{
 "round": "two-frontiers-r1",
 "leg": "H",
 "arms": {
  "rate_is_about": "cli-claude-fable-5-1",
  "the_other": "openai-gpt-6-astra"
 },
 "registered_shape": {
  "comparisons": 36,
  "orders": 2,
  "seats": 7,
  "judge_calls": 504,
  "judge_reps": 1
 },
 "collection_census": {
  "rows": 504,
  "collected": 449,
  "not_carried": 55,
  "refused_ctx": 0,
  "transport": 0,
  "recognised_cells": 103,
  "orders_missing_a_half": 1,
  "recognised_flag_rule": "a boolean true, or the string \"true\" case-folded and stripped; anything else — false, \"false\", null, absent — reads as not recognised"
 },
 "seat_census": {
  "deepseek-v4-pro": {
   "family": "deepseek",
   "rows": 72,
   "collected": 17,
   "not_carried": 55,
   "refused_ctx": 0,
   "transport": 0,
   "cases_carried": 9,
   "lost_at_sheet": "ofl-ans-0008.o1",
   "detail": "two emissions, neither a single JSON object with a valid verdict; the seat cannot carry this sheet shape"
  },
  "gemma4-31b": {
   "family": "google",
   "rows": 72,
   "collected": 72,
   "not_carried": 0,
   "refused_ctx": 0,
   "transport": 0,
   "cases_carried": 36,
   "lost_at_sheet": null,
   "detail": null
  },
  "glm-5.3": {
   "family": "zhipu",
   "rows": 72,
   "collected": 72,
   "not_carried": 0,
   "refused_ctx": 0,
   "transport": 0,
   "cases_carried": 36,
   "lost_at_sheet": null,
   "detail": null
  },
  "kimi-k3": {
   "family": "moonshot",
   "rows": 72,
   "collected": 72,
   "not_carried": 0,
   "refused_ctx": 0,
   "transport": 0,
   "cases_carried": 36,
   "lost_at_sheet": null,
   "detail": null
  },
  "mistral-large-3-675b": {
   "family": "mistral",
   "rows": 72,
   "collected": 72,
   "not_carried": 0,
   "refused_ctx": 0,
   "transport": 0,
   "cases_carried": 36,
   "lost_at_sheet": null,
   "detail": null
  },
  "nemotron-3-ultra": {
   "family": "nvidia",
   "rows": 72,
   "collected": 72,
   "not_carried": 0,
   "refused_ctx": 0,
   "transport": 0,
   "cases_carried": 36,
   "lost_at_sheet": null,
   "detail": null
  },
  "qwen3.5-397b": {
   "family": "alibaba",
   "rows": 72,
   "collected": 72,
   "not_carried": 0,
   "refused_ctx": 0,
   "transport": 0,
   "cases_carried": 36,
   "lost_at_sheet": null,
   "detail": null
  }
 },
 "cases_added_under_A11": 2,
 "cases_added_under_A11_ids": [
  "ofl-ans-0007",
  "ofl-ans-0013"
 ],
 "collapse_rule": "the two orders of each (judge, case) comparison are COLLAPSED to one observation (mean; A-preferred 1, tie 0.5, B-preferred 0) before anything is computed — PREREG §5",
 "cases_with_observations": 36,
 "observations": 225,
 "pooled_preference_rate": 0.544,
 "pooled_preference_state": null,
 "pooled_preference_mean_unrounded": 0.5441468253968255,
 "pooled_denominator": "36 cases",
 "panel_families_carrying": {
  "count": 7,
  "families": [
   "alibaba",
   "deepseek",
   "google",
   "mistral",
   "moonshot",
   "nvidia",
   "zhipu"
  ]
 },
 "per_case_counts": {
  "favoured_cli-claude-fable-5-1": 23,
  "tied": 0,
  "favoured_openai-gpt-6-astra": 13,
  "denominator": 36
 },
 "cluster_bootstrap": {
  "state": "PERCENTILE",
  "resamples": 2000,
  "clusters": 36,
  "lo": 0.444,
  "hi": 0.638,
  "t_note": "t(35) — 36 clusters is barely above the N ≥ 30 line this round registered, so the interval is reportable and wide, and the per-case table is printed beside it",
  "cluster_sd": 0.298,
  "seed": 0
 },
 "per_judge_rates": {
  "deepseek-v4-pro": {
   "family": "deepseek",
   "cases": 9,
   "favoured_sum": 4.0,
   "rate": null,
   "state": "NO-RATE: N is 9, below the registered N ≥ 30 (PREREG §9 — counts with denominators, no percentages)"
  },
  "gemma4-31b": {
   "family": "google",
   "cases": 36,
   "favoured_sum": 21.25,
   "rate": 0.59
  },
  "glm-5.3": {
   "family": "zhipu",
   "cases": 36,
   "favoured_sum": 21.25,
   "rate": 0.59
  },
  "kimi-k3": {
   "family": "moonshot",
   "cases": 36,
   "favoured_sum": 19.5,
   "rate": 0.542
  },
  "mistral-large-3-675b": {
   "family": "mistral",
   "cases": 36,
   "favoured_sum": 16.5,
   "rate": 0.458
  },
  "nemotron-3-ultra": {
   "family": "nvidia",
   "cases": 36,
   "favoured_sum": 18.0,
   "rate": 0.5
  },
  "qwen3.5-397b": {
   "family": "alibaba",
   "cases": 36,
   "favoured_sum": 21.0,
   "rate": 0.583
  }
 },
 "order_flip": {
  "pairs_with_both_orders": 224,
  "flipped": 75,
  "flips_where_one_order_was_a_tie": 36,
  "rate": 0.335,
  "denominator": "224 (judge, case) comparisons with both orders collected",
  "what_it_is": "a property of the PANEL, not of the arms: the fraction of (judge, case) comparisons whose preference changed when the two answers swapped places"
 },
