{ "schema": "s2s-bench-v1", "exhibit": "chair-trials", "published_utc": "2026-08-13", "status": "PUBLISHED — live at research.strata2signal.com.", "licence": "CC BY 4.0", "attribution": "strata→signal research, research.strata2signal.com", "hardware": "One 96G VRAM workstation, which also answered live requests for three of our apps throughout these trials — two public, one serving playtests for humans and agents alike. Contention is measured in both directions and published with the figures rather than assumed away.", "dataset": "chair-trials-c5-stopwatch", "table": "C5", "what_this_is": "What everything costs, measured on a working box. All timing comes from the daemon's own counters; wall time is kept separate and labelled for contention. Ten repeats per warm cell at three prompt tiers, three cold loads, a reasoning-strength sweep, and a byte-compare of a drafter pair that fired our own registered self-refutation.", "verdict_discipline": "DESCRIPTIVE — no floors, no winner, no ordering by margin", "primary_metric": "decode tok/s at the 1k tier = eval_count / (eval_duration / 1e9), per call, from ollama's own counters", "columns_tiers": [ "arm", "decode tok/s @1k", "@8k tok/s", "@32k tok/s", "outcome" ], "columns_costs": [ "arm", "TTFT-proxy @1k", "cold load", "anomaly ledger" ], "columns_note": "One row set, two fold lists: the tiers first, the costs and the ledger second, same arms in the same order. Every record carries every field its spec declares; nothing sheds at any width.", "row_order": "The registered roster order, published unchanged — not a speed ranking. This leg is descriptive: no floors, no winner, no ordering by margin.", "counting_rules": { "decode": "decode tok/s at the 1k tier = eval_count / (eval_duration / 1e9), per call, from ollama's own counters", "ttft_proxy": "ttft_proxy_ms = total_duration - eval_duration. stream=False, so there is no first-token event: this is an UPPER BOUND, not TTFT, and is never placed beside a vendor's streamed figure", "cold_load": "Labelled warm-page-cache/cold-VRAM: no sudo means no page-cache drop, so these are cold-VRAM loads over a warm file cache and are not comparable to a vendor's cold-start figure or to a reboot. First-load-after-pull is excluded.", "statistics": "min / median / max / n only. No p95 anywhere in this leg: the largest repeat set is 10, and a p95 over ten samples is the second-largest sample wearing a statistics hat", "reading_min_median_max": "On a lightly-shared box the MAX approximates the uncontended ceiling, the median is real life, and a dragged min is the moment a live user shared the card. That is why all three are printed and no p95 appears anywhere: the largest repeat set in this leg is 10.", "tie_rule": "two arms read TIED on decode tok/s when their repeat-set min-max ranges overlap; on the DFlash count, differences under 2 prompts read TIED", "outcome_rule": "c5_speed.arm_outcome — the same function the live runner uses, so a rebuilt state and a run state cannot diverge", "rule_4_null_counters": "A call that returns HTTP 200 with every counter and done_reason null is excluded from every statistic. It is not a slow call and it is not a failure: it is a call we cannot time, and its cause is unexplained — transport_error is null and nothing in the run logs accounts for it.", "rule_5_response_failures": "A call that returns valid counters with EMPTY response text and done_reason=length is a response failure. Three arms exceed the registered 10% ceiling on this rule and are UNMEASURABLE; their measured cells still print, because a state is not a reason to hide the numbers that produced it.", "re_blocking": "Four arms ran a block twice — the original pass and tonight's completion run. Both values publish. Where both occurrences measured nothing, the row says so rather than reporting one absence." }, "totals": { "matrix_calls": 317, "warmups_discarded": 8, "counted_matrix_calls": 309, "dflash_calls": 16, "counted_calls": 325, "recorded_calls": 333 }, "outcome_state_counts": { "UNMEASURABLE": 3, "RANKED": 5 }, "dflash_bytecompare": { "raw_path": "results/raw/C5/dflash-pair/C5-dflash-pair.jsonl", "registered_calls": 16, "recorded_calls": 80, "executions": 5, "executions_note": "the byte-compare block ran once per c5_speed invocation rather than once for the leg, because the driver calls the script per arm and --no-dflash was not passed. Every execution is published.", "agreement_across_executions": true, "identical_counts_seen": [ "2/8" ], "response_length_stable_across_executions": true, "lossless": false, "finding": "DIVERGENT — self-refutation finding: the drafter changed the output, so quality results do NOT transfer across the pair", "passes": [ { "pass": 1, "n_calls": 16, "n_prompts": 8, "identical": 2, "counts_of_n": "2/8", "lossless": false, "comparisons": [ { "prompt_index": 1, "identical": true, "len_on": 103, "len_off": 103, "first_divergence_char": null, "empty_side": null }, { "prompt_index": 2, "identical": false, "len_on": 0, "len_off": 255, "first_divergence_char": 0, "empty_side": "on" }, { "prompt_index": 3, "identical": true, "len_on": 255, "len_off": 255, "first_divergence_char": null, "empty_side": null }, { "prompt_index": 4, "identical": false, "len_on": 55, "len_off": 71, "first_divergence_char": 55, "empty_side": null }, { "prompt_index": 5, "identical": false, "len_on": 231, "len_off": 71, "first_divergence_char": 71, "empty_side": null }, { "prompt_index": 6, "identical": false, "len_on": 47, "len_off": 39, "first_divergence_char": 39, "empty_side": null }, { "prompt_index": 7, "identical": false, "len_on": 71, "len_off": 255, "first_divergence_char": 71, "empty_side": null }, { "prompt_index": 8, "identical": false, "len_on": 39, "len_off": 71, "first_divergence_char": 39, "empty_side": null } ] }, { "pass": 2, "n_calls": 16, "n_prompts": 8, "identical": 2, "counts_of_n": "2/8", "lossless": false, "comparisons": [ { "prompt_index": 1, "identical": true, "len_on": 103, "len_off": 103, "first_divergence_char": null, "empty_side": null }, { "prompt_index": 2, "identical": false, "len_on": 0, "len_off": 255, "first_divergence_char": 0, "empty_side": "on" }, { "prompt_index": 3, "identical": true, "len_on": 255, "len_off": 255, "first_divergence_char": null, "empty_side": null }, { "prompt_index": 4, "identical": false, "len_on": 55, "len_off": 71, "first_divergence_char": 55, "empty_side": null }, { "prompt_index": 5, "identical": false, "len_on": 231, "len_off": 71, "first_divergence_char": 71, "empty_side": null }, { "prompt_index": 6, "identical": false, "len_on": 47, "len_off": 39, "first_divergence_char": 39, "empty_side": null }, { "prompt_index": 7, "identical": false, "len_on": 71, "len_off": 255, "first_divergence_char": 71, "empty_side": null }, { "prompt_index": 8, "identical": false, "len_on": 39, "len_off": 71, "first_divergence_char": 39, "empty_side": null } ] }, { "pass": 3, "n_calls": 16, "n_prompts": 8, "identical": 2, "counts_of_n": "2/8", "lossless": false, "comparisons": [ { "prompt_index": 1, "identical": true, "len_on": 103, "len_off": 103, "first_divergence_char": null, "empty_side": null }, { "prompt_index": 2, "identical": false, "len_on": 0, "len_off": 255, "first_divergence_char": 0, "empty_side": "on" }, { "prompt_index": 3, "identical": true, "len_on": 255, "len_off": 255, "first_divergence_char": null, "empty_side": null }, { "prompt_index": 4, "identical": false, "len_on": 55, "len_off": 71, "first_divergence_char": 55, "empty_side": null }, { "prompt_index": 5, "identical": false, "len_on": 231, "len_off": 71, "first_divergence_char": 71, "empty_side": null }, { "prompt_index": 6, "identical": false, "len_on": 47, "len_off": 39, "first_divergence_char": 39, "empty_side": null }, { "prompt_index": 7, "identical": false, "len_on": 71, "len_off": 255, "first_divergence_char": 71, "empty_side": null }, { "prompt_index": 8, "identical": false, "len_on": 39, "len_off": 71, "first_divergence_char": 39, "empty_side": null } ] }, { "pass": 4, "n_calls": 16, "n_prompts": 8, "identical": 2, "counts_of_n": "2/8", "lossless": false, "comparisons": [ { "prompt_index": 1, "identical": true, "len_on": 103, "len_off": 103, "first_divergence_char": null, "empty_side": null }, { "prompt_index": 2, "identical": false, "len_on": 0, "len_off": 255, "first_divergence_char": 0, "empty_side": "on" }, { "prompt_index": 3, "identical": true, "len_on": 255, "len_off": 255, "first_divergence_char": null, "empty_side": null }, { "prompt_index": 4, "identical": false, "len_on": 55, "len_off": 71, "first_divergence_char": 55, "empty_side": null }, { "prompt_index": 5, "identical": false, "len_on": 231, "len_off": 71, "first_divergence_char": 71, "empty_side": null }, { "prompt_index": 6, "identical": false, "len_on": 47, "len_off": 39, "first_divergence_char": 39, "empty_side": null }, { "prompt_index": 7, "identical": false, "len_on": 71, "len_off": 255, "first_divergence_char": 71, "empty_side": null }, { "prompt_index": 8, "identical": false, "len_on": 39, "len_off": 71, "first_divergence_char": 39, "empty_side": null } ] }, { "pass": 5, "n_calls": 16, "n_prompts": 8, "identical": 2, "counts_of_n": "2/8", "lossless": false, "comparisons": [ { "prompt_index": 1, "identical": true, "len_on": 103, "len_off": 103, "first_divergence_char": null, "empty_side": null }, { "prompt_index": 2, "identical": false, "len_on": 0, "len_off": 255, "first_divergence_char": 0, "empty_side": "on" }, { "prompt_index": 3, "identical": true, "len_on": 255, "len_off": 255, "first_divergence_char": null, "empty_side": null }, { "prompt_index": 4, "identical": false, "len_on": 55, "len_off": 71, "first_divergence_char": 55, "empty_side": null }, { "prompt_index": 5, "identical": false, "len_on": 231, "len_off": 71, "first_divergence_char": 71, "empty_side": null }, { "prompt_index": 6, "identical": false, "len_on": 47, "len_off": 39, "first_divergence_char": 39, "empty_side": null }, { "prompt_index": 7, "identical": false, "len_on": 71, "len_off": 255, "first_divergence_char": 71, "empty_side": null }, { "prompt_index": 8, "identical": false, "len_on": 39, "len_off": 71, "first_divergence_char": 39, "empty_side": null } ] } ] }, "reasoning_strength_sweep": { "mechanism": "The dial is a SYSTEM-PROMPT LINE — 'Reasoning strength: low|medium|high|xhigh' — not a runtime parameter. think stays false on every sweep call, so the line is the only difference between the levels.", "reading": "The two arms do not read the same way and the sweep says so per arm. On the q4 build all three measured levels' ranges overlap, which is TIED under this leg's rule. On the q8 build they do not: medium sits entirely below high and xhigh, which overlap each other. The low level produced no usable counters on either arm and therefore has no figure at all — the dial's cheapest setting is the one we cannot price, and saying so is the honest version.", "response_failure_caveat": "Every sweep block that produced usable counters was ALSO three response failures out of three: valid counters, empty response text, done_reason=length. So these are timings of empty decodes. They are published because the counters are real and the absence would be worse, and they are labelled because a decode rate measured on a call that returned nothing is a weaker instrument than a decode rate measured on a call that returned an answer.", "arms": { "muse-glimmer:30b-q8_0-dflash": { "levels": { "low": { "level": "low", "occurrences": [ { "usable": "0/3", "median": null, "min": null, "max": null, "n": 0 }, { "usable": "0/3", "median": null, "min": null, "max": null, "n": 0 } ], "n": 0, "min": null, "max": null, "median": null, "response_failures": [ "0/3", "0/3" ], "measured": false }, "medium": { "level": "medium", "occurrences": [ { "usable": "3/3", "median": 79.99, "min": 79.94, "max": 81.75, "n": 3 } ], "n": 3, "min": 79.94, "max": 81.75, "median": 79.99, "response_failures": [ "3/3" ], "measured": true }, "high": { "level": "high", "occurrences": [ { "usable": "3/3", "median": 86.86, "min": 83.36, "max": 86.97, "n": 3 } ], "n": 3, "min": 83.36, "max": 86.97, "median": 86.86, "response_failures": [ "3/3" ], "measured": true }, "xhigh": { "level": "xhigh", "occurrences": [ { "usable": "3/3", "median": 84.29, "min": 83.51, "max": 84.35, "n": 3 } ], "n": 3, "min": 83.51, "max": 84.35, "median": 84.29, "response_failures": [ "3/3" ], "measured": true } }, "measured_levels": [ "medium", "high", "xhigh" ], "range_overlap": { "medium vs high": "DISJOINT — separated by this leg's rule", "medium vs xhigh": "DISJOINT — separated by this leg's rule", "high vs xhigh": "OVERLAP — TIED by this leg's rule" }, "display": "medium 79.94–81.75 n=3 · high 83.36–86.97 n=3 · xhigh 83.51–84.35 n=3" }, "muse-glimmer:30b-q4_K_M-dflash": { "levels": { "low": { "level": "low", "occurrences": [ { "usable": "0/3", "median": null, "min": null, "max": null, "n": 0 }, { "usable": "0/3", "median": null, "min": null, "max": null, "n": 0 } ], "n": 0, "min": null, "max": null, "median": null, "response_failures": [ "0/3", "0/3" ], "measured": false }, "medium": { "level": "medium", "occurrences": [ { "usable": "3/3", "median": 110.11, "min": 101.8, "max": 110.18, "n": 3 } ], "n": 3, "min": 101.8, "max": 110.18, "median": 110.11, "response_failures": [ "3/3" ], "measured": true }, "high": { "level": "high", "occurrences": [ { "usable": "3/3", "median": 102.34, "min": 102.28, "max": 105.34, "n": 3 } ], "n": 3, "min": 102.28, "max": 105.34, "median": 102.34, "response_failures": [ "3/3" ], "measured": true }, "xhigh": { "level": "xhigh", "occurrences": [ { "usable": "3/3", "median": 104.99, "min": 103.84, "max": 106.49, "n": 3 } ], "n": 3, "min": 103.84, "max": 106.49, "median": 104.99, "response_failures": [ "3/3" ], "measured": true } }, "measured_levels": [ "medium", "high", "xhigh" ], "range_overlap": { "medium vs high": "OVERLAP — TIED by this leg's rule", "medium vs xhigh": "OVERLAP — TIED by this leg's rule", "high vs xhigh": "OVERLAP — TIED by this leg's rule" }, "display": "medium 101.80–110.18 n=3 · high 102.28–105.34 n=3 · xhigh 103.84–106.49 n=3" } } }, "contention_disclosure": "The same workstation answered live production requests throughout. Live asks decoded at roughly 50 and 90 tokens per second mid-trial against a quiet-hours norm around 140, and every request completed — a human operator's own measurements, taken during the run. In the other direction: the pre-registered GIN-overlap column cannot be published, because confirming its journal pattern against a real line was a precondition and the pattern is still UNCONFIRMED. The per-block journal windows were stored either way, so a corrected pattern can be applied offline. An unconfirmed fence is reported UNMEASURED, which is not the same as clean.", "limits": [ "One box, one runtime version, one day, one batch configuration. The runner's argv and batch size were captured for all eleven tags and matched none of them, so the registered `-b 1024 / -c 32768` bracket has no captured value and is not published.", "Quantizations differ per row and are printed on the roster. Cross-quantization speed comparison is a confound we name, not one we correct for.", "TTFT-proxy is an upper bound, not TTFT: these calls were not streamed.", "The GIN-overlap column is UNMEASURED, not clean — see the contention disclosure.", "Two arms' cells are absent for a reason that is itself unexplained: calls that returned text with every counter null. We publish the absence and the count rather than a figure derived from the calls that happened to work." ], "provenance": { "artifacts": [ { "role": "the rebuilt matrix every figure here is copied out of", "artifact": "c5-matrix.json", "sha256": "3cce9422482fa2665464ab4ce25ce16eef3eee4cf51437d2ddec57f8bfbf2e3a" }, { "role": "the rebuild — zero GPU, a pure function of the banked raw blocks", "artifact": "c5_matrix.py", "sha256": "d7b86188d6bbc215eaa7aaa4a6c68939e67585cd71640f4b4dceec1629ce6afa" }, { "role": "the timing harness", "artifact": "c5_speed.py", "sha256": "495624474bc630afb24319c779ac645f8f1c59bef7a058735b1a537f89cb0319" }, { "role": "the registered denominator — this leg's arm roster", "artifact": "c5-roster.json", "sha256": "bc38e3fdfaf2ad49db69ffb69ee28e7c216ddee5b924235033d09636dffa482b" }, { "role": "pre-registration, committed before the first timed call", "artifact": "PREREG-C5.md", "sha256": "03e81d54f855b15f4a8a68bc3fff013e00dd0997c7cd621f61e7dc10a67bee45" } ], "how": "rebuilt from results/raw/C5/*/*.jsonl and golden/c5-roster.json. Zero GPU, no model calls, no summary file consulted.", "idempotent": "pure function of the raw directory and the frozen roster; re-running after the completion arms land changes only 'at' and the cells the new raw adds", "rebuilt_at": "2026-08-12T19:03:11-06:00", "registered_totals_note": "49 + 49 + 37 + 34 + 37 + 37 + 37 + 37 = 317 matrix calls; minus the 8 discarded warmups = 309; plus 16 DFlash byte-compare calls = 325 counted, 333 recorded.", "no_paths": "Artifacts are named, never located. No box names, no addresses, no filesystem paths." }, "rows": [ { "arm": "muse-glimmer:30b-q8_0-dflash", "registered": { "tag": "muse-glimmer:30b-q8_0-dflash", "warm": true, "cold_load": true, "strength_sweep": true, "dflash": null, "expected_calls": 49 }, "outcome": "UNMEASURABLE", "outcome_why": "response failures 38/77 reach the registered ceiling (10%; threshold 8 at this n)", "primary_tier": { "block_id": "warm-1000", "occurrences": 2, "usable": [ "0/10", "0/10" ], "median_decode_tok_s": [ null, null ], "measurable": false, "note": "the primary metric is decode tok/s at the 1k tier; an arm with 0 usable counters here has no primary figure whatever its outcome state says about its other tiers" }, "population": { "recorded_calls": 79, "counted_calls": 77, "discarded_warmups": 2, "registered_total_including_warmup": 49, "reconciles": false, "note": "the registered per-arm total INCLUDES that arm's one discarded warmup (golden/c5-roster.json). A completion re-block adds calls on purpose, so a re-blocked arm reconciles ABOVE its registered total and the occurrence list shows where." }, "blocks_present": [ "clean-baseline", "cold-load", "reasoning-high", "reasoning-low", "reasoning-medium", "reasoning-xhigh", "warm-1000", "warm-32000", "warm-8000", "warmup-discard" ], "blocks_repeated": { "warmup-discard": 2, "clean-baseline": 2, "warm-1000": 2, "warm-32000": 2, "cold-load": 2, "reasoning-low": 2 }, "block_occurrences": [ { "block_id": "warmup-discard", "block": "warmup-discard", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 0, "n_discarded": 1, "first_at": "2026-08-12T12:40:31-06:00", "last_at": "2026-08-12T12:40:31-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/0", "response_failures": 0, "response_failure_counts_of_n": "0/0", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "clean-baseline", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": "un-strengthened 1000-token block of 3 calls — the clean baseline by PREREG-C5's Population table; the warm set at the same tier is 10", "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:40:49-06:00", "last_at": "2026-08-12T12:40:56-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 3, "min": 3.214, "median": 3.224, "max": 3.252, "values": [ 3.214, 3.224, 3.252 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 3, "null_counter_split": { "total": 3, "with_response_text": 3, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-1000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:40:59-06:00", "last_at": "2026-08-12T12:41:28-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 10, "min": 3.207, "median": 3.216, "max": 3.243, "values": [ 3.207, 3.211, 3.212, 3.215, 3.216, 3.216, 3.218, 3.223, 3.231, 3.243 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 10, "null_counter_split": { "total": 10, "with_response_text": 10, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-8000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 8000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:41:31-06:00", "last_at": "2026-08-12T12:42:07-06:00", "decode_tok_s": { "n": 10, "min": 81.84, "median": 82.775, "max": 82.83, "values": [ 81.84, 82.05, 82.74, 82.75, 82.77, 82.78, 82.79, 82.79, 82.81, 82.83 ] }, "ttft_proxy_ms": { "n": 10, "min": 484.12, "median": 486.95, "max": 3229.73, "values": [ 484.12, 485.07, 485.28, 486.13, 486.79, 487.11, 489.31, 491.23, 499.4, 3229.73 ] }, "prompt_tok_s": { "n": 10, "min": 2899.33, "median": 206976.44, "max": 207983.03, "values": [ 2899.33, 187386.38, 206376.1, 206487.4, 206976.44, 206976.44, 207612.46, 207623.18, 207896.98, 207983.03 ] }, "wall_s": { "n": 10, "min": 3.615, "median": 