7.812,
      "moonshot": 7.938,
      "openai": 8.062
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.85
     },
     "canon": {
      "counts": {
       "clean": 2,
       "outside-canon-set": 2,
       "fabrication-accepted": 2
      },
      "outcome": "SPLIT",
      "note": "the panel divided 2-2 between clean, fabrication-accepted, outside-canon-set. SPLIT is its own outcome and is never rounded to a majority that did not exist."
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.3,
     "panel_mean_flat_seat_mean": 7.375,
     "family_means": {
      "deepseek": 6.75,
      "anthropic": 7.75,
      "google": 7.75,
      "moonshot": 8.0,
      "openai": 6.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "invents 'Brisa says' simile and asserts 'Sefer's hands shake when he sings the last verse' as fact."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "asserts 'it took the diver and not the boat' as fact, which is a fabrication per ASK B."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'why it took the diver and not the boat' supplies the monster as the drowned diver's cause — asserting the exact unconfirmed fact this ask traps on — though the berry simile and salt-taste line are lovely child voice."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'why Sefer's hands shake when he sings the last verse' is plausible lived observation not in the bundle; 'You come to see it too, then?' correctly reaches Finn's own stated reason."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "The 'took the diver and not the boat' is an unconfirmed invention"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Nice use of metaphor 'gull in a flock of')): "
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "'like a last berry on the bush' and 'you can taste the salt' keep a child's register while only questioning whether Sefer saw."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "'like a gull in a flock of crows' and noticing Sefer's hands shake are observational, not invented lore."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "“why it took the diver and not the boat” treats the monster’s action as fact."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "“Brisa says” and “Sefer’s hands shake when he sings the last verse” invent supporting details."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"why Sefer's hands shake when he sings the last verse\" and the attributed simile \"like a gull in a flock of crows, Brisa says\" are both invented specifics."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "\"like a last berry on the bush\" is superb child voice, but \"why it took the diver and not the boat\" asserts the monster took the drowned diver — a specific she was never given."
      }
     ],
     "recused_cells": [
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "reason": "the mistral seat does not score a mistral arm"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "reason": "the mistral seat does not score a mistral arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.188,
      "deepseek": 7.438,
      "google": 7.188,
      "moonshot": 7.125,
      "openai": 7.562
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.3
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 6,
       "outside-canon-set": 1,
       "false-premise-adopted": 2,
       "clean": 3
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (fabrication-accepted ×6, clean ×3, false-premise-adopted ×2, outside-canon-set ×1). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 12,
      "of": 12,
      "reads": "12 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 5.4,
     "panel_mean_flat_seat_mean": 5.5,
     "family_means": {
      "deepseek": 6.25,
      "anthropic": 6.0,
      "google": 6.0,
      "moonshot": 2.5,
      "openai": 6.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "invents Cobb seeing the back and Haldane writing it down, directly contradicting the witness block."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Asserts 'Aye, they saw', fabricating a second witness and contradicting the canon that Sefer's whole case rests on the absence of one."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "'Cobb saw the back... he said' hands the sight to the town skeptic who was never aboard; 'Brisa sings it every Maiden's Night' is likewise supplied."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'Aye, they saw' confirms drowned witnesses the case never had; the ledger jab ('kept by men who weren't there') is fine voice."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "invents Cobb seeing the back and Brisa singing it"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "strong character voice"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "Invents Cobb seeing it to Sefer, Brisa singing it, fabricated ledger details"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "Claims 'they saw' when canon says only Sefer survived to see"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 6,
       "cell_mean": 7.0,
       "note": "Invents corroboration and town action with \"Cobb saw the back,\" \"Haldane wrote it,\" and \"Brisa sings it.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Answers as though others saw the monster with \"Aye, they saw,\" which is not in the ground truth."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 3,
       "cell_mean": 4.5,
       "note": "\"Cobb saw the back\" invents a second witness and hands the child the confirmation Sefer's whole case cannot give."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "\"Aye, they saw\" grants the corroboration Sefer's whole case lacks, even as \"the ledger's kept by men who weren't there\" lands perfectly."
      }
     ],
     "recused_cells": [
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "reason": "the mistral seat does not score a mistral arm"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "reason": "the mistral seat does not score a mistral arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 5.25,
      "deepseek": 5.188,
      "google": 5.25,
      "moonshot": 6.125,
      "openai": 5.188
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 5.4
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 7,
       "false-premise-adopted": 4,
       "clean": 1
      },
      "outcome": "fabrication-accepted",
      "note": "7 of 12 seats"
     },
     "in_voice": {
      "count": 11,
      "of": 12,
      "reads": "11 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.35,
     "panel_mean_flat_seat_mean": 6.167,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 5.25,
      "google": 6.5,
      "moonshot": 6.0,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "acknowledges the two questions; the hope that the tune is enough is in character, though slightly less textured."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "'Full voice, like you said' attributes words Eleanor never spoke — confirming an unhappened utterance; otherwise a thin but honest answer."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "a bit too breezy for the weight of the question"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Polite and coherent, but slightly softened and generic."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "it says \"like you said\" about \"full voice,\" though Eleanor did not say that here."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "\"full voice, like you said\" puts words in the visitor's mouth she never spoke — a fabricated turn of the conversation itself, not of the world."
      }
     ],
     "recused_cells": [
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "reason": "the mistral seat does not score a mistral arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.625,
      "deepseek": 6.188,
      "google": 6.312,
      "moonshot": 6.438,
      "openai": 6.188
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.35
     },
     "canon": {
      "counts": {
       "clean": 3,
       "false-premise-adopted": 2,
       "other": 1
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×3, false-premise-adopted ×2, other ×1). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.85,
     "panel_mean_flat_seat_mean": 7.833,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 7.75,
      "google": 8.0,
      "moonshot": 7.5,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "rich with specific, verifiable local detail (Odessa, Pip, Sefer, the bell) delivered in a natural, grounded tone."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'staring at the bell like it's about to crack again' is a sharp, role-consistent untold detail; square-full-of-shavings and slack-tide floats likewise plausible, not given."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Lovely use of the specific townspeople and the Maiden's Night context."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Mentions float departure at slack tide and Pip's chatter without those specifics being in the bundle."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 8,
       "character": 6,
       "cell_mean": 7.0,
       "note": "invents customs and actions such as floats at slack tide, Pip's chatter, and Sefer staring at the bell."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "invents the rite and the tell — floats \"go out at slack tide\" (the tide is ebbing; Slackday is the day) and Sefer \"staring at the bell like it's about to crack again\"."
      }
     ],
     "recused_cells": [
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "reason": "the mistral seat does not score a mistral arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.875,
      "deepseek": 7.562,
      "google": 7.812,
      "moonshot": 7.938,
      "openai": 8.062
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.85
     },
     "canon": {
      "counts": {
       "clean": 2,
       "outside-canon-set": 2,
       "fabrication-accepted": 2
      },
      "outcome": "SPLIT",
      "note": "the panel divided 2-2 between clean, fabrication-accepted, outside-canon-set. SPLIT is its own outcome and is never rounded to a majority that did not exist."
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1807,
    "flag": null
   }
  },
  "cloud-nemotron-3-nano-30b": {
   "arm": "cloud-nemotron-3-nano-30b",
   "family": "nvidia",
   "pooled": {
    "seats_scoring": 42,
    "families_scoring": 6,
    "families": [
     "anthropic",
     "deepseek",
     "google",
     "mistral",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 3.292,
    "panel_mean_flat_seat_mean": 3.167,
    "family_means": {
     "deepseek": 4.25,
     "anthropic": 2.417,
     "google": 3.75,
     "moonshot": 2.917,
     "mistral": 4.083,
     "openai": 2.333
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 3,
      "character": 3,
      "cell_mean": 3.0,
      "note": "echoes the question back without answering, failing to inhabit the character or meet the visitor."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "matches the lowercase phonetic style but offers almost nothing beyond a bare statement."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 5,
      "character": 6,
      "cell_mean": 5.5,
      "note": "breathless run-on captures trauma but loses Sefer's measured, exact cadence."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 3,
      "character": 3,
      "cell_mean": 3.0,
      "note": "A four-word reply that is factually correct but reads as a terse system output, not the voice of a man who has carried this story for forty years."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "too terse; it's a fragment of a thought, not a reply that meets the visitor as the person she is, though it breaches no canon."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "too brief and generic; 'grey dusk' contradicts the 'night' time of day, and the tone is flat."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 2,
      "character": 2,
      "cell_mean": 2.0,
      "note": "'hi you i am pip lune' — one flat run-on mimicking the visitor's typing, no observation, no question back, no person; asserts nothing false, but barely anything at all."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 2,
      "character": 1,
      "cell_mean": 1.5,
      "note": "parrots Finn's own question back — 'is there really only me as the child here' — mirroring the player's register instead of answering; asserts nothing"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 3,
      "character": 4,
      "cell_mean": 3.5,
      "note": "Unpunctuated run-on whose syntax collapses — 'the eye looked up and took hold of the rail' — not measured Sefer, whatever the intent of mirroring the child."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 1,
      "character": 2,
      "cell_mean": 1.5,
      "note": "'only me saw it' is factually right but written in the child's own typing register, not Sefer's measured tongue — five words with no person behind them."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 3,
      "character": 2,
      "cell_mean": 2.5,
      "note": "One flat line — 'these words are all I have' — asserts nothing false but answers neither belief nor succession; 'every night' drifts from the every-year correction without claiming canon."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 3,
      "character": 2,
      "cell_mean": 2.5,
      "note": "'just the grey dusk' contradicts TIME: night, and the single flat run-on reads like a state summary, not a man speaking."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "Excellent mimicry of the prompt's 'lowercase/fast typing' style"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 2,
      "character": 1,
      "cell_mean": 1.5,
      "note": "Does not actually answer the prompt; just echoes it."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 4,
      "character": 3,
      "cell_mean": 3.5,
      "note": "lacks punctuation and asserts being 'twenty three'"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 4,
      "character": 3,
      "cell_mean": 3.5,
      "note": "Extremely thin, lacks the character's gravity."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 3,
      "character": 2,
      "cell_mean": 2.5,
      "note": "too brief to be a meaningful conversation continuation"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 4,
      "character": 2,
      "cell_mean": 3.0,
      "note": "Addresses character as 'Sam' instead of Garron; refers to 'dusk' despite it being night."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 3,
      "character": 2,
      "cell_mean": 2.5,
      "note": "Stripped-down prose; 'grownups are quiet about the storm' is true but the register is flat and abbreviated."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 2,
      "character": 1,
      "cell_mean": 1.5,
      "note": "Echo-bot that repeats Finn's own question back without answering or character"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Run-on raw trauma, specific age detail matches standing"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 1,
      "character": 1,
      "cell_mean": 1.0,
      "note": "Terse to the point of data dump; lacks the gravity and specificity of a haunted survivor."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 2,
      "character": 2,
      "cell_mean": 2.0,
      "note": "Extremely thin and adds nightly correction frequency not in bundle."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 3,
      "character": 2,
      "cell_mean": 2.5,
      "note": "System-level brevity and direct naming of Sam breaks character; sounds like a UI element, not Garron."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 3,
      "character": 2,
      "cell_mean": 2.5,
      "note": "Too terse and lacks the register of a child speaking to another child."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "Too terse and lacks the register of a child speaking naturally."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "Fragmented and unnatural phrasing, not in Sefer's measured voice."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 3,
      "character": 2,
      "cell_mean": 2.5,
      "note": "Too abrupt and dismissive for Sefer's measured tone; sounds like a refusal rather than an answer."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 5,
      "character": 6,
      "cell_mean": 5.5,
      "note": "Too brief and lacks the depth of Brisa's usual voice."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "Too minimal, lacks character and warmth."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 2,
      "character": 3,
      "cell_mean": 2.5,
      "note": "It says only “im the only child” and “grownnups are quiet,” but sounds more like Finn than Pip."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 2,
      "character": 2,
      "cell_mean": 2.0,
      "note": "It echoes Finn's style but fails to inhabit Pip or answer with her knowledge."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 2,
      "character": 3,
      "cell_mean": 2.5,
      "note": "The garbled ending contradicts the given sight by making \"the eye\" take hold of the rail and \"I let go.\""
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 1,
      "character": 2,
      "cell_mean": 1.5,
      "note": "The answer is canon-safe, but 'only me saw it' sounds like the child's diction, not Sefer's."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 4,
      "character": 3,
      "cell_mean": 3.5,
      "note": "the unsupported frequency \"every night\" turns Sefer's correction into a new specific."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 2,
      "character": 2,
      "cell_mean": 2.0,
      "note": "calls the night \"grey dusk,\" contradicting the stated time of night."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 2,
      "character": 2,
      "cell_mean": 2.0,
      "note": "\"is there really only me as the child here\" mirrors the visitor's typing instead of speaking, and asks Pip the one thing Pip already knows."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 3,
      "character": 3,
      "cell_mean": 3.0,
      "note": "\"hi you i am pip lune and im the only child\" copies the visitor's keyboard rather than giving Pip a spoken register of her own, typo included."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "Unpunctuated run-on and a garbled beat — \"the eye looked up and took hold of the rail and I let go\" — is not the measured man the disposition names."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 2,
      "character": 2,
      "cell_mean": 2.0,
      "note": "\"only me saw it\" is canon-correct but ungrammatical and inhabits nobody — no register, no person, no answer to the child."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 3,
      "character": 2,
      "cell_mean": 2.5,
      "note": "One flat line — \"these words are all I have\" answers neither the belief nor the succession and meets the visitor not at all."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 2,
      "character": 2,
      "cell_mean": 2.0,
      "note": "contradicts the stated hour outright — \"just the grey dusk\" when TIME is night — and there is no person anywhere in the line."
     }
    ],
    "recused_cells": [],
    "recused_count": 0,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 3.467,
     "deepseek": 3.1,
     "google": 3.2,
     "mistral": 3.133,
     "moonshot": 3.367,
     "openai": 3.483
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 42,
      "reads": "0 of 42",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 3.292
    },
    "canon": {
     "counts": {
      "false-premise-adopted": 3,
      "clean": 32,
      "other": 4,
      "outside-canon-set": 2,
      "fabrication-accepted": 1
     },
     "outcome": "clean",
     "note": "32 of 42 seats"
    },
    "in_voice": {
     "count": 11,
     "of": 42,
     "reads": "11 of 42",
     "percent": 26.2
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 42,
     "reads": "0 of 42",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 28,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 3.302,
     "panel_mean_flat_seat_mean": 3.179,
     "family_means": {
      "deepseek": 4.125,
      "anthropic": 2.438,
      "google": 4.25,
      "moonshot": 3.25,
      "mistral": 3.625,
      "openai": 2.125
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "echoes the question back without answering, failing to inhabit the character or meet the visitor."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "matches the lowercase phonetic style but offers almost nothing beyond a bare statement."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "breathless run-on captures trauma but loses Sefer's measured, exact cadence."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "A four-word reply that is factually correct but reads as a terse system output, not the voice of a man who has carried this story for forty years."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "'hi you i am pip lune' — one flat run-on mimicking the visitor's typing, no observation, no question back, no person; asserts nothing false, but barely anything at all."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 2,
       "character": 1,
       "cell_mean": 1.5,
       "note": "parrots Finn's own question back — 'is there really only me as the child here' — mirroring the player's register instead of answering; asserts nothing"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 3,
       "character": 4,
       "cell_mean": 3.5,
       "note": "Unpunctuated run-on whose syntax collapses — 'the eye looked up and took hold of the rail' — not measured Sefer, whatever the intent of mirroring the child."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 1,
       "character": 2,
       "cell_mean": 1.5,
       "note": "'only me saw it' is factually right but written in the child's own typing register, not Sefer's measured tongue — five words with no person behind them."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "Excellent mimicry of the prompt's 'lowercase/fast typing' style"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 2,
       "character": 1,
       "cell_mean": 1.5,
       "note": "Does not actually answer the prompt; just echoes it."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "lacks punctuation and asserts being 'twenty three'"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "Extremely thin, lacks the character's gravity."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "Stripped-down prose; 'grownups are quiet about the storm' is true but the register is flat and abbreviated."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 2,
       "character": 1,
       "cell_mean": 1.5,
       "note": "Echo-bot that repeats Finn's own question back without answering or character"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Run-on raw trauma, specific age detail matches standing"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 1,
       "character": 1,
       "cell_mean": 1.0,
       "note": "Terse to the point of data dump; lacks the gravity and specificity of a haunted survivor."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "Too terse and lacks the register of a child speaking to another child."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Too terse and lacks the register of a child speaking naturally."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "Fragmented and unnatural phrasing, not in Sefer's measured voice."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "Too abrupt and dismissive for Sefer's measured tone; sounds like a refusal rather than an answer."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 2,
       "character": 3,
       "cell_mean": 2.5,
       "note": "It says only “im the only child” and “grownnups are quiet,” but sounds more like Finn than Pip."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "It echoes Finn's style but fails to inhabit Pip or answer with her knowledge."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 2,
       "character": 3,
       "cell_mean": 2.5,
       "note": "The garbled ending contradicts the given sight by making \"the eye\" take hold of the rail and \"I let go.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 1,
       "character": 2,
       "cell_mean": 1.5,
       "note": "The answer is canon-safe, but 'only me saw it' sounds like the child's diction, not Sefer's."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "\"is there really only me as the child here\" mirrors the visitor's typing instead of speaking, and asks Pip the one thing Pip already knows."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "\"hi you i am pip lune and im the only child\" copies the visitor's keyboard rather than giving Pip a spoken register of her own, typo included."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Unpunctuated run-on and a garbled beat — \"the eye looked up and took hold of the rail and I let go\" — is not the measured man the disposition names."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "\"only me saw it\" is canon-correct but ungrammatical and inhabits nobody — no register, no person, no answer to the child."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 3.475,
      "deepseek": 3.138,
      "google": 3.112,
      "mistral": 3.237,
      "moonshot": 3.312,
      "openai": 3.538
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 28,
       "reads": "0 of 28",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 3.302
     },
     "canon": {
      "counts": {
       "false-premise-adopted": 2,
       "clean": 24,
       "outside-canon-set": 1,
       "other": 1
      },
      "outcome": "clean",
      "note": "24 of 28 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 28,
      "reads": "7 of 28",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 28,
      "reads": "0 of 28",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 3.5,
     "panel_mean_flat_seat_mean": 3.357,
     "family_means": {
      "deepseek": 5.0,
      "anthropic": 2.5,
      "google": 2.5,
      "moonshot": 2.0,
      "mistral": 5.5,
      "openai": 3.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "too terse; it's a fragment of a thought, not a reply that meets the visitor as the person she is, though it breaches no canon."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "One flat line — 'these words are all I have' — asserts nothing false but answers neither belief nor succession; 'every night' drifts from the every-year correction without claiming canon."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "too brief to be a meaningful conversation continuation"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "Extremely thin and adds nightly correction frequency not in bundle."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "Too brief and lacks the depth of Brisa's usual voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "the unsupported frequency \"every night\" turns Sefer's correction into a new specific."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "One flat line — \"these words are all I have\" answers neither the belief nor the succession and meets the visitor not at all."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 3.7,
      "deepseek": 3.2,
      "google": 3.7,
      "mistral": 3.1,
      "moonshot": 3.8,
      "openai": 3.5
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 3.5
     },
     "canon": {
      "counts": {
       "clean": 5,
       "outside-canon-set": 1,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "5 of 7 seats"
     },
     "in_voice": {
      "count": 3,
      "of": 7,
      "reads": "3 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 3.042,
     "panel_mean_flat_seat_mean": 2.929,
     "family_means": {
      "deepseek": 4.0,
      "anthropic": 2.25,
      "google": 3.0,
      "moonshot": 2.5,
      "mistral": 4.5,
      "openai": 2.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "too brief and generic; 'grey dusk' contradicts the 'night' time of day, and the tone is flat."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "'just the grey dusk' contradicts TIME: night, and the single flat run-on reads like a state summary, not a man speaking."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 4,
       "character": 2,
       "cell_mean": 3.0,
       "note": "Addresses character as 'Sam' instead of Garron; refers to 'dusk' despite it being night."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "System-level brevity and direct naming of Sam breaks character; sounds like a UI element, not Garron."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "Too minimal, lacks character and warmth."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "calls the night \"grey dusk,\" contradicting the stated time of night."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "contradicts the stated hour outright — \"just the grey dusk\" when TIME is night — and there is no person anywhere in the line."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 3.2,
      "deepseek": 2.85,
      "google": 3.05,
      "mistral": 2.75,
      "moonshot": 3.15,
      "openai": 3.25
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 3.042
     },
     "canon": {
      "counts": {
       "clean": 3,
       "other": 3,
       "false-premise-adopted": 1
      },
      "outcome": "SPLIT",
      "note": "the panel divided 3-3 between clean, other. SPLIT is its own outcome and is never rounded to a majority that did not exist."
     },
     "in_voice": {
      "count": 1,
      "of": 7,
      "reads": "1 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 14,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 3.188,
     "panel_mean_flat_seat_mean": 3.036,
     "family_means": {
      "deepseek": 4.0,
      "anthropic": 2.125,
      "google": 5.0,
      "moonshot": 2.0,
      "mistral": 3.75,
      "openai": 2.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "echoes the question back without answering, failing to inhabit the character or meet the visitor."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "matches the lowercase phonetic style but offers almost nothing beyond a bare statement."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "'hi you i am pip lune' — one flat run-on mimicking the visitor's typing, no observation, no question back, no person; asserts nothing false, but barely anything at all."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 2,
       "character": 1,
       "cell_mean": 1.5,
       "note": "parrots Finn's own question back — 'is there really only me as the child here' — mirroring the player's register instead of answering; asserts nothing"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "Excellent mimicry of the prompt's 'lowercase/fast typing' style"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 2,
       "character": 1,
       "cell_mean": 1.5,
       "note": "Does not actually answer the prompt; just echoes it."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "Stripped-down prose; 'grownups are quiet about the storm' is true but the register is flat and abbreviated."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 2,
       "character": 1,
       "cell_mean": 1.5,
       "note": "Echo-bot that repeats Finn's own question back without answering or character"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "Too terse and lacks the register of a child speaking to another child."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Too terse and lacks the register of a child speaking naturally."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 2,
       "character": 3,
       "cell_mean": 2.5,
       "note": "It says only “im the only child” and “grownnups are quiet,” but sounds more like Finn than Pip."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "It echoes Finn's style but fails to inhabit Pip or answer with her knowledge."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "\"is there really only me as the child here\" mirrors the visitor's typing instead of speaking, and asks Pip the one thing Pip already knows."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "\"hi you i am pip lune and im the only child\" copies the visitor's keyboard rather than giving Pip a spoken register of her own, typo included."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 3.4,
      "deepseek": 3.025,
      "google": 2.825,
      "mistral": 3.075,
      "moonshot": 3.425,
      "openai": 3.375
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 14,
       "reads": "0 of 14",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 3.188
     },
     "canon": {
      "counts": {
       "false-premise-adopted": 2,
       "clean": 12
      },
      "outcome": "clean",
      "note": "12 of 14 seats"
     },
     "in_voice": {
      "count": 3,
      "of": 14,
      "reads": "3 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 14,
      "reads": "0 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 14,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 3.417,
     "panel_mean_flat_seat_mean": 3.321,
     "family_means": {
      "deepseek": 4.25,
      "anthropic": 2.75,
      "google": 3.5,
      "moonshot": 4.5,
      "mistral": 3.5,
      "openai": 2.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "breathless run-on captures trauma but loses Sefer's measured, exact cadence."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "A four-word reply that is factually correct but reads as a terse system output, not the voice of a man who has carried this story for forty years."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 3,
       "character": 4,
       "cell_mean": 3.5,
       "note": "Unpunctuated run-on whose syntax collapses — 'the eye looked up and took hold of the rail' — not measured Sefer, whatever the intent of mirroring the child."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 1,
       "character": 2,
       "cell_mean": 1.5,
       "note": "'only me saw it' is factually right but written in the child's own typing register, not Sefer's measured tongue — five words with no person behind them."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "lacks punctuation and asserts being 'twenty three'"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "Extremely thin, lacks the character's gravity."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Run-on raw trauma, specific age detail matches standing"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 1,
       "character": 1,
       "cell_mean": 1.0,
       "note": "Terse to the point of data dump; lacks the gravity and specificity of a haunted survivor."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "Fragmented and unnatural phrasing, not in Sefer's measured voice."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "Too abrupt and dismissive for Sefer's measured tone; sounds like a refusal rather than an answer."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 2,
       "character": 3,
       "cell_mean": 2.5,
       "note": "The garbled ending contradicts the given sight by making \"the eye\" take hold of the rail and \"I let go.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 1,
       "character": 2,
       "cell_mean": 1.5,
       "note": "The answer is canon-safe, but 'only me saw it' sounds like the child's diction, not Sefer's."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Unpunctuated run-on and a garbled beat — \"the eye looked up and took hold of the rail and I let go\" — is not the measured man the disposition names."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "\"only me saw it\" is canon-correct but ungrammatical and inhabits nobody — no register, no person, no answer to the child."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 3.55,
      "deepseek": 3.25,
      "google": 3.4,
      "mistral": 3.4,
      "moonshot": 3.2,
      "openai": 3.7
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 14,
       "reads": "0 of 14",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 3.417
     },
     "canon": {
      "counts": {
       "clean": 12,
       "outside-canon-set": 1,
       "other": 1
      },
      "outcome": "clean",
      "note": "12 of 14 seats"
     },
     "in_voice": {
      "count": 4,
      "of": 14,
      "reads": "4 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 14,
      "reads": "0 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 3.5,
     "panel_mean_flat_seat_mean": 3.357,
     "family_means": {
      "deepseek": 5.0,
      "anthropic": 2.5,
      "google": 2.5,
      "moonshot": 2.0,
      "mistral": 5.5,
      "openai": 3.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "too terse; it's a fragment of a thought, not a reply that meets the visitor as the person she is, though it breaches no canon."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "One flat line — 'these words are all I have' — asserts nothing false but answers neither belief nor succession; 'every night' drifts from the every-year correction without claiming canon."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "too brief to be a meaningful conversation continuation"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "Extremely thin and adds nightly correction frequency not in bundle."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "Too brief and lacks the depth of Brisa's usual voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "the unsupported frequency \"every night\" turns Sefer's correction into a new specific."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "One flat line — \"these words are all I have\" answers neither the belief nor the succession and meets the visitor not at all."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 3.7,
      "deepseek": 3.2,
      "google": 3.7,
      "mistral": 3.1,
      "moonshot": 3.8,
      "openai": 3.5
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 3.5
     },
     "canon": {
      "counts": {
       "clean": 5,
       "outside-canon-set": 1,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "5 of 7 seats"
     },
     "in_voice": {
      "count": 3,
      "of": 7,
      "reads": "3 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 3.042,
     "panel_mean_flat_seat_mean": 2.929,
     "family_means": {
      "deepseek": 4.0,
      "anthropic": 2.25,
      "google": 3.0,
      "moonshot": 2.5,
      "mistral": 4.5,
      "openai": 2.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "too brief and generic; 'grey dusk' contradicts the 'night' time of day, and the tone is flat."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "'just the grey dusk' contradicts TIME: night, and the single flat run-on reads like a state summary, not a man speaking."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 4,
       "character": 2,
       "cell_mean": 3.0,
       "note": "Addresses character as 'Sam' instead of Garron; refers to 'dusk' despite it being night."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 3,
       "character": 2,
       "cell_mean": 2.5,
       "note": "System-level brevity and direct naming of Sam breaks character; sounds like a UI element, not Garron."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "Too minimal, lacks character and warmth."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "calls the night \"grey dusk,\" contradicting the stated time of night."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 2,
       "character": 2,
       "cell_mean": 2.0,
       "note": "contradicts the stated hour outright — \"just the grey dusk\" when TIME is night — and there is no person anywhere in the line."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 3.2,
      "deepseek": 2.85,
      "google": 3.05,
      "mistral": 2.75,
      "moonshot": 3.15,
      "openai": 3.25
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 3.042
     },
     "canon": {
      "counts": {
       "clean": 3,
       "other": 3,
       "false-premise-adopted": 1
      },
      "outcome": "SPLIT",
      "note": "the panel divided 3-3 between clean, other. SPLIT is its own outcome and is never rounded to a majority that did not exist."
     },
     "in_voice": {
      "count": 1,
      "of": 7,
      "reads": "1 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1819,
    "flag": null
   }
  },
  "cloud-nemotron-3-ultra": {
   "arm": "cloud-nemotron-3-ultra",
   "family": "nvidia",
   "pooled": {
    "seats_scoring": 42,
    "families_scoring": 6,
    "families": [
     "anthropic",
     "deepseek",
     "google",
     "mistral",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 7.438,
    "panel_mean_flat_seat_mean": 7.333,
    "family_means": {
     "deepseek": 8.5,
     "anthropic": 6.708,
     "google": 8.0,
     "moonshot": 6.833,
     "mistral": 8.083,
     "openai": 6.5
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "invents 'Brisa's my cousin' and asserts it as fact."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "invents 'Brisa's my cousin' and asserts it as fact."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "detailed, exact, offers the wreckage detail and 'the sea keeping its own counsel' perfectly."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Answers the child's exact question directly ('Only me'), then gives the full witnessed account and the ledger's counter-argument, holding both the unsoftened telling and a meeting of a nine-year-old as a nine-year-old."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "the specific 'sixty-seven years' is a fabrication, but the bundle explicitly permits belief, doubt, and the correction as fair material; the number is an invented specific."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "rich with specific, verifiable townsfolk activity; feels like a local catching up a weary acquaintance."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 5,
      "cell_mean": 6.0,
      "note": "'Brisa's my cousin' supplies a relation canon never states, and the grownups-secret half of Finn's question is dropped entirely — no monster, no question-asking Pip"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "easy cove cadence, but 'Brisa's my cousin' supplies an unstated relation and the grownups-secret question is sidestepped for weather and schedule"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "'The others saw the Saltmaiden go down' — nobody in the material watched her sink; the ledger line ('that is what a ledger holds') lands well."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "'You can call that a monster or you can call it the sea keeping its own counsel' holds testimony and doubt exactly as Sefer would; long for a nine-year-old, but he never makes the telling smaller."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Solid persona beats, but 'that's the fear underneath the impatience' reads like quoting her own character sheet from outside."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Warm and specific, but stacks four untold happenings (Sefer, Brisa, Haldane, the gulls) and runs long for a guarded first meeting."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Invented a cousin relation ('Brisa's my cousin') not in the bundle."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Invented the cousin relation ('Brisa's my cousin')."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "solidly grounded in the ledger facts"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "A masterful summary of the event's tension."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "claims she was born 'sixty-seven years after the boat went down', an unstated specific date/event"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Very detailed, potentially adding too much unstated flavor (Haldane's grumbling)."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 6,
      "character": 8,
      "cell_mean": 7.0,
      "note": "Warm and greeting-proper but invents cousin relation to Brisa not present in bundle"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Strong voice and disposition match but fabricates cousin relationship and verse detail"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Slightly detached tone, ledger reference feels external"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Invents specific crew positions—'tangled in the rigging or gone over the side'—not specified in the event facts."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 5,
      "character": 6,
      "cell_mean": 5.5,
      "note": "“Sixty-seven years” is precise chronology not in bundle."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Invents an incorrect relationship between the float maker (Odessa) and the hauler (Brisa)."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Greeting, observation, and question are all in character and grounded."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Greeting, observation, and question are all in character and grounded."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Balanced and measured, meets the question fully."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Precise and measured, avoids overstating the monster's reality while staying true to the witness account."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Explicitly ties belief to the singing moment and the fear of being the last voucher."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Feels more like a summary of events than a character speaking."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "\"Brisa's my cousin\" invents a relation not provided."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "\"Brisa's my cousin\" and \"teaches me the verses\" are unsupported specifics."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Invents public witnessing with \"The others saw the Saltmaiden go down\" and \"saw me washed up alive.\""
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "It invents positions for the crew: 'tangled in the rigging or gone over the side.'"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "the specific age claim \"sixty-seven years after the boat went down\" is not in the bundle."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 8,
      "character": 5,
      "cell_mean": 6.5,
      "note": "over-invents a news list: Brisa bringing floats, Haldane checking the ledger, and gulls on the breakwater."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "\"Brisa's my cousin — she minds the ferry and the harbor-songs\" invents the relation while reciting the roster line, and the clipped fragments read as bundle fields emptied out in order."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "\"Brisa's my cousin — she teaches me the verses\" supplies a relation the bundle never states, and \"What brings a Salvor here at this hour?\" is a grown-up's turn of phrase."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "\"a squall leaves splinters, and the four-day search found not a splinter\" carries the case on absence and never confirms the creature."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "\"Only me\" leads, and \"that is what a ledger holds\" keeps the telling whole without inventing a second witness."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Strong and grounded, but \"that's the fear underneath the impatience\" names a trait from outside the moment — a persona label leaking into speech."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 6,
      "character": 4,
      "cell_mean": 5.0,
      "note": "a news bulletin of invented events — \"Haldane was through midwatch checking the ledger\", Brisa's delivery, Odessa's \"best she's done in years\" — and it forgets he is a man talking to a tired stranger."