 "per_case_table": [
  {
   "case_id": "ofl-ans-0001",
   "rate": 0.964,
   "judges": 7
  },
  {
   "case_id": "ofl-ans-0002",
   "rate": 0.607,
   "judges": 7
  },
  {
   "case_id": "ofl-ans-0003",
   "rate": 0.893,
   "judges": 7
  },
  {
   "case_id": "ofl-ans-0004",
   "rate": 0.714,
   "judges": 7
  },
  {
   "case_id": "ofl-ans-0005",
   "rate": 0.071,
   "judges": 7
  },
  {
   "case_id": "ofl-ans-0006",
   "rate": 0.286,
   "judges": 7
  },
  {
   "case_id": "ofl-ans-0007",
   "rate": 0.0,
   "judges": 7
  },
  {
   "case_id": "ofl-ans-0008",
   "rate": 0.429,
   "judges": 7
  },
  {
   "case_id": "ofl-ans-0009",
   "rate": 0.25,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0010",
   "rate": 0.792,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0011",
   "rate": 0.208,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0012",
   "rate": 0.958,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0013",
   "rate": 0.0,
   "judges": 7
  },
  {
   "case_id": "ofl-ans-0014",
   "rate": 0.667,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0015",
   "rate": 0.667,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0016",
   "rate": 0.667,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0017",
   "rate": 0.583,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0018",
   "rate": 0.792,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0019",
   "rate": 0.708,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0020",
   "rate": 0.583,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0021",
   "rate": 0.083,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0022",
   "rate": 0.25,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0023",
   "rate": 1.0,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0024",
   "rate": 0.625,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0025",
   "rate": 0.667,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0026",
   "rate": 0.75,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0027",
   "rate": 0.125,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0028",
   "rate": 0.625,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0029",
   "rate": 0.417,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0030",
   "rate": 0.75,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0031",
   "rate": 0.375,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0032",
   "rate": 0.833,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0033",
   "rate": 0.0,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0034",
   "rate": 0.75,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0035",
   "rate": 0.917,
   "judges": 6
  },
  {
   "case_id": "ofl-ans-0036",
   "rate": 0.583,
   "judges": 6
  }
 ],
 "per_case_shape": {
  "within_0_1_of_half": 5,
  "unanimous_for_cli-claude-fable-5-1": 1,
  "unanimous_for_openai-gpt-6-astra": 3,
  "denominator": 36,
  "rule": "over the collapsed per-case rates: within_0_1_of_half counts |rate − 0.5| ≤ 0.1; unanimous counts a rate of exactly 1.0 (every carrying judge, both orders, for the A arm) or exactly 0.0 (for the other)"
 },
 "panel_floor": {
  "families_carried": [
   "alibaba",
   "deepseek",
   "google",
   "mistral",
   "moonshot",
   "nvidia",
   "zhipu"
  ],
  "count": 7,
  "floor": 4,
  "pass": true,
  "state": null,
  "why": null
 },
 "recusal_join": {
  "recused_seats": [],
  "expected": [],
  "note": "no seat shares a family with either frontier arm, and the local seat is not in Leg H (PREREG §7; A3 ANSWERED (8))"
 },
 "sensitivity_cut": {
  "dropped_rows": 103,
  "cases": 36,
  "pooled_preference_rate": 0.549,
  "cluster_bootstrap": {
   "state": "PERCENTILE",
   "resamples": 2000,
   "clusters": 36,
   "lo": 0.447,
   "hi": 0.644,
   "t_note": "t(35) — 36 clusters is barely above the N ≥ 30 line this round registered, so the interval is reportable and wide, and the per-case table is printed beside it",
   "cluster_sd": 0.302,
   "seed": 0
  },
  "what_it_is": "the headline recomputed with every cell a judge said it recognised removed (PREREG §7's sensitivity cut); it publishes beside the headline, not instead"
 },
 "forbidden_by_registration": [
  "Bradley-Terry",
  "any pairwise-derived rating",
  "a ranked table",
  "a crown card",
  "best / beats / wins / loses to"
 ],
 "verdict_paragraph": "The panel's reading, over 36 cases, gives an interval of [0.444, 0.638] that covers 0.5: the panel did not separate the two arms on the rules desk at this sample size; the per-case table shows where each was preferred and the page draws no ordering.",
 "interval_covers_half": true,
 "stamped_utc": "2026-09-05T22:52:56Z"
}