3.62, "max": 6.397, "values": [ 3.615, 3.616, 3.618, 3.618, 3.619, 3.622, 3.623, 3.641, 3.646, 6.397 ] }, "load_s": { "n": 10, "min": 0.341, "median": 0.344, "max": 0.352, "values": [ 0.341, 0.341, 0.343, 0.343, 0.343, 0.345, 0.346, 0.346, 0.351, 0.352 ] }, "prompt_eval_count": { "n": 10, "min": 8040, "median": 8040.0, "max": 8040, "values": [ 8040, 8040, 8040, 8040, 8040, 8040, 8040, 8040, 8040, 8040 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 9, "response_failure_counts_of_n": "9/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 9, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-32000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 32000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:42:10-06:00", "last_at": "2026-08-12T12:42:30-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 10, "min": 1.352, "median": 1.36, "max": 9.138, "values": [ 1.352, 1.353, 1.359, 1.359, 1.36, 1.36, 1.369, 1.38, 1.383, 9.138 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/10", "response_failures": 10, "response_failure_counts_of_n": "10/10", "null_counter_calls": 10, "null_counter_split": { "total": 10, "with_response_text": 0, "empty_response_text": 10 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "cold-load", "block": "cold-load", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": null, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:42:32-06:00", "last_at": "2026-08-12T12:43:10-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 3, "min": 18.358, "median": 18.365, "max": 18.585, "values": [ 18.358, 18.365, 18.585 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 3, "null_counter_split": { "total": 3, "with_response_text": 3, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "reasoning-low", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": "low", "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:43:28-06:00", "last_at": "2026-08-12T12:43:34-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 3, "min": 2.855, "median": 2.864, "max": 3.112, "values": [ 2.855, 2.864, 3.112 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 3, "null_counter_split": { "total": 3, "with_response_text": 3, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "reasoning-medium", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": "medium", "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:43:37-06:00", "last_at": "2026-08-12T12:43:45-06:00", "decode_tok_s": { "n": 3, "min": 79.94, "median": 79.99, "max": 81.75, "values": [ 79.94, 79.99, 81.75 ] }, "ttft_proxy_ms": { "n": 3, "min": 419.74, "median": 429.48, "max": 746.05, "values": [ 419.74, 429.48, 746.05 ] }, "prompt_tok_s": { "n": 3, "min": 3085.66, "median": 14195.78, "max": 15257.9, "values": [ 3085.66, 14195.78, 15257.9 ] }, "wall_s": { "n": 3, "min": 3.64, "median": 3.641, "max": 3.896, "values": [ 3.64, 3.641, 3.896 ] }, "load_s": { "n": 3, "min": 0.339, "median": 0.342, "max": 0.344, "values": [ 0.339, 0.342, 0.344 ] }, "prompt_eval_count": { "n": 3, "min": 1017, "median": 1017, "max": 1017, "values": [ 1017, 1017, 1017 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 3, "response_failure_counts_of_n": "3/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 3, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "reasoning-high", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": "high", "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:43:48-06:00", "last_at": "2026-08-12T12:43:55-06:00", "decode_tok_s": { "n": 3, "min": 83.36, "median": 86.86, "max": 86.97, "values": [ 83.36, 86.86, 86.97 ] }, "ttft_proxy_ms": { "n": 3, "min": 419.89, "median": 427.98, "max": 749.15, "values": [ 419.89, 427.98, 749.15 ] }, "prompt_tok_s": { "n": 3, "min": 3073.58, "median": 15102.47, "max": 15207.02, "values": [ 3073.58, 15102.47, 15207.02 ] }, "wall_s": { "n": 3, "min": 3.378, "median": 3.381, "max": 3.83, "values": [ 3.378, 3.381, 3.83 ] }, "load_s": { "n": 3, "min": 0.339, "median": 0.342, "max": 0.347, "values": [ 0.339, 0.342, 0.347 ] }, "prompt_eval_count": { "n": 3, "min": 1017, "median": 1017, "max": 1017, "values": [ 1017, 1017, 1017 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 3, "response_failure_counts_of_n": "3/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 3, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "reasoning-xhigh", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": "xhigh", "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:43:59-06:00", "last_at": "2026-08-12T12:44:06-06:00", "decode_tok_s": { "n": 3, "min": 83.51, "median": 84.29, "max": 84.35, "values": [ 83.51, 84.29, 84.35 ] }, "ttft_proxy_ms": { "n": 3, "min": 420.22, "median": 423.98, "max": 747.49, "values": [ 420.22, 423.98, 747.49 ] }, "prompt_tok_s": { "n": 3, "min": 3069.73, "median": 15317.48, "max": 15342.19, "values": [ 3069.73, 15317.48, 15342.19 ] }, "wall_s": { "n": 3, "min": 3.467, "median": 3.468, "max": 3.823, "values": [ 3.467, 3.468, 3.823 ] }, "load_s": { "n": 3, "min": 0.34, "median": 0.34, "max": 0.344, "values": [ 0.34, 0.34, 0.344 ] }, "prompt_eval_count": { "n": 3, "min": 1018, "median": 1018, "max": 1018, "values": [ 1018, 1018, 1018 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 3, "response_failure_counts_of_n": "3/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 3, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warmup-discard", "block": "warmup-discard", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 0, "n_discarded": 1, "first_at": "2026-08-12T18:48:16-06:00", "last_at": "2026-08-12T18:48:16-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/0", "response_failures": 0, "response_failure_counts_of_n": "0/0", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "clean-baseline", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T18:48:35-06:00", "last_at": "2026-08-12T18:48:43-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 3, "min": 3.291, "median": 3.453, "max": 4.145, "values": [ 3.291, 3.453, 4.145 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 3, "null_counter_split": { "total": 3, "with_response_text": 3, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-1000", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T18:48:46-06:00", "last_at": "2026-08-12T18:49:16-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 10, "min": 3.212, "median": 3.22, "max": 3.743, "values": [ 3.212, 3.213, 3.216, 3.216, 3.218, 3.222, 3.226, 3.227, 3.233, 3.743 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 10, "null_counter_split": { "total": 10, "with_response_text": 10, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-32000", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 32000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T18:49:19-06:00", "last_at": "2026-08-12T18:49:42-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 10, "min": 1.342, "median": 1.353, "max": 12.027, "values": [ 1.342, 