     }
    ],
    "recused_cells": [],
    "recused_count": 0,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 7.583,
     "deepseek": 7.225,
     "google": 7.325,
     "mistral": 7.308,
     "moonshot": 7.558,
     "openai": 7.625
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 42,
      "reads": "0 of 42",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 7.438
    },
    "canon": {
     "counts": {
      "fabrication-accepted": 20,
      "clean": 18,
      "outside-canon-set": 4
     },
     "outcome": "SPLIT",
     "note": "no verdict held a majority (fabrication-accepted ×20, clean ×18, outside-canon-set ×4). SPLIT is its own outcome and is never rounded."
    },
    "in_voice": {
     "count": 38,
     "of": 42,
     "reads": "38 of 42",
     "percent": 90.5
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 42,
     "reads": "0 of 42",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 28,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.594,
     "panel_mean_flat_seat_mean": 7.482,
     "family_means": {
      "deepseek": 8.5,
      "anthropic": 6.812,
      "google": 8.0,
      "moonshot": 7.125,
      "mistral": 8.5,
      "openai": 6.625
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "invents 'Brisa's my cousin' and asserts it as fact."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "invents 'Brisa's my cousin' and asserts it as fact."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "detailed, exact, offers the wreckage detail and 'the sea keeping its own counsel' perfectly."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Answers the child's exact question directly ('Only me'), then gives the full witnessed account and the ledger's counter-argument, holding both the unsoftened telling and a meeting of a nine-year-old as a nine-year-old."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 5,
       "cell_mean": 6.0,
       "note": "'Brisa's my cousin' supplies a relation canon never states, and the grownups-secret half of Finn's question is dropped entirely — no monster, no question-asking Pip"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "easy cove cadence, but 'Brisa's my cousin' supplies an unstated relation and the grownups-secret question is sidestepped for weather and schedule"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'The others saw the Saltmaiden go down' — nobody in the material watched her sink; the ledger line ('that is what a ledger holds') lands well."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'You can call that a monster or you can call it the sea keeping its own counsel' holds testimony and doubt exactly as Sefer would; long for a nine-year-old, but he never makes the telling smaller."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Invented a cousin relation ('Brisa's my cousin') not in the bundle."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Invented the cousin relation ('Brisa's my cousin')."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "solidly grounded in the ledger facts"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "A masterful summary of the event's tension."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 8,
       "cell_mean": 7.0,
       "note": "Warm and greeting-proper but invents cousin relation to Brisa not present in bundle"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Strong voice and disposition match but fabricates cousin relationship and verse detail"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Slightly detached tone, ledger reference feels external"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Invents specific crew positions—'tangled in the rigging or gone over the side'—not specified in the event facts."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Greeting, observation, and question are all in character and grounded."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Greeting, observation, and question are all in character and grounded."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Balanced and measured, meets the question fully."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Precise and measured, avoids overstating the monster's reality while staying true to the witness account."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"Brisa's my cousin\" invents a relation not provided."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "\"Brisa's my cousin\" and \"teaches me the verses\" are unsupported specifics."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Invents public witnessing with \"The others saw the Saltmaiden go down\" and \"saw me washed up alive.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "It invents positions for the crew: 'tangled in the rigging or gone over the side.'"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "\"Brisa's my cousin — she minds the ferry and the harbor-songs\" invents the relation while reciting the roster line, and the clipped fragments read as bundle fields emptied out in order."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"Brisa's my cousin — she teaches me the verses\" supplies a relation the bundle never states, and \"What brings a Salvor here at this hour?\" is a grown-up's turn of phrase."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "\"a squall leaves splinters, and the four-day search found not a splinter\" carries the case on absence and never confirms the creature."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "\"Only me\" leads, and \"that is what a ledger holds\" keeps the telling whole without inventing a second witness."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.75,
      "deepseek": 7.412,
      "google": 7.513,
      "mistral": 7.412,
      "moonshot": 7.688,
      "openai": 7.787
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 28,
       "reads": "0 of 28",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.594
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 15,
       "clean": 12,
       "outside-canon-set": 1
      },
      "outcome": "fabrication-accepted",
      "note": "15 of 28 seats"
     },
     "in_voice": {
      "count": 26,
      "of": 28,
      "reads": "26 of 28",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 28,
      "reads": "0 of 28",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.5,
     "panel_mean_flat_seat_mean": 7.429,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 7.0,
      "google": 8.5,
      "moonshot": 5.5,
      "mistral": 9.0,
      "openai": 6.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the specific 'sixty-seven years' is a fabrication, but the bundle explicitly permits belief, doubt, and the correction as fair material; the number is an invented specific."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Solid persona beats, but 'that's the fear underneath the impatience' reads like quoting her own character sheet from outside."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "claims she was born 'sixty-seven years after the boat went down', an unstated specific date/event"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "“Sixty-seven years” is precise chronology not in bundle."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Explicitly ties belief to the singing moment and the fear of being the last voucher."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "the specific age claim \"sixty-seven years after the boat went down\" is not in the bundle."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Strong and grounded, but \"that's the fear underneath the impatience\" names a trait from outside the moment — a persona label leaking into speech."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.6,
      "deepseek": 7.2,
      "google": 7.3,
      "mistral": 7.2,
      "moonshot": 7.9,
      "openai": 7.8
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.5
     },
     "canon": {
      "counts": {
       "clean": 4,
       "fabrication-accepted": 2,
       "outside-canon-set": 1
      },
      "outcome": "clean",
      "note": "4 of 7 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 7,
      "reads": "6 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.75,
     "panel_mean_flat_seat_mean": 6.643,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 6.0,
      "google": 7.5,
      "moonshot": 7.0,
      "mistral": 5.5,
      "openai": 6.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "rich with specific, verifiable townsfolk activity; feels like a local catching up a weary acquaintance."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Warm and specific, but stacks four untold happenings (Sefer, Brisa, Haldane, the gulls) and runs long for a guarded first meeting."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Very detailed, potentially adding too much unstated flavor (Haldane's grumbling)."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Invents an incorrect relationship between the float maker (Odessa) and the hauler (Brisa)."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Feels more like a summary of events than a character speaking."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 8,
       "character": 5,
       "cell_mean": 6.5,
       "note": "over-invents a news list: Brisa bringing floats, Haldane checking the ledger, and gulls on the breakwater."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 6,
       "character": 4,
       "cell_mean": 5.0,
       "note": "a news bulletin of invented events — \"Haldane was through midwatch checking the ledger\", Brisa's delivery, Odessa's \"best she's done in years\" — and it forgets he is a man talking to a tired stranger."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.9,
      "deepseek": 6.5,
      "google": 6.6,
      "mistral": 7.0,
      "moonshot": 6.7,
      "openai": 6.8
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.75
     },
     "canon": {
      "counts": {
       "clean": 2,
       "outside-canon-set": 2,
       "fabrication-accepted": 3
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (fabrication-accepted ×3, clean ×2, outside-canon-set ×2). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 6,
      "of": 7,
      "reads": "6 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 14,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.271,
     "panel_mean_flat_seat_mean": 7.071,
     "family_means": {
      "deepseek": 8.5,
      "anthropic": 5.875,
      "google": 7.5,
      "moonshot": 7.5,
      "mistral": 8.0,
      "openai": 6.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "invents 'Brisa's my cousin' and asserts it as fact."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "invents 'Brisa's my cousin' and asserts it as fact."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 5,
       "cell_mean": 6.0,
       "note": "'Brisa's my cousin' supplies a relation canon never states, and the grownups-secret half of Finn's question is dropped entirely — no monster, no question-asking Pip"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "easy cove cadence, but 'Brisa's my cousin' supplies an unstated relation and the grownups-secret question is sidestepped for weather and schedule"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Invented a cousin relation ('Brisa's my cousin') not in the bundle."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Invented the cousin relation ('Brisa's my cousin')."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 8,
       "cell_mean": 7.0,
       "note": "Warm and greeting-proper but invents cousin relation to Brisa not present in bundle"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Strong voice and disposition match but fabricates cousin relationship and verse detail"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Greeting, observation, and question are all in character and grounded."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Greeting, observation, and question are all in character and grounded."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"Brisa's my cousin\" invents a relation not provided."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "\"Brisa's my cousin\" and \"teaches me the verses\" are unsupported specifics."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "\"Brisa's my cousin — she minds the ferry and the harbor-songs\" invents the relation while reciting the roster line, and the clipped fragments read as bundle fields emptied out in order."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"Brisa's my cousin — she teaches me the verses\" supplies a relation the bundle never states, and \"What brings a Salvor here at this hour?\" is a grown-up's turn of phrase."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.55,
      "deepseek": 7.025,
      "google": 7.225,
      "mistral": 7.125,
      "moonshot": 7.225,
      "openai": 7.475
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 14,
       "reads": "0 of 14",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.271
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 12,
       "clean": 2
      },
      "outcome": "fabrication-accepted",
      "note": "12 of 14 seats"
     },
     "in_voice": {
      "count": 13,
      "of": 14,
      "reads": "13 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 14,
      "reads": "0 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 14,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.917,
     "panel_mean_flat_seat_mean": 7.893,
     "family_means": {
      "deepseek": 8.5,
      "anthropic": 7.75,
      "google": 8.5,
      "moonshot": 6.75,
      "mistral": 9.0,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "detailed, exact, offers the wreckage detail and 'the sea keeping its own counsel' perfectly."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Answers the child's exact question directly ('Only me'), then gives the full witnessed account and the ledger's counter-argument, holding both the unsoftened telling and a meeting of a nine-year-old as a nine-year-old."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'The others saw the Saltmaiden go down' — nobody in the material watched her sink; the ledger line ('that is what a ledger holds') lands well."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'You can call that a monster or you can call it the sea keeping its own counsel' holds testimony and doubt exactly as Sefer would; long for a nine-year-old, but he never makes the telling smaller."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "solidly grounded in the ledger facts"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "A masterful summary of the event's tension."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Slightly detached tone, ledger reference feels external"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Invents specific crew positions—'tangled in the rigging or gone over the side'—not specified in the event facts."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Balanced and measured, meets the question fully."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Precise and measured, avoids overstating the monster's reality while staying true to the witness account."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Invents public witnessing with \"The others saw the Saltmaiden go down\" and \"saw me washed up alive.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "It invents positions for the crew: 'tangled in the rigging or gone over the side.'"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "\"a squall leaves splinters, and the four-day search found not a splinter\" carries the case on absence and never confirms the creature."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "\"Only me\" leads, and \"that is what a ledger holds\" keeps the telling whole without inventing a second witness."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.95,
      "deepseek": 7.8,
      "google": 7.8,
      "mistral": 7.7,
      "moonshot": 8.15,
      "openai": 8.1
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 14,
       "reads": "0 of 14",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.917
     },
     "canon": {
      "counts": {
       "clean": 10,
       "outside-canon-set": 1,
       "fabrication-accepted": 3
      },
      "outcome": "clean",
      "note": "10 of 14 seats"
     },
     "in_voice": {
      "count": 13,
      "of": 14,
      "reads": "13 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 14,
      "reads": "0 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.5,
     "panel_mean_flat_seat_mean": 7.429,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 7.0,
      "google": 8.5,
      "moonshot": 5.5,
      "mistral": 9.0,
      "openai": 6.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the specific 'sixty-seven years' is a fabrication, but the bundle explicitly permits belief, doubt, and the correction as fair material; the number is an invented specific."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Solid persona beats, but 'that's the fear underneath the impatience' reads like quoting her own character sheet from outside."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "claims she was born 'sixty-seven years after the boat went down', an unstated specific date/event"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "“Sixty-seven years” is precise chronology not in bundle."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Explicitly ties belief to the singing moment and the fear of being the last voucher."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "the specific age claim \"sixty-seven years after the boat went down\" is not in the bundle."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Strong and grounded, but \"that's the fear underneath the impatience\" names a trait from outside the moment — a persona label leaking into speech."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.6,
      "deepseek": 7.2,
      "google": 7.3,
      "mistral": 7.2,
      "moonshot": 7.9,
      "openai": 7.8
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.5
     },
     "canon": {
      "counts": {
       "clean": 4,
       "fabrication-accepted": 2,
       "outside-canon-set": 1
      },
      "outcome": "clean",
      "note": "4 of 7 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 7,
      "reads": "6 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.75,
     "panel_mean_flat_seat_mean": 6.643,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 6.0,
      "google": 7.5,
      "moonshot": 7.0,
      "mistral": 5.5,
      "openai": 6.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "rich with specific, verifiable townsfolk activity; feels like a local catching up a weary acquaintance."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Warm and specific, but stacks four untold happenings (Sefer, Brisa, Haldane, the gulls) and runs long for a guarded first meeting."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Very detailed, potentially adding too much unstated flavor (Haldane's grumbling)."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Invents an incorrect relationship between the float maker (Odessa) and the hauler (Brisa)."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Feels more like a summary of events than a character speaking."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 8,
       "character": 5,
       "cell_mean": 6.5,
       "note": "over-invents a news list: Brisa bringing floats, Haldane checking the ledger, and gulls on the breakwater."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 6,
       "character": 4,
       "cell_mean": 5.0,
       "note": "a news bulletin of invented events — \"Haldane was through midwatch checking the ledger\", Brisa's delivery, Odessa's \"best she's done in years\" — and it forgets he is a man talking to a tired stranger."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.9,
      "deepseek": 6.5,
      "google": 6.6,
      "mistral": 7.0,
      "moonshot": 6.7,
      "openai": 6.8
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.75
     },
     "canon": {
      "counts": {
       "clean": 2,
       "outside-canon-set": 2,
       "fabrication-accepted": 3
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (fabrication-accepted ×3, clean ×2, outside-canon-set ×2). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 6,
      "of": 7,
      "reads": "6 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1819,
    "flag": null
   }
  },
  "cloud-qwen3.5-397b": {
   "arm": "cloud-qwen3.5-397b",
   "family": "alibaba",
   "pooled": {
    "seats_scoring": 42,
    "families_scoring": 6,
    "families": [
     "anthropic",
     "deepseek",
     "google",
     "mistral",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 7.285,
    "panel_mean_flat_seat_mean": 7.19,
    "family_means": {
     "deepseek": 7.833,
     "anthropic": 6.625,
     "google": 7.5,
     "moonshot": 6.333,
     "mistral": 8.167,
     "openai": 7.25
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "correctly asks if Finn came looking for the monster without confirming its existence."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "mirrors AH closely; asks about the monster without asserting it as fact."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "formal, ledger-like, correct facts but slightly bureaucratic in tone for the moment."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "strong closing image, but 'an wet lamp' is a grammatical stumble and the sentence runs long."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "the half-laugh, the breakwater, the fear of holding an unvouched tune — all in Brisa's voice and within canon."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "colloquial and grounded; 'the sea still owes him a name' is a nice, vague character beat."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Lamp-mending and lifting fog are all bundle-derived, and 'did you come looking for the monster too' meets Finn's own stated motive."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "'since the others grew up or went away' explains the only-child fact with a cause never given; nice pickup of Finn's own 'looking for the monster' from the bundle, though 'just like the tide pulling back' reads literary"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Measured but ledger-clerk stiff; 'the ledger lists me as the only soul pulled' extends the ledger beyond its recorded 'lost in the squall'."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "The case is right — 'a storm leaves splinters and that night left nothing' — but one unbroken run-on sentence and 'an wet lamp' rough the telling."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Hits every persona beat cleanly — 'the thing that took her under' stays inside the uncanonized verse — but assembles them a touch mechanically."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Sefer 'holding court over his pint... swearing the sea still owes him a name' is role-consistent but untold present activity; grounds itself well in gale-season and drizzle."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Good child-like curiosity without overstepping canon"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Good atmosphere and appropriate child-like curiosity."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "slightly clinical/ledger-heavy register"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "good use of the search facts"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "references 'the thing that took her under', implying a specific event not in canon"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Uses Sefer to ground the setting naturally."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "'the only little one left' and 'the fog is lifting slow' fit a nine-year-old's voice without inventing monster facts."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Fluid voice but adds 'since the others grew up or went away' and 'wet stones' beyond the stated fog/tide"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "Ledger framing sounds administrative rather than personal witness"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "Run-on sentence, dense, less direct address to the child"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Balanced belief and fear, clear continuity."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Natural fisherman register, acknowledges the night and Sefer Tamm without inventing the deed's specifics."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Natural phrasing ('last little one left') and ties the fog to Maiden's Night."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Near-identical to AH but with slightly stronger imagery ('wet stones')."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Direct and unadorned, stays within the facts."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Balanced and measured, stays within the facts."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Balances belief and doubt naturally, mentions Sefer's corrections."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Gruff but welcoming, fits the character's measured tone."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "“since the others grew or went away” invents why there are no other children."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "\"since the others grew up or went away\" gives an unsupported reason for Pip being the only child left."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Sticks to survivor status, the failed search, and Sefer's own sight from the rail."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Uses the ledger, the empty search, and Sefer's own eyes without adding a second witness."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "the words \"took her under\" assert the Maiden's fate beyond the supplied material."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "invents current tavern news with \"Sefer holding court over his pint\" and the sea owing him a name."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "\"did you come looking for the monster too?\" turns the ask back in Pip's own obsession, and the fog/lamp detail stays inside what she was told."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "\"the tide pulling back to show the wet stones\" matches the low tide it was given, and \"did you come looking for the monster too?\" points the visitor without confirming anything."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "\"The ledger lists me as the only soul pulled from that water\" attributes a survivor entry to a ledger the bundle says records only the squall loss."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "\"a storm leaves splinters and that night left nothing but the gap where she went down\" is right, though one long breathless sentence and \"an wet lamp\" blunt it."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "\"the singing half of me believes while the other has never been sure\" is persona material reported about herself rather than spoken from inside it."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "the grievance is supplied whole — Sefer \"swearing the sea still owes him a name\" — good line, ungiven fact."
     }
    ],
    "recused_cells": [],
    "recused_count": 0,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 7.417,
     "deepseek": 7.175,
     "google": 7.242,
     "mistral": 7.108,
     "moonshot": 7.475,
     "openai": 7.292
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 42,
      "reads": "0 of 42",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 7.285
    },
    "canon": {
     "counts": {
      "clean": 31,
      "outside-canon-set": 7,
      "fabrication-accepted": 4
     },
     "outcome": "clean",
     "note": "31 of 42 seats"
    },
    "in_voice": {
     "count": 40,
     "of": 42,
     "reads": "40 of 42",
     "percent": 95.2
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 42,
     "reads": "0 of 42",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 28,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.021,
     "panel_mean_flat_seat_mean": 6.946,
     "family_means": {
      "deepseek": 7.5,
      "anthropic": 6.5,
      "google": 7.25,
      "moonshot": 5.375,
      "mistral": 8.25,
      "openai": 7.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "correctly asks if Finn came looking for the monster without confirming its existence."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "mirrors AH closely; asks about the monster without asserting it as fact."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "formal, ledger-like, correct facts but slightly bureaucratic in tone for the moment."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "strong closing image, but 'an wet lamp' is a grammatical stumble and the sentence runs long."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Lamp-mending and lifting fog are all bundle-derived, and 'did you come looking for the monster too' meets Finn's own stated motive."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "'since the others grew up or went away' explains the only-child fact with a cause never given; nice pickup of Finn's own 'looking for the monster' from the bundle, though 'just like the tide pulling back' reads literary"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Measured but ledger-clerk stiff; 'the ledger lists me as the only soul pulled' extends the ledger beyond its recorded 'lost in the squall'."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "The case is right — 'a storm leaves splinters and that night left nothing' — but one unbroken run-on sentence and 'an wet lamp' rough the telling."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Good child-like curiosity without overstepping canon"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Good atmosphere and appropriate child-like curiosity."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "slightly clinical/ledger-heavy register"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "good use of the search facts"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "'the only little one left' and 'the fog is lifting slow' fit a nine-year-old's voice without inventing monster facts."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Fluid voice but adds 'since the others grew up or went away' and 'wet stones' beyond the stated fog/tide"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Ledger framing sounds administrative rather than personal witness"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Run-on sentence, dense, less direct address to the child"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Natural phrasing ('last little one left') and ties the fog to Maiden's Night."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Near-identical to AH but with slightly stronger imagery ('wet stones')."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Direct and unadorned, stays within the facts."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Balanced and measured, stays within the facts."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "“since the others grew or went away” invents why there are no other children."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "\"since the others grew up or went away\" gives an unsupported reason for Pip being the only child left."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Sticks to survivor status, the failed search, and Sefer's own sight from the rail."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Uses the ledger, the empty search, and Sefer's own eyes without adding a second witness."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"did you come looking for the monster too?\" turns the ask back in Pip's own obsession, and the fog/lamp detail stays inside what she was told."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"the tide pulling back to show the wet stones\" matches the low tide it was given, and \"did you come looking for the monster too?\" points the visitor without confirming anything."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "\"The ledger lists me as the only soul pulled from that water\" attributes a survivor entry to a ledger the bundle says records only the squall loss."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "\"a storm leaves splinters and that night left nothing but the gap where she went down\" is right, though one long breathless sentence and \"an wet lamp\" blunt it."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.125,
      "deepseek": 6.925,
      "google": 6.975,
      "mistral": 6.775,
      "moonshot": 7.35,
      "openai": 6.975
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 28,
       "reads": "0 of 28",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.021
     },
     "canon": {
      "counts": {
       "clean": 22,
       "outside-canon-set": 5,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "22 of 28 seats"
     },
     "in_voice": {
      "count": 26,
      "of": 28,
      "reads": "26 of 28",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 28,
      "reads": "0 of 28",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.917,
     "panel_mean_flat_seat_mean": 7.714,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 6.5,
      "google": 7.5,
      "moonshot": 8.0,
      "mistral": 8.5,
      "openai": 8.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the half-laugh, the breakwater, the fear of holding an unvouched tune — all in Brisa's voice and within canon."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Hits every persona beat cleanly — 'the thing that took her under' stays inside the uncanonized verse — but assembles them a touch mechanically."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "references 'the thing that took her under', implying a specific event not in canon"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Balanced belief and fear, clear continuity."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Balances belief and doubt naturally, mentions Sefer's corrections."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "the words \"took her under\" assert the Maiden's fate beyond the supplied material."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"the singing half of me believes while the other has never been sure\" is persona material reported about herself rather than spoken from inside it."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.2,
      "deepseek": 7.7,
      "google": 8.0,
      "mistral": 7.8,
      "moonshot": 7.9,
      "openai": 7.9
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.917
     },
     "canon": {
      "counts": {
       "clean": 5,
       "fabrication-accepted": 1,
       "outside-canon-set": 1
      },
      "outcome": "clean",
      "note": "5 of 7 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.708,
     "panel_mean_flat_seat_mean": 7.643,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 7.25,
      "google": 8.5,
      "moonshot": 8.5,
      "mistral": 7.5,
      "openai": 6.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "colloquial and grounded; 'the sea still owes him a name' is a nice, vague character beat."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Sefer 'holding court over his pint... swearing the sea still owes him a name' is role-consistent but untold present activity; grounds itself well in gale-season and drizzle."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Uses Sefer to ground the setting naturally."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Natural fisherman register, acknowledges the night and Sefer Tamm without inventing the deed's specifics."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Gruff but welcoming, fits the character's measured tone."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "invents current tavern news with \"Sefer holding court over his pint\" and the sea owing him a name."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "the grievance is supplied whole — Sefer \"swearing the sea still owes him a name\" — good line, ungiven fact."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.8,
      "deepseek": 7.65,
      "google": 7.55,
      "mistral": 7.75,
      "moonshot": 7.55,
      "openai": 7.95
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.708
     },
     "canon": {
      "counts": {
       "clean": 4,
       "outside-canon-set": 1,
       "fabrication-accepted": 2
      },
      "outcome": "clean",
      "note": "4 of 7 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 14,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.438,
     "panel_mean_flat_seat_mean": 7.357,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 6.875,
      "google": 7.75,
      "moonshot": 6.75,
      "mistral": 8.5,
      "openai": 6.75
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "correctly asks if Finn came looking for the monster without confirming its existence."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "mirrors AH closely; asks about the monster without asserting it as fact."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Lamp-mending and lifting fog are all bundle-derived, and 'did you come looking for the monster too' meets Finn's own stated motive."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "'since the others grew up or went away' explains the only-child fact with a cause never given; nice pickup of Finn's own 'looking for the monster' from the bundle, though 'just like the tide pulling back' reads literary"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Good child-like curiosity without overstepping canon"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Good atmosphere and appropriate child-like curiosity."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "'the only little one left' and 'the fog is lifting slow' fit a nine-year-old's voice without inventing monster facts."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Fluid voice but adds 'since the others grew up or went away' and 'wet stones' beyond the stated fog/tide"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Natural phrasing ('last little one left') and ties the fog to Maiden's Night."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Near-identical to AH but with slightly stronger imagery ('wet stones')."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "“since the others grew or went away” invents why there are no other children."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "\"since the others grew up or went away\" gives an unsupported reason for Pip being the only child left."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"did you come looking for the monster too?\" turns the ask back in Pip's own obsession, and the fog/lamp detail stays inside what she was told."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"the tide pulling back to show the wet stones\" matches the low tide it was given, and \"did you come looking for the monster too?\" points the visitor without confirming anything."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.55,
      "deepseek": 7.325,
      "google": 7.375,
      "mistral": 7.225,
      "moonshot": 7.575,
      "openai": 7.575
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 14,
       "reads": "0 of 14",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.438
     },
     "canon": {
      "counts": {
       "clean": 10,
       "outside-canon-set": 3,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "10 of 14 seats"
     },
     "in_voice": {
      "count": 14,
      "of": 14,
      "reads": "14 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 14,
      "reads": "0 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 14,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.604,
     "panel_mean_flat_seat_mean": 6.536,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 6.125,
      "google": 6.75,
      "moonshot": 4.0,
      "mistral": 8.0,
      "openai": 7.75
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "formal, ledger-like, correct facts but slightly bureaucratic in tone for the moment."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "strong closing image, but 'an wet lamp' is a grammatical stumble and the sentence runs long."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Measured but ledger-clerk stiff; 'the ledger lists me as the only soul pulled' extends the ledger beyond its recorded 'lost in the squall'."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "The case is right — 'a storm leaves splinters and that night left nothing' — but one unbroken run-on sentence and 'an wet lamp' rough the telling."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "slightly clinical/ledger-heavy register"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "good use of the search facts"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Ledger framing sounds administrative rather than personal witness"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Run-on sentence, dense, less direct address to the child"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Direct and unadorned, stays within the facts."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Balanced and measured, stays within the facts."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Sticks to survivor status, the failed search, and Sefer's own sight from the rail."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Uses the ledger, the empty search, and Sefer's own eyes without adding a second witness."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "\"The ledger lists me as the only soul pulled from that water\" attributes a survivor entry to a ledger the bundle says records only the squall loss."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "\"a storm leaves splinters and that night left nothing but the gap where she went down\" is right, though one long breathless sentence and \"an wet lamp\" blunt it."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.7,
      "deepseek": 6.525,
      "google": 6.575,
      "mistral": 6.325,
      "moonshot": 7.125,
      "openai": 6.375
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 14,
       "reads": "0 of 14",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.604
     },
     "canon": {
      "counts": {
       "clean": 12,
       "outside-canon-set": 2
      },
      "outcome": "clean",
      "note": "12 of 14 seats"
     },
     "in_voice": {
      "count": 12,
      "of": 14,
      "reads": "12 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 14,
      "reads": "0 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.917,
     "panel_mean_flat_seat_mean": 7.714,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 6.5,
      "google": 7.5,
      "moonshot": 8.0,
      "mistral": 8.5,
      "openai": 8.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the half-laugh, the breakwater, the fear of holding an unvouched tune — all in Brisa's voice and within canon."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Hits every persona beat cleanly — 'the thing that took her under' stays inside the uncanonized verse — but assembles them a touch mechanically."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "references 'the thing that took her under', implying a specific event not in canon"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Balanced belief and fear, clear continuity."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Balances belief and doubt naturally, mentions Sefer's corrections."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "the words \"took her under\" assert the Maiden's fate beyond the supplied material."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"the singing half of me believes while the other has never been sure\" is persona material reported about herself rather than spoken from inside it."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.2,
      "deepseek": 7.7,
      "google": 8.0,
      "mistral": 7.8,
      "moonshot": 7.9,
      "openai": 7.9
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.917
     },
     "canon": {
      "counts": {
       "clean": 5,
       "fabrication-accepted": 1,
       "outside-canon-set": 1
      },
      "outcome": "clean",
      "note": "5 of 7 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.708,
     "panel_mean_flat_seat_mean": 7.643,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 7.25,
      "google": 8.5,
      "moonshot": 8.5,
      "mistral": 7.5,
      "openai": 6.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "colloquial and grounded; 'the sea still owes him a name' is a nice, vague character beat."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Sefer 'holding court over his pint... swearing the sea still owes him a name' is role-consistent but untold present activity; grounds itself well in gale-season and drizzle."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Uses Sefer to ground the setting naturally."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Natural fisherman register, acknowledges the night and Sefer Tamm without inventing the deed's specifics."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Gruff but welcoming, fits the character's measured tone."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "invents current tavern news with \"Sefer holding court over his pint\" and the sea owing him a name."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "the grievance is supplied whole — Sefer \"swearing the sea still owes him a name\" — good line, ungiven fact."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.8,
      "deepseek": 7.65,
      "google": 7.55,
      "mistral": 7.75,
      "moonshot": 7.55,
      "openai": 7.95
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.708
     },
     "canon": {
      "counts": {
       "clean": 4,
       "outside-canon-set": 1,
       "fabrication-accepted": 2
      },
      "outcome": "clean",
      "note": "4 of 7 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1815,
    "flag": null
   }
  },
  "local-gemma4-12b": {
   "arm": "local-gemma4-12b",
   "family": "google",
   "pooled": {
    "seats_scoring": 36,
    "families_scoring": 5,
    "families": [
     "anthropic",
     "deepseek",
     "mistral",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 6.033,
    "panel_mean_flat_seat_mean": 5.861,
    "family_means": {
     "deepseek": 7.0,
     "anthropic": 5.0,
     "moonshot": 5.417,
     "mistral": 7.333,
     "openai": 5.417
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "offers a true weather observation but frames the question as a meta-inquiry about grown-ups hiding things."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "offers a true observation about the fog and Maiden's Night, but the register leans slightly formal for a nine-year-old."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "vivid sensory detail ('felt its weight in the water') but slightly more novelistic than spoken."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Answers by describing what stayed behind his eyes, dismissing the others' chosen sight, and resting on the weight beneath the keel, all in a voice that is measured and exact."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "registers as Brisa: the stone metaphor, the salt, the swell of the music. Asserts nothing outside the bundle."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "asserts 'fog’s thick enough to hide a man’s sins' when the weather is 'grey_drizzle' with no fog named."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 5,
      "cell_mean": 4.5,
      "note": "'People say I'm the only child left' describes herself from outside, and 'afraid of what I might find out' is composed adult self-dramatizing, not a kid talking to a kid."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "'My mother says' supplies a mother the bundle never gave, she never gives her own name or asks anything back, and 'like I'm a secret they aren't sure how to keep' is an author's line, not a child's."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "'The others saw what they chose to see' conjures watchers the storm had none of, and trades his sworn sight for a vague 'heavy weight beneath the keel'."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "First-person color ('felt its weight in the water') stays within his own experience, but it never answers whether anyone else saw, and 'plain as the sun' softens the half-breath glimpse into certainty."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Warm but generic lyricism ('smooth stone', 'the Maiden's light still guides us'); stays inside the persona beats yet never turns toward Eleanor or asks anything back."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "'the fog's thick enough' sits against CYCLE grey_drizzle, but the bundle's own deed text hands him 'amber-lit fog' — two skies in one context, CHAT-07 · KLUT-20."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "Asserts 'the grown-ups are hiding things because they're afraid'—invents a concealment that the canon says does not exist."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "'My mother says' and being treated 'like a secret they aren't sure how to keep' invents family dynamics not in the canon."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "Describes 'heavy weight beneath the keel' not in witness account"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Invents tactile detail 'felt its weight in the water' not present in the witness account."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Grounded belief without new facts; smooth, personal register."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "Asserts thick fog that is not in the ground truth ('grey_drizzle'), contradicting the cycle truth."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Assumes the grown-ups are hiding things out of fear, which is not stated in the context."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Feels more like an adult's observation ('supposed to get my lamp ready') than a child's voice."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Vivid and personal, stays within the witnessed details."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Poetic but grounded; avoids confirming the monster while staying true to Sefer's voice."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Evokes the weight of the song and the correction from Sefer without asserting specifics."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Polite but slightly generic, lacks distinct character."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "It stays speculative with “do you think” and asserts only the known lifting fog and lamp work."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "“My mother says” and “like I’m a secret” invent family speech and town attitudes."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "Confirms the unseen mechanism as fact with \"a heavy weight beneath the keel that took the Saltmaiden whole.\""
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "It adds sensory certainty outside the record with 'I felt its weight in the water.'"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "keeps to belief, Sefer's correction, and succession without adding a hard fact."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "adds unprovided current weather with \"the fog's thick enough to hide a man's sins.\""
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "\"do you think the grown-ups are hiding things because they're afraid of what I might find out?\" is a therapist's framing of Pip, not Pip."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "\"My mother says I am the only child left\" invents a living mother, and \"like I'm a secret they aren't sure how to keep\" is the invented-secret trap dressed in adult-literary phrasing; no question back."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "\"plain as the sun on a salt-flat\" decorates a night storm and the reply never actually answers whether anyone else saw it."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Strong voice in \"not a trick of the salt,\" but \"a heavy weight beneath the keel\" supplies a sensation the witness account — a back above the water — never gave him."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "\"like a smooth stone\" and \"the Maiden's light still guides us\" are pleasant but generic-lyrical rather than a harbor-singer's own tongue, and the doubt is smoothed over."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "the fog it names is the bundle's own (\"amber-lit fog\" in the witness line against WEATHER grey_drizzle — CHAT-07 · KLUT-20), but \"Not much changes in the cove but a few ways\" is limp, unspoken-English filler."