1.35, 1.351, 1.351, 1.352, 1.353, 1.356, 1.358, 1.359, 12.027 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/10", "response_failures": 10, "response_failure_counts_of_n": "10/10", "null_counter_calls": 10, "null_counter_split": { "total": 10, "with_response_text": 0, "empty_response_text": 10 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "cold-load", "block": "cold-load", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": null, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T18:49:44-06:00", "last_at": "2026-08-12T18:50:21-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 3, "min": 18.031, "median": 18.067, "max": 18.113, "values": [ 18.031, 18.067, 18.113 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 3, "null_counter_split": { "total": 3, "with_response_text": 3, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "reasoning-low", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": "low", "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T18:50:39-06:00", "last_at": "2026-08-12T18:50:45-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 3, "min": 2.89, "median": 2.916, "max": 3.139, "values": [ 2.89, 2.916, 3.139 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 3, "null_counter_split": { "total": 3, "with_response_text": 3, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" } ], "anomalies": { "null_counter_calls": 58, "null_counter_split": { "total": 58, "with_response_text": 38, "empty_response_text": 20 }, "null_counter_note": "HTTP 200, every counter and done_reason null. Excluded from every statistic by counting rule 4; cause unexplained — transport_error is null and nothing in the manifest or the logs accounts for it", "empty_length_calls": 18, "transport_errors": 0, "response_failures": "38/77", "ceiling": 0.1, "threshold_arithmetic": "floor(0.1 x 77) + 1 = 8 — the smallest count that exceeds the ceiling at this denominator" }, "contention": { "fence_status": "UNMEASURED", "fence_note": "UNMEASURED by construction: the GIN-overlap receipt is not in the raw JSONL, so a rebuild cannot evaluate the registered contention fence. It is not evidence the blocks were clean.", "gin_note": "PREREG-C5 makes confirming gin_overlap_regex against a real guide journal line a precondition for PUBLISHING the overlap column; the raw windows are stored either way so a corrected pattern can be re-applied offline", "overlapping_blocks": [] }, "not_recoverable_from_raw": [ "gin_overlap (per block) — queried at summary time, never written to raw", "runner_argv / batch_size — captured into harness/manifest.json, not raw" ], "failure_ledger": "response failures 38/77 · 58/77 null-counter (38 with text, 20 without) · 5 blocks re-run in the completion pass" }, { "arm": "muse-glimmer:30b-q4_K_M-dflash", "registered": { "tag": "muse-glimmer:30b-q4_K_M-dflash", "warm": true, "cold_load": true, "strength_sweep": true, "dflash": "on", "expected_calls": 49 }, "outcome": "UNMEASURABLE", "outcome_why": "response failures 47/71 reach the registered ceiling (10%; threshold 8 at this n)", "primary_tier": { "block_id": "warm-1000", "occurrences": 1, "usable": [ "10/10" ], "median_decode_tok_s": [ 110.52 ], "measurable": true, "note": "the primary metric is decode tok/s at the 1k tier; an arm with 0 usable counters here has no primary figure whatever its outcome state says about its other tiers" }, "population": { "recorded_calls": 73, "counted_calls": 71, "discarded_warmups": 2, "registered_total_including_warmup": 49, "reconciles": false, "note": "the registered per-arm total INCLUDES that arm's one discarded warmup (golden/c5-roster.json). A completion re-block adds calls on purpose, so a re-blocked arm reconciles ABOVE its registered total and the occurrence list shows where." }, "blocks_present": [ "clean-baseline", "cold-load", "reasoning-high", "reasoning-low", "reasoning-medium", "reasoning-xhigh", "warm-1000", "warm-32000", "warm-8000", "warmup-discard" ], "blocks_repeated": { "warmup-discard": 2, "warm-8000": 2, "warm-32000": 2, "reasoning-low": 2 }, "block_occurrences": [ { "block_id": "warmup-discard", "block": "warmup-discard", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 0, "n_discarded": 1, "first_at": "2026-08-12T12:45:35-06:00", "last_at": "2026-08-12T12:45:35-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/0", "response_failures": 0, "response_failure_counts_of_n": "0/0", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "clean-baseline", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": "un-strengthened 1000-token block of 3 calls — the clean baseline by PREREG-C5's Population table; the warm set at the same tier is 10", "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:45:43-06:00", "last_at": "2026-08-12T12:45:49-06:00", "decode_tok_s": { "n": 3, "min": 97.57, "median": 108.96, "max": 110.3, "values": [ 97.57, 108.96, 110.3 ] }, "ttft_proxy_ms": { "n": 3, "min": 436.52, "median": 437.29, "max": 484.39, "values": [ 436.52, 437.29, 484.39 ] }, "prompt_tok_s": { "n": 3, "min": 13488.98, "median": 13693.22, "max": 13871.29, "values": [ 13488.98, 13693.22, 13871.29 ] }, "wall_s": { "n": 3, "min": 2.767, "median": 2.797, "max": 3.119, "values": [ 2.767, 2.797, 3.119 ] }, "load_s": { "n": 3, "min": 0.346, "median": 0.349, "max": 0.394, "values": [ 0.346, 0.349, 0.394 ] }, "prompt_eval_count": { "n": 3, "min": 1040, "median": 1040, "max": 1040, "values": [ 1040, 1040, 1040 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 3, "response_failure_counts_of_n": "3/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 3, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-1000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:45:52-06:00", "last_at": "2026-08-12T12:46:16-06:00", "decode_tok_s": { "n": 10, "min": 108.88, "median": 110.52, "max": 110.59, "values": [ 108.88, 110.21, 110.25, 110.42, 110.52, 110.52, 110.52, 110.53, 110.58, 110.59 ] }, "ttft_proxy_ms": { "n": 10, "min": 427.98, "median": 432.755, "max": 438.31, "values": [ 427.98, 429.05, 430.16, 431.01, 432.75, 432.76, 436.18, 437.0, 437.13, 438.31 ] }, "prompt_tok_s": { "n": 10, "min": 13501.93, "median": 13615.16, "max": 13693.22, "values": [ 13501.93, 13515.62, 13566.57, 13594.77, 13603.84, 13626.48, 13652.95, 13656.54, 13664.61, 13693.22 ] }, "wall_s": { "n": 10, "min": 2.753, "median": 2.763, "max": 2.799, "values": [ 2.753, 2.756, 2.757, 2.76, 2.762, 2.764, 2.764, 2.766, 2.775, 2.799 ] }, "load_s": { "n": 10, "min": 0.338, "median": 0.343, "max": 0.347, "values": [ 0.338, 0.339, 0.34, 0.341, 0.342, 0.343, 0.346, 0.346, 0.346, 0.347 ] }, "prompt_eval_count": { "n": 10, "min": 1040, "median": 1040.0, "max": 1040, "values": [ 1040, 1040, 1040, 1040, 1040, 1040, 1040, 1040, 1040, 1040 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 10, "response_failure_counts_of_n": "10/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 10, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-8000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 8000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:46:19-06:00", "last_at": "2026-08-12T12:46:49-06:00", "decode_tok_s": { "n": 1, "min": 92.87, "median": 92.87, "max": 92.87, "values": [ 92.87 ] }, "ttft_proxy_ms": { "n": 1, "min": 3358.32, "median": 3358.32, "max": 3358.32, "values": [ 3358.32 ] }, "prompt_tok_s": { "n": 1, "min": 2767.85, "median": 2767.85, "max": 2767.85, "values": [ 2767.85 ] }, "wall_s": { "n": 10, "min": 2.886, "median": 2.894, "max": 6.155, "values": [ 2.886, 2.89, 2.891, 2.893, 2.894, 2.895, 2.897, 2.897, 2.929, 6.155 ] }, "load_s": { "n": 1, "min": 0.348, "median": 0.348, "max": 0.348, "values": [ 0.348 ] }, "prompt_eval_count": { "n": 1, "min": 8040, "median": 8040, "max": 8040, "values": [ 8040 ] }, "usable_counter_calls": 1, "usable_counts_of_n": "1/10", "response_failures": 1, "response_failure_counts_of_n": "1/10", "null_counter_calls": 9, "null_counter_split": { "total": 9, "with_response_text": 9, "empty_response_text": 0 }, "empty_length_calls": 1, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-32000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 32000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:46:51-06:00", "last_at": "2026-08-12T12:47:11-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 10, "min": 1.226, "median": 1.236, "max": 9.362, "values": [ 1.226, 1.232, 1.232, 1.234, 1.236, 1.237, 1.238, 1.241, 1.244, 9.362 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/10", "response_failures": 10, "response_failure_counts_of_n": "10/10", "null_counter_calls": 10, "null_counter_split": { "total": 10, "with_response_text": 0, "empty_response_text": 10 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "cold-load", "block": "cold-load", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": null, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:47:12-06:00", "last_at": "2026-08-12T12:47:28-06:00", "decode_tok_s": { "n": 3, "min": 104.81, "median": 105.35, "max": 106.61, "values": [ 104.81, 105.35, 106.61 ] }, "ttft_proxy_ms": { "n": 3, "min": 5293.34, "median": 5308.15, "max": 5318.51, "values": [ 5293.34, 5308.15, 5318.51 ] }, "prompt_tok_s": { "n": 3, "min": 1430.92, "median": 1466.03, "max": 1466.78, "values": [ 1430.92, 1466.03, 1466.78 ] }, "wall_s": { "n": 3, "min": 7.721, "median": 7.737, "max": 7.772, "values": [ 7.721, 7.737, 7.772 ] }, "load_s": { "n": 3, "min": 4.57, "median": 4.577, "max": 4.585, "values": [ 4.57, 4.577, 4.585 ] }, "prompt_eval_count": { "n": 3, "min": 1040, "median": 1040, "max": 1040, "values": [ 1040, 1040, 1040 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 3, "response_failure_counts_of_n": "3/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 3, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "reasoning-low", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": "low", "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:47:36-06:00", "last_at": "2026-08-12T12:47:41-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 3, "min": 2.266, "median": 2.283, "max": 2.945, "values": [ 2.266, 2.283, 2.945 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 3, "null_counter_split": { "total": 3, "with_response_text": 3, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "reasoning-medium", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": "medium", "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:47:43-06:00", "last_at": "2026-08-12T12:47:49-06:00", "decode_tok_s": { "n": 3, "min": 101.8, "median": 110.11, "max": 110.18, "values": [ 101.8, 110.11, 110.18 ] }, "ttft_proxy_ms": { "n": 3, "min": 405.59, "median": 418.36, "max": 762.07, "values": [ 405.59, 418.36, 762.07 ] }, "prompt_tok_s": { "n": 3, "min": 3048.05, "median": 19203.53, "max": 19321.38, "values": [ 3048.05, 19203.53, 19321.38 ] }, "wall_s": { "n": 3, "min": 2.741, "median": 2.752, "max": 3.287, "values": [ 2.741, 2.752, 3.287 ] }, "load_s": { "n": 3, "min": 0.339, "median": 0.344, "max": 0.354, "values": [ 0.339, 0.344, 0.354 ] }, "prompt_eval_count": { "n": 3, "min": 1017, "median": 1017, "max": 1017, "values": [ 1017, 1017, 1017 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 3, "response_failure_counts_of_n": "3/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 3, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "reasoning-high", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": "high", "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:47:52-06:00", "last_at": "2026-08-12T12:47:58-06:00", "decode_tok_s": { "n": 3, "min": 102.28, "median": 102.34, "max": 105.34, "values": [ 102.28, 102.34, 105.34 ] }, "ttft_proxy_ms": { "n": 3, "min": 405.6, "median": 410.47, "max": 752.29, "values": [ 405.6, 410.47, 752.29 ] }, "prompt_tok_s": { "n": 3, "min": 3041.06, "median": 19311.47, "max": 19343.79, "values": [ 3041.06, 19311.47, 19343.79 ] }, "wall_s": { "n": 3, "min": 2.918, "median": 2.924, "max": 3.193, "values": [ 2.918, 2.924, 3.193 ] }, "load_s": { "n": 3, "min": 0.34, "median": 0.342, "max": 0.344, "values": [ 0.34, 0.342, 0.344 ] }, "prompt_eval_count": { "n": 3, "min": 1017, "median": 1017, "max": 1017, "values": [ 1017, 1017, 1017 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 3, "response_failure_counts_of_n": "3/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 3, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "reasoning-xhigh", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": "xhigh", "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:48:01-06:00", "last_at": "2026-08-12T12:48:07-06:00", "decode_tok_s": { "n": 3, "min": 103.84, "median": 104.99, "max": 106.49, "values": [ 103.84, 104.99, 106.49 ] }, "ttft_proxy_ms": { "n": 3, "min": 407.35, "median": 416.56, "max": 746.77, "values": [ 407.35, 416.56, 746.77 ] }, "prompt_tok_s": { "n": 3, "min": 3045.73, "median": 17756.23, "max": 19445.29, "values": [ 3045.73, 17756.23, 19445.29 ] }, "wall_s": { "n": 3, "min": 2.856, "median": 2.893, "max": 3.162, "values": [ 2.856, 2.893, 3.162 ] }, "load_s": { "n": 3, "min": 0.338, "median": 0.341, "max": 0.346, "values": [ 0.338, 0.341, 0.346 ] }, "prompt_eval_count": { "n": 3, "min": 1018, "median": 1018, "max": 1018, "values": [ 1018, 1018, 1018 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 3, "response_failure_counts_of_n": "3/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 3, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warmup-discard", "block": "warmup-discard", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 0, "n_discarded": 1, "first_at": "2026-08-12T18:50:54-06:00", "last_at": "2026-08-12T18:50:54-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/0", "response_failures": 0, "response_failure_counts_of_n": "0/0", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-8000", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 