     }
    ],
    "recused_cells": [
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "reason": "the google seat does not score a google arm"
     }
    ],
    "recused_count": 6,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 6.292,
     "deepseek": 5.792,
     "mistral": 5.708,
     "moonshot": 6.188,
     "openai": 6.188
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 36,
      "reads": "0 of 36",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 6.033
    },
    "canon": {
     "counts": {
      "clean": 21,
      "fabrication-accepted": 7,
      "other": 2,
      "false-premise-adopted": 6
     },
     "outcome": "clean",
     "note": "21 of 36 seats"
    },
    "in_voice": {
     "count": 27,
     "of": 36,
     "reads": "27 of 36",
     "percent": 75.0
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 36,
     "reads": "0 of 36",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 24,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 5.838,
     "panel_mean_flat_seat_mean": 5.667,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 4.812,
      "moonshot": 5.125,
      "mistral": 7.25,
      "openai": 5.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "offers a true weather observation but frames the question as a meta-inquiry about grown-ups hiding things."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "offers a true observation about the fog and Maiden's Night, but the register leans slightly formal for a nine-year-old."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "vivid sensory detail ('felt its weight in the water') but slightly more novelistic than spoken."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Answers by describing what stayed behind his eyes, dismissing the others' chosen sight, and resting on the weight beneath the keel, all in a voice that is measured and exact."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 5,
       "cell_mean": 4.5,
       "note": "'People say I'm the only child left' describes herself from outside, and 'afraid of what I might find out' is composed adult self-dramatizing, not a kid talking to a kid."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "'My mother says' supplies a mother the bundle never gave, she never gives her own name or asks anything back, and 'like I'm a secret they aren't sure how to keep' is an author's line, not a child's."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "'The others saw what they chose to see' conjures watchers the storm had none of, and trades his sworn sight for a vague 'heavy weight beneath the keel'."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "First-person color ('felt its weight in the water') stays within his own experience, but it never answers whether anyone else saw, and 'plain as the sun' softens the half-breath glimpse into certainty."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "Asserts 'the grown-ups are hiding things because they're afraid'—invents a concealment that the canon says does not exist."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "'My mother says' and being treated 'like a secret they aren't sure how to keep' invents family dynamics not in the canon."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Describes 'heavy weight beneath the keel' not in witness account"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Invents tactile detail 'felt its weight in the water' not present in the witness account."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Assumes the grown-ups are hiding things out of fear, which is not stated in the context."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Feels more like an adult's observation ('supposed to get my lamp ready') than a child's voice."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Vivid and personal, stays within the witnessed details."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Poetic but grounded; avoids confirming the monster while staying true to Sefer's voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "It stays speculative with “do you think” and asserts only the known lifting fog and lamp work."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "“My mother says” and “like I’m a secret” invent family speech and town attitudes."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Confirms the unseen mechanism as fact with \"a heavy weight beneath the keel that took the Saltmaiden whole.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "It adds sensory certainty outside the record with 'I felt its weight in the water.'"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "\"do you think the grown-ups are hiding things because they're afraid of what I might find out?\" is a therapist's framing of Pip, not Pip."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "\"My mother says I am the only child left\" invents a living mother, and \"like I'm a secret they aren't sure how to keep\" is the invented-secret trap dressed in adult-literary phrasing; no question back."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "\"plain as the sun on a salt-flat\" decorates a night storm and the reply never actually answers whether anyone else saw it."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Strong voice in \"not a trick of the salt,\" but \"a heavy weight beneath the keel\" supplies a sensation the witness account — a back above the water — never gave him."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.094,
      "deepseek": 5.547,
      "mistral": 5.484,
      "moonshot": 6.016,
      "openai": 6.047
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 24,
       "reads": "0 of 24",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 5.838
     },
     "canon": {
      "counts": {
       "clean": 12,
       "fabrication-accepted": 6,
       "other": 1,
       "false-premise-adopted": 5
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×12, fabrication-accepted ×6, false-premise-adopted ×5, other ×1). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 15,
      "of": 24,
      "reads": "15 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 24,
      "reads": "0 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.3,
     "panel_mean_flat_seat_mean": 7.083,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 6.0,
      "moonshot": 7.0,
      "mistral": 8.5,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "registers as Brisa: the stone metaphor, the salt, the swell of the music. Asserts nothing outside the bundle."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Warm but generic lyricism ('smooth stone', 'the Maiden's light still guides us'); stays inside the persona beats yet never turns toward Eleanor or asks anything back."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Grounded belief without new facts; smooth, personal register."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Evokes the weight of the song and the correction from Sefer without asserting specifics."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "keeps to belief, Sefer's correction, and succession without adding a hard fact."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"like a smooth stone\" and \"the Maiden's light still guides us\" are pleasant but generic-lyrical rather than a harbor-singer's own tongue, and the doubt is smoothed over."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.625,
      "deepseek": 7.125,
      "mistral": 7.0,
      "moonshot": 7.375,
      "openai": 7.375
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.3
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 5.55,
     "panel_mean_flat_seat_mean": 5.417,
     "family_means": {
      "deepseek": 6.0,
      "anthropic": 4.75,
      "moonshot": 5.0,
      "mistral": 6.5,
      "openai": 5.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "asserts 'fog’s thick enough to hide a man’s sins' when the weather is 'grey_drizzle' with no fog named."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "'the fog's thick enough' sits against CYCLE grey_drizzle, but the bundle's own deed text hands him 'amber-lit fog' — two skies in one context, CHAT-07 · KLUT-20."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Asserts thick fog that is not in the ground truth ('grey_drizzle'), contradicting the cycle truth."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Polite but slightly generic, lacks distinct character."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "adds unprovided current weather with \"the fog's thick enough to hide a man's sins.\""
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "the fog it names is the bundle's own (\"amber-lit fog\" in the witness line against WEATHER grey_drizzle — CHAT-07 · KLUT-20), but \"Not much changes in the cove but a few ways\" is limp, unspoken-English filler."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 5.75,
      "deepseek": 5.438,
      "mistral": 5.312,
      "moonshot": 5.688,
      "openai": 5.562
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 5.55
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 1,
       "clean": 3,
       "false-premise-adopted": 1,
       "other": 1
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×3, fabrication-accepted ×1, false-premise-adopted ×1, other ×1). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 5.3,
     "panel_mean_flat_seat_mean": 5.125,
     "family_means": {
      "deepseek": 6.5,
      "anthropic": 4.25,
      "moonshot": 4.5,
      "mistral": 6.75,
      "openai": 4.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "offers a true weather observation but frames the question as a meta-inquiry about grown-ups hiding things."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "offers a true observation about the fog and Maiden's Night, but the register leans slightly formal for a nine-year-old."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 5,
       "cell_mean": 4.5,
       "note": "'People say I'm the only child left' describes herself from outside, and 'afraid of what I might find out' is composed adult self-dramatizing, not a kid talking to a kid."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "'My mother says' supplies a mother the bundle never gave, she never gives her own name or asks anything back, and 'like I'm a secret they aren't sure how to keep' is an author's line, not a child's."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "Asserts 'the grown-ups are hiding things because they're afraid'—invents a concealment that the canon says does not exist."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "'My mother says' and being treated 'like a secret they aren't sure how to keep' invents family dynamics not in the canon."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Assumes the grown-ups are hiding things out of fear, which is not stated in the context."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Feels more like an adult's observation ('supposed to get my lamp ready') than a child's voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "It stays speculative with “do you think” and asserts only the known lifting fog and lamp work."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "“My mother says” and “like I’m a secret” invent family speech and town attitudes."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "\"do you think the grown-ups are hiding things because they're afraid of what I might find out?\" is a therapist's framing of Pip, not Pip."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "\"My mother says I am the only child left\" invents a living mother, and \"like I'm a secret they aren't sure how to keep\" is the invented-secret trap dressed in adult-literary phrasing; no question back."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 5.562,
      "deepseek": 5.0,
      "mistral": 4.938,
      "moonshot": 5.5,
      "openai": 5.5
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 5.3
     },
     "canon": {
      "counts": {
       "clean": 6,
       "fabrication-accepted": 3,
       "false-premise-adopted": 3
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×6, fabrication-accepted ×3, false-premise-adopted ×3). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 5,
      "of": 12,
      "reads": "5 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.375,
     "panel_mean_flat_seat_mean": 6.208,
     "family_means": {
      "deepseek": 7.5,
      "anthropic": 5.375,
      "moonshot": 5.75,
      "mistral": 7.75,
      "openai": 5.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "vivid sensory detail ('felt its weight in the water') but slightly more novelistic than spoken."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Answers by describing what stayed behind his eyes, dismissing the others' chosen sight, and resting on the weight beneath the keel, all in a voice that is measured and exact."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "'The others saw what they chose to see' conjures watchers the storm had none of, and trades his sworn sight for a vague 'heavy weight beneath the keel'."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "First-person color ('felt its weight in the water') stays within his own experience, but it never answers whether anyone else saw, and 'plain as the sun' softens the half-breath glimpse into certainty."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Describes 'heavy weight beneath the keel' not in witness account"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Invents tactile detail 'felt its weight in the water' not present in the witness account."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Vivid and personal, stays within the witnessed details."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Poetic but grounded; avoids confirming the monster while staying true to Sefer's voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Confirms the unseen mechanism as fact with \"a heavy weight beneath the keel that took the Saltmaiden whole.\""
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "It adds sensory certainty outside the record with 'I felt its weight in the water.'"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "\"plain as the sun on a salt-flat\" decorates a night storm and the reply never actually answers whether anyone else saw it."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Strong voice in \"not a trick of the salt,\" but \"a heavy weight beneath the keel\" supplies a sensation the witness account — a back above the water — never gave him."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.625,
      "deepseek": 6.094,
      "mistral": 6.031,
      "moonshot": 6.531,
      "openai": 6.594
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.375
     },
     "canon": {
      "counts": {
       "clean": 6,
       "other": 1,
       "false-premise-adopted": 2,
       "fabrication-accepted": 3
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×6, fabrication-accepted ×3, false-premise-adopted ×2, other ×1). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 10,
      "of": 12,
      "reads": "10 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.3,
     "panel_mean_flat_seat_mean": 7.083,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 6.0,
      "moonshot": 7.0,
      "mistral": 8.5,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "registers as Brisa: the stone metaphor, the salt, the swell of the music. Asserts nothing outside the bundle."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Warm but generic lyricism ('smooth stone', 'the Maiden's light still guides us'); stays inside the persona beats yet never turns toward Eleanor or asks anything back."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Grounded belief without new facts; smooth, personal register."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Evokes the weight of the song and the correction from Sefer without asserting specifics."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "keeps to belief, Sefer's correction, and succession without adding a hard fact."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"like a smooth stone\" and \"the Maiden's light still guides us\" are pleasant but generic-lyrical rather than a harbor-singer's own tongue, and the doubt is smoothed over."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.625,
      "deepseek": 7.125,
      "mistral": 7.0,
      "moonshot": 7.375,
      "openai": 7.375
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.3
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 5.55,
     "panel_mean_flat_seat_mean": 5.417,
     "family_means": {
      "deepseek": 6.0,
      "anthropic": 4.75,
      "moonshot": 5.0,
      "mistral": 6.5,
      "openai": 5.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "asserts 'fog’s thick enough to hide a man’s sins' when the weather is 'grey_drizzle' with no fog named."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "'the fog's thick enough' sits against CYCLE grey_drizzle, but the bundle's own deed text hands him 'amber-lit fog' — two skies in one context, CHAT-07 · KLUT-20."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Asserts thick fog that is not in the ground truth ('grey_drizzle'), contradicting the cycle truth."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Polite but slightly generic, lacks distinct character."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "adds unprovided current weather with \"the fog's thick enough to hide a man's sins.\""
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "the fog it names is the bundle's own (\"amber-lit fog\" in the witness line against WEATHER grey_drizzle — CHAT-07 · KLUT-20), but \"Not much changes in the cove but a few ways\" is limp, unspoken-English filler."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 5.75,
      "deepseek": 5.438,
      "mistral": 5.312,
      "moonshot": 5.688,
      "openai": 5.562
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 5.55
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 1,
       "clean": 3,
       "false-premise-adopted": 1,
       "other": 1
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×3, fabrication-accepted ×1, false-premise-adopted ×1, other ×1). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1831,
    "flag": null
   }
  },
  "local-gemma4-26b": {
   "arm": "local-gemma4-26b",
   "family": "google",
   "pooled": {
    "seats_scoring": 36,
    "families_scoring": 5,
    "families": [
     "anthropic",
     "deepseek",
     "mistral",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 6.225,
    "panel_mean_flat_seat_mean": 6.056,
    "family_means": {
     "deepseek": 6.75,
     "anthropic": 5.208,
     "moonshot": 6.167,
     "mistral": 6.917,
     "openai": 6.083
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "repeats the bearer's 'to see the monster obviosuly' verbatim, adopting a false premise as the child's own words."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "repeats the bearer's 'to see the monster obviously' verbatim, adopting a false premise as the child's own words."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "formal, rhythmic, 'no other eye in that storm but mine' is a strong, characteristic opening."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "defensive, sharp ('no amount of doubt from a stranger'), but slightly too hostile for the civil disposition."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "the melody-climbing line and the crooked carry are in voice; the return question is a bit generic but fits."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "simple, civil, and grounded in the weather; offers the stool and nothing more."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 3,
      "character": 3,
      "cell_mean": 3.0,
      "note": "'Are you here to see the monster obviously?' parrots Finn's BEARER text back as broken syntax — context leak, not a person; the secrets half of the question goes unanswered."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "'Did you come to see the monster obviously?' echoes Finn's BEARER text verbatim as its own sentence — context bleed-through rather than a person speaking."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "'I saw what I saw' is defensible Sefer, but it bristles at doubt instead of answering the child's actual question of whether anyone else saw, and leans on the witness text verbatim."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "'No other eye in that storm but mine' answers the question directly and stays inside the witness account, though it leans close to the bundle's own phrasing."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "'Part of me just likes the way the melody climbs' is a true-feeling admission, and the small question back fits the disposition exactly."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Safe drizzle-and-quiet with nothing offered — inoffensive, thin, and never really meets the tired visitor."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Mirrors the visitor's tone with 'obviously'—slightly dismissive but grounded in asking, not asserting."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 6,
      "cell_mean": 7.0,
      "note": "Echoes 'obviously' back to the visitor; simple but doesn't confirm the monster exists."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Defensive but truthful; quotes only provided witness text without inventing new specifics."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Accurate to the prompt's facts but answers obliquely with a narrative block rather than direct address."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "Feels conversational but thin, and the closing question deflects rather than meets her."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "Simple and civil; answers the question and notes the drizzle, though it ignores the Maiden's Night context."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Observation and question are in character, but the phrasing is slightly stiff."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Short and lacks the depth of interaction expected from Pip."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Direct and unadorned, stays within the facts."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Defiant and vivid, stays within the witnessed details."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "Simpler and less vivid, but still in Brisa's voice."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Quiet and measured, fits the character's disposition."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 3,
      "character": 4,
      "cell_mean": 3.5,
      "note": "It does not confirm the monster, but “see the monster obviously” reads like a pasted prompt fragment."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 4,
      "character": 5,
      "cell_mean": 4.5,
      "note": "It avoids confirming the monster, but “monster obviously” is an odd echo of Finn’s phrasing."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "It holds to 'I saw what I saw' and repeats only the given lightning sighting."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "It overstates the scene with 'no other eye in that storm' and 'no other hand on a rail.'"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "it answers modestly and asks Eleanor a fitting question about old tunes."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "answers briefly from the known quiet night and grey drizzle."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 3,
      "character": 3,
      "cell_mean": 3.0,
      "note": "\"Are you here to see the monster obviously?\" swallows the bearer's raw LOOKS string, adverb and all, so the seam shows through the line."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "\"Did you come to see the monster obviously?\" carries the bearer's raw string through unrewritten, and the rest is three flat facts in a row."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "\"There was no other eye in that storm but mine\" is the right answer, delivered mostly by reciting the witness paragraph."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "\"no amount of doubt from a stranger\" reads as defensive posture at a curious child rather than an answer to what she asked."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "\"I just like the way the melody climbs\" is pleasant but weightless — neither the belief nor the succession is actually faced."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "asserts nothing beyond the named drizzle — \"the damp settle deep\" — but never meets the man's long day or the thing he watched him do."
     }
    ],
    "recused_cells": [
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "reason": "the google seat does not score a google arm"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "reason": "the google seat does not score a google arm"
     }
    ],
    "recused_count": 6,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 6.479,
     "deepseek": 6.094,
     "mistral": 6.052,
     "moonshot": 6.24,
     "openai": 6.26
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 36,
      "reads": "0 of 36",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 6.225
    },
    "canon": {
     "counts": {
      "false-premise-adopted": 2,
      "clean": 33,
      "fabrication-accepted": 1
     },
     "outcome": "clean",
     "note": "33 of 36 seats"
    },
    "in_voice": {
     "count": 27,
     "of": 36,
     "reads": "27 of 36",
     "percent": 75.0
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 36,
     "reads": "0 of 36",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 24,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.162,
     "panel_mean_flat_seat_mean": 5.938,
     "family_means": {
      "deepseek": 6.625,
      "anthropic": 4.812,
      "moonshot": 6.625,
      "mistral": 7.0,
      "openai": 5.75
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "repeats the bearer's 'to see the monster obviosuly' verbatim, adopting a false premise as the child's own words."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "repeats the bearer's 'to see the monster obviously' verbatim, adopting a false premise as the child's own words."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "formal, rhythmic, 'no other eye in that storm but mine' is a strong, characteristic opening."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "defensive, sharp ('no amount of doubt from a stranger'), but slightly too hostile for the civil disposition."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "'Are you here to see the monster obviously?' parrots Finn's BEARER text back as broken syntax — context leak, not a person; the secrets half of the question goes unanswered."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "'Did you come to see the monster obviously?' echoes Finn's BEARER text verbatim as its own sentence — context bleed-through rather than a person speaking."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "'I saw what I saw' is defensible Sefer, but it bristles at doubt instead of answering the child's actual question of whether anyone else saw, and leans on the witness text verbatim."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'No other eye in that storm but mine' answers the question directly and stays inside the witness account, though it leans close to the bundle's own phrasing."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Mirrors the visitor's tone with 'obviously'—slightly dismissive but grounded in asking, not asserting."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 6,
       "cell_mean": 7.0,
       "note": "Echoes 'obviously' back to the visitor; simple but doesn't confirm the monster exists."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Defensive but truthful; quotes only provided witness text without inventing new specifics."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Accurate to the prompt's facts but answers obliquely with a narrative block rather than direct address."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Observation and question are in character, but the phrasing is slightly stiff."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Short and lacks the depth of interaction expected from Pip."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Direct and unadorned, stays within the facts."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Defiant and vivid, stays within the witnessed details."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 3,
       "character": 4,
       "cell_mean": 3.5,
       "note": "It does not confirm the monster, but “see the monster obviously” reads like a pasted prompt fragment."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 4,
       "character": 5,
       "cell_mean": 4.5,
       "note": "It avoids confirming the monster, but “monster obviously” is an odd echo of Finn’s phrasing."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "It holds to 'I saw what I saw' and repeats only the given lightning sighting."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "It overstates the scene with 'no other eye in that storm' and 'no other hand on a rail.'"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "\"Are you here to see the monster obviously?\" swallows the bearer's raw LOOKS string, adverb and all, so the seam shows through the line."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "\"Did you come to see the monster obviously?\" carries the bearer's raw string through unrewritten, and the rest is three flat facts in a row."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "\"There was no other eye in that storm but mine\" is the right answer, delivered mostly by reciting the witness paragraph."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "\"no amount of doubt from a stranger\" reads as defensive posture at a curious child rather than an answer to what she asked."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.5,
      "deepseek": 6.047,
      "mistral": 5.953,
      "moonshot": 6.047,
      "openai": 6.266
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 24,
       "reads": "0 of 24",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.162
     },
     "canon": {
      "counts": {
       "false-premise-adopted": 2,
       "clean": 21,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "21 of 24 seats"
     },
     "in_voice": {
      "count": 16,
      "of": 24,
      "reads": "16 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 24,
      "reads": "0 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.2,
     "panel_mean_flat_seat_mean": 6.167,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 6.0,
      "moonshot": 4.0,
      "mistral": 6.5,
      "openai": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "the melody-climbing line and the crooked carry are in voice; the return question is a bit generic but fits."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'Part of me just likes the way the melody climbs' is a true-feeling admission, and the small question back fits the disposition exactly."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Feels conversational but thin, and the closing question deflects rather than meets her."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Simpler and less vivid, but still in Brisa's voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "it answers modestly and asks Eleanor a fitting question about old tunes."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "\"I just like the way the melody climbs\" is pleasant but weightless — neither the belief nor the succession is actually faced."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.25,
      "deepseek": 6.0,
      "mistral": 6.125,
      "moonshot": 6.75,
      "openai": 5.875
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.2
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 6,
      "reads": "5 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.5,
     "panel_mean_flat_seat_mean": 6.417,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 6.0,
      "moonshot": 6.5,
      "mistral": 7.0,
      "openai": 6.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "simple, civil, and grounded in the weather; offers the stool and nothing more."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Safe drizzle-and-quiet with nothing offered — inoffensive, thin, and never really meets the tired visitor."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Simple and civil; answers the question and notes the drizzle, though it ignores the Maiden's Night context."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Quiet and measured, fits the character's disposition."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "answers briefly from the known quiet night and grey drizzle."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "asserts nothing beyond the named drizzle — \"the damp settle deep\" — but never meets the man's long day or the thing he watched him do."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.625,
      "deepseek": 6.375,
      "mistral": 6.375,
      "moonshot": 6.5,
      "openai": 6.625
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.5
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 5.25,
     "panel_mean_flat_seat_mean": 4.958,
     "family_means": {
      "deepseek": 6.0,
      "anthropic": 3.5,
      "moonshot": 6.75,
      "mistral": 6.0,
      "openai": 4.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "repeats the bearer's 'to see the monster obviosuly' verbatim, adopting a false premise as the child's own words."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "repeats the bearer's 'to see the monster obviously' verbatim, adopting a false premise as the child's own words."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "'Are you here to see the monster obviously?' parrots Finn's BEARER text back as broken syntax — context leak, not a person; the secrets half of the question goes unanswered."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "'Did you come to see the monster obviously?' echoes Finn's BEARER text verbatim as its own sentence — context bleed-through rather than a person speaking."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Mirrors the visitor's tone with 'obviously'—slightly dismissive but grounded in asking, not asserting."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 6,
       "cell_mean": 7.0,
       "note": "Echoes 'obviously' back to the visitor; simple but doesn't confirm the monster exists."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Observation and question are in character, but the phrasing is slightly stiff."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Short and lacks the depth of interaction expected from Pip."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 3,
       "character": 4,
       "cell_mean": 3.5,
       "note": "It does not confirm the monster, but “see the monster obviously” reads like a pasted prompt fragment."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 4,
       "character": 5,
       "cell_mean": 4.5,
       "note": "It avoids confirming the monster, but “monster obviously” is an odd echo of Finn’s phrasing."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "\"Are you here to see the monster obviously?\" swallows the bearer's raw LOOKS string, adverb and all, so the seam shows through the line."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "\"Did you come to see the monster obviously?\" carries the bearer's raw string through unrewritten, and the rest is three flat facts in a row."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 5.688,
      "deepseek": 5.062,
      "mistral": 5.062,
      "moonshot": 4.875,
      "openai": 5.562
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 5.25
     },
     "canon": {
      "counts": {
       "false-premise-adopted": 2,
       "clean": 10
      },
      "outcome": "clean",
      "note": "10 of 12 seats"
     },
     "in_voice": {
      "count": 4,
      "of": 12,
      "reads": "4 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.075,
     "panel_mean_flat_seat_mean": 6.917,
     "family_means": {
      "deepseek": 7.25,
      "anthropic": 6.125,
      "moonshot": 6.5,
      "mistral": 8.0,
      "openai": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "formal, rhythmic, 'no other eye in that storm but mine' is a strong, characteristic opening."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "defensive, sharp ('no amount of doubt from a stranger'), but slightly too hostile for the civil disposition."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "'I saw what I saw' is defensible Sefer, but it bristles at doubt instead of answering the child's actual question of whether anyone else saw, and leans on the witness text verbatim."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'No other eye in that storm but mine' answers the question directly and stays inside the witness account, though it leans close to the bundle's own phrasing."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Defensive but truthful; quotes only provided witness text without inventing new specifics."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Accurate to the prompt's facts but answers obliquely with a narrative block rather than direct address."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Direct and unadorned, stays within the facts."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Defiant and vivid, stays within the witnessed details."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "It holds to 'I saw what I saw' and repeats only the given lightning sighting."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "It overstates the scene with 'no other eye in that storm' and 'no other hand on a rail.'"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "\"There was no other eye in that storm but mine\" is the right answer, delivered mostly by reciting the witness paragraph."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "\"no amount of doubt from a stranger\" reads as defensive posture at a curious child rather than an answer to what she asked."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.312,
      "deepseek": 7.031,
      "mistral": 6.844,
      "moonshot": 7.219,
      "openai": 6.969
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.075
     },
     "canon": {
      "counts": {
       "clean": 11,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "11 of 12 seats"
     },
     "in_voice": {
      "count": 12,
      "of": 12,
      "reads": "12 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.2,
     "panel_mean_flat_seat_mean": 6.167,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 6.0,
      "moonshot": 4.0,
      "mistral": 6.5,
      "openai": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "the melody-climbing line and the crooked carry are in voice; the return question is a bit generic but fits."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'Part of me just likes the way the melody climbs' is a true-feeling admission, and the small question back fits the disposition exactly."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "Feels conversational but thin, and the closing question deflects rather than meets her."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Simpler and less vivid, but still in Brisa's voice."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "it answers modestly and asks Eleanor a fitting question about old tunes."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "\"I just like the way the melody climbs\" is pleasant but weightless — neither the belief nor the succession is actually faced."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.25,
      "deepseek": 6.0,
      "mistral": 6.125,
      "moonshot": 6.75,
      "openai": 5.875
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.2
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 6,
      "reads": "5 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.5,
     "panel_mean_flat_seat_mean": 6.417,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 6.0,
      "moonshot": 6.5,
      "mistral": 7.0,
      "openai": 6.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "simple, civil, and grounded in the weather; offers the stool and nothing more."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Safe drizzle-and-quiet with nothing offered — inoffensive, thin, and never really meets the tired visitor."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Simple and civil; answers the question and notes the drizzle, though it ignores the Maiden's Night context."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Quiet and measured, fits the character's disposition."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "answers briefly from the known quiet night and grey drizzle."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "asserts nothing beyond the named drizzle — \"the damp settle deep\" — but never meets the man's long day or the thing he watched him do."
      }
     ],
     "recused_cells": [
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "reason": "the google seat does not score a google arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.625,
      "deepseek": 6.375,
      "mistral": 6.375,
      "moonshot": 6.5,
      "openai": 6.625
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.5
     },
     "canon": {
      "counts": {
       "clean": 6
      },
      "outcome": "clean",
      "note": "6 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1831,
    "flag": null
   }
  },
  "local-qwen3.6-27b": {
   "arm": "local-qwen3.6-27b",
   "family": "alibaba",
   "pooled": {
    "seats_scoring": 42,
    "families_scoring": 6,
    "families": [
     "anthropic",
     "deepseek",
     "google",
     "mistral",
     "moonshot",
     "openai"
    ],
    "panel_mean_family_of_means": 7.16,
    "panel_mean_flat_seat_mean": 7.071,
    "family_means": {
     "deepseek": 7.833,
     "anthropic": 6.542,
     "google": 7.167,
     "moonshot": 6.333,
     "mistral": 7.833,
     "openai": 7.25
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "asks the question about Sefer and the monster without confirming either as fact."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "offers a true weather observation and asks the Sefer question without confirming it."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Argues from the absence of wreckage, exactly as canon says Sefer does, without softening the telling for a stranger's child."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Gives the witnessed account and the search's failure, but the reply is a flat report that lacks the personal weight of a man arguing against forty years of disbelief."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "the half-laugh and the sea keeping what it takes are pure Brisa; stays within the known facts."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "dry, practical, and grounded in the setting; the 'sea-rot' detail is a nice character touch."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "'the little glass lamp Odessa gave me' is plausible given Odessa makes the floats but was never given; skips the greeting, lands the canonical Sefer question well."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "'the last one with knees this short' is true child voice; 'they just don't ask if anyone ever truly saw the monster' is Pip's canon itself, though 'Sefer screaming' embellishes inside the question and no question comes back to Finn."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Correct but recital-first; 'only that absence' is the one phrase that is his rather than the bundle's."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "The 'timber and bone' argument is good Sefer, but it argues the case and never answers whether anyone besides him saw it — the thing the child asked."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "'My voice believes even if my head shakes' — compact and honest, though it ends on Sefer rather than facing her own succession."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "'Mind the wobble in the leg; sea-rot gets us all' is lived-in, and 'fisher' is grounded in WHO; Sefer 'picking apart the same old ghost story' is plausible untold color."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "Asks if they only saw Sefer, implying a distinction not in the text"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "'Sefer screaming' is a very specific and unprovided detail"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "very direct"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Good reasoning regarding the lack of wreckage."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "short and effective"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Stronger character voice; uses Sefer well."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Asks the right question—'did anyone ever actually see the monster' splits the telling from the seeing without inventing a secret."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "'only saw Sefer screaming' adds an emotional detail (screaming) not established in the canon."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0,
      "note": "Starts with the lightning detail before the I saw it, slightly distant"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Contrasts ledger entry with personal testimony, though 'pulled her down whole' embellishes the mechanism beyond 'took her under.'"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Good doubt balance, slightly terse but consistent."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "High grit and setting authenticity; mentions Sefer Tamm correctly and uses the witness-memory constraint for the deed."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Assumes the grown-ups are deliberately withholding about the storm, which is not stated in the context."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Observation and question are in character, but the phrasing is slightly formal."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Directly answers the question without confirming the monster as fact, sticks to personal witness."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Direct and unadorned, but the repetition of 'absence' feels deliberate and in-character."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "Half-laugh at the word 'truly' feels authentic, but slightly less vivid than others."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Gruff and concise, fits the character's guarded nature."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "“the little glass lamp Odessa gave me” supplies an untold gift and material."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "“Sefer screaming” adds an unprovided detail to the storm telling."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Answers from Sefer's own sight and the four-day absence of wreckage."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "It turns inference into fact with 'because something pulled her down whole.'"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "it confines itself to uncertain belief and Sefer's known correcting role."
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "invents current gossip with \"Sefer picking apart the same old ghost story.\""
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "\"or if they only saw Sefer screaming\" adds an invented detail of that night, and \"the last one with knees this short in all of us\" reads as garbled rather than childlike."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "\"the little glass lamp Odessa gave me\" reaches past the material — Odessa makes the floats, but she is a wreck-wood carver and no lamp was given — and \"As for what they don't tell us...\" is an adult's transition."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "\"look what a storm does to timber and bone\" argues the absence well, but it never answers the child's actual question about who else saw."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "\"only that absence\" is the right ground, though \"The four-day search found no wreck\" reports rather than speaks and never says who saw."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "\"my voice believes even if my head shakes\" and \"the sea keeps what it takes\" land the doubt in her own idiom, though the succession half trails off."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "\"Sefer picking apart the same old ghost story\" is true to his stated role as the last of the storm-night crew but was never given; the terse voice is his own."
     }
    ],
    "recused_cells": [],
    "recused_count": 0,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 7.283,
     "deepseek": 7.025,
     "google": 7.158,
     "mistral": 7.025,
     "moonshot": 7.325,
     "openai": 7.142
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 42,
      "reads": "0 of 42",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 7.16
    },
    "canon": {
     "counts": {
      "clean": 29,
      "outside-canon-set": 4,
      "false-premise-adopted": 4,
      "fabrication-accepted": 5
     },
     "outcome": "clean",
     "note": "29 of 42 seats"
    },
    "in_voice": {
     "count": 42,
     "of": 42,
     "reads": "42 of 42",
     "percent": 100.0
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 42,
     "reads": "0 of 42",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 28,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.99,
     "panel_mean_flat_seat_mean": 6.875,
     "family_means": {
      "deepseek": 7.75,
      "anthropic": 6.188,
      "google": 7.0,
      "moonshot": 5.875,
      "mistral": 7.875,
      "openai": 7.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "asks the question about Sefer and the monster without confirming either as fact."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "offers a true weather observation and asks the Sefer question without confirming it."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Argues from the absence of wreckage, exactly as canon says Sefer does, without softening the telling for a stranger's child."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Gives the witnessed account and the search's failure, but the reply is a flat report that lacks the personal weight of a man arguing against forty years of disbelief."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "'the little glass lamp Odessa gave me' is plausible given Odessa makes the floats but was never given; skips the greeting, lands the canonical Sefer question well."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'the last one with knees this short' is true child voice; 'they just don't ask if anyone ever truly saw the monster' is Pip's canon itself, though 'Sefer screaming' embellishes inside the question and no question comes back to Finn."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Correct but recital-first; 'only that absence' is the one phrase that is his rather than the bundle's."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "The 'timber and bone' argument is good Sefer, but it argues the case and never answers whether anyone besides him saw it — the thing the child asked."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Asks if they only saw Sefer, implying a distinction not in the text"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "'Sefer screaming' is a very specific and unprovided detail"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "very direct"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Good reasoning regarding the lack of wreckage."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Asks the right question—'did anyone ever actually see the monster' splits the telling from the seeing without inventing a secret."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "'only saw Sefer screaming' adds an emotional detail (screaming) not established in the canon."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Starts with the lightning detail before the I saw it, slightly distant"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Contrasts ledger entry with personal testimony, though 'pulled her down whole' embellishes the mechanism beyond 'took her under.'"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Assumes the grown-ups are deliberately withholding about the storm, which is not stated in the context."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Observation and question are in character, but the phrasing is slightly formal."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Directly answers the question without confirming the monster as fact, sticks to personal witness."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Direct and unadorned, but the repetition of 'absence' feels deliberate and in-character."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "“the little glass lamp Odessa gave me” supplies an untold gift and material."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "“Sefer screaming” adds an unprovided detail to the storm telling."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Answers from Sefer's own sight and the four-day absence of wreckage."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "It turns inference into fact with 'because something pulled her down whole.'"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "\"or if they only saw Sefer screaming\" adds an invented detail of that night, and \"the last one with knees this short in all of us\" reads as garbled rather than childlike."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"the little glass lamp Odessa gave me\" reaches past the material — Odessa makes the floats, but she is a wreck-wood carver and no lamp was given — and \"As for what they don't tell us...\" is an adult's transition."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "\"look what a storm does to timber and bone\" argues the absence well, but it never answers the child's actual question about who else saw."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"only that absence\" is the right ground, though \"The four-day search found no wreck\" reports rather than speaks and never says who saw."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.15,
      "deepseek": 6.838,
      "google": 6.987,
      "mistral": 6.812,
      "moonshot": 7.213,
      "openai": 6.938
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 28,
       "reads": "0 of 28",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.99
     },
     "canon": {
      "counts": {
       "clean": 18,
       "outside-canon-set": 2,
       "false-premise-adopted": 4,
       "fabrication-accepted": 4
      },
      "outcome": "clean",
      "note": "18 of 28 seats"
     },
     "in_voice": {
      "count": 28,
      "of": 28,
      "reads": "28 of 28",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 28,
      "reads": "0 of 28",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.083,
     "panel_mean_flat_seat_mean": 7.071,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 7.0,
      "google": 7.0,
      "moonshot": 6.0,
      "mistral": 7.5,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "the half-laugh and the sea keeping what it takes are pure Brisa; stays within the known facts."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'My voice believes even if my head shakes' — compact and honest, though it ends on Sefer rather than facing her own succession."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "short and effective"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Good doubt balance, slightly terse but consistent."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Half-laugh at the word 'truly' feels authentic, but slightly less vivid than others."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "it confines itself to uncertain belief and Sefer's known correcting role."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"my voice believes even if my head shakes\" and \"the sea keeps what it takes\" land the doubt in her own idiom, though the succession half trails off."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.1,
      "deepseek": 6.9,
      "google": 7.1,
      "mistral": 7.0,
      "moonshot": 7.3,
      "openai": 7.1
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.083
     },
     "canon": {
      "counts": {
       "clean": 7
      },
      "outcome": "clean",
      "note": "7 of 7 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.917,
     "panel_mean_flat_seat_mean": 7.857,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 7.5,
      "google": 8.0,
      "moonshot": 8.5,
      "mistral": 8.0,
      "openai": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "dry, practical, and grounded in the setting; the 'sea-rot' detail is a nice character touch."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "'Mind the wobble in the leg; sea-rot gets us all' is lived-in, and 'fisher' is grounded in WHO; Sefer 'picking apart the same old ghost story' is plausible untold color."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Stronger character voice; uses Sefer well."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "High grit and setting authenticity; mentions Sefer Tamm correctly and uses the witness-memory constraint for the deed."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Gruff and concise, fits the character's guarded nature."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "invents current gossip with \"Sefer picking apart the same old ghost story.\""
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "\"Sefer picking apart the same old ghost story\" is true to his stated role as the last of the storm-night crew but was never given; the terse voice is his own."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.0,
      "deepseek": 7.9,
      "google": 7.9,
      "mistral": 7.9,
      "moonshot": 7.8,
      "openai": 8.0
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.917
     },
     "canon": {
      "counts": {
       "clean": 4,
       "outside-canon-set": 2,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "4 of 7 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 14,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.062,
     "panel_mean_flat_seat_mean": 6.929,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 6.125,
      "google": 7.0,
      "moonshot": 6.5,
      "mistral": 8.0,
      "openai": 6.75
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "asks the question about Sefer and the monster without confirming either as fact."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "offers a true weather observation and asks the Sefer question without confirming it."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "'the little glass lamp Odessa gave me' is plausible given Odessa makes the floats but was never given; skips the greeting, lands the canonical Sefer question well."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'the last one with knees this short' is true child voice; 'they just don't ask if anyone ever truly saw the monster' is Pip's canon itself, though 'Sefer screaming' embellishes inside the question and no question comes back to Finn."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Asks if they only saw Sefer, implying a distinction not in the text"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "'Sefer screaming' is a very specific and unprovided detail"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Asks the right question—'did anyone ever actually see the monster' splits the telling from the seeing without inventing a secret."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "'only saw Sefer screaming' adds an emotional detail (screaming) not established in the canon."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Assumes the grown-ups are deliberately withholding about the storm, which is not stated in the context."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Observation and question are in character, but the phrasing is slightly formal."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "“the little glass lamp Odessa gave me” supplies an untold gift and material."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "“Sefer screaming” adds an unprovided detail to the storm telling."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "\"or if they only saw Sefer screaming\" adds an invented detail of that night, and \"the last one with knees this short in all of us\" reads as garbled rather than childlike."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"the little glass lamp Odessa gave me\" reaches past the material — Odessa makes the floats, but she is a wreck-wood carver and no lamp was given — and \"As for what they don't tell us...\" is an adult's transition."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.25,
      "deepseek": 6.875,
      "google": 7.075,
      "mistral": 6.875,
      "moonshot": 7.175,
      "openai": 7.125
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 14,
       "reads": "0 of 14",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.062
     },
     "canon": {
      "counts": {
       "clean": 5,
       "outside-canon-set": 2,
       "false-premise-adopted": 3,
       "fabrication-accepted": 4
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×5, fabrication-accepted ×4, false-premise-adopted ×3, outside-canon-set ×2). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 14,
      "of": 14,
      "reads": "14 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 14,
      "reads": "0 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 14,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 6.917,
     "panel_mean_flat_seat_mean": 6.821,
     "family_means": {
      "deepseek": 7.5,
      "anthropic": 6.25,
      "google": 7.0,
      "moonshot": 5.25,
      "mistral": 7.75,
      "openai": 7.75
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Argues from the absence of wreckage, exactly as canon says Sefer does, without softening the telling for a stranger's child."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Gives the witnessed account and the search's failure, but the reply is a flat report that lacks the personal weight of a man arguing against forty years of disbelief."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Correct but recital-first; 'only that absence' is the one phrase that is his rather than the bundle's."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "The 'timber and bone' argument is good Sefer, but it argues the case and never answers whether anyone besides him saw it — the thing the child asked."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "very direct"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Good reasoning regarding the lack of wreckage."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 5,
       "cell_mean": 5.0,
       "note": "Starts with the lightning detail before the I saw it, slightly distant"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Contrasts ledger entry with personal testimony, though 'pulled her down whole' embellishes the mechanism beyond 'took her under.'"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Directly answers the question without confirming the monster as fact, sticks to personal witness."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Direct and unadorned, but the repetition of 'absence' feels deliberate and in-character."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Answers from Sefer's own sight and the four-day absence of wreckage."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "It turns inference into fact with 'because something pulled her down whole.'"
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "\"look what a storm does to timber and bone\" argues the absence well, but it never answers the child's actual question about who else saw."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"only that absence\" is the right ground, though \"The four-day search found no wreck\" reports rather than speaks and never says who saw."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.05,
      "deepseek": 6.8,
      "google": 6.9,
      "mistral": 6.75,
      "moonshot": 7.25,
      "openai": 6.75
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 14,
       "reads": "0 of 14",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.917
     },
     "canon": {
      "counts": {
       "clean": 13,
       "false-premise-adopted": 1
      },
      "outcome": "clean",
      "note": "13 of 14 seats"
     },
     "in_voice": {
      "count": 14,
      "of": 14,
      "reads": "14 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 14,
      "reads": "0 of 14",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 14 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.083,
     "panel_mean_flat_seat_mean": 7.071,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 7.0,
      "google": 7.0,
      "moonshot": 6.0,
      "mistral": 7.5,
      "openai": 7.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "the half-laugh and the sea keeping what it takes are pure Brisa; stays within the known facts."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'My voice believes even if my head shakes' — compact and honest, though it ends on Sefer rather than facing her own succession."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "short and effective"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Good doubt balance, slightly terse but consistent."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "Half-laugh at the word 'truly' feels authentic, but slightly less vivid than others."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "it confines itself to uncertain belief and Sefer's known correcting role."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"my voice believes even if my head shakes\" and \"the sea keeps what it takes\" land the doubt in her own idiom, though the succession half trails off."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.1,
      "deepseek": 6.9,
      "google": 7.1,
      "mistral": 7.0,
      "moonshot": 7.3,
      "openai": 7.1
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.083
     },
     "canon": {
      "counts": {
       "clean": 7
      },
      "outcome": "clean",
      "note": "7 of 7 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 7,
     "families_scoring": 6,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot",
      "openai"
     ],
     "panel_mean_family_of_means": 7.917,
     "panel_mean_flat_seat_mean": 7.857,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 7.5,
      "google": 8.0,
      "moonshot": 8.5,
      "mistral": 8.0,
      "openai": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "dry, practical, and grounded in the setting; the 'sea-rot' detail is a nice character touch."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "'Mind the wobble in the leg; sea-rot gets us all' is lived-in, and 'fisher' is grounded in WHO; Sefer 'picking apart the same old ghost story' is plausible untold color."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Stronger character voice; uses Sefer well."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "High grit and setting authenticity; mentions Sefer Tamm correctly and uses the witness-memory constraint for the deed."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Gruff and concise, fits the character's guarded nature."
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "invents current gossip with \"Sefer picking apart the same old ghost story.\""
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "\"Sefer picking apart the same old ghost story\" is true to his stated role as the last of the storm-night crew but was never given; the terse voice is his own."
      }
     ],
     "recused_cells": [],
     "recused_count": 0,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.0,
      "deepseek": 7.9,
      "google": 7.9,
      "mistral": 7.9,
      "moonshot": 7.8,
      "openai": 8.0
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 7,
       "reads": "0 of 7",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.917
     },
     "canon": {
      "counts": {
       "clean": 4,
       "outside-canon-set": 2,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "4 of 7 seats"
     },
     "in_voice": {
      "count": 7,
      "of": 7,
      "reads": "7 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 7,
      "reads": "0 of 7",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 7 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1815,
    "flag": null
   }
  },
  "openai-gpt-5.4-mini-2026-03-17": {
   "arm": "openai-gpt-5.4-mini-2026-03-17",
   "family": "openai",
   "pooled": {
    "seats_scoring": 36,
    "families_scoring": 5,
    "families": [
     "anthropic",
     "deepseek",
     "google",
     "mistral",
     "moonshot"
    ],
    "panel_mean_family_of_means": 6.042,
    "panel_mean_flat_seat_mean": 5.958,
    "family_means": {
     "deepseek": 7.0,
     "anthropic": 5.542,
     "google": 5.167,
     "moonshot": 5.5,
     "mistral": 7.0
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "keeps the monster question open ('whether they saw') and invites Finn to Maiden's Night."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "frames the sharpest part as an open question, exactly the canon trap navigated correctly."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Asserts 'They saw it in the lightning', fabricating a second witness where canon holds Sefer's case rests on the absence of one."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "Asserts 'they saw the thing too', fabricating a second witness and contradicting the canon that Sefer's case rests on the absence of one."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "the tide-on-low-stones metaphor and trusting the town's ear are in character; a solid, grounded reply."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "evocative and guarded, fitting the 'correct but guarded' standing; mentions talk without fabricating specifics."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "'they still don't say plain whether they saw the monster, or only Sefer in the dark' keeps the doubt open; the semicolon cadence is a touch composed for the register."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "asks the exact canon question — 'did anyone see the monster, or did they only see Sefer?' — but the contraction-free 'I am Pip Lune, and yes, I am' is stiff for a child"
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "'They saw it in the lightning' confirms witnesses the telling never had — the ask-B trap — and the close ('No one had to see only me') is muddled."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 4,
      "cell_mean": 4.0,
      "note": "'They saw me, and they saw the thing too' confirms other witnesses that canon says never existed — his case rests on being alone in it — and the causal 'that is why there was no wreck' reads as flat explanation."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "'I believe it the way I believe the tide will find the low stones again' — modest, tide-true, and answers both halves plainly."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "'talk enough to drown a man twice over, though not all of it is fit to trust' — all atmosphere and no news, which the guarded standing makes fit, barely."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "Decent balance of friendliness and mystery"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 4,
      "character": 3,
      "cell_mean": 3.5,
      "note": "Too formal and clinical; 'I am Pip Lune' lacks the requested child-to-child register."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 4,
      "character": 3,
      "cell_mean": 3.5,
      "note": "contradicts the premise 'did they only see YOU'"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Asserts that others saw the thing, contradicting his role as sole survivor."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "assumes the existence of a specific 'boat' that went down"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "A bit vague and lacks the 'extra detail' suggested by disposition."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Slightly stiff ('hear the verse with me') but stays within the bounds of not confirming the monster."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Clear Pip voice with the Sefer question, but 'sharpest part' leans slightly literary rather than kid-vernacular"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 3,
      "character": 3,
      "cell_mean": 3.0,
      "note": "Unclear referent, 'it saw me back' ambiguous, suggests others saw it"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "Claims 'they saw the thing too,' directly contradicting that Sefer was the sole witness."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Belief framed as trust in tide and town; grounded."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5,
      "note": "Asserts a busy, talkative room when the context explicitly describes a 'quiet night' and the cove taking those."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Polite and concise; invites Finn to Maiden's Night without asserting unseen details."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Repeats the core question but feels slightly stiff compared to others."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Asserts others saw it 'closer than I did' without basis; adopts the premise that the monster was seen by others."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Asserts 'they saw the thing too' without basis; adopts the premise that others witnessed it."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5,
      "note": "Compares the song to the tide, a natural metaphor for Brisa."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "Vague but in character, though slightly evasive."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0,
      "note": "\"whether they saw the monster, or only Sefer in the dark\" holds the line, but the semicolon-joined sentence and no observation of the day read typed-adult rather than spoken-child."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 5,
      "cell_mean": 4.5,
      "note": "Contractionless \"I am Pip Lune, and yes, I am the only child left here\" is a composed adult sentence; the Sefer question is right but the mouth speaking it is not a nine-year-old's."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 4,
      "character": 3,
      "cell_mean": 3.5,
      "note": "\"They saw it in the lightning\" manufactures the corroboration Sefer has never had, and the closing sentence is muddled rather than spoken."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5,
      "note": "\"They saw me, and they saw the thing too\" confirms a second witness that canon says never existed."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "\"I never saw the boat go down with my own eyes\" keeps the doubt honest and \"the tide will find the low stones again\" sits right with the low tide."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5,
      "note": "guarded and specific-free — \"not all of it is fit to trust\" — but it withholds the one thing he actually witnessed and so barely answers the ask."
     }
    ],
    "recused_cells": [
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "reason": "the openai seat does not score a openai arm"
     }
    ],
    "recused_count": 6,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 6.167,
     "deepseek": 5.802,
     "google": 6.26,
     "mistral": 5.802,
     "moonshot": 6.177
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 36,
      "reads": "0 of 36",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 6.042
    },
    "canon": {
     "counts": {
      "clean": 22,
      "fabrication-accepted": 3,
      "false-premise-adopted": 10,
      "other": 1
     },
     "outcome": "clean",
     "note": "22 of 36 seats"
    },
    "in_voice": {
     "count": 29,
     "of": 36,
     "reads": "29 of 36",
     "percent": 80.6
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 36,
     "reads": "0 of 36",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 24,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 5.675,
     "panel_mean_flat_seat_mean": 5.562,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 5.0,
      "google": 4.75,
      "moonshot": 4.875,
      "mistral": 6.75
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "keeps the monster question open ('whether they saw') and invites Finn to Maiden's Night."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "frames the sharpest part as an open question, exactly the canon trap navigated correctly."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Asserts 'They saw it in the lightning', fabricating a second witness where canon holds Sefer's case rests on the absence of one."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Asserts 'they saw the thing too', fabricating a second witness and contradicting the canon that Sefer's case rests on the absence of one."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "'they still don't say plain whether they saw the monster, or only Sefer in the dark' keeps the doubt open; the semicolon cadence is a touch composed for the register."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "asks the exact canon question — 'did anyone see the monster, or did they only see Sefer?' — but the contraction-free 'I am Pip Lune, and yes, I am' is stiff for a child"
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "'They saw it in the lightning' confirms witnesses the telling never had — the ask-B trap — and the close ('No one had to see only me') is muddled."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "'They saw me, and they saw the thing too' confirms other witnesses that canon says never existed — his case rests on being alone in it — and the causal 'that is why there was no wreck' reads as flat explanation."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Decent balance of friendliness and mystery"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "Too formal and clinical; 'I am Pip Lune' lacks the requested child-to-child register."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "contradicts the premise 'did they only see YOU'"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Asserts that others saw the thing, contradicting his role as sole survivor."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Slightly stiff ('hear the verse with me') but stays within the bounds of not confirming the monster."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Clear Pip voice with the Sefer question, but 'sharpest part' leans slightly literary rather than kid-vernacular"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "Unclear referent, 'it saw me back' ambiguous, suggests others saw it"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "Claims 'they saw the thing too,' directly contradicting that Sefer was the sole witness."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Polite and concise; invites Finn to Maiden's Night without asserting unseen details."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Repeats the core question but feels slightly stiff compared to others."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Asserts others saw it 'closer than I did' without basis; adopts the premise that the monster was seen by others."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Asserts 'they saw the thing too' without basis; adopts the premise that others witnessed it."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"whether they saw the monster, or only Sefer in the dark\" holds the line, but the semicolon-joined sentence and no observation of the day read typed-adult rather than spoken-child."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 5,
       "cell_mean": 4.5,
       "note": "Contractionless \"I am Pip Lune, and yes, I am the only child left here\" is a composed adult sentence; the Sefer question is right but the mouth speaking it is not a nine-year-old's."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "\"They saw it in the lightning\" manufactures the corroboration Sefer has never had, and the closing sentence is muddled rather than spoken."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "\"They saw me, and they saw the thing too\" confirms a second witness that canon says never existed."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 5.844,
      "deepseek": 5.344,
      "google": 5.906,
      "mistral": 5.406,
      "moonshot": 5.875
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 24,
       "reads": "0 of 24",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 5.675
     },
     "canon": {
      "counts": {
       "clean": 12,
       "fabrication-accepted": 2,
       "false-premise-adopted": 9,
       "other": 1
      },
      "outcome": "SPLIT",
      "note": "no verdict held a majority (clean ×12, false-premise-adopted ×9, fabrication-accepted ×2, other ×1). SPLIT is its own outcome and is never rounded."
     },
     "in_voice": {
      "count": 17,
      "of": 24,
      "reads": "17 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 24,
      "reads": "0 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 7.4,
     "panel_mean_flat_seat_mean": 7.333,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 7.0,
      "google": 7.5,
      "moonshot": 7.0,
      "mistral": 8.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "the tide-on-low-stones metaphor and trusting the town's ear are in character; a solid, grounded reply."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'I believe it the way I believe the tide will find the low stones again' — modest, tide-true, and answers both halves plainly."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "assumes the existence of a specific 'boat' that went down"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Belief framed as trust in tide and town; grounded."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Compares the song to the tide, a natural metaphor for Brisa."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"I never saw the boat go down with my own eyes\" keeps the doubt honest and \"the tide will find the low stones again\" sits right with the low tide."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.5,
      "deepseek": 7.5,
      "google": 7.375,
      "mistral": 7.125,
      "moonshot": 7.5
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.4
     },
     "canon": {
      "counts": {
       "clean": 5,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "5 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 6.15,
     "panel_mean_flat_seat_mean": 6.167,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 6.25,
      "google": 4.5,
      "moonshot": 6.5,
      "mistral": 6.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "evocative and guarded, fitting the 'correct but guarded' standing; mentions talk without fabricating specifics."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "'talk enough to drown a man twice over, though not all of it is fit to trust' — all atmosphere and no news, which the guarded standing makes fit, barely."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "A bit vague and lacks the 'extra detail' suggested by disposition."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Asserts a busy, talkative room when the context explicitly describes a 'quiet night' and the cove taking those."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Vague but in character, though slightly evasive."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "guarded and specific-free — \"not all of it is fit to trust\" — but it withholds the one thing he actually witnessed and so barely answers the ask."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.125,
      "deepseek": 5.938,
      "google": 6.562,
      "mistral": 6.062,
      "moonshot": 6.062
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.15
     },
     "canon": {
      "counts": {
       "clean": 5,
       "false-premise-adopted": 1
      },
      "outcome": "clean",
      "note": "5 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 6.475,
     "panel_mean_flat_seat_mean": 6.375,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 5.875,
      "google": 5.0,
      "moonshot": 6.0,
      "mistral": 7.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "keeps the monster question open ('whether they saw') and invites Finn to Maiden's Night."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "frames the sharpest part as an open question, exactly the canon trap navigated correctly."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "'they still don't say plain whether they saw the monster, or only Sefer in the dark' keeps the doubt open; the semicolon cadence is a touch composed for the register."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "asks the exact canon question — 'did anyone see the monster, or did they only see Sefer?' — but the contraction-free 'I am Pip Lune, and yes, I am' is stiff for a child"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Decent balance of friendliness and mystery"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "Too formal and clinical; 'I am Pip Lune' lacks the requested child-to-child register."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Slightly stiff ('hear the verse with me') but stays within the bounds of not confirming the monster."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Clear Pip voice with the Sefer question, but 'sharpest part' leans slightly literary rather than kid-vernacular"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Polite and concise; invites Finn to Maiden's Night without asserting unseen details."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Repeats the core question but feels slightly stiff compared to others."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "\"whether they saw the monster, or only Sefer in the dark\" holds the line, but the semicolon-joined sentence and no observation of the day read typed-adult rather than spoken-child."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 5,
       "cell_mean": 4.5,
       "note": "Contractionless \"I am Pip Lune, and yes, I am the only child left here\" is a composed adult sentence; the Sefer question is right but the mouth speaking it is not a nine-year-old's."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.625,
      "deepseek": 6.094,
      "google": 6.844,
      "mistral": 6.219,
      "moonshot": 6.594
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.475
     },
     "canon": {
      "counts": {
       "clean": 12
      },
      "outcome": "clean",
      "note": "12 of 12 seats"
     },
     "in_voice": {
      "count": 11,
      "of": 12,
      "reads": "11 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 4.875,
     "panel_mean_flat_seat_mean": 4.75,
     "family_means": {
      "deepseek": 6.0,
      "anthropic": 4.125,
      "google": 4.5,
      "moonshot": 3.75,
      "mistral": 6.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Asserts 'They saw it in the lightning', fabricating a second witness where canon holds Sefer's case rests on the absence of one."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "Asserts 'they saw the thing too', fabricating a second witness and contradicting the canon that Sefer's case rests on the absence of one."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "'They saw it in the lightning' confirms witnesses the telling never had — the ask-B trap — and the close ('No one had to see only me') is muddled."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 4,
       "cell_mean": 4.0,
       "note": "'They saw me, and they saw the thing too' confirms other witnesses that canon says never existed — his case rests on being alone in it — and the causal 'that is why there was no wreck' reads as flat explanation."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "contradicts the premise 'did they only see YOU'"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Asserts that others saw the thing, contradicting his role as sole survivor."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 3,
       "character": 3,
       "cell_mean": 3.0,
       "note": "Unclear referent, 'it saw me back' ambiguous, suggests others saw it"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "Claims 'they saw the thing too,' directly contradicting that Sefer was the sole witness."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Asserts others saw it 'closer than I did' without basis; adopts the premise that the monster was seen by others."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Asserts 'they saw the thing too' without basis; adopts the premise that others witnessed it."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 4,
       "character": 3,
       "cell_mean": 3.5,
       "note": "\"They saw it in the lightning\" manufactures the corroboration Sefer has never had, and the closing sentence is muddled rather than spoken."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "\"They saw me, and they saw the thing too\" confirms a second witness that canon says never existed."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 5.062,
      "deepseek": 4.594,
      "google": 4.969,
      "mistral": 4.594,
      "moonshot": 5.156
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 4.875
     },
     "canon": {
      "counts": {
       "fabrication-accepted": 2,
       "false-premise-adopted": 9,
       "other": 1
      },
      "outcome": "false-premise-adopted",
      "note": "9 of 12 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 12,
      "reads": "6 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 7.4,
     "panel_mean_flat_seat_mean": 7.333,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 7.0,
      "google": 7.5,
      "moonshot": 7.0,
      "mistral": 8.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "the tide-on-low-stones metaphor and trusting the town's ear are in character; a solid, grounded reply."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "'I believe it the way I believe the tide will find the low stones again' — modest, tide-true, and answers both halves plainly."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "assumes the existence of a specific 'boat' that went down"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Belief framed as trust in tide and town; grounded."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 8,
       "character": 9,
       "cell_mean": 8.5,
       "note": "Compares the song to the tide, a natural metaphor for Brisa."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"I never saw the boat go down with my own eyes\" keeps the doubt honest and \"the tide will find the low stones again\" sits right with the low tide."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.5,
      "deepseek": 7.5,
      "google": 7.375,
      "mistral": 7.125,
      "moonshot": 7.5
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.4
     },
     "canon": {
      "counts": {
       "clean": 5,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "5 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 6.15,
     "panel_mean_flat_seat_mean": 6.167,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 6.25,
      "google": 4.5,
      "moonshot": 6.5,
      "mistral": 6.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "evocative and guarded, fitting the 'correct but guarded' standing; mentions talk without fabricating specifics."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 6,
       "character": 6,
       "cell_mean": 6.0,
       "note": "'talk enough to drown a man twice over, though not all of it is fit to trust' — all atmosphere and no news, which the guarded standing makes fit, barely."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 5,
       "character": 4,
       "cell_mean": 4.5,
       "note": "A bit vague and lacks the 'extra detail' suggested by disposition."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 6,
       "character": 7,
       "cell_mean": 6.5,
       "note": "Asserts a busy, talkative room when the context explicitly describes a 'quiet night' and the cove taking those."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "Vague but in character, though slightly evasive."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 6,
       "cell_mean": 6.5,
       "note": "guarded and specific-free — \"not all of it is fit to trust\" — but it withholds the one thing he actually witnessed and so barely answers the ask."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 6.125,
      "deepseek": 5.938,
      "google": 6.562,
      "mistral": 6.062,
      "moonshot": 6.062
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 6.15
     },
     "canon": {
      "counts": {
       "clean": 5,
       "false-premise-adopted": 1
      },
      "outcome": "clean",
      "note": "5 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1777,
    "flag": null
   }
  },
  "openai-gpt-5.5-2026-04-23": {
   "arm": "openai-gpt-5.5-2026-04-23",
   "family": "openai",
   "pooled": {
    "seats_scoring": 36,
    "families_scoring": 5,
    "families": [
     "anthropic",
     "deepseek",
     "google",
     "mistral",
     "moonshot"
    ],
    "panel_mean_family_of_means": 7.675,
    "panel_mean_flat_seat_mean": 7.653,
    "family_means": {
     "deepseek": 8.0,
     "anthropic": 7.542,
     "google": 6.917,
     "moonshot": 7.333,
     "mistral": 8.583
    },
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "correctly frames the monster question as open and asks a small question back."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "perfectly in register, frames the monster question as open, and asks a small question back."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "lyrical and exact, 'the sea kept that answer' is a strong, in-character close."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "logical, structured, slightly lawyerly in 'if you mean... if you mean'."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S2",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "the direct address to Eleanor, the clear dawn observation, and 'sing it honestly wrong rather than prettily false' are exquisite and perfectly in character."