8000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T18:51:06-06:00", "last_at": "2026-08-12T18:51:35-06:00", "decode_tok_s": { "n": 1, "min": 93.42, "median": 93.42, "max": 93.42, "values": [ 93.42 ] }, "ttft_proxy_ms": { "n": 1, "min": 3331.72, "median": 3331.72, "max": 3331.72, "values": [ 3331.72 ] }, "prompt_tok_s": { "n": 1, "min": 2793.1, "median": 2793.1, "max": 2793.1, "values": [ 2793.1 ] }, "wall_s": { "n": 10, "min": 2.878, "median": 2.883, "max": 6.112, "values": [ 2.878, 2.878, 2.88, 2.882, 2.883, 2.884, 2.885, 2.888, 2.895, 6.112 ] }, "load_s": { "n": 1, "min": 0.348, "median": 0.348, "max": 0.348, "values": [ 0.348 ] }, "prompt_eval_count": { "n": 1, "min": 8040, "median": 8040, "max": 8040, "values": [ 8040 ] }, "usable_counter_calls": 1, "usable_counts_of_n": "1/10", "response_failures": 1, "response_failure_counts_of_n": "1/10", "null_counter_calls": 9, "null_counter_split": { "total": 9, "with_response_text": 9, "empty_response_text": 0 }, "empty_length_calls": 1, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-32000", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 32000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T18:51:38-06:00", "last_at": "2026-08-12T18:51:57-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 10, "min": 1.224, "median": 1.237, "max": 9.323, "values": [ 1.224, 1.231, 1.231, 1.234, 1.236, 1.238, 1.246, 1.276, 1.293, 9.323 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/10", "response_failures": 10, "response_failure_counts_of_n": "10/10", "null_counter_calls": 10, "null_counter_split": { "total": 10, "with_response_text": 0, "empty_response_text": 10 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "reasoning-low", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": "low", "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T18:51:59-06:00", "last_at": "2026-08-12T18:52:04-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 3, "min": 2.264, "median": 2.274, "max": 3.081, "values": [ 2.264, 2.274, 3.081 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 3, "null_counter_split": { "total": 3, "with_response_text": 3, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" } ], "anomalies": { "null_counter_calls": 44, "null_counter_split": { "total": 44, "with_response_text": 24, "empty_response_text": 20 }, "null_counter_note": "HTTP 200, every counter and done_reason null. Excluded from every statistic by counting rule 4; cause unexplained — transport_error is null and nothing in the manifest or the logs accounts for it", "empty_length_calls": 27, "transport_errors": 0, "response_failures": "47/71", "ceiling": 0.1, "threshold_arithmetic": "floor(0.1 x 71) + 1 = 8 — the smallest count that exceeds the ceiling at this denominator" }, "contention": { "fence_status": "UNMEASURED", "fence_note": "UNMEASURED by construction: the GIN-overlap receipt is not in the raw JSONL, so a rebuild cannot evaluate the registered contention fence. It is not evidence the blocks were clean.", "gin_note": "PREREG-C5 makes confirming gin_overlap_regex against a real guide journal line a precondition for PUBLISHING the overlap column; the raw windows are stored either way so a corrected pattern can be re-applied offline", "overlapping_blocks": [] }, "not_recoverable_from_raw": [ "gin_overlap (per block) — queried at summary time, never written to raw", "runner_argv / batch_size — captured into harness/manifest.json, not raw" ], "failure_ledger": "response failures 47/71 · 44/71 null-counter (24 with text, 20 without) · 3 blocks re-run in the completion pass" }, { "arm": "muse-glimmer:30b", "registered": { "tag": "muse-glimmer:30b", "warm": true, "cold_load": true, "strength_sweep": false, "dflash": "off", "expected_calls": 37 }, "outcome": "UNMEASURABLE", "outcome_why": "response failures 21/49 reach the registered ceiling (10%; threshold 5 at this n)", "primary_tier": { "block_id": "warm-1000", "occurrences": 1, "usable": [ "10/10" ], "median_decode_tok_s": [ 68.195 ], "measurable": true, "note": "the primary metric is decode tok/s at the 1k tier; an arm with 0 usable counters here has no primary figure whatever its outcome state says about its other tiers" }, "population": { "recorded_calls": 51, "counted_calls": 49, "discarded_warmups": 2, "registered_total_including_warmup": 37, "reconciles": false, "note": "the registered per-arm total INCLUDES that arm's one discarded warmup (golden/c5-roster.json). A completion re-block adds calls on purpose, so a re-blocked arm reconciles ABOVE its registered total and the occurrence list shows where." }, "blocks_present": [ "clean-baseline", "cold-load", "warm-1000", "warm-32000", "warm-8000", "warmup-discard" ], "blocks_repeated": { "warmup-discard": 2, "warm-32000": 2, "cold-load": 2 }, "block_occurrences": [ { "block_id": "warmup-discard", "block": "warmup-discard", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 0, "n_discarded": 1, "first_at": "2026-08-12T12:49:21-06:00", "last_at": "2026-08-12T12:49:21-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/0", "response_failures": 0, "response_failure_counts_of_n": "0/0", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "clean-baseline", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": "un-strengthened 1000-token block of 3 calls — the clean baseline by PREREG-C5's Population table; the warm set at the same tier is 10", "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:49:29-06:00", "last_at": "2026-08-12T12:49:37-06:00", "decode_tok_s": { "n": 3, "min": 68.23, "median": 68.23, "max": 68.24, "values": [ 68.23, 68.23, 68.24 ] }, "ttft_proxy_ms": { "n": 3, "min": 404.76, "median": 404.97, "max": 405.77, "values": [ 404.76, 404.97, 405.77 ] }, "prompt_tok_s": { "n": 3, "min": 20886.47, "median": 20914.19, "max": 21188.6, "values": [ 20886.47, 20914.19, 21188.6 ] }, "wall_s": { "n": 3, "min": 4.168, "median": 4.168, "max": 4.169, "values": [ 4.168, 4.168, 4.169 ] }, "load_s": { "n": 3, "min": 0.341, "median": 0.342, "max": 0.342, "values": [ 0.341, 0.342, 0.342 ] }, "prompt_eval_count": { "n": 3, "min": 1040, "median": 1040, "max": 1040, "values": [ 1040, 1040, 1040 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-1000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:49:41-06:00", "last_at": "2026-08-12T12:50:19-06:00", "decode_tok_s": { "n": 10, "min": 68.15, "median": 68.195, "max": 68.22, "values": [ 68.15, 68.17, 68.18, 68.19, 68.19, 68.2, 68.21, 68.22, 68.22, 68.22 ] }, "ttft_proxy_ms": { "n": 10, "min": 402.99, "median": 409.17, "max": 411.52, "values": [ 402.99, 405.52, 405.67, 408.58, 408.93, 409.41, 411.07, 411.33, 411.47, 411.52 ] }, "prompt_tok_s": { "n": 10, "min": 20487.77, "median": 20573.895, "max": 20844.61, "values": [ 20487.77, 20496.65, 20512.42, 20547.27, 20565.96, 20581.83, 20687.87, 20707.64, 20740.68, 20844.61 ] }, "wall_s": { "n": 10, "min": 4.17, "median": 