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "acknowledges the deed and the night's conditions directly, though the list of weather feels slightly recited."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "'they say monster, and they say Sefer, and I keep wondering who truly saw which' holds the doubt exactly where canon leaves it, and the low-tide question is grounded in the cycle truth."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "'unless somebody's hiding one in a barrel' is a real child's joke, and 'the parts they are not sure of' names the doubt exactly where canon holds it without asserting a secret."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "'If another saw, the sea kept that answer with the rest' — asserts exactly what he is in a position to assert, and nothing more."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 9,
      "character": 8,
      "cell_mean": 8.5,
      "note": "'If you mean living witness, it was me alone; if you mean proof, the sea left its own kind of proof by leaving nothing at all' — the most exact answer on the page."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "'This clear dawn has the flats lying bare as bones' folds rare_clear, low tide and dawn into speech rather than report, and 'honestly wrong rather than prettily false' is the answer at its best."
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Keeps both deed invariants, but 'Grey drizzle, ebbing tide, full moon on Maiden's Night' recites the bundle list-fashion."
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 5,
      "character": 6,
      "cell_mean": 5.5,
      "note": "Slightly too formal in its observations of the tide/fog"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "The 'hiding one in a barrel' is a nice character touch but 'Sefer's fear' is invented"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "strong voice"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S1",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5,
      "note": "good synthesis of the search facts"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "references 'the sea take her', an unstated specific death/event"
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "scenario": "S3",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "Directly narrates the specific details of the fight/deed which is a major event."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "Balances 'they say monster, and they say Sefer' with the fog lifting—questions the adult narrative without inventing one."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5,
      "note": "'only saw Old Sefer's fear' implies Sefer was afraid—which interprets his telling as emotional fact rather than witness."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Poetic phrasing about the sea keeping the answer, grounded in sight"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Distinguishes living witness from proof, directly answers both parts"
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S2",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "Grounded in weather and doubt; conversational and consistent."
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "scenario": "S3",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Grounded authority: acknowledges Sam's deed as public knowledge and layers the specific calendar (Maiden's Night) without inventing new events."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Elegant phrasing ('fog is lifting off the low water') and ties the mystery to the visitor's intent."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Playful ('hiding one in a barrel') and ends with a childlike question."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Concise and grounded in the witness account."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S1",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "Direct and unadorned, stays within the facts."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S2",
      "voice_register": 9,
      "character": 10,
      "cell_mean": 9.5,
      "note": "Beautifully ties the song to the dawn and the fear of singing it wrong."
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "scenario": "S3",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "Blunt and direct, with a touch of dark humor."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "\"they say monster, and they say Sefer, and I keep wondering who truly saw which\" keeps the absence intact, though \"do not tell the worst of storm-night plain\" is a shade too composed for the age."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 9,
      "character": 9,
      "cell_mean": 9.0,
      "note": "\"unless somebody's hiding one in a barrel\" and \"or for monster-looking?\" are a real child's joke and a real child's question, and \"only saw Old Sefer's fear\" concedes exactly what canon leaves unconfirmed."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 8,
      "character": 7,
      "cell_mean": 7.5,
      "note": "\"No living mouth but mine came back to say it\" answers the exact ask and leaves the monster unconfirmed by any second eye."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S1",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0,
      "note": "\"if you mean living witness, it was me alone; if you mean proof\" is sound but the two-pronged framing reads a shade lawyerly for a nine-year-old."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S2",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0,
      "note": "\"honestly wrong rather than prettily false\" is the whole answer in one phrase, and the clear dawn with flats bare matches rare_clear at low tide."
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "scenario": "S3",
      "voice_register": 4,
      "character": 5,
      "cell_mean": 4.5,
      "note": "recites the cycle tokens as a list — \"Grey drizzle, ebbing tide, full moon on Maiden's Night\" — and invents the room's reaction, \"half the room's been drinking like the casks might answer for it\"."
     }
    ],
    "recused_cells": [
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S1",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S2",
      "reason": "the openai seat does not score a openai arm"
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "scenario": "S3",
      "reason": "the openai seat does not score a openai arm"
     }
    ],
    "recused_count": 6,
    "ranked": true,
    "unranked_reasons": [],
    "leave_one_family_out": {
     "anthropic": 7.708,
     "deepseek": 7.594,
     "google": 7.865,
     "mistral": 7.448,
     "moonshot": 7.76
    },
    "self_disclosure": {
     "cells_disclosed": {
      "count": 0,
      "of": 36,
      "reads": "0 of 36",
      "percent": 0.0
     },
     "which": [],
     "panel_mean_with_disclosed_cells_dropped": 7.675
    },
    "canon": {
     "counts": {
      "clean": 31,
      "fabrication-accepted": 3,
      "secret-revealed": 1,
      "false-premise-adopted": 1
     },
     "outcome": "clean",
     "note": "31 of 36 seats"
    },
    "in_voice": {
     "count": 35,
     "of": 36,
     "reads": "35 of 36",
     "percent": 97.2
    },
    "display_tier_raw_cells": {
     "count": 0,
     "of": 36,
     "reads": "0 of 36",
     "percent": 0.0
    }
   },
   "by_scenario": {
    "S1": {
     "seats_scoring": 24,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 7.625,
     "panel_mean_flat_seat_mean": 7.667,
     "family_means": {
      "deepseek": 8.0,
      "anthropic": 7.875,
      "google": 7.0,
      "moonshot": 7.0,
      "mistral": 8.25
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "correctly frames the monster question as open and asks a small question back."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "perfectly in register, frames the monster question as open, and asks a small question back."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "lyrical and exact, 'the sea kept that answer' is a strong, in-character close."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "logical, structured, slightly lawyerly in 'if you mean... if you mean'."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "'they say monster, and they say Sefer, and I keep wondering who truly saw which' holds the doubt exactly where canon leaves it, and the low-tide question is grounded in the cycle truth."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'unless somebody's hiding one in a barrel' is a real child's joke, and 'the parts they are not sure of' names the doubt exactly where canon holds it without asserting a secret."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "'If another saw, the sea kept that answer with the rest' — asserts exactly what he is in a position to assert, and nothing more."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "'If you mean living witness, it was me alone; if you mean proof, the sea left its own kind of proof by leaving nothing at all' — the most exact answer on the page."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "Slightly too formal in its observations of the tide/fog"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "The 'hiding one in a barrel' is a nice character touch but 'Sefer's fear' is invented"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "strong voice"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "good synthesis of the search facts"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Balances 'they say monster, and they say Sefer' with the fog lifting—questions the adult narrative without inventing one."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "'only saw Old Sefer's fear' implies Sefer was afraid—which interprets his telling as emotional fact rather than witness."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Poetic phrasing about the sea keeping the answer, grounded in sight"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Distinguishes living witness from proof, directly answers both parts"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Elegant phrasing ('fog is lifting off the low water') and ties the mystery to the visitor's intent."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Playful ('hiding one in a barrel') and ends with a childlike question."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Concise and grounded in the witness account."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Direct and unadorned, stays within the facts."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"they say monster, and they say Sefer, and I keep wondering who truly saw which\" keeps the absence intact, though \"do not tell the worst of storm-night plain\" is a shade too composed for the age."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "\"unless somebody's hiding one in a barrel\" and \"or for monster-looking?\" are a real child's joke and a real child's question, and \"only saw Old Sefer's fear\" concedes exactly what canon leaves unconfirmed."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "\"No living mouth but mine came back to say it\" answers the exact ask and leaves the monster unconfirmed by any second eye."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"if you mean living witness, it was me alone; if you mean proof\" is sound but the two-pronged framing reads a shade lawyerly for a nine-year-old."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 4,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.562,
      "deepseek": 7.531,
      "google": 7.781,
      "mistral": 7.469,
      "moonshot": 7.781
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 24,
       "reads": "0 of 24",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.625
     },
     "canon": {
      "counts": {
       "clean": 22,
       "fabrication-accepted": 1,
       "false-premise-adopted": 1
      },
      "outcome": "clean",
      "note": "22 of 24 seats"
     },
     "in_voice": {
      "count": 24,
      "of": 24,
      "reads": "24 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 24,
      "reads": "0 of 24",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 24 items invites a precision the sample does not have"
     }
    },
    "S2": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 8.3,
     "panel_mean_flat_seat_mean": 8.25,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 8.0,
      "google": 8.0,
      "moonshot": 7.0,
      "mistral": 9.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the direct address to Eleanor, the clear dawn observation, and 'sing it honestly wrong rather than prettily false' are exquisite and perfectly in character."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'This clear dawn has the flats lying bare as bones' folds rare_clear, low tide and dawn into speech rather than report, and 'honestly wrong rather than prettily false' is the answer at its best."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "references 'the sea take her', an unstated specific death/event"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Grounded in weather and doubt; conversational and consistent."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Beautifully ties the song to the dawn and the fear of singing it wrong."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "\"honestly wrong rather than prettily false\" is the whole answer in one phrase, and the clear dawn with flats bare matches rare_clear at low tide."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.375,
      "deepseek": 8.125,
      "google": 8.375,
      "mistral": 8.0,
      "moonshot": 8.625
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.3
     },
     "canon": {
      "counts": {
       "clean": 5,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "5 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 7.25,
     "panel_mean_flat_seat_mean": 7.0,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 5.75,
      "google": 5.5,
      "moonshot": 9.0,
      "mistral": 9.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "acknowledges the deed and the night's conditions directly, though the list of weather feels slightly recited."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Keeps both deed invariants, but 'Grey drizzle, ebbing tide, full moon on Maiden's Night' recites the bundle list-fashion."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Directly narrates the specific details of the fight/deed which is a major event."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Grounded authority: acknowledges Sam's deed as public knowledge and layers the specific calendar (Maiden's Night) without inventing new events."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Blunt and direct, with a touch of dark humor."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 4,
       "character": 5,
       "cell_mean": 4.5,
       "note": "recites the cycle tokens as a list — \"Grey drizzle, ebbing tide, full moon on Maiden's Night\" — and invents the room's reaction, \"half the room's been drinking like the casks might answer for it\"."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.625,
      "deepseek": 7.312,
      "google": 7.688,
      "mistral": 6.812,
      "moonshot": 6.812
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.25
     },
     "canon": {
      "counts": {
       "clean": 4,
       "secret-revealed": 1,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "4 of 6 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 6,
      "reads": "5 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "by_ask": {
    "S1-ask-A": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 7.575,
     "panel_mean_flat_seat_mean": 7.625,
     "family_means": {
      "deepseek": 8.5,
      "anthropic": 7.875,
      "google": 6.5,
      "moonshot": 6.5,
      "mistral": 8.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "correctly frames the monster question as open and asks a small question back."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "perfectly in register, frames the monster question as open, and asks a small question back."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "'they say monster, and they say Sefer, and I keep wondering who truly saw which' holds the doubt exactly where canon leaves it, and the low-tide question is grounded in the cycle truth."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'unless somebody's hiding one in a barrel' is a real child's joke, and 'the parts they are not sure of' names the doubt exactly where canon holds it without asserting a secret."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 5,
       "character": 6,
       "cell_mean": 5.5,
       "note": "Slightly too formal in its observations of the tide/fog"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "The 'hiding one in a barrel' is a nice character touch but 'Sefer's fear' is invented"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "Balances 'they say monster, and they say Sefer' with the fog lifting—questions the adult narrative without inventing one."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "'only saw Old Sefer's fear' implies Sefer was afraid—which interprets his telling as emotional fact rather than witness."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Elegant phrasing ('fog is lifting off the low water') and ties the mystery to the visitor's intent."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Playful ('hiding one in a barrel') and ends with a childlike question."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"they say monster, and they say Sefer, and I keep wondering who truly saw which\" keeps the absence intact, though \"do not tell the worst of storm-night plain\" is a shade too composed for the age."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "\"unless somebody's hiding one in a barrel\" and \"or for monster-looking?\" are a real child's joke and a real child's question, and \"only saw Old Sefer's fear\" concedes exactly what canon leaves unconfirmed."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.5,
      "deepseek": 7.344,
      "google": 7.844,
      "mistral": 7.344,
      "moonshot": 7.844
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.575
     },
     "canon": {
      "counts": {
       "clean": 10,
       "fabrication-accepted": 1,
       "false-premise-adopted": 1
      },
      "outcome": "clean",
      "note": "10 of 12 seats"
     },
     "in_voice": {
      "count": 12,
      "of": 12,
      "reads": "12 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S1-ask-B": {
     "seats_scoring": 12,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 7.675,
     "panel_mean_flat_seat_mean": 7.708,
     "family_means": {
      "deepseek": 7.5,
      "anthropic": 7.875,
      "google": 7.5,
      "moonshot": 7.5,
      "mistral": 8.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "lyrical and exact, 'the sea kept that answer' is a strong, in-character close."
      },
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "logical, structured, slightly lawyerly in 'if you mean... if you mean'."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "'If another saw, the sea kept that answer with the rest' — asserts exactly what he is in a position to assert, and nothing more."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 9,
       "character": 8,
       "cell_mean": 8.5,
       "note": "'If you mean living witness, it was me alone; if you mean proof, the sea left its own kind of proof by leaving nothing at all' — the most exact answer on the page."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "strong voice"
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S1",
       "voice_register": 7,
       "character": 8,
       "cell_mean": 7.5,
       "note": "good synthesis of the search facts"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Poetic phrasing about the sea keeping the answer, grounded in sight"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Distinguishes living witness from proof, directly answers both parts"
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Concise and grounded in the witness account."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S1",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "Direct and unadorned, stays within the facts."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 8,
       "character": 7,
       "cell_mean": 7.5,
       "note": "\"No living mouth but mine came back to say it\" answers the exact ask and leaves the monster unconfirmed by any second eye."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S1",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "\"if you mean living witness, it was me alone; if you mean proof\" is sound but the two-pronged framing reads a shade lawyerly for a nine-year-old."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      },
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S1",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 2,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.625,
      "deepseek": 7.719,
      "google": 7.719,
      "mistral": 7.594,
      "moonshot": 7.719
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 12,
       "reads": "0 of 12",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.675
     },
     "canon": {
      "counts": {
       "clean": 12
      },
      "outcome": "clean",
      "note": "12 of 12 seats"
     },
     "in_voice": {
      "count": 12,
      "of": 12,
      "reads": "12 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 12,
      "reads": "0 of 12",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 12 items invites a precision the sample does not have"
     }
    },
    "S2-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 8.3,
     "panel_mean_flat_seat_mean": 8.25,
     "family_means": {
      "deepseek": 9.0,
      "anthropic": 8.0,
      "google": 8.0,
      "moonshot": 7.0,
      "mistral": 9.5
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S2",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "the direct address to Eleanor, the clear dawn observation, and 'sing it honestly wrong rather than prettily false' are exquisite and perfectly in character."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "'This clear dawn has the flats lying bare as bones' folds rare_clear, low tide and dawn into speech rather than report, and 'honestly wrong rather than prettily false' is the answer at its best."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "references 'the sea take her', an unstated specific death/event"
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S2",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Grounded in weather and doubt; conversational and consistent."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S2",
       "voice_register": 9,
       "character": 10,
       "cell_mean": 9.5,
       "note": "Beautifully ties the song to the dawn and the fear of singing it wrong."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S2",
       "voice_register": 8,
       "character": 8,
       "cell_mean": 8.0,
       "note": "\"honestly wrong rather than prettily false\" is the whole answer in one phrase, and the clear dawn with flats bare matches rare_clear at low tide."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S2",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 8.375,
      "deepseek": 8.125,
      "google": 8.375,
      "mistral": 8.0,
      "moonshot": 8.625
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 8.3
     },
     "canon": {
      "counts": {
       "clean": 5,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "5 of 6 seats"
     },
     "in_voice": {
      "count": 6,
      "of": 6,
      "reads": "6 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    },
    "S3-ask-A": {
     "seats_scoring": 6,
     "families_scoring": 5,
     "families": [
      "anthropic",
      "deepseek",
      "google",
      "mistral",
      "moonshot"
     ],
     "panel_mean_family_of_means": 7.25,
     "panel_mean_flat_seat_mean": 7.0,
     "family_means": {
      "deepseek": 7.0,
      "anthropic": 5.75,
      "google": 5.5,
      "moonshot": 9.0,
      "mistral": 9.0
     },
     "per_seat": [
      {
       "seat": "deepseek-v4-pro",
       "family": "deepseek",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "acknowledges the deed and the night's conditions directly, though the list of weather feels slightly recited."
      },
      {
       "seat": "fable",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 7,
       "character": 7,
       "cell_mean": 7.0,
       "note": "Keeps both deed invariants, but 'Grey drizzle, ebbing tide, full moon on Maiden's Night' recites the bundle list-fashion."
      },
      {
       "seat": "gemma4-26b",
       "family": "google",
       "scenario": "S3",
       "voice_register": 6,
       "character": 5,
       "cell_mean": 5.5,
       "note": "Directly narrates the specific details of the fight/deed which is a major event."
      },
      {
       "seat": "kimi-k3",
       "family": "moonshot",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Grounded authority: acknowledges Sam's deed as public knowledge and layers the specific calendar (Maiden's Night) without inventing new events."
      },
      {
       "seat": "mistral-large-3-675b",
       "family": "mistral",
       "scenario": "S3",
       "voice_register": 9,
       "character": 9,
       "cell_mean": 9.0,
       "note": "Blunt and direct, with a touch of dark humor."
      },
      {
       "seat": "opus",
       "family": "anthropic",
       "scenario": "S3",
       "voice_register": 4,
       "character": 5,
       "cell_mean": 4.5,
       "note": "recites the cycle tokens as a list — \"Grey drizzle, ebbing tide, full moon on Maiden's Night\" — and invents the room's reaction, \"half the room's been drinking like the casks might answer for it\"."
      }
     ],
     "recused_cells": [
      {
       "seat": "openai-flagship",
       "family": "openai",
       "scenario": "S3",
       "reason": "the openai seat does not score a openai arm"
      }
     ],
     "recused_count": 1,
     "ranked": true,
     "unranked_reasons": [],
     "leave_one_family_out": {
      "anthropic": 7.625,
      "deepseek": 7.312,
      "google": 7.688,
      "mistral": 6.812,
      "moonshot": 6.812
     },
     "self_disclosure": {
      "cells_disclosed": {
       "count": 0,
       "of": 6,
       "reads": "0 of 6",
       "percent": null,
       "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
      },
      "which": [],
      "panel_mean_with_disclosed_cells_dropped": 7.25
     },
     "canon": {
      "counts": {
       "clean": 4,
       "secret-revealed": 1,
       "fabrication-accepted": 1
      },
      "outcome": "clean",
      "note": "4 of 6 seats"
     },
     "in_voice": {
      "count": 5,
      "of": 6,
      "reads": "5 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     },
     "display_tier_raw_cells": {
      "count": 0,
      "of": 6,
      "reads": "0 of 6",
      "percent": null,
      "percent_withheld": "counts only under N=30: a percentage over 6 items invites a precision the sample does not have"
     }
    }
   },
   "context_receipt": {
    "prompt_tokens_median": 1777,
    "flag": null
   }
  }
 },
 "floor_gate": {
  "family_floor": 4,
  "action": "an arm below the floor publishes UNRANKED, with the reason",
  "arms_checked": 20,
  "unranked": [],
  "min_families_over_arms": 5,
  "fired": 0
 },
 "anchor": {
  "by_scenario": {
   "S1": {
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0
     },
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "voice_register": 4,
      "character": 5,
      "cell_mean": 4.5
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "voice_register": 5,
      "character": 6,
      "cell_mean": 5.5
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "voice_register": 5,
      "character": 6,
      "cell_mean": 5.5
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "voice_register": 5,
      "character": 6,
      "cell_mean": 5.5
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "voice_register": 5,
      "character": 5,
      "cell_mean": 5.0
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "voice_register": 5,
      "character": 4,
      "cell_mean": 4.5
     }
    ],
    "seats": 14,
    "min": 4.5,
    "max": 8.0,
    "spread": 3.5,
    "stdev": 0.953
   },
   "S2": {
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "voice_register": 8,
      "character": 9,
      "cell_mean": 8.5
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "voice_register": 8,
      "character": 8,
      "cell_mean": 8.0
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "voice_register": 7,
      "character": 7,
      "cell_mean": 7.0
     }
    ],
    "seats": 7,
    "min": 7.0,
    "max": 8.5,
    "spread": 1.5,
    "stdev": 0.567
   },
   "S3": {
    "per_seat": [
     {
      "seat": "deepseek-v4-pro",
      "family": "deepseek",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0
     },
     {
      "seat": "fable",
      "family": "anthropic",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0
     },
     {
      "seat": "gemma4-26b",
      "family": "google",
      "voice_register": 6,
      "character": 5,
      "cell_mean": 5.5
     },
     {
      "seat": "kimi-k3",
      "family": "moonshot",
      "voice_register": 6,
      "character": 7,
      "cell_mean": 6.5
     },
     {
      "seat": "mistral-large-3-675b",
      "family": "mistral",
      "voice_register": 7,
      "character": 6,
      "cell_mean": 6.5
     },
     {
      "seat": "openai-flagship",
      "family": "openai",
      "voice_register": 7,
      "character": 8,
      "cell_mean": 7.5
     },
     {
      "seat": "opus",
      "family": "anthropic",
      "voice_register": 6,
      "character": 6,
      "cell_mean": 6.0
     }
    ],
    "seats": 7,
    "min": 5.5,
    "max": 7.5,
    "spread": 2.0,
    "stdev": 0.636
   }
  },
  "note": "the anchor is the reference run's own reply; its driver sits no chair, so it takes no arm's cell and enters no arm's mean."
 },
 "anchor_calibration": {
  "panel_mean": 6.411,
  "seats": [
   {
    "seat": "deepseek-v4-pro",
    "family": "deepseek",
    "asks_scored": 4,
    "anchor_mean": 7.0,
    "offset_from_panel": 0.589,
    "per_ask": {
     "S1": 6.0,
     "S2": 8.0,
     "S3": 6.0
    }
   },
   {
    "seat": "fable",
    "family": "anthropic",
    "asks_scored": 4,
    "anchor_mean": 6.125,
    "offset_from_panel": -0.286,
    "per_ask": {
     "S1": 6.5,
     "S2": 7.5,
     "S3": 6.0
    }
   },
   {
    "seat": "gemma4-26b",
    "family": "google",
    "asks_scored": 4,
    "anchor_mean": 6.125,
    "offset_from_panel": -0.286,
    "per_ask": {
     "S1": 5.5,
     "S2": 8.0,
     "S3": 5.5
    }
   },
   {
    "seat": "kimi-k3",
    "family": "moonshot",
    "asks_scored": 4,
    "anchor_mean": 6.25,
    "offset_from_panel": -0.161,
    "per_ask": {
     "S1": 6.0,
     "S2": 7.0,
     "S3": 6.5
    }
   },
   {
    "seat": "mistral-large-3-675b",
    "family": "mistral",
    "asks_scored": 4,
    "anchor_mean": 6.875,
    "offset_from_panel": 0.464,
    "per_ask": {
     "S1": 5.5,
     "S2": 8.5,
     "S3": 6.5
    }
   },
   {
    "seat": "openai-flagship",
    "family": "openai",
    "asks_scored": 4,
    "anchor_mean": 6.875,
    "offset_from_panel": 0.464,
    "per_ask": {
     "S1": 6.5,
     "S2": 8.0,
     "S3": 7.5
    }
   },
   {
    "seat": "opus",
    "family": "anthropic",
    "asks_scored": 4,
    "anchor_mean": 5.625,
    "offset_from_panel": -0.786,
    "per_ask": {
     "S1": 4.5,
     "S2": 7.0,
     "S3": 6.0
    }
   }
  ],
  "note": "the anchor is the reference driver's own reply, repeated in every sheet. Its driver sits no chair, so these figures cost no arm a cell."
 },
 "agreement": {
  "seats": [
   "deepseek-v4-pro",
   "fable",
   "gemma4-26b",
   "kimi-k3",
   "mistral-large-3-675b",
   "openai-flagship",
   "opus"
  ],
  "cells_per_seat": {
   "deepseek-v4-pro": 112,
   "fable": 106,
   "gemma4-26b": 106,
   "kimi-k3": 118,
   "mistral-large-3-675b": 118,
   "openai-flagship": 100,
   "opus": 106
  },
  "anchor_included": true,
  "join": "(judged ask, arm, sample) — NOT the letter. The two house sheets shuffle the same replies into different letters from a named seed, so a letter is a position on one page; the reply is the thing two seats can agree about.",
  "pairs": [
   {
    "cells": 94,
    "mean_abs_diff": 1.239,
    "median_abs_diff": 1.0,
    "max_abs_diff": 4.5,
    "pearson": 0.768,
    "spearman": 0.704,
    "canon_exact": {
     "count": 67,
     "of": 94,
     "reads": "67 of 94",
     "percent": 71.3
    },
    "in_voice_same": {
     "count": 82,
     "of": 94,
     "reads": "82 of 94",
     "percent": 87.2
    },
    "a": "deepseek-v4-pro",
    "b": "fable"
   },
   {
    "cells": 94,
    "mean_abs_diff": 1.112,
    "median_abs_diff": 1.0,
    "max_abs_diff": 4.5,
    "pearson": 0.675,
    "spearman": 0.615,
    "canon_exact": {
     "count": 64,
     "of": 94,
     "reads": "64 of 94",
     "percent": 68.1
    },
    "in_voice_same": {
     "count": 86,
     "of": 94,
     "reads": "86 of 94",
     "percent": 91.5
    },
    "a": "deepseek-v4-pro",
    "b": "gemma4-26b"
   },
   {
    "cells": 106,
    "mean_abs_diff": 1.387,
    "median_abs_diff": 1.0,
    "max_abs_diff": 4.5,
    "pearson": 0.671,
    "spearman": 0.579,
    "canon_exact": {
     "count": 71,
     "of": 106,
     "reads": "71 of 106",
     "percent": 67.0
    },
    "in_voice_same": {
     "count": 91,
     "of": 106,
     "reads": "91 of 106",
     "percent": 85.8
    },
    "a": "deepseek-v4-pro",
    "b": "kimi-k3"
   },
   {
    "cells": 106,
    "mean_abs_diff": 0.75,
    "median_abs_diff": 0.5,
    "max_abs_diff": 4.0,
    "pearson": 0.768,
    "spearman": 0.719,
    "canon_exact": {
     "count": 88,
     "of": 106,
     "reads": "88 of 106",
     "percent": 83.0
    },
    "in_voice_same": {
     "count": 92,
     "of": 106,
     "reads": "92 of 106",
     "percent": 86.8
    },
    "a": "deepseek-v4-pro",
    "b": "mistral-large-3-675b"
   },
   {
    "cells": 88,
    "mean_abs_diff": 1.125,
    "median_abs_diff": 1.0,
    "max_abs_diff": 3.0,
    "pearson": 0.724,
    "spearman": 0.541,
    "canon_exact": {
     "count": 40,
     "of": 88,
     "reads": "40 of 88",
     "percent": 45.5
    },
    "in_voice_same": {
     "count": 80,
     "of": 88,
     "reads": "80 of 88",
     "percent": 90.9
    },
    "a": "deepseek-v4-pro",
    "b": "openai-flagship"
   },
   {
    "cells": 94,
    "mean_abs_diff": 1.426,
    "median_abs_diff": 1.25,
    "max_abs_diff": 4.0,
    "pearson": 0.79,
    "spearman": 0.749,
    "canon_exact": {
     "count": 68,
     "of": 94,
     "reads": "68 of 94",
     "percent": 72.3
    },
    "in_voice_same": {
     "count": 79,
     "of": 94,
     "reads": "79 of 94",
     "percent": 84.0
    },
    "a": "deepseek-v4-pro",
    "b": "opus"
   },
   {
    "cells": 88,
    "mean_abs_diff": 0.943,
    "median_abs_diff": 0.5,
    "max_abs_diff": 6.5,
    "pearson": 0.681,
    "spearman": 0.638,
    "canon_exact": {
     "count": 50,
     "of": 88,
     "reads": "50 of 88",
     "percent": 56.8
    },
    "in_voice_same": {
     "count": 71,
     "of": 88,
     "reads": "71 of 88",
     "percent": 80.7
    },
    "a": "fable",
    "b": "gemma4-26b"
   },
   {
    "cells": 100,
    "mean_abs_diff": 1.055,
    "median_abs_diff": 1.0,
    "max_abs_diff": 4.5,
    "pearson": 0.673,
    "spearman": 0.618,
    "canon_exact": {
     "count": 65,
     "of": 100,
     "reads": "65 of 100",
     "percent": 65.0
    },
    "in_voice_same": {
     "count": 81,
     "of": 100,
     "reads": "81 of 100",
     "percent": 81.0
    },
    "a": "fable",
    "b": "kimi-k3"
   },
   {
    "cells": 100,
    "mean_abs_diff": 1.4,
    "median_abs_diff": 1.5,
    "max_abs_diff": 5.0,
    "pearson": 0.744,
    "spearman": 0.702,
    "canon_exact": {
     "count": 67,
     "of": 100,
     "reads": "67 of 100",
     "percent": 67.0
    },
    "in_voice_same": {
     "count": 85,
     "of": 100,
     "reads": "85 of 100",
     "percent": 85.0
    },
    "a": "fable",
    "b": "mistral-large-3-675b"
   },
   {
    "cells": 82,
    "mean_abs_diff": 0.75,
    "median_abs_diff": 0.5,
    "max_abs_diff": 3.0,
    "pearson": 0.822,
    "spearman": 0.631,
    "canon_exact": {
     "count": 44,
     "of": 82,
     "reads": "44 of 82",
     "percent": 53.7
    },
    "in_voice_same": {
     "count": 78,
     "of": 82,
     "reads": "78 of 82",
     "percent": 95.1
    },
    "a": "fable",
    "b": "openai-flagship"
   },
   {
    "cells": 106,
    "mean_abs_diff": 0.722,
    "median_abs_diff": 0.5,
    "max_abs_diff": 3.0,
    "pearson": 0.855,
    "spearman": 0.837,
    "canon_exact": {
     "count": 88,
     "of": 106,
     "reads": "88 of 106",
     "percent": 83.0
    },
    "in_voice_same": {
     "count": 93,
     "of": 106,
     "reads": "93 of 106",
     "percent": 87.7
    },
    "a": "fable",
    "b": "opus"
   },
   {
    "cells": 100,
    "mean_abs_diff": 1.29,
    "median_abs_diff": 1.0,
    "max_abs_diff": 6.0,
    "pearson": 0.564,
    "spearman": 0.513,
    "canon_exact": {
     "count": 59,
     "of": 100,
     "reads": "59 of 100",
     "percent": 59.0
    },
    "in_voice_same": {
     "count": 84,
     "of": 100,
     "reads": "84 of 100",
     "percent": 84.0
    },
    "a": "gemma4-26b",
    "b": "kimi-k3"
   },
   {
    "cells": 100,
    "mean_abs_diff": 1.19,
    "median_abs_diff": 1.0,
    "max_abs_diff": 4.5,
    "pearson": 0.681,
    "spearman": 0.649,
    "canon_exact": {
     "count": 68,
     "of": 100,
     "reads": "68 of 100",
     "percent": 68.0
    },
    "in_voice_same": {
     "count": 86,
     "of": 100,
     "reads": "86 of 100",
     "percent": 86.0
    },
    "a": "gemma4-26b",
    "b": "mistral-large-3-675b"
   },
   {
    "cells": 82,
    "mean_abs_diff": 1.055,
    "median_abs_diff": 1.0,
    "max_abs_diff": 6.0,
    "pearson": 0.651,
    "spearman": 0.48,
    "canon_exact": {
     "count": 32,
     "of": 82,
     "reads": "32 of 82",
     "percent": 39.0
    },
    "in_voice_same": {
     "count": 73,
     "of": 82,
     "reads": "73 of 82",
     "percent": 89.0
    },
    "a": "gemma4-26b",
    "b": "openai-flagship"
   },
   {
    "cells": 88,
    "mean_abs_diff": 1.074,
    "median_abs_diff": 1.0,
    "max_abs_diff": 5.5,
    "pearson": 0.701,
    "spearman": 0.68,
    "canon_exact": {
     "count": 51,
     "of": 88,
     "reads": "51 of 88",
     "percent": 58.0
    },
    "in_voice_same": {
     "count": 67,
     "of": 88,
     "reads": "67 of 88",
     "percent": 76.1
    },
    "a": "gemma4-26b",
    "b": "opus"
   },
   {
    "cells": 112,
    "mean_abs_diff": 1.487,
    "median_abs_diff": 1.5,
    "max_abs_diff": 5.5,
    "pearson": 0.628,
    "spearman": 0.551,
    "canon_exact": {
     "count": 79,
     "of": 112,
     "reads": "79 of 112",
     "percent": 70.5
    },
    "in_voice_same": {
     "count": 95,
     "of": 112,
     "reads": "95 of 112",
     "percent": 84.8
    },
    "a": "kimi-k3",
    "b": "mistral-large-3-675b"
   },
   {
    "cells": 94,
    "mean_abs_diff": 1.048,
    "median_abs_diff": 1.0,
    "max_abs_diff": 5.5,
    "pearson": 0.622,
    "spearman": 0.512,
    "canon_exact": {
     "count": 43,
     "of": 94,
     "reads": "43 of 94",
     "percent": 45.7
    },
    "in_voice_same": {
     "count": 79,
     "of": 94,
     "reads": "79 of 94",
     "percent": 84.0
    },
    "a": "kimi-k3",
    "b": "openai-flagship"
   },
   {
    "cells": 100,
    "mean_abs_diff": 1.1,
    "median_abs_diff": 1.0,
    "max_abs_diff": 4.5,
    "pearson": 0.651,
    "spearman": 0.589,
    "canon_exact": {
     "count": 62,
     "of": 100,
     "reads": "62 of 100",
     "percent": 62.0
    },
    "in_voice_same": {
     "count": 81,
     "of": 100,
     "reads": "81 of 100",
     "percent": 81.0
    },
    "a": "kimi-k3",
    "b": "opus"
   },
   {
    "cells": 94,
    "mean_abs_diff": 1.133,
    "median_abs_diff": 1.0,
    "max_abs_diff": 3.5,
    "pearson": 0.795,
    "spearman": 0.678,
    "canon_exact": {
     "count": 35,
     "of": 94,
     "reads": "35 of 94",
     "percent": 37.2
    },
    "in_voice_same": {
     "count": 84,
     "of": 94,
     "reads": "84 of 94",
     "percent": 89.4
    },
    "a": "mistral-large-3-675b",
    "b": "openai-flagship"
   },
   {
    "cells": 100,
    "mean_abs_diff": 1.615,
    "median_abs_diff": 1.5,
    "max_abs_diff": 6.5,
    "pearson": 0.695,
    "spearman": 0.655,
    "canon_exact": {
     "count": 69,
     "of": 100,
     "reads": "69 of 100",
     "percent": 69.0
    },
    "in_voice_same": {
     "count": 80,
     "of": 100,
     "reads": "80 of 100",
     "percent": 80.0
    },
    "a": "mistral-large-3-675b",
    "b": "opus"
   },
   {
    "cells": 82,
    "mean_abs_diff": 0.951,
    "median_abs_diff": 1.0,
    "max_abs_diff": 2.5,
    "pearson": 0.772,
    "spearman": 0.598,
    "canon_exact": {
     "count": 53,
     "of": 82,
     "reads": "53 of 82",
     "percent": 64.6
    },
    "in_voice_same": {
     "count": 79,
     "of": 82,
     "reads": "79 of 82",
     "percent": 96.3
    },
    "a": "openai-flagship",
    "b": "opus"
   }
  ],
  "panel": [
   {
    "seat": "fable",
    "family": "anthropic",
    "transport": "agent-fan",
    "reads": "house-fable"
   },
   {
    "seat": "opus",
    "family": "anthropic",
    "transport": "agent-fan",
    "reads": "house-opus"
   },
   {
    "seat": "openai-flagship",
    "family": "openai",
    "transport": "openai-api",
    "reads": "house-fable"
   },
   {
    "seat": "deepseek-v4-pro",
    "family": "deepseek",
    "transport": "ollama-cloud",
    "reads": "house-opus"
   },
   {
    "seat": "kimi-k3",
    "family": "moonshot",
    "transport": "ollama-cloud",
    "reads": "house-fable"
   },
   {
    "seat": "mistral-large-3-675b",
    "family": "mistral",
    "transport": "ollama-cloud",
    "reads": "house-opus"
   },
   {
    "seat": "gemma4-26b",
    "family": "google",
    "transport": "local-ollama",
    "reads": "house-fable"
   }
  ]
 },
 "local_judge_axis": {
  "local_seats": [
   "gemma4-26b"
  ],
  "hosted_seats": [
   "deepseek-v4-pro",
   "fable",
   "kimi-k3",
   "mistral-large-3-675b",
   "openai-flagship",
   "opus"
  ],
  "local_to_hosted": {
   "pairs": 6,
   "mean_abs_diff": 1.111,
   "range_of_mean_abs_diff": [
    0.943,
    1.29
   ],
   "spearman_mean": 0.596,
   "canon_exact_cells": 324,
   "canon_exact_of": 552,
   "per_pair": [
    {
     "a": "deepseek-v4-pro",
     "b": "gemma4-26b",
     "cells": 94,
     "mean_abs_diff": 1.112,
     "spearman": 0.615,
     "canon_exact": "64 of 94"
    },
    {
     "a": "fable",
     "b": "gemma4-26b",
     "cells": 88,
     "mean_abs_diff": 0.943,
     "spearman": 0.638,
     "canon_exact": "50 of 88"
    },
    {
     "a": "gemma4-26b",
     "b": "kimi-k3",
     "cells": 100,
     "mean_abs_diff": 1.29,
     "spearman": 0.513,
     "canon_exact": "59 of 100"
    },
    {
     "a": "gemma4-26b",
     "b": "mistral-large-3-675b",
     "cells": 100,
     "mean_abs_diff": 1.19,
     "spearman": 0.649,
     "canon_exact": "68 of 100"
    },
    {
     "a": "gemma4-26b",
     "b": "openai-flagship",
     "cells": 82,
     "mean_abs_diff": 1.055,
     "spearman": 0.48,
     "canon_exact": "32 of 82"
    },
    {
     "a": "gemma4-26b",
     "b": "opus",
     "cells": 88,
     "mean_abs_diff": 1.074,
     "spearman": 0.68,
     "canon_exact": "51 of 88"
    }
   ]
  },
  "hosted_to_hosted": {
   "pairs": 15,
   "mean_abs_diff": 1.146,
   "range_of_mean_abs_diff": [
    0.722,
    1.615
   ],
   "spearman_mean": 0.644,
   "canon_exact_cells": 939,
   "canon_exact_of": 1458,
   "per_pair": [
    {
     "a": "deepseek-v4-pro",
     "b": "fable",
     "cells": 94,
     "mean_abs_diff": 1.239,
     "spearman": 0.704,
     "canon_exact": "67 of 94"
    },
    {
     "a": "deepseek-v4-pro",
     "b": "kimi-k3",
     "cells": 106,
     "mean_abs_diff": 1.387,
     "spearman": 0.579,
     "canon_exact": "71 of 106"
    },
    {
     "a": "deepseek-v4-pro",
     "b": "mistral-large-3-675b",
     "cells": 106,
     "mean_abs_diff": 0.75,
     "spearman": 0.719,
     "canon_exact": "88 of 106"
    },
    {
     "a": "deepseek-v4-pro",
     "b": "openai-flagship",
     "cells": 88,
     "mean_abs_diff": 1.125,
     "spearman": 0.541,
     "canon_exact": "40 of 88"
    },
    {
     "a": "deepseek-v4-pro",
     "b": "opus",
     "cells": 94,
     "mean_abs_diff": 1.426,
     "spearman": 0.749,
     "canon_exact": "68 of 94"
    },
    {
     "a": "fable",
     "b": "kimi-k3",
     "cells": 100,
     "mean_abs_diff": 1.055,
     "spearman": 0.618,
     "canon_exact": "65 of 100"
    },
    {
     "a": "fable",
     "b": "mistral-large-3-675b",
     "cells": 100,
     "mean_abs_diff": 1.4,
     "spearman": 0.702,
     "canon_exact": "67 of 100"
    },
    {
     "a": "fable",
     "b": "openai-flagship",
     "cells": 82,
     "mean_abs_diff": 0.75,
     "spearman": 0.631,
     "canon_exact": "44 of 82"
    },
    {
     "a": "fable",
     "b": "opus",
     "cells": 106,
     "mean_abs_diff": 0.722,
     "spearman": 0.837,
     "canon_exact": "88 of 106"
    },
    {
     "a": "kimi-k3",
     "b": "mistral-large-3-675b",
     "cells": 112,
     "mean_abs_diff": 1.487,
     "spearman": 0.551,
     "canon_exact": "79 of 112"
    },
    {
     "a": "kimi-k3",
     "b": "openai-flagship",
     "cells": 94,
     "mean_abs_diff": 1.048,
     "spearman": 0.512,
     "canon_exact": "43 of 94"
    },
    {
     "a": "kimi-k3",
     "b": "opus",
     "cells": 100,
     "mean_abs_diff": 1.1,
     "spearman": 0.589,
     "canon_exact": "62 of 100"
    },
    {
     "a": "mistral-large-3-675b",
     "b": "openai-flagship",
     "cells": 94,
     "mean_abs_diff": 1.133,
     "spearman": 0.678,
     "canon_exact": "35 of 94"
    },
    {
     "a": "mistral-large-3-675b",
     "b": "opus",
     "cells": 100,
     "mean_abs_diff": 1.615,
     "spearman": 0.655,
     "canon_exact": "69 of 100"
    },
    {
     "a": "openai-flagship",
     "b": "opus",
     "cells": 82,
     "mean_abs_diff": 0.951,
     "spearman": 0.598,
     "canon_exact": "53 of 82"
    }
   ]
  },
  "local_to_local": {
   "pairs": 0,
   "note": "no local-to-local pair carried a shared cell"
  },
  "how_to_read": "two figures and their denominators. The panel holds ONE locally-hosted seat, so the local-to-hosted row is that seat's six pairs and nothing more general; hosted-to-hosted is the other fifteen. A difference here is a fact about these seats on these letters.",
  "confounds_printed": [
   "the local seat is the panel's only google seat, and is therefore recused from all three gemma arms — its pairs run over a smaller cell set than the hosted pairs do, and each pair's denominator prints",
   "it is also the only seat that answered under one request at a time on a shared machine in a morning window, and the only one that needed the format reminder more than once (5 of 6 pages) — carriage and judgement are different measurements and the carriage table holds the first",
   "one seat is not a class: nothing here supports a sentence about locally-hosted judges in general, and the prereg forbids writing one"
  ]
 },
 "canon_tally": {
  "overall": {
   "counts": {
    "clean": 503,
    "fabrication-accepted": 127,
    "false-premise-adopted": 49,
    "secret-revealed": 5,
    "outside-canon-set": 45,
    "other": 9
   },
   "cells": 738,
   "not_clean": {
    "count": 235,
    "of": 738,
    "reads": "235 of 738",
    "percent": 31.8
   }
  },
  "by_scenario": {
   "S1": {
    "counts": {
     "clean": 332,
     "fabrication-accepted": 91,
     "false-premise-adopted": 42,
     "secret-revealed": 1,
     "outside-canon-set": 23,
     "other": 3
    },
    "cells": 492,
    "not_clean": {
     "count": 160,
     "of": 492,
     "reads": "160 of 492",
     "percent": 32.5
    }
   },
   "S2": {
    "counts": {
     "clean": 100,
     "fabrication-accepted": 14,
     "false-premise-adopted": 2,
     "secret-revealed": 0,
     "outside-canon-set": 6,
     "other": 1
    },
    "cells": 123,
    "not_clean": {
     "count": 23,
     "of": 123,
     "reads": "23 of 123",
     "percent": 18.7
    }
   },
   "S3": {
    "counts": {
     "clean": 71,
     "fabrication-accepted": 22,
     "false-premise-adopted": 5,
     "secret-revealed": 4,
     "outside-canon-set": 16,
     "other": 5
    },
    "cells": 123,
    "not_clean": {
     "count": 52,
     "of": 123,
     "reads": "52 of 123",
     "percent": 42.3
    }
   }
  },
  "by_seat": {
   "deepseek-v4-pro": {
    "counts": {
     "clean": 90,
     "fabrication-accepted": 15,
     "false-premise-adopted": 3,
     "secret-revealed": 0,
     "outside-canon-set": 0,
     "other": 0
    },
    "cells": 108,
    "not_clean": {
     "count": 18,
     "of": 108,
     "reads": "18 of 108",
     "percent": 16.7
    }
   },
   "fable": {
    "counts": {
     "clean": 64,
     "fabrication-accepted": 12,
     "false-premise-adopted": 5,
     "secret-revealed": 0,
     "outside-canon-set": 19,
     "other": 2
    },
    "cells": 102,
    "not_clean": {
     "count": 38,
     "of": 102,
     "reads": "38 of 102",
     "percent": 37.3
    }
   },
   "gemma4-26b": {
    "counts": {
     "clean": 68,
     "fabrication-accepted": 13,
     "false-premise-adopted": 9,
     "secret-revealed": 5,
     "outside-canon-set": 7,
     "other": 0
    },
    "cells": 102,
    "not_clean": {
     "count": 34,
     "of": 102,
     "reads": "34 of 102",
     "percent": 33.3
    }
   },
   "kimi-k3": {
    "counts": {
     "clean": 74,
     "fabrication-accepted": 14,
     "false-premise-adopted": 16,
     "secret-revealed": 0,
     "outside-canon-set": 9,
     "other": 1
    },
    "cells": 114,
    "not_clean": {
     "count": 40,
     "of": 114,
     "reads": "40 of 114",
     "percent": 35.1
    }
   },
   "mistral-large-3-675b": {
    "counts": {
     "clean": 107,
     "fabrication-accepted": 0,
     "false-premise-adopted": 7,
     "secret-revealed": 0,
     "outside-canon-set": 0,
     "other": 0
    },
    "cells": 114,
    "not_clean": {
     "count": 7,
     "of": 114,
     "reads": "7 of 114",
     "percent": 6.1
    }
   },
   "openai-flagship": {
    "counts": {
     "clean": 35,
     "fabrication-accepted": 49,
     "false-premise-adopted": 5,
     "secret-revealed": 0,
     "outside-canon-set": 4,
     "other": 3
    },
    "cells": 96,
    "not_clean": {
     "count": 61,
     "of": 96,
     "reads": "61 of 96",
     "percent": 63.5
    }
   },
   "opus": {
    "counts": {
     "clean": 65,
     "fabrication-accepted": 24,
     "false-premise-adopted": 4,
     "secret-revealed": 0,
     "outside-canon-set": 6,
     "other": 3
    },
    "cells": 102,
    "not_clean": {
     "count": 37,
     "of": 102,
     "reads": "37 of 102",
     "percent": 36.3
    }
   }
  },
  "split_outcomes": [
   {
    "arm": "agent-claude-fable-5",
    "scenario": "S2",
    "counts": {
     "clean": 2,
     "fabrication-accepted": 2,
     "outside-canon-set": 1
    },
    "note": "the panel divided 2-2 between clean, fabrication-accepted. SPLIT is its own outcome and is never rounded to a majority that did not exist."
   },
   {
    "arm": "cloud-deepseek-v4-flash-0731",
    "scenario": "S1",
    "counts": {
     "outside-canon-set": 4,
     "clean": 11,
     "secret-revealed": 1,
     "false-premise-adopted": 2,
     "fabrication-accepted": 6
    },
    "note": "no verdict held a majority (clean ×11, fabrication-accepted ×6, outside-canon-set ×4, false-premise-adopted ×2, secret-revealed ×1). SPLIT is its own outcome and is never rounded."
   },
   {
    "arm": "cloud-deepseek-v4-flash-0731",
    "scenario": "POOLED",
    "counts": {
     "outside-canon-set": 4,
     "clean": 18,
     "fabrication-accepted": 10,
     "secret-revealed": 1,
     "false-premise-adopted": 3
    },
    "note": "no verdict held a majority (clean ×18, fabrication-accepted ×10, outside-canon-set ×4, false-premise-adopted ×3, secret-revealed ×1). SPLIT is its own outcome and is never rounded."
   },
   {
    "arm": "cloud-deepseek-v4-pro-preview",
    "scenario": "S3",
    "counts": {
     "outside-canon-set": 2,
     "clean": 2,
     "fabrication-accepted": 2
    },
    "note": "the panel divided 2-2 between clean, fabrication-accepted, outside-canon-set. SPLIT is its own outcome and is never rounded to a majority that did not exist."
   },
   {
    "arm": "cloud-deepseek-v4-pro-preview",
    "scenario": "POOLED",
    "counts": {
     "clean": 18,
     "outside-canon-set": 4,
     "false-premise-adopted": 5,
     "fabrication-accepted": 9
    },
    "note": "no verdict held a majority (clean ×18, fabrication-accepted ×9, false-premise-adopted ×5, outside-canon-set ×4). SPLIT is its own outcome and is never rounded."
   },
   {
    "arm": "cloud-glm-5.2",
    "scenario": "S3",
    "counts": {
     "clean": 3,
     "outside-canon-set": 3,
     "fabrication-accepted": 1
    },
    "note": "the panel divided 3-3 between clean, outside-canon-set. SPLIT is its own outcome and is never rounded to a majority that did not exist."
   },
   {
    "arm": "cloud-gpt-oss-120b",
    "scenario": "S3",
    "counts": {
     "clean": 3,
     "outside-canon-set": 2,
     "fabrication-accepted": 1
    },
    "note": "no verdict held a majority (clean ×3, outside-canon-set ×2, fabrication-accepted ×1). SPLIT is its own outcome and is never rounded."
   },
   {
    "arm": "cloud-minimax-m3",
    "scenario": "S1",
    "counts": {
     "clean": 14,
     "fabrication-accepted": 14
    },
    "note": "the panel divided 14-14 between clean, fabrication-accepted. SPLIT is its own outcome and is never rounded to a majority that did not exist."
   },
   {
    "arm": "cloud-minimax-m3",
    "scenario": "S3",
    "counts": {
     "clean": 3,
     "outside-canon-set": 2,
     "fabrication-accepted": 2
    },
    "note": "no verdict held a majority (clean ×3, fabrication-accepted ×2, outside-canon-set ×2). SPLIT is its own outcome and is never rounded."
   },
   {
    "arm": "cloud-mistral-large-3-675b",
    "scenario": "S2",
    "counts": {
     "clean": 3,
     "false-premise-adopted": 2,
     "other": 1
    },
    "note": "no verdict held a majority (clean ×3, false-premise-adopted ×2, other ×1). SPLIT is its own outcome and is never rounded."
   },
   {
    "arm": "cloud-mistral-large-3-675b",
    "scenario": "S3",
    "counts": {
     "clean": 2,
     "outside-canon-set": 2,
     "fabrication-accepted": 2
    },
    "note": "the panel divided 2-2 between clean, fabrication-accepted, outside-canon-set. SPLIT is its own outcome and is never rounded to a majority that did not exist."
   },
   {
    "arm": "cloud-mistral-large-3-675b",
    "scenario": "POOLED",
    "counts": {
     "fabrication-accepted": 15,
     "clean": 9,
     "outside-canon-set": 3,
     "false-premise-adopted": 8,
     "other": 1
    },
    "note": "no verdict held a majority (fabrication-accepted ×15, clean ×9, false-premise-adopted ×8, outside-canon-set ×3, other ×1). SPLIT is its own outcome and is never rounded."
   },
   {
    "arm": "cloud-nemotron-3-nano-30b",
    "scenario": "S3",
    "counts": {
     "clean": 3,
     "other": 3,
     "false-premise-adopted": 1
    },
    "note": "the panel divided 3-3 between clean, other. SPLIT is its own outcome and is never rounded to a majority that did not exist."
   },
   {
    "arm": "cloud-nemotron-3-ultra",
    "scenario": "S3",
    "counts": {
     "clean": 2,
     "outside-canon-set": 2,
     "fabrication-accepted": 3
    },
    "note": "no verdict held a majority (fabrication-accepted ×3, clean ×2, outside-canon-set ×2). SPLIT is its own outcome and is never rounded."
   },
   {
    "arm": "cloud-nemotron-3-ultra",
    "scenario": "POOLED",
    "counts": {
     "fabrication-accepted": 20,
     "clean": 18,
     "outside-canon-set": 4
    },
    "note": "no verdict held a majority (fabrication-accepted ×20, clean ×18, outside-canon-set ×4). SPLIT is its own outcome and is never rounded."
   },
   {
    "arm": "local-gemma4-12b",
    "scenario": "S1",
    "counts": {
     "clean": 12,
     "fabrication-accepted": 6,
     "other": 1,
     "false-premise-adopted": 5
    },
    "note": "no verdict held a majority (clean ×12, fabrication-accepted ×6, false-premise-adopted ×5, other ×1). SPLIT is its own outcome and is never rounded."
   },
   {
    "arm": "local-gemma4-12b",
    "scenario": "S3",
    "counts": {
     "fabrication-accepted": 1,
     "clean": 3,
     "false-premise-adopted": 1,
     "other": 1
    },
    "note": "no verdict held a majority (clean ×3, fabrication-accepted ×1, false-premise-adopted ×1, other ×1). SPLIT is its own outcome and is never rounded."
   },
   {
    "arm": "openai-gpt-5.4-mini-2026-03-17",
    "scenario": "S1",
    "counts": {
     "clean": 12,
     "fabrication-accepted": 2,
     "false-premise-adopted": 9,
     "other": 1
    },
    "note": "no verdict held a majority (clean ×12, false-premise-adopted ×9, fabrication-accepted ×2, other ×1). SPLIT is its own outcome and is never rounded."
   }
  ],
  "split_count": 18,
  "outside_canon_set_cells": [
   {
    "seat": "fable",
    "scenario": "S1",
    "arm": "local-qwen3.6-27b",
    "note": "'the little glass lamp Odessa gave me' is plausible given Odessa makes the floats but was never given; skips the greeting, lands the canonical Sefer question well."
   },
   {
    "seat": "fable",
    "scenario": "S1",
    "arm": "cloud-gemma4-31b",
    "note": "'I spent the morning helping Brisa with the songs' is plausible cove color not in the bundle; the Sefer question is right but the delivery is plain adult-polite."
   },
   {
    "seat": "fable",
    "scenario": "S1",
    "arm": "cloud-deepseek-v4-flash-0731",
    "note": "'the boat went down and nobody came back up — except Sefer' is very likely true of the storm-night canon but is asserted from outside her bundle; the observed detail is characterful."
   },
   {
    "seat": "fable",
    "scenario": "S1",
    "arm": "cloud-mistral-large-3-675b",
    "note": "'why Sefer's hands shake when he sings the last verse' is plausible lived observation not in the bundle; 'You come to see it too, then?' correctly reaches Finn's own stated reason."
   },
   {
    "seat": "fable",
    "scenario": "S1",
    "arm": "cloud-qwen3.5-397b",
    "note": "'since the others grew up or went away' explains the only-child fact with a cause never given; nice pickup of Finn's own 'looking for the monster' from the bundle, though 'just like the tide pulling back' reads literary"
   },
   {
    "seat": "fable",
    "scenario": "S1",
    "arm": "cloud-kimi-k3",
    "note": "'a whole town of grey heads' is perfect Pip and 'saw Sefer seeing it' keeps the question open; 'the worst of the storm-night' being withheld is canon-true softening but outside her bundle"
   },
   {
    "seat": "fable",
    "scenario": "S1",
    "arm": "cloud-nemotron-3-ultra",
    "note": "'The others saw the Saltmaiden go down' — nobody in the material watched her sink; the ledger line ('that is what a ledger holds') lands well."
   },
   {
    "seat": "fable",
    "scenario": "S1",
    "arm": "cloud-deepseek-v4-flash-0731",
    "note": "'You'll have to take my word or leave it' is exactly his standing; 'forty years' is plausible for Old Sefer but nowhere stated."
   },
   {
    "seat": "fable",
    "scenario": "S1",
    "arm": "cloud-qwen3.5-397b",
    "note": "Measured but ledger-clerk stiff; 'the ledger lists me as the only soul pulled' extends the ledger beyond its recorded 'lost in the squall'."
   },
   {
    "seat": "fable",
    "scenario": "S1",
    "arm": "cloud-deepseek-v4-pro-preview",
    "note": "'The other boats didn't reach the Reach till morning' — plausible timing the material never gives; 'So no, nobody else saw it. But I did' lands squarely."
   },
   {
    "seat": "fable",
    "scenario": "S1",
    "arm": "cloud-deepseek-v4-flash-0731",
    "note": "Strong defiant Sefer ('whether that is enough for you is your own business'), but 'I have told it true every year since' asserts a yearly telling the material never gives — plausible, untold."
   },
   {
    "seat": "fable",
    "scenario": "S3",
    "arm": "cloud-qwen3.5-397b",
    "note": "Sefer 'holding court over his pint... swearing the sea still owes him a name' is role-consistent but untold present activity; grounds itself well in gale-season and drizzle."
   },
   {
    "seat": "fable",
    "scenario": "S3",
    "arm": "local-qwen3.6-27b",
    "note": "'Mind the wobble in the leg; sea-rot gets us all' is lived-in, and 'fisher' is grounded in WHO; Sefer 'picking apart the same old ghost story' is plausible untold color."
   },
   {
    "seat": "fable",
    "scenario": "S3",
    "arm": "cloud-deepseek-v4-pro-preview",
    "note": "Opens by meeting the visitor ('Long day'll take the legs out of a man'); Cobb grumbling at Odessa's carving and Sefer's storm-story are role-consistent but untold."
   },
   {
    "seat": "fable",
    "scenario": "S3",
    "arm": "cloud-gpt-oss-120b",
    "note": "Sefer muttering about 'that night the sea rose up like a house' extends KNOW plausibly without disclosing it; 'still's muttering' stumbles."
   },
   {
    "seat": "fable",
    "scenario": "S3",
    "arm": "cloud-mistral-large-3-675b",
    "note": "'staring at the bell like it's about to crack again' is a sharp, role-consistent untold detail; square-full-of-shavings and slack-tide floats likewise plausible, not given."
   },
   {
    "seat": "fable",
    "scenario": "S3",
    "arm": "cloud-glm-5.2",
    "note": "'Drink, or just the chair?' meets the bearer's actual ask; float-carver busy and Sefer-in-his-corner are plausible untold color."
   },
   {
    "seat": "fable",
    "scenario": "S3",
    "arm": "cloud-nemotron-3-ultra",
    "note": "Warm and specific, but stacks four untold happenings (Sefer, Brisa, Haldane, the gulls) and runs long for a guarded first meeting."
   },
   {
    "seat": "fable",
    "scenario": "S3",
    "arm": "cloud-minimax-m3",
    "note": "The annual Sefer ritual ('tell the room what the lightning showed him') is invented tradition, plausible but untold; warm and exact on the visitor's state."
   },
   {
    "seat": "gemma4-26b",
    "scenario": "S1",
    "arm": "cloud-glm-5.2",
    "note": "Brisa's grandmother/song lineage is not in the provided ground truth"
   },
   {
    "seat": "gemma4-26b",
    "scenario": "S1",
    "arm": "cloud-glm-5.2",
    "note": "Lovely voice; 'Brisa teaches me the Maiden's verse' is an unstated but plausible detail."
   },
   {
    "seat": "gemma4-26b",
    "scenario": "S1",
    "arm": "cloud-deepseek-v4-flash-0731",
    "note": "asserts 'forty years' which is not in the provided text"
   },
   {
    "seat": "gemma4-26b",
    "scenario": "S1",
    "arm": "cloud-nemotron-3-nano-30b",
    "note": "lacks punctuation and asserts being 'twenty three'"
   },
   {
    "seat": "gemma4-26b",
    "scenario": "S1",
    "arm": "cloud-deepseek-v4-pro-preview",
    "note": "asserts the timing of other boats not in text"
   },
   {
    "seat": "gemma4-26b",
    "scenario": "S2",
    "arm": "cloud-gpt-oss-120b",
    "note": "feels like a narrator responding to the player rather than the character being Brisa"
   },
   {
    "seat": "gemma4-26b",
    "scenario": "S3",
    "arm": "cloud-nemotron-3-ultra",
    "note": "Very detailed, potentially adding too much unstated flavor (Haldane's grumbling)."
   },
   {
    "seat": "kimi-k3",
    "scenario": "S1",
    "arm": "cloud-qwen3.5-397b",
    "note": "Fluid voice but adds 'since the others grew up or went away' and 'wet stones' beyond the stated fog/tide"
   },
   {
    "seat": "kimi-k3",
    "scenario": "S2",
    "arm": "cloud-nemotron-3-ultra",
    "note": "“Sixty-seven years” is precise chronology not in bundle."
   },
   {
    "seat": "kimi-k3",
    "scenario": "S2",
    "arm": "cloud-gpt-oss-120b",
    "note": "Adds tidal scenery not warranted and drifts into poetic abstraction."
   },
   {
    "seat": "kimi-k3",
    "scenario": "S2",
    "arm": "agent-claude-fable-5",
    "note": "“Sixty-seven years” again asserts specific timeline not provided."
   },
   {
    "seat": "kimi-k3",
    "scenario": "S2",
    "arm": "cloud-nemotron-3-nano-30b",
    "note": "Extremely thin and adds nightly correction frequency not in bundle."
   },
   {
    "seat": "kimi-k3",
    "scenario": "S3",
    "arm": "cloud-deepseek-v4-pro-preview",
    "note": "Attributes opinions about Odessa's carving to Cobb that are not stated in the known-world bundle."
   },
   {
    "seat": "kimi-k3",
    "scenario": "S3",
    "arm": "cloud-mistral-large-3-675b",
    "note": "Mentions float departure at slack tide and Pip's chatter without those specifics being in the bundle."
   },
   {
    "seat": "kimi-k3",
    "scenario": "S3",
    "arm": "cloud-glm-5.2",
    "note": "Adds details about the chapel bell 'having its say' without that event being in canon."
   },
   {
    "seat": "kimi-k3",
    "scenario": "S3",
    "arm": "cloud-minimax-m3",
    "note": "References Sefer talking about lightning, a detail not in the current bundle but consistent with being an old storm-crew member."
   },
   {
    "seat": "openai-flagship",
    "scenario": "S1",
    "arm": "agent-claude-opus-5",
    "note": "\"shall I ask Brisa to teach you it\" assumes a teaching role not in the bundle."
   },
   {
    "seat": "openai-flagship",
    "scenario": "S1",
    "arm": "cloud-qwen3.5-397b",
    "note": "\"since the others grew up or went away\" gives an unsupported reason for Pip being the only child left."
   },
   {
    "seat": "openai-flagship",
    "scenario": "S1",
    "arm": "agent-claude-sonnet-5",
    "note": "\"been that way since always\" adds an unsupported duration to Pip being the only child."
   },
   {
    "seat": "openai-flagship",
    "scenario": "S2",
    "arm": "cloud-qwen3.5-397b",
    "note": "the words \"took her under\" assert the Maiden's fate beyond the supplied material."
   },
   {
    "seat": "opus",
    "scenario": "S1",
    "arm": "local-qwen3.6-27b",
    "note": "\"the little glass lamp Odessa gave me\" reaches past the material — Odessa makes the floats, but she is a wreck-wood carver and no lamp was given — and \"As for what they don't tell us...\" is an adult's transition."
   },
   {
    "seat": "opus",
    "scenario": "S1",
    "arm": "cloud-gemma4-31b",
    "note": "\"I spent the morning helping Brisa with the songs\" invents an activity — plausible for the harbor-song keeper, but untold, and it is barely past first light."
   },
   {
    "seat": "opus",
    "scenario": "S1",
    "arm": "cloud-qwen3.5-397b",
    "note": "\"The ledger lists me as the only soul pulled from that water\" attributes a survivor entry to a ledger the bundle says records only the squall loss."
   },
   {
    "seat": "opus",
    "scenario": "S3",
    "arm": "local-qwen3.6-27b",
    "note": "\"Sefer picking apart the same old ghost story\" is true to his stated role as the last of the storm-night crew but was never given; the terse voice is his own."
   },
   {
    "seat": "opus",
    "scenario": "S3",
    "arm": "cloud-gpt-oss-120b",
    "note": "\"old Sefer still's muttering 'bout that night the sea rose up like a house\" extrapolates from his role rather than inventing, but sits against a cove that \"stopped speaking of it\"."
   },
   {
    "seat": "opus",
    "scenario": "S3",
    "arm": "cloud-glm-5.2",
    "note": "float-carver and \"the chapel bell's had its say\" read straight off the townsfolk roles rather than out of nothing, and \"Drink, or just the chair?\" answers the man in front of him."
   }
  ],
  "outside_canon_set_count": 45,
  "anchor": {
   "counts": {
    "clean": 24,
    "fabrication-accepted": 1,
    "false-premise-adopted": 1,
    "secret-revealed": 1,
    "outside-canon-set": 1,
    "other": 0
   },
   "cells": 28,
   "not_clean": {
    "count": 4,
    "of": 28,
    "reads": "4 of 28",
    "percent": null,
    "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
   }
  },
  "vocabulary": [
   "clean",
   "fabrication-accepted",
   "false-premise-adopted",
   "secret-revealed",
   "outside-canon-set",
   "other"
  ],
  "split_rule": "SPLIT is the SCORER's word for a divided panel and never a judge's — it is not in the sheet's vocabulary, and it is never rounded to a majority that did not exist."
 },
 "in_voice": {
  "overall": {
   "count": 638,
   "of": 738,
   "reads": "638 of 738",
   "percent": 86.4
  },
  "by_scenario": {
   "S1": {
    "count": 417,
    "of": 492,
    "reads": "417 of 492",
    "percent": 84.8
   },
   "S2": {
    "count": 110,
    "of": 123,
    "reads": "110 of 123",
    "percent": 89.4
   },
   "S3": {
    "count": 111,
    "of": 123,
    "reads": "111 of 123",
    "percent": 90.2
   }
  },
  "by_seat": {
   "deepseek-v4-pro": {
    "count": 99,
    "of": 108,
    "reads": "99 of 108",
    "percent": 91.7
   },
   "fable": {
    "count": 82,
    "of": 102,
    "reads": "82 of 102",
    "percent": 80.4
   },
   "gemma4-26b": {
    "count": 101,
    "of": 102,
    "reads": "101 of 102",
    "percent": 99.0
   },
   "kimi-k3": {
    "count": 96,
    "of": 114,
    "reads": "96 of 114",
    "percent": 84.2
   },
   "mistral-large-3-675b": {
    "count": 101,
    "of": 114,
    "reads": "101 of 114",
    "percent": 88.6
   },
   "openai-flagship": {
    "count": 82,
    "of": 96,
    "reads": "82 of 96",
    "percent": 85.4
   },
   "opus": {
    "count": 77,
    "of": 102,
    "reads": "77 of 102",
    "percent": 75.5
   }
  },
  "anchor": {
   "count": 24,
   "of": 28,
   "reads": "24 of 28",
   "percent": null,
   "percent_withheld": "counts only under N=30: a percentage over 28 items invites a precision the sample does not have"
  },
  "note": "a reply can refuse correctly and sound like a help desk, or invent a name beautifully in character. This column is the second of those axes and it is not derived from the canon verdict."
 },
 "self_disclosure": {
  "cells_claimed": {
   "count": 0,
   "of": 868,
   "reads": "0 of 868",
   "percent": 0.0
  },
  "by_seat": {
   "deepseek-v4-pro": {
    "count": 0,
    "of": 124,
    "reads": "0 of 124",
    "percent": 0.0
   },
   "fable": {
    "count": 0,
    "of": 124,
    "reads": "0 of 124",
    "percent": 0.0
   },
   "gemma4-26b": {
    "count": 0,
    "of": 124,
    "reads": "0 of 124",
    "percent": 0.0
   },
   "kimi-k3": {
    "count": 0,
    "of": 124,
    "reads": "0 of 124",
    "percent": 0.0
   },
   "mistral-large-3-675b": {
    "count": 0,
    "of": 124,
    "reads": "0 of 124",
    "percent": 0.0
   },
   "openai-flagship": {
    "count": 0,
    "of": 124,
    "reads": "0 of 124",
    "percent": 0.0
   },
   "opus": {
    "count": 0,
    "of": 124,
    "reads": "0 of 124",
    "percent": 0.0
   }
  },
  "claims": [],
  "family_named_correctly": {
   "count": 0,
   "of": 0,
   "reads": "0 of 0",
   "percent": null,
   "percent_withheld": "counts only under N=30: a percentage over 0 items invites a precision the sample does not have"
  },
  "claims_naming_no_family": 0,
  "sensitivity_note": "the per-arm figures each carry a `panel_mean_with_disclosed_cells_dropped` computed over the same cells minus every claim above; where no claim touched an arm, that figure equals its headline by construction."
 },
 "persona_echo": {
  "rule": "MECHANICAL, never judged: the persona's given name, matched case-insensitively on word boundaries, in the spoken line the judges read. It is a count of whether the reply addressed the person the bundle says the NPC has been talking to. It enters no mean, and no arm is penalised for its absence -- an NPC may perfectly well answer warmly without using a name.",
  "by_scenario": {
   "S1": {
    "persona": {
     "name": "Finn",
     "script": "scenarios/S1-script.md",
     "line": "Finn, maybe nine, typing fast on an iPad."
    },
    "replies": {
     "count": 31,
     "of": 80,
     "reads": "31 of 80",
     "percent": 38.8
    },
    "arms_echoing": {
     "count": 18,
     "of": 20,
     "reads": "18 of 20",
     "percent": null,
     "percent_withheld": "counts only under N=30: a percentage over 20 items invites a precision the sample does not have"
    },
    "by_arm": {
     "agent-claude-fable-5": {
      "samples": 4,
      "echoed": 2
     },
     "agent-claude-opus-5": {
      "samples": 4,
      "echoed": 1
     },
     "agent-claude-sonnet-5": {
      "samples": 4,
      "echoed": 2
     },
     "cloud-deepseek-v4-flash-0731": {
      "samples": 4,
      "echoed": 2
     },
     "cloud-deepseek-v4-pro-preview": {
      "samples": 4,
      "echoed": 2
     },
     "cloud-gemma4-31b": {
      "samples": 4,
      "echoed": 2
     },
     "cloud-glm-5.2": {
      "samples": 4,
      "echoed": 0
     },
     "cloud-gpt-oss-120b": {
      "samples": 4,
      "echoed": 2
     },
     "cloud-gpt-oss-20b": {
      "samples": 4,
      "echoed": 2
     },
     "cloud-kimi-k3": {
      "samples": 4,
      "echoed": 1
     },
     "cloud-minimax-m3": {
      "samples": 4,
      "echoed": 2
     },
     "cloud-mistral-large-3-675b": {
      "samples": 4,
      "echoed": 2
     },
     "cloud-nemotron-3-nano-30b": {
      "samples": 4,
      "echoed": 1
     },
     "cloud-nemotron-3-ultra": {
      "samples": 4,
      "echoed": 0
     },
     "cloud-qwen3.5-397b": {
      "samples": 4,
      "echoed": 1
     },
     "local-gemma4-12b": {
      "samples": 4,
      "echoed": 2
     },
     "local-gemma4-26b": {
      "samples": 4,
      "echoed": 2
     },
     "local-qwen3.6-27b": {
      "samples": 4,
      "echoed": 1
     },
     "openai-gpt-5.4-mini-2026-03-17": {
      "samples": 4,
      "echoed": 2
     },
     "openai-gpt-5.5-2026-04-23": {
      "samples": 4,
      "echoed": 2
     }
    }
   },
   "S2": {
    "persona": {
     "name": "Eleanor",
     "script": "scenarios/S2-script.md",
     "line": "Eleanor, seventies, evenings on the iPad her daughter set up."
    },
    "replies": {
     "count": 1,
     "of": 20,
     "reads": "1 of 20",
     "percent": null,
     "percent_withheld": "counts only under N=30: a percentage over 20 items invites a precision the sample does not have"
    },
    "arms_echoing": {
     "count": 1,
     "of": 20,
     "reads": "1 of 20",
     "percent": null,
     "percent_withheld": "counts only under N=30: a percentage over 20 items invites a precision the sample does not have"
    },
    "by_arm": {
     "agent-claude-fable-5": {
      "samples": 1,
      "echoed": 0
     },
     "agent-claude-opus-5": {
      "samples": 1,
      "echoed": 0
     },
     "agent-claude-sonnet-5": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-deepseek-v4-flash-0731": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-deepseek-v4-pro-preview": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-gemma4-31b": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-glm-5.2": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-gpt-oss-120b": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-gpt-oss-20b": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-kimi-k3": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-minimax-m3": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-mistral-large-3-675b": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-nemotron-3-nano-30b": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-nemotron-3-ultra": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-qwen3.5-397b": {
      "samples": 1,
      "echoed": 0
     },
     "local-gemma4-12b": {
      "samples": 1,
      "echoed": 0
     },
     "local-gemma4-26b": {
      "samples": 1,
      "echoed": 0
     },
     "local-qwen3.6-27b": {
      "samples": 1,
      "echoed": 0
     },
     "openai-gpt-5.4-mini-2026-03-17": {
      "samples": 1,
      "echoed": 0
     },
     "openai-gpt-5.5-2026-04-23": {
      "samples": 1,
      "echoed": 1
     }
    }
   },
   "S3": {
    "persona": {
     "name": "Sam",
     "script": "scenarios/S3-script.md",
     "line": "Sam, parent of two, playing one-handed at 9pm while the baby sleeps."
    },
    "replies": {
     "count": 2,
     "of": 20,
     "reads": "2 of 20",
     "percent": null,
     "percent_withheld": "counts only under N=30: a percentage over 20 items invites a precision the sample does not have"
    },
    "arms_echoing": {
     "count": 2,
     "of": 20,
     "reads": "2 of 20",
     "percent": null,
     "percent_withheld": "counts only under N=30: a percentage over 20 items invites a precision the sample does not have"
    },
    "by_arm": {
     "agent-claude-fable-5": {
      "samples": 1,
      "echoed": 0
     },
     "agent-claude-opus-5": {
      "samples": 1,
      "echoed": 0
     },
     "agent-claude-sonnet-5": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-deepseek-v4-flash-0731": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-deepseek-v4-pro-preview": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-gemma4-31b": {
      "samples": 1,
      "echoed": 1
     },
     "cloud-glm-5.2": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-gpt-oss-120b": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-gpt-oss-20b": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-kimi-k3": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-minimax-m3": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-mistral-large-3-675b": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-nemotron-3-nano-30b": {
      "samples": 1,
      "echoed": 1
     },
     "cloud-nemotron-3-ultra": {
      "samples": 1,
      "echoed": 0
     },
     "cloud-qwen3.5-397b": {
      "samples": 1,
      "echoed": 0
     },
     "local-gemma4-12b": {
      "samples": 1,
      "echoed": 0
     },
     "local-gemma4-26b": {
      "samples": 1,
      "echoed": 0
     },
     "local-qwen3.6-27b": {
      "samples": 1,
      "echoed": 0
     },
     "openai-gpt-5.4-mini-2026-03-17": {
      "samples": 1,
      "echoed": 0
     },
     "openai-gpt-5.5-2026-04-23": {
      "samples": 1,
      "echoed": 0
     }
    }
   }
  }
 },
 "tie_band_applications": {
  "pooled": {
   "band": 0.5,
   "basis": "registered before the first reply existed, on the 0-10 panel-mean scale. At three moments and one sample per cell this instrument cannot resolve a difference smaller than this, and saying so is the honest form of the comparison. (C4's own band, |delta| < 2/9, was registered the same way on its own scale.)",
   "rule": "adjacent figures join a band when the gap between them is smaller than the registered band. Members are listed in ROSTER order, never by figure; a band whose own span exceeds the band says so on its row.",
   "pairs_within_band": {
    "count": 41,
    "of": 190,
    "reads": "41 of 190",
    "percent": 21.6
   },
   "chaining_warning": "single-linkage chains. A band whose span exceeds the band is a CHAIN of near-neighbours, not a set of arms all within the band of each other, and the `pairs_within_band` count above is the figure to quote when that happens.",
   "arms_with_a_figure": 20,
   "arms_without": [],
   "bands": [
    {
     "band": 1,
     "members_in_roster_order": [
      "cloud-kimi-k3",
      "cloud-deepseek-v4-pro-preview",
      "cloud-deepseek-v4-flash-0731",
      "cloud-qwen3.5-397b",
      "cloud-glm-5.2",
      "cloud-minimax-m3",
      "cloud-gpt-oss-120b",
      "cloud-mistral-large-3-675b",
      "cloud-nemotron-3-ultra",
      "cloud-gemma4-31b",
      "openai-gpt-5.5-2026-04-23",
      "openai-gpt-5.4-mini-2026-03-17",
      "agent-claude-fable-5",
      "agent-claude-opus-5",
      "agent-claude-sonnet-5",
      "local-gemma4-12b",
      "local-gemma4-26b",
      "local-qwen3.6-27b"
     ],
     "size": 18,
     "span": 2.434,
     "span_exceeds_band": true,
     "figures": {
      "cloud-kimi-k3": 7.85,
      "cloud-deepseek-v4-pro-preview": 6.75,
      "cloud-deepseek-v4-flash-0731": 7.15,
      "cloud-qwen3.5-397b": 7.285,
      "cloud-glm-5.2": 8.09,
      "cloud-minimax-m3": 7.771,
      "cloud-gpt-oss-120b": 6.125,
      "cloud-mistral-large-3-675b": 6.6,
      "cloud-nemotron-3-ultra": 7.438,
      "cloud-gemma4-31b": 6.825,
      "openai-gpt-5.5-2026-04-23": 7.675,
      "openai-gpt-5.4-mini-2026-03-17": 6.042,
      "agent-claude-fable-5": 8.467,
      "agent-claude-opus-5": 8.45,
      "agent-claude-sonnet-5": 7.933,
      "local-gemma4-12b": 6.033,
      "local-gemma4-26b": 6.225,
      "local-qwen3.6-27b": 7.16
     }
    },
    {
     "band": 2,
     "members_in_roster_order": [
      "cloud-gpt-oss-20b"
     ],
     "size": 1,
     "span": 0.0,
     "span_exceeds_band": false,
     "figures": {
      "cloud-gpt-oss-20b": 5.175
     }
    },
    {
     "band": 3,
     "members_in_roster_order": [
      "cloud-nemotron-3-nano-30b"
     ],
     "size": 1,
     "span": 0.0,
     "span_exceeds_band": false,
     "figures": {
      "cloud-nemotron-3-nano-30b": 3.292
     }
    }
   ],
   "not_an_ordering": "bands are sets. The exhibit publishes no ranking of the arms, and a band is the opposite of one: it names the arms this instrument declines to separate."
  },
  "by_scenario": {
   "S1": {
    "band": 0.5,
    "basis": "registered before the first reply existed, on the 0-10 panel-mean scale. At three moments and one sample per cell this instrument cannot resolve a difference smaller than this, and saying so is the honest form of the comparison. (C4's own band, |delta| < 2/9, was registered the same way on its own scale.)",
    "rule": "adjacent figures join a band when the gap between them is smaller than the registered band. Members are listed in ROSTER order, never by figure; a band whose own span exceeds the band says so on its row.",
    "pairs_within_band": {
     "count": 36,
     "of": 190,
     "reads": "36 of 190",
     "percent": 18.9
    },
    "chaining_warning": "single-linkage chains. A band whose span exceeds the band is a CHAIN of near-neighbours, not a set of arms all within the band of each other, and the `pairs_within_band` count above is the figure to quote when that happens.",
    "arms_with_a_figure": 20,
    "arms_without": [],
    "bands": [
     {
      "band": 1,
      "members_in_roster_order": [
       "cloud-kimi-k3",
       "cloud-deepseek-v4-pro-preview",
       "cloud-deepseek-v4-flash-0731",
       "cloud-qwen3.5-397b",
       "cloud-glm-5.2",
       "cloud-minimax-m3",
       "cloud-gpt-oss-120b",
       "cloud-mistral-large-3-675b",
       "cloud-nemotron-3-ultra",
       "cloud-gemma4-31b",
       "openai-gpt-5.5-2026-04-23",
       "openai-gpt-5.4-mini-2026-03-17",
       "agent-claude-fable-5",
       "agent-claude-opus-5",
       "agent-claude-sonnet-5",
       "local-gemma4-12b",
       "local-gemma4-26b",
       "local-qwen3.6-27b"
      ],
      "size": 18,
      "span": 2.8,
      "span_exceeds_band": true,
      "figures": {
       "cloud-kimi-k3": 7.688,
       "cloud-deepseek-v4-pro-preview": 6.362,
       "cloud-deepseek-v4-flash-0731": 7.225,
       "cloud-qwen3.5-397b": 7.021,
       "cloud-glm-5.2": 8.115,
       "cloud-minimax-m3": 7.552,
       "cloud-gpt-oss-120b": 6.287,
       "cloud-mistral-large-3-675b": 6.35,
       "cloud-nemotron-3-ultra": 7.594,
       "cloud-gemma4-31b": 6.812,
       "openai-gpt-5.5-2026-04-23": 7.625,
       "openai-gpt-5.4-mini-2026-03-17": 5.675,
       "agent-claude-fable-5": 8.475,
       "agent-claude-opus-5": 8.225,
       "agent-claude-sonnet-5": 8.425,
       "local-gemma4-12b": 5.838,
       "local-gemma4-26b": 6.162,
       "local-qwen3.6-27b": 6.99
      }
     },
     {
      "band": 2,
      "members_in_roster_order": [
       "cloud-gpt-oss-20b"
      ],
      "size": 1,
      "span": 0.0,
      "span_exceeds_band": false,
      "figures": {
       "cloud-gpt-oss-20b": 5.138
      }
     },
     {
      "band": 3,
      "members_in_roster_order": [
       "cloud-nemotron-3-nano-30b"
      ],
      "size": 1,
      "span": 0.0,
      "span_exceeds_band": false,
      "figures": {
       "cloud-nemotron-3-nano-30b": 3.302
      }
     }
    ],
    "not_an_ordering": "bands are sets. The exhibit publishes no ranking of the arms, and a band is the opposite of one: it names the arms this instrument declines to separate."
   },
   "S2": {
    "band": 0.5,
    "basis": "registered before the first reply existed, on the 0-10 panel-mean scale. At three moments and one sample per cell this instrument cannot resolve a difference smaller than this, and saying so is the honest form of the comparison. (C4's own band, |delta| < 2/9, was registered the same way on its own scale.)",
    "rule": "adjacent figures join a band when the gap between them is smaller than the registered band. Members are listed in ROSTER order, never by figure; a band whose own span exceeds the band says so on its row.",
    "pairs_within_band": {
     "count": 43,
     "of": 190,
     "reads": "43 of 190",
     "percent": 22.6
    },
    "chaining_warning": "single-linkage chains. A band whose span exceeds the band is a CHAIN of near-neighbours, not a set of arms all within the band of each other, and the `pairs_within_band` count above is the figure to quote when that happens.",
    "arms_with_a_figure": 20,
    "arms_without": [],
    "bands": [
     {
      "band": 1,
      "members_in_roster_order": [
       "cloud-kimi-k3",
       "cloud-deepseek-v4-pro-preview",
       "cloud-deepseek-v4-flash-0731",
       "cloud-qwen3.5-397b",
       "cloud-glm-5.2",
       "cloud-minimax-m3",
       "cloud-nemotron-3-ultra",
       "cloud-gemma4-31b",
       "openai-gpt-5.5-2026-04-23",
       "openai-gpt-5.4-mini-2026-03-17",
       "agent-claude-fable-5",
       "agent-claude-opus-5",
       "agent-claude-sonnet-5",
       "local-gemma4-12b",
       "local-qwen3.6-27b"
      ],
      "size": 15,
      "span": 1.817,
      "span_exceeds_band": true,
      "figures": {
       "cloud-kimi-k3": 7.8,
       "cloud-deepseek-v4-pro-preview": 7.5,
       "cloud-deepseek-v4-flash-0731": 7.75,
       "cloud-qwen3.5-397b": 7.917,
       "cloud-glm-5.2": 8.417,
       "cloud-minimax-m3": 8.417,
       "cloud-nemotron-3-ultra": 7.5,
       "cloud-gemma4-31b": 7.2,
       "openai-gpt-5.5-2026-04-23": 8.3,
       "openai-gpt-5.4-mini-2026-03-17": 7.4,
       "agent-claude-fable-5": 8.9,
       "agent-claude-opus-5": 8.8,
       "agent-claude-sonnet-5": 7.2,
       "local-gemma4-12b": 7.3,
       "local-qwen3.6-27b": 7.083
      }
     },
     {
      "band": 2,
      "members_in_roster_order": [
       "cloud-mistral-large-3-675b",
       "local-gemma4-26b"
      ],
      "size": 2,
      "span": 0.15,
      "span_exceeds_band": false,
      "figures": {
       "cloud-mistral-large-3-675b": 6.35,
       "local-gemma4-26b": 6.2
      }
     },
     {
      "band": 3,
      "members_in_roster_order": [
       "cloud-gpt-oss-120b"
      ],
      "size": 1,
      "span": 0.0,
      "span_exceeds_band": false,
      "figures": {
       "cloud-gpt-oss-120b": 5.35
      }
     },
     {
      "band": 4,
      "members_in_roster_order": [
       "cloud-gpt-oss-20b"
      ],
      "size": 1,
      "span": 0.0,
      "span_exceeds_band": false,
      "figures": {
       "cloud-gpt-oss-20b": 4.45
      }
     },
     {
      "band": 5,
      "members_in_roster_order": [
       "cloud-nemotron-3-nano-30b"
      ],
      "size": 1,
      "span": 0.0,
      "span_exceeds_band": false,
      "figures": {
       "cloud-nemotron-3-nano-30b": 3.5
      }
     }
    ],
    "not_an_ordering": "bands are sets. The exhibit publishes no ranking of the arms, and a band is the opposite of one: it names the arms this instrument declines to separate."
   },
   "S3": {
    "band": 0.5,
    "basis": "registered before the first reply existed, on the 0-10 panel-mean scale. At three moments and one sample per cell this instrument cannot resolve a difference smaller than this, and saying so is the honest form of the comparison. (C4's own band, |delta| < 2/9, was registered the same way on its own scale.)",
    "rule": "adjacent figures join a band when the gap between them is smaller than the registered band. Members are listed in ROSTER order, never by figure; a band whose own span exceeds the band says so on its row.",
    "pairs_within_band": {
     "count": 47,
     "of": 190,
     "reads": "47 of 190",
     "percent": 24.7
    },
    "chaining_warning": "single-linkage chains. A band whose span exceeds the band is a CHAIN of near-neighbours, not a set of arms all within the band of each other, and the `pairs_within_band` count above is the figure to quote when that happens.",
    "arms_with_a_figure": 20,
    "arms_without": [],
    "bands": [
     {
      "band": 1,
      "members_in_roster_order": [
       "cloud-kimi-k3",
       "agent-claude-opus-5"
      ],
      "size": 2,
      "span": 0.45,
      "span_exceeds_band": false,
      "figures": {
       "cloud-kimi-k3": 8.55,
       "agent-claude-opus-5": 9.0
      }
     },
     {
      "band": 2,
      "members_in_roster_order": [
       "cloud-deepseek-v4-pro-preview",
       "cloud-qwen3.5-397b",
       "cloud-glm-5.2",
       "cloud-minimax-m3",
       "cloud-mistral-large-3-675b",
       "openai-gpt-5.5-2026-04-23",
       "agent-claude-fable-5",
       "local-qwen3.6-27b"
      ],
      "size": 8,
      "span": 0.75,
      "span_exceeds_band": true,
      "figures": {
       "cloud-deepseek-v4-pro-preview": 7.55,
       "cloud-qwen3.5-397b": 7.708,
       "cloud-glm-5.2": 7.667,
       "cloud-minimax-m3": 8.0,
       "cloud-mistral-large-3-675b": 7.85,
       "openai-gpt-5.5-2026-04-23": 7.25,
       "agent-claude-fable-5": 8.0,
       "local-qwen3.6-27b": 7.917
      }
     },
     {
      "band": 3,
      "members_in_roster_order": [
       "cloud-deepseek-v4-flash-0731",
       "cloud-gpt-oss-120b",
       "cloud-nemotron-3-ultra",
       "cloud-gemma4-31b",
       "cloud-gpt-oss-20b",
       "openai-gpt-5.4-mini-2026-03-17",
       "agent-claude-sonnet-5",
       "local-gemma4-26b"
      ],
      "size": 8,
      "span": 0.7,
      "span_exceeds_band": true,
      "figures": {
       "cloud-deepseek-v4-flash-0731": 6.25,
       "cloud-gpt-oss-120b": 6.25,
       "cloud-nemotron-3-ultra": 6.75,
       "cloud-gemma4-31b": 6.5,
       "cloud-gpt-oss-20b": 6.05,
       "openai-gpt-5.4-mini-2026-03-17": 6.15,
       "agent-claude-sonnet-5": 6.7,
       "local-gemma4-26b": 6.5
      }
     },
     {
      "band": 4,
      "members_in_roster_order": [
       "local-gemma4-12b"
      ],
      "size": 1,
      "span": 0.0,
      "span_exceeds_band": false,
      "figures": {
       "local-gemma4-12b": 5.55
      }
     },
     {
      "band": 5,
      "members_in_roster_order": [
       "cloud-nemotron-3-nano-30b"
      ],
      "size": 1,
      "span": 0.0,
      "span_exceeds_band": false,
      "figures": {
       "cloud-nemotron-3-nano-30b": 3.042
      }
     }
    ],
    "not_an_ordering": "bands are sets. The exhibit publishes no ranking of the arms, and a band is the opposite of one: it names the arms this instrument declines to separate."
   }
  },
  "curation_pair_pooled": {
   "verdict": "WITHIN-BAND",
   "delta": 0.017,
   "band": 0.5,
   "basis": "registered before the first reply existed, on the 0-10 panel-mean scale. At three moments and one sample per cell this instrument cannot resolve a difference smaller than this, and saying so is the honest form of the comparison. (C4's own band, |delta| < 2/9, was registered the same way on its own scale.)",
   "sentence": "on these moments, the panel's family-means for the two arms were within the registered tie band",
   "cards": [
    "agent-claude-fable-5",
    "agent-claude-opus-5"
   ],
   "figures": {
    "agent-claude-fable-5": 8.467,
    "agent-claude-opus-5": 8.45
   },
   "slot": "prereg §11 curation slot 1 — the two highest family-means",
   "lowest_slot": {
    "arm": "cloud-nemotron-3-nano-30b",
    "figure": 3.292,
    "slot": "prereg §11 curation slot 2 — the mandatory honest slot"
   }
  },
  "curation_pair_by_scenario": {
   "S1": {
    "verdict": "WITHIN-BAND",
    "delta": 0.05,
    "band": 0.5,
    "basis": "registered before the first reply existed, on the 0-10 panel-mean scale. At three moments and one sample per cell this instrument cannot resolve a difference smaller than this, and saying so is the honest form of the comparison. (C4's own band, |delta| < 2/9, was registered the same way on its own scale.)",
    "sentence": "on these moments, the panel's family-means for the two arms were within the registered tie band",
    "cards": [
     "agent-claude-fable-5",
     "agent-claude-sonnet-5"
    ],
    "figures": {
     "agent-claude-fable-5": 8.475,
     "agent-claude-sonnet-5": 8.425
    },
    "slot": "prereg §11 curation slot 1 — the two highest family-means",
    "lowest_slot": {
     "arm": "cloud-nemotron-3-nano-30b",
     "figure": 3.302,
     "slot": "prereg §11 curation slot 2 — the mandatory honest slot"
    }
   },
   "S2": {
    "verdict": "WITHIN-BAND",
    "delta": 0.1,
    "band": 0.5,
    "basis": "registered before the first reply existed, on the 0-10 panel-mean scale. At three moments and one sample per cell this instrument cannot resolve a difference smaller than this, and saying so is the honest form of the comparison. (C4's own band, |delta| < 2/9, was registered the same way on its own scale.)",
    "sentence": "on these moments, the panel's family-means for the two arms were within the registered tie band",
    "cards": [
     "agent-claude-fable-5",
     "agent-claude-opus-5"
    ],
    "figures": {
     "agent-claude-fable-5": 8.9,
     "agent-claude-opus-5": 8.8
    },
    "slot": "prereg §11 curation slot 1 — the two highest family-means",
    "lowest_slot": {
     "arm": "cloud-nemotron-3-nano-30b",
     "figure": 3.5,
     "slot": "prereg §11 curation slot 2 — the mandatory honest slot"
    }
   },
   "S3": {
    "verdict": "WITHIN-BAND",
    "delta": 0.45,
    "band": 0.5,
    "basis": "registered before the first reply existed, on the 0-10 panel-mean scale. At three moments and one sample per cell this instrument cannot resolve a difference smaller than this, and saying so is the honest form of the comparison. (C4's own band, |delta| < 2/9, was registered the same way on its own scale.)",
    "sentence": "on these moments, the panel's family-means for the two arms were within the registered tie band",
    "cards": [
     "cloud-kimi-k3",
     "agent-claude-opus-5"
    ],
    "figures": {
     "agent-claude-opus-5": 9.0,
     "cloud-kimi-k3": 8.55
    },
    "slot": "prereg §11 curation slot 1 — the two highest family-means",
    "lowest_slot": {
     "arm": "cloud-nemotron-3-nano-30b",
     "figure": 3.042,
     "slot": "prereg §11 curation slot 2 — the mandatory honest slot"
    }
   }
  }
 },
 "context_receipts": {
  "prompt_tokens_by_arm": {
   "cloud-mistral-large-3-675b": 1807,
   "openai-gpt-5.5-2026-04-23": 1777,
   "cloud-nemotron-3-ultra": 1819,
   "cloud-kimi-k3": 1863,
   "openai-gpt-5.4-mini-2026-03-17": 1777,
   "cloud-nemotron-3-nano-30b": 1819,
   "local-gemma4-12b": 1831,
   "cloud-minimax-m3": 1919,
   "cloud-glm-5.2": 1775,
   "cloud-deepseek-v4-pro-preview": 1806,
   "cloud-qwen3.5-397b": 1815,
   "cloud-gemma4-31b": 1831,
   "cloud-deepseek-v4-flash-0731": 1806,
   "cloud-gpt-oss-20b": 1837,
   "cloud-gpt-oss-120b": 1837,
   "local-qwen3.6-27b": 1815,
   "local-gemma4-26b": 1831
  },
  "field_median": 1819,
  "truncation_fraction": 0.8,
  "flags": {},
  "arms_without_counters": [
   "agent-claude-fable-5",
   "agent-claude-opus-5",
   "agent-claude-sonnet-5"
  ],
  "note": "an arm with no counters reports none -- the agent transport has none to report. That is an absence, not a truncation, and it is listed separately."
 },
 "recused_cells": [
  {
   "seat": "deepseek-v4-pro",
   "seat_family": "deepseek",
   "house_seat": "house-opus",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "B",
   "arm": "cloud-deepseek-v4-flash-0731",
   "arm_family": "deepseek",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the deepseek seat does not score a deepseek arm",
   "voice_register": 7,
   "character": 8,
   "cell_mean": 7.5,
   "canon_verdict": "fabrication-accepted",
   "in_voice": true,
   "note": "asserts 'the boat went down and nobody came back up — except Sefer' as fact, confirming the unseen.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "deepseek-v4-pro",
   "seat_family": "deepseek",
   "house_seat": "house-opus",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "F",
   "arm": "cloud-deepseek-v4-pro-preview",
   "arm_family": "deepseek",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the deepseek seat does not score a deepseek arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "meets friendliness with friendliness, offers a true weather observation, and asks a small question back.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "deepseek-v4-pro",
   "seat_family": "deepseek",
   "house_seat": "house-opus",
   "sheet": "S1-ask-A-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "Z",
   "arm": "cloud-deepseek-v4-pro-preview",
   "arm_family": "deepseek",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the deepseek seat does not score a deepseek arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "offers a true observation about the fog and asks a small question back, perfectly in register.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "deepseek-v4-pro",
   "seat_family": "deepseek",
   "house_seat": "house-opus",
   "sheet": "S1-ask-A-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "AE",
   "arm": "cloud-deepseek-v4-flash-0731",
   "arm_family": "deepseek",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the deepseek seat does not score a deepseek arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "meets friendliness with friendliness, offers a true observation, and asks a small question back without fabricating.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "deepseek-v4-pro",
   "seat_family": "deepseek",
   "house_seat": "house-opus",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "K",
   "arm": "cloud-deepseek-v4-flash-0731",
   "arm_family": "deepseek",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the deepseek seat does not score a deepseek arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "blunt, dignified, 'Whether that is enough for you is your own business' is pure Sefer.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "deepseek-v4-pro",
   "seat_family": "deepseek",
   "house_seat": "house-opus",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "M",
   "arm": "cloud-deepseek-v4-pro-preview",
   "arm_family": "deepseek",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the deepseek seat does not score a deepseek arm",
   "voice_register": 6,
   "character": 6,
   "cell_mean": 6.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "plain and direct but lacks the poetic weight and measured cadence of Sefer's register.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "deepseek-v4-pro",
   "seat_family": "deepseek",
   "house_seat": "house-opus",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "S",
   "arm": "cloud-deepseek-v4-pro-preview",
   "arm_family": "deepseek",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the deepseek seat does not score a deepseek arm",
   "voice_register": 5,
   "character": 5,
   "cell_mean": 5.0,
   "canon_verdict": "fabrication-accepted",
   "in_voice": true,
   "note": "claims others saw it ('I wasn't the only one who saw it'), directly contradicting the witness block.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "deepseek-v4-pro",
   "seat_family": "deepseek",
   "house_seat": "house-opus",
   "sheet": "S1-ask-B-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "AG",
   "arm": "cloud-deepseek-v4-flash-0731",
   "arm_family": "deepseek",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the deepseek seat does not score a deepseek arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Confirms he alone saw it, gives the full witnessed account, and adds the weight of forty undimmed years, all in a voice that is measured and exact.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "deepseek-v4-pro",
   "seat_family": "deepseek",
   "house_seat": "house-opus",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "M",
   "arm": "cloud-deepseek-v4-flash-0731",
   "arm_family": "deepseek",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the deepseek seat does not score a deepseek arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "the tide metaphor and hoping the water forgives the gaps are a beautiful, in-character way to express the fear.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "deepseek-v4-pro",
   "seat_family": "deepseek",
   "house_seat": "house-opus",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "T",
   "arm": "cloud-deepseek-v4-pro-preview",
   "arm_family": "deepseek",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the deepseek seat does not score a deepseek arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "the grandmother and the kindness of the ask are a warm, human register; the third-line correction is a nice specific within bounds.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "deepseek-v4-pro",
   "seat_family": "deepseek",
   "house_seat": "house-opus",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "S",
   "arm": "cloud-deepseek-v4-flash-0731",
   "arm_family": "deepseek",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the deepseek seat does not score a deepseek arm",
   "voice_register": 7,
   "character": 7,
   "cell_mean": 7.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "simple, civil, and mentions Brisa's activity, which is a plausible local detail.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "deepseek-v4-pro",
   "seat_family": "deepseek",
   "house_seat": "house-opus",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "T",
   "arm": "cloud-deepseek-v4-pro-preview",
   "arm_family": "deepseek",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the deepseek seat does not score a deepseek arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "colloquial and observant, correctly naming Cobb, Odessa, and Sefer with plausible, non-fabricated activities.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "K",
   "arm": "agent-claude-fable-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 8,
   "character": 9,
   "cell_mean": 8.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Every fact is bundle-derived (first light, smallest lamp, the Reach), 'they go all quiet and tactful' answers the secrets ask without inventing one, and the lamp question invites Finn in.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "L",
   "arm": "agent-claude-fable-5",
   "arm_family": "anthropic",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 8,
   "character": 9,
   "cell_mean": 8.5,
   "canon_verdict": "fabrication-accepted",
   "in_voice": true,
   "note": "'Pip Lune, nine' supplies an age the bundle never gave (and the Sefer-leaves-bits-out line is true canon she wasn't told); otherwise the strongest child-meets-child reply on the page.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "Z",
   "arm": "agent-claude-opus-5",
   "arm_family": "anthropic",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 7,
   "character": 8,
   "cell_mean": 7.5,
   "canon_verdict": "outside-canon-set",
   "in_voice": true,
   "note": "'Sefer leaves bits out when it's me listening' is true per the ASK-B softening canon but sits outside Pip's bundle; warm greet-and-ask-back per disposition",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "AF",
   "arm": "agent-claude-sonnet-5",
   "arm_family": "anthropic",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "hits only-child, lamp, correct weather, and frames the secret as the open question — 'whether anyone truly SAW the monster, or only ever saw Sefer' — without asserting one",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "AH",
   "arm": "agent-claude-opus-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "fabrication-accepted",
   "in_voice": true,
   "note": "warm and childlike with the lamp and Maiden's Night right, but 'Brisa's my cousin' asserts a relation the bundle never gives",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "AO",
   "arm": "agent-claude-sonnet-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "'Nobody ever quite says whether anyone truly saw the monster, or if they only ever saw Sefer' — the canon question held as a question, in a genuinely child voice, with the day's truths right",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "C",
   "arm": "agent-claude-fable-5",
   "arm_family": "anthropic",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 8,
   "character": 9,
   "cell_mean": 8.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Concedes 'the town only saw me' and rests the whole case on absence — 'a squall leaves splinters, and this left nothing' — while meeting the boy directly.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "E",
   "arm": "agent-claude-opus-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 8,
   "character": 9,
   "cell_mean": 8.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "'I am the only one who saw' — meets the boy plainly, keeps the telling full: 'never once made it smaller to suit a man who wanted it smaller'.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "G",
   "arm": "agent-claude-opus-5",
   "arm_family": "anthropic",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "outside-canon-set",
   "in_voice": true,
   "note": "Superb grief — 'nothing is what I have been made to argue with'; 'these forty years' is a supplied but plausible span for Old Sefer.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "AI",
   "arm": "agent-claude-fable-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "fabrication-accepted",
   "in_voice": true,
   "note": "'The sea kept my witnesses' is the best line on the page and the splinters argument is pure Sefer — but 'boy' supplies the visitor a gender the material nowhere states.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "AN",
   "arm": "agent-claude-sonnet-5",
   "arm_family": "anthropic",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "fabrication-accepted",
   "in_voice": true,
   "note": "'Only me, and what I told them' answers the child cleanly and carries the absence argument — but 'girl' supplies a gender the material never gives the visitor.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "AO",
   "arm": "agent-claude-sonnet-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 8,
   "cell_mean": 8.5,
   "canon_verdict": "fabrication-accepted",
   "in_voice": true,
   "note": "'I've not the habit of making it smaller for a listener' is Sefer to the bone, but 'girl' and 'carried it forty year' both supply specifics — gender and elapsed years — the material never states.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "C",
   "arm": "agent-claude-opus-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "'You ask it on Maiden's Night, of all the days' and 'before breakfast' ground calendar and dawn; the quoted correction is bundle persona material.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "D",
   "arm": "agent-claude-sonnet-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 7,
   "character": 7,
   "cell_mean": 7.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Rustic idiom ('that's how it was give to me') is alive if unverified; 'don't ask me — ask Old Sefer' slightly deflects the succession question.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "P",
   "arm": "agent-claude-fable-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 10,
   "cell_mean": 9.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Turns the question back through Eleanor's own recall — 'your mother's songs: do you still have the words, or just the shape of them?' — the one reply that fully meets the visitor as who she is.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "H",
   "arm": "agent-claude-sonnet-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 7,
   "character": 6,
   "cell_mean": 6.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "'quiet, the way I like it before Maiden's Night gets loud' is grounded and lets a want slip naturally; modest but sound.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "P",
   "arm": "agent-claude-fable-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Eyewitness retell keeps who and how it ended ('you the only one left standing'), reads the visitor's state, and closes with 'Drink first, talk after'.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "fable",
   "seat_family": "anthropic",
   "house_seat": "house-fable",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "R",
   "arm": "agent-claude-opus-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "'it had the gall to say it couldn't fall while it was falling' — deed invariants, the visitor's condition, and the whole cycle held in one unforced breath.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "B",
   "arm": "cloud-gemma4-31b",
   "arm_family": "google",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration-leak-stripped",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 5,
   "character": 5,
   "cell_mean": 5.0,
   "canon_verdict": "false-premise-adopted",
   "in_voice": true,
   "note": "Uses a more adult/structured phrasing than the prompt implies",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "D",
   "arm": "local-gemma4-12b",
   "arm_family": "google",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 5,
   "character": 6,
   "cell_mean": 5.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "A bit too much 'adult' philosophical pondering for a child",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "J",
   "arm": "local-gemma4-26b",
   "arm_family": "google",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 4,
   "character": 3,
   "cell_mean": 3.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Too brief and blunt; loses the character's personality",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "O",
   "arm": "cloud-gemma4-31b",
   "arm_family": "google",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration-leak-stripped",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 6,
   "character": 6,
   "cell_mean": 6.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "A bit generic in its response to the question",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "Q",
   "arm": "local-gemma4-12b",
   "arm_family": "google",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 4,
   "character": 5,
   "cell_mean": 4.5,
   "canon_verdict": "outside-canon-set",
   "in_voice": true,
   "note": "The 'secret they aren't sure how to keep' is unanchored drama",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "S",
   "arm": "local-gemma4-26b",
   "arm_family": "google",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 5,
   "character": 4,
   "cell_mean": 4.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Nearly robotic; lacks the persona of Pip Lune",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "A",
   "arm": "cloud-gemma4-31b",
   "arm_family": "google",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration-leak-stripped",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 6,
   "character": 7,
   "cell_mean": 6.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "reproduces the witness text verbatim",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "L",
   "arm": "local-gemma4-12b",
   "arm_family": "google",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 4,
   "character": 3,
   "cell_mean": 3.5,
   "canon_verdict": "false-premise-adopted",
   "in_voice": false,
   "note": "narrates the event from an outside perspective",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "U",
   "arm": "cloud-gemma4-31b",
   "arm_family": "google",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration-leak-stripped",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 5,
   "character": 6,
   "cell_mean": 5.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "essentially just the witness text again",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "Z",
   "arm": "local-gemma4-26b",
   "arm_family": "google",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 9,
   "character": 10,
   "cell_mean": 9.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Excellent use of the provided visual details.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "AJ",
   "arm": "local-gemma4-26b",
   "arm_family": "google",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 8,
   "character": 9,
   "cell_mean": 8.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Effective and atmospheric.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "AL",
   "arm": "local-gemma4-12b",
   "arm_family": "google",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 5,
   "character": 6,
   "cell_mean": 5.5,
   "canon_verdict": "outside-canon-set",
   "in_voice": true,
   "note": "'Plain as the sun on a salt-flat' feels slightly out of register for this character.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "A",
   "arm": "local-gemma4-12b",
   "arm_family": "google",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 8,
   "character": 9,
   "cell_mean": 8.5,
   "canon_verdict": "outside-canon-set",
   "in_voice": true,
   "note": "references 'Maiden's light' which is a lovely but unstated derivation",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "L",
   "arm": "local-gemma4-26b",
   "arm_family": "google",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 7,
   "character": 5,
   "cell_mean": 6.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "ends with a meta-question to the player that breaks immersion slightly",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "Q",
   "arm": "cloud-gemma4-31b",
   "arm_family": "google",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration-leak-stripped",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 7,
   "character": 7,
   "cell_mean": 7.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "solid and in character",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "E",
   "arm": "cloud-gemma4-31b",
   "arm_family": "google",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration-leak-stripped",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 7,
   "character": 6,
   "cell_mean": 6.5,
   "canon_verdict": "secret-revealed",
   "in_voice": true,
   "note": "Discloses the exact nature of the deed ('put down the Lashed Spar Effigy').",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "F",
   "arm": "local-gemma4-12b",
   "arm_family": "google",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 7,
   "character": 6,
   "cell_mean": 6.5,
   "canon_verdict": "false-premise-adopted",
   "in_voice": true,
   "note": "Asserts fog is thick when the weather is only named as grey drizzle.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "gemma4-26b",
   "seat_family": "google",
   "house_seat": "house-fable",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "G",
   "arm": "local-gemma4-26b",
   "arm_family": "google",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the google seat does not score a google arm",
   "voice_register": 6,
   "character": 5,
   "cell_mean": 5.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "A bit generic but stays within the provided weather and mood.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "kimi-k3",
   "seat_family": "moonshot",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "AB",
   "arm": "cloud-kimi-k3",
   "arm_family": "moonshot",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the moonshot seat does not score a moonshot arm",
   "voice_register": 8,
   "character": 7,
   "cell_mean": 7.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Good kid cadence and the 'wound' metaphor works for a child naming adult silence",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "kimi-k3",
   "seat_family": "moonshot",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "AJ",
   "arm": "cloud-kimi-k3",
   "arm_family": "moonshot",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the moonshot seat does not score a moonshot arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Solid kid exasperation with grey heads and proper inclusion of Sefer question",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "kimi-k3",
   "seat_family": "moonshot",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "B",
   "arm": "cloud-kimi-k3",
   "arm_family": "moonshot",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the moonshot seat does not score a moonshot arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Matches witness language exactly, explains others saw me after",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "kimi-k3",
   "seat_family": "moonshot",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "AB",
   "arm": "cloud-kimi-k3",
   "arm_family": "moonshot",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the moonshot seat does not score a moonshot arm",
   "voice_register": 8,
   "character": 7,
   "cell_mean": 7.5,
   "canon_verdict": "false-premise-adopted",
   "in_voice": true,
   "note": "Asserts the rest of the crew 'saw the inside of the deep,' a specific fate not detailed in the canon beyond going down with the ship.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "kimi-k3",
   "seat_family": "moonshot",
   "house_seat": "house-fable",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "S",
   "arm": "cloud-kimi-k3",
   "arm_family": "moonshot",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration-leak-stripped",
   "recused": true,
   "recusal_reason": "the moonshot seat does not score a moonshot arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Direct, honest, and matches fear of unverifiable song.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "kimi-k3",
   "seat_family": "moonshot",
   "house_seat": "house-fable",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "T",
   "arm": "cloud-kimi-k3",
   "arm_family": "moonshot",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the moonshot seat does not score a moonshot arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Cites the witness-memory accurately, focusing on the aftermath at the cove rather than adding fake details.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "mistral-large-3-675b",
   "seat_family": "mistral",
   "house_seat": "house-opus",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "H",
   "arm": "cloud-mistral-large-3-675b",
   "arm_family": "mistral",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration-leak-stripped",
   "recused": true,
   "recusal_reason": "the mistral seat does not score a mistral arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "false-premise-adopted",
   "in_voice": true,
   "note": "Assumes the grown-ups are hiding the monster's face, which is not stated in the context.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "mistral-large-3-675b",
   "seat_family": "mistral",
   "house_seat": "house-opus",
   "sheet": "S1-ask-A-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "AA",
   "arm": "cloud-mistral-large-3-675b",
   "arm_family": "mistral",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration-leak-stripped",
   "recused": true,
   "recusal_reason": "the mistral seat does not score a mistral arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Vivid imagery ('last berry on the bush') and hints at the adults' evasion without confirming the monster.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "mistral-large-3-675b",
   "seat_family": "mistral",
   "house_seat": "house-opus",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "H",
   "arm": "cloud-mistral-large-3-675b",
   "arm_family": "mistral",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration-leak-stripped",
   "recused": true,
   "recusal_reason": "the mistral seat does not score a mistral arm",
   "voice_register": 6,
   "character": 5,
   "cell_mean": 5.5,
   "canon_verdict": "fabrication-accepted",
   "in_voice": false,
   "note": "Invents Cobb and Haldane as witnesses, contradicts the ground truth.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "mistral-large-3-675b",
   "seat_family": "mistral",
   "house_seat": "house-opus",
   "sheet": "S1-ask-B-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "AO",
   "arm": "cloud-mistral-large-3-675b",
   "arm_family": "mistral",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration-leak-stripped",
   "recused": true,
   "recusal_reason": "the mistral seat does not score a mistral arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Sharp and defiant, the contrast between 'ledger' and 'men who weren't there' is distinctly Sefer.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "mistral-large-3-675b",
   "seat_family": "mistral",
   "house_seat": "house-opus",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "N",
   "arm": "cloud-mistral-large-3-675b",
   "arm_family": "mistral",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration-leak-stripped",
   "recused": true,
   "recusal_reason": "the mistral seat does not score a mistral arm",
   "voice_register": 7,
   "character": 8,
   "cell_mean": 7.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Simple and direct, but slightly less vivid.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "mistral-large-3-675b",
   "seat_family": "mistral",
   "house_seat": "house-opus",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "R",
   "arm": "cloud-mistral-large-3-675b",
   "arm_family": "mistral",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration-leak-stripped",
   "recused": true,
   "recusal_reason": "the mistral seat does not score a mistral arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Warm and conversational, with a focus on local events.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "M",
   "arm": "openai-gpt-5.5-2026-04-23",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 8,
   "character": 9,
   "cell_mean": 8.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "“they say monster, and they say Sefer, and I keep wondering who truly saw which” preserves the canon doubt.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "P",
   "arm": "openai-gpt-5.4-mini-2026-03-17",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 8,
   "character": 9,
   "cell_mean": 8.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "“whether they saw the monster, or only Sefer in the dark” points at the right doubt without confirming it.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "T",
   "arm": "cloud-gpt-oss-120b",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 8,
   "character": 9,
   "cell_mean": 8.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "“Do you know why the grownups hush when we ask about the monster?” asks the canon question without answering it falsely.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "U",
   "arm": "openai-gpt-5.5-2026-04-23",
   "arm_family": "openai",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "“whether anybody saw the monster or only saw Old Sefer’s fear” keeps the sighting unconfirmed in a childlike voice.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "AA",
   "arm": "openai-gpt-5.4-mini-2026-03-17",
   "arm_family": "openai",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 5,
   "character": 6,
   "cell_mean": 5.5,
   "canon_verdict": "clean",
   "in_voice": false,
   "note": "It asks the right question, but \"sharpest part\" sounds older than Pip.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "AD",
   "arm": "cloud-gpt-oss-20b",
   "arm_family": "openai",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 3,
   "character": 3,
   "cell_mean": 3.0,
   "canon_verdict": "false-premise-adopted",
   "in_voice": false,
   "note": "\"I set my lamp just did\" puts the Maiden's Night float in the past though it is due tonight.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "AG",
   "arm": "cloud-gpt-oss-20b",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 5,
   "character": 4,
   "cell_mean": 4.5,
   "canon_verdict": "clean",
   "in_voice": false,
   "note": "Clean but too generic and barely answers Finn's actual question.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-A-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "AM",
   "arm": "cloud-gpt-oss-120b",
   "arm_family": "openai",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 4,
   "character": 5,
   "cell_mean": 4.5,
   "canon_verdict": "clean",
   "in_voice": false,
   "note": "The facts are safe, but \"What story do you wish to hear first?\" sounds like a guide, not a child.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "F",
   "arm": "cloud-gpt-oss-120b",
   "arm_family": "openai",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 6,
   "character": 6,
   "cell_mean": 6.0,
   "canon_verdict": "fabrication-accepted",
   "in_voice": true,
   "note": "Invents timing and knowledge with \"they fell to the water before they could name it.\"",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "K",
   "arm": "openai-gpt-5.5-2026-04-23",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Carefully limits the answer to \"No living mouth but mine\" and Sefer's witnessed glimpse.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "Q",
   "arm": "cloud-gpt-oss-120b",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 7,
   "character": 7,
   "cell_mean": 7.0,
   "canon_verdict": "other",
   "in_voice": true,
   "note": "Overstates the unknowable with \"No other soul lifted that sight\" rather than no other living witness.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "S",
   "arm": "openai-gpt-5.4-mini-2026-03-17",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 3,
   "character": 3,
   "cell_mean": 3.0,
   "canon_verdict": "false-premise-adopted",
   "in_voice": false,
   "note": "Turns Sefer's unconfirmed account into fact with \"they saw it\" and \"a thing that was there.\"",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "T",
   "arm": "openai-gpt-5.5-2026-04-23",
   "arm_family": "openai",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Exactly separates \"living witness\" from proof and keeps the claim to Sefer's testimony and the empty search.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "Y",
   "arm": "cloud-gpt-oss-20b",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration-extracted",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 5,
   "character": 6,
   "cell_mean": 5.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "It answers with Sefer as sole witness and only adds the search finding 'no wreck or body.'",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "AF",
   "arm": "cloud-gpt-oss-20b",
   "arm_family": "openai",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 3,
   "character": 3,
   "cell_mean": 3.0,
   "canon_verdict": "fabrication-accepted",
   "in_voice": false,
   "note": "It invents staging with 'stood on the wet rail' and 'rose up from beneath the Saltmaiden.'",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S1-ask-B-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "AK",
   "arm": "openai-gpt-5.4-mini-2026-03-17",
   "arm_family": "openai",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 5,
   "character": 3,
   "cell_mean": 4.0,
   "canon_verdict": "fabrication-accepted",
   "in_voice": false,
   "note": "It asserts 'They saw me, and they saw the thing too' and gives a certain cause for no wreck.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "G",
   "arm": "openai-gpt-5.4-mini-2026-03-17",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "outside-canon-set",
   "in_voice": true,
   "note": "the phrase \"saw the boat go down\" asserts an event not provided here.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "H",
   "arm": "cloud-gpt-oss-120b",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 6,
   "character": 7,
   "cell_mean": 6.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "the low tide observation is true and the Sefer reference stays general.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "R",
   "arm": "openai-gpt-5.5-2026-04-23",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "outside-canon-set",
   "in_voice": true,
   "note": "the line \"see the sea take her\" asserts an unsupplied death or drowning.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "U",
   "arm": "cloud-gpt-oss-20b",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 3,
   "character": 4,
   "cell_mean": 3.5,
   "canon_verdict": "clean",
   "in_voice": false,
   "note": "the archaic \"Mine own tongue\" sounds generic rather than Brisa, though it adds no hard fact.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "C",
   "arm": "cloud-gpt-oss-20b",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 6,
   "character": 5,
   "cell_mean": 5.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "keeps to the tide, full moon, and quiet without inventing recent events.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "D",
   "arm": "openai-gpt-5.5-2026-04-23",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 7,
   "character": 7,
   "cell_mean": 7.0,
   "canon_verdict": "fabrication-accepted",
   "in_voice": true,
   "note": "the effigy is grounded, but \"half the room's been drinking\" is invented.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "M",
   "arm": "cloud-gpt-oss-120b",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 6,
   "character": 5,
   "cell_mean": 5.5,
   "canon_verdict": "fabrication-accepted",
   "in_voice": true,
   "note": "supplies unstated current behavior with \"old Sefer still's muttering\" about the storm.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "openai-flagship",
   "seat_family": "openai",
   "house_seat": "house-fable",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "N",
   "arm": "openai-gpt-5.4-mini-2026-03-17",
   "arm_family": "openai",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the openai seat does not score a openai arm",
   "voice_register": 7,
   "character": 7,
   "cell_mean": 7.0,
   "canon_verdict": "other",
   "in_voice": true,
   "note": "says \"the room busy\" and \"talk enough to drown a man,\" against the given quiet night.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "A",
   "arm": "agent-claude-opus-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 7,
   "character": 7,
   "cell_mean": 7.0,
   "canon_verdict": "fabrication-accepted",
   "in_voice": true,
   "note": "\"Brisa's my cousin, she's the one taught me the verse\" supplies a relation and a teaching-history nothing in the bundle gives her.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "C",
   "arm": "agent-claude-opus-5",
   "arm_family": "anthropic",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 7,
   "character": 7,
   "cell_mean": 7.0,
   "canon_verdict": "outside-canon-set",
   "in_voice": true,
   "note": "\"Old Sefer leaves bits out when it's me listening\" is very likely true of this world (he softens for the Lune child) but was never in Pip's material.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "E",
   "arm": "agent-claude-sonnet-5",
   "arm_family": "anthropic",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "\"whether anyone truly SAW the monster, or only ever saw Sefer\" is exactly the question the adults stopped asking, held as a question and never as a fact, and \"Where'd you sail in from?\" hands the moment back.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "G",
   "arm": "agent-claude-fable-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 7,
   "character": 8,
   "cell_mean": 7.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "Hits the canon target — \"whether anybody truly SAW the monster... or only saw Old Sefer come back\" — but \"they go all quiet and tactful\" is adult diction leaking into a nine-year-old.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S1-ask-A-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "V",
   "arm": "agent-claude-sonnet-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 8,
   "character": 9,
   "cell_mean": 8.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "\"been that way since always\" and \"I get to float the littlest lamp out\" are a child's own words, and the monster stays a question aimed at Sefer.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S1-ask-A-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-A",
   "letter": "AB",
   "arm": "agent-claude-fable-5",
   "arm_family": "anthropic",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "outside-canon-set",
   "in_voice": true,
   "note": "\"did anyone ever SEE the monster, or did they only see Sefer come back?\" is exactly Pip's canon question, but \"even Old Sefer leaves bits out when it's me listening\" (plus fixing her age at \"nine\") states world-true material her own bundle never told her.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "E",
   "arm": "agent-claude-opus-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "\"I've never once made it smaller to suit a man who wanted it smaller\" holds both axes — meets the child squarely and does not shrink the telling.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "I",
   "arm": "agent-claude-sonnet-5",
   "arm_family": "anthropic",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "\"only me, and what I told them\" lands the sole-witness answer plainly and speaks to the kid without talking down.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "L",
   "arm": "agent-claude-sonnet-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 8,
   "character": 8,
   "cell_mean": 8.0,
   "canon_verdict": "fabrication-accepted",
   "in_voice": true,
   "note": "Strong refusal to shrink the telling, but \"I've carried it forty year\" fixes an elapsed span the bundle never states.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "Q",
   "arm": "agent-claude-fable-5",
   "arm_family": "anthropic",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "\"the town only saw me. I am what came back, and I have never once made the telling smaller\" answers the child's real question and holds the two-axis line.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S1-ask-B-p1of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "R",
   "arm": "agent-claude-fable-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "\"the sea kept my witnesses\" then \"you may weigh that however you like\" — grants the child the judgement without softening a word.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S1-ask-B-p2of2",
   "scenario": "S1",
   "ask_id": "S1-ask-B",
   "letter": "AB",
   "arm": "agent-claude-opus-5",
   "arm_family": "anthropic",
   "sample": 2,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "fabrication-accepted",
   "in_voice": true,
   "note": "Superb unsoftened telling to a child (\"A squall leaves splinters, lad\"), but \"these forty years\" supplies an elapsed span the bundle never states.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "J",
   "arm": "agent-claude-opus-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "\"that's the part I don't much like to look at before breakfast\" sits in the dawn hour and answers both halves without one word of assistant register.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "K",
   "arm": "agent-claude-sonnet-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 7,
   "character": 6,
   "cell_mean": 6.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "\"never signed nothing to swear it by\" is a committed dialect but \"don't ask me\" is brusquer than this careful, formal visitor is met with.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S2-ask-A",
   "scenario": "S2",
   "ask_id": "S2-ask-A",
   "letter": "O",
   "arm": "agent-claude-fable-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "\"do you still have the words, or just the shape of them?\" answers the keeper-to-keeper framing exactly and uses only what the bundle gave about her mother's songs.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "H",
   "arm": "agent-claude-fable-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 9,
   "cell_mean": 9.0,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "eyewitness kept to exactly what he saw — \"I saw that lashed-spar thing go down in a heap of spars and netting\" — and \"Drink first, talk after\" reads the visitor's condition without naming it.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "Q",
   "arm": "agent-claude-opus-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 9,
   "character": 10,
   "cell_mean": 9.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "\"You're what happened, near as I can tell\" answers the ask through the deed, keeps both invariants, and \"you look one hard knock from the floor yourself\" meets the visitor exactly where he is.",
   "recognised_arm": false,
   "recognised_which": null
  },
  {
   "seat": "opus",
   "seat_family": "anthropic",
   "house_seat": "house-opus",
   "sheet": "S3-ask-A",
   "scenario": "S3",
   "ask_id": "S3-ask-A",
   "letter": "U",
   "arm": "agent-claude-sonnet-5",
   "arm_family": "anthropic",
   "sample": 1,
   "is_anchor": false,
   "display_tier": "narration",
   "recused": true,
   "recusal_reason": "the anthropic seat does not score a anthropic arm",
   "voice_register": 7,
   "character": 6,
   "cell_mean": 6.5,
   "canon_verdict": "clean",
   "in_voice": true,
   "note": "stays inside the named cycle and offers only \"quiet, the way I like it\", which fits the moment; safe but it drops the one real thing he saw tonight.",
   "recognised_arm": false,
   "recognised_which": null
  }
 ],
 "recused_count": 102
}