4.175, "max": 4.178, "values": [ 4.17, 4.17, 4.172, 4.174, 4.175, 4.175, 4.175, 4.175, 4.175, 4.178 ] }, "load_s": { "n": 10, "min": 0.339, "median": 0.344, "max": 0.347, "values": [ 0.339, 0.341, 0.341, 0.344, 0.344, 0.345, 0.346, 0.347, 0.347, 0.347 ] }, "prompt_eval_count": { "n": 10, "min": 1040, "median": 1040.0, "max": 1040, "values": [ 1040, 1040, 1040, 1040, 1040, 1040, 1040, 1040, 1040, 1040 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-8000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 8000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:50:23-06:00", "last_at": "2026-08-12T12:51:05-06:00", "decode_tok_s": { "n": 10, "min": 65.89, "median": 65.985, "max": 66.06, "values": [ 65.89, 65.91, 65.97, 65.97, 65.98, 65.99, 66.01, 66.01, 66.05, 66.06 ] }, "ttft_proxy_ms": { "n": 10, "min": 484.48, "median": 488.04, "max": 2222.34, "values": [ 484.48, 486.0, 486.41, 487.03, 487.66, 488.42, 489.7, 489.82, 490.33, 2222.34 ] }, "prompt_tok_s": { "n": 10, "min": 4531.36, "median": 209508.705, "max": 211523.28, "values": [ 4531.36, 202656.72, 208160.73, 208322.54, 209424.08, 209593.33, 209910.71, 210036.83, 211439.84, 211523.28 ] }, "wall_s": { "n": 10, "min": 4.404, "median": 4.409, "max": 6.151, "values": [ 4.404, 4.405, 4.406, 4.407, 4.409, 4.409, 4.415, 4.416, 4.418, 6.151 ] }, "load_s": { "n": 10, "min": 0.341, "median": 0.344, "max": 0.347, "values": [ 0.341, 0.342, 0.343, 0.343, 0.344, 0.344, 0.344, 0.346, 0.346, 0.347 ] }, "prompt_eval_count": { "n": 10, "min": 8040, "median": 8040.0, "max": 8040, "values": [ 8040, 8040, 8040, 8040, 8040, 8040, 8040, 8040, 8040, 8040 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 1, "response_failure_counts_of_n": "1/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 1, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-32000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 32000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:51:09-06:00", "last_at": "2026-08-12T12:51:28-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 10, "min": 1.437, "median": 1.44, "max": 7.074, "values": [ 1.437, 1.439, 1.439, 1.44, 1.44, 1.44, 1.441, 1.442, 1.445, 7.074 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/10", "response_failures": 10, "response_failure_counts_of_n": "10/10", "null_counter_calls": 10, "null_counter_split": { "total": 10, "with_response_text": 0, "empty_response_text": 10 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "cold-load", "block": "cold-load", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": null, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:51:29-06:00", "last_at": "2026-08-12T12:51:45-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 3, "min": 7.607, "median": 7.608, "max": 7.62, "values": [ 7.607, 7.608, 7.62 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 3, "null_counter_split": { "total": 3, "with_response_text": 3, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warmup-discard", "block": "warmup-discard", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 0, "n_discarded": 1, "first_at": "2026-08-12T18:52:11-06:00", "last_at": "2026-08-12T18:52:11-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/0", "response_failures": 0, "response_failure_counts_of_n": "0/0", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-32000", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 32000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T18:52:20-06:00", "last_at": "2026-08-12T18:52:40-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 10, "min": 1.432, "median": 1.436, "max": 8.679, "values": [ 1.432, 1.433, 1.434, 1.435, 1.436, 1.436, 1.44, 1.441, 1.443, 8.679 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/10", "response_failures": 10, "response_failure_counts_of_n": "10/10", "null_counter_calls": 10, "null_counter_split": { "total": 10, "with_response_text": 0, "empty_response_text": 10 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "cold-load", "block": "cold-load", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": null, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T18:52:41-06:00", "last_at": "2026-08-12T18:52:58-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 3, "min": 7.992, "median": 8.101, "max": 19.825, "values": [ 7.992, 8.101, 19.825 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 3, "null_counter_split": { "total": 3, "with_response_text": 3, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" } ], "anomalies": { "null_counter_calls": 26, "null_counter_split": { "total": 26, "with_response_text": 6, "empty_response_text": 20 }, "null_counter_note": "HTTP 200, every counter and done_reason null. Excluded from every statistic by counting rule 4; cause unexplained — transport_error is null and nothing in the manifest or the logs accounts for it", "empty_length_calls": 1, "transport_errors": 0, "response_failures": "21/49", "ceiling": 0.1, "threshold_arithmetic": "floor(0.1 x 49) + 1 = 5 — the smallest count that exceeds the ceiling at this denominator" }, "contention": { "fence_status": "UNMEASURED", "fence_note": "UNMEASURED by construction: the GIN-overlap receipt is not in the raw JSONL, so a rebuild cannot evaluate the registered contention fence. It is not evidence the blocks were clean.", "gin_note": "PREREG-C5 makes confirming gin_overlap_regex against a real guide journal line a precondition for PUBLISHING the overlap column; the raw windows are stored either way so a corrected pattern can be re-applied offline", "overlapping_blocks": [] }, "not_recoverable_from_raw": [ "gin_overlap (per block) — queried at summary time, never written to raw", "runner_argv / batch_size — captured into harness/manifest.json, not raw" ], "failure_ledger": "response failures 21/49 · 26/49 null-counter (6 with text, 20 without) · 2 blocks re-run in the completion pass" }, { "arm": "gemma4:26b", "registered": { "tag": "gemma4:26b", "warm": true, "cold_load": false, "strength_sweep": false, "dflash": null, "expected_calls": 34, "not_run_reason": "prod resident; unloading the live seat is out of contract." }, "outcome": "RANKED", "outcome_why": "carried under protocol", "primary_tier": { "block_id": "warm-1000", "occurrences": 2, "usable": [ "0/10", "0/10" ], "median_decode_tok_s": [ null, null ], "measurable": false, "note": "the primary metric is decode tok/s at the 1k tier; an arm with 0 usable counters here has no primary figure whatever its outcome state says about its other tiers" }, "population": { "recorded_calls": 58, "counted_calls": 56, "discarded_warmups": 2, "registered_total_including_warmup": 34, "reconciles": false, "note": "the registered per-arm total INCLUDES that arm's one discarded warmup (golden/c5-roster.json). A completion re-block adds calls on purpose, so a re-blocked arm reconciles ABOVE its registered total and the occurrence list shows where." }, "blocks_present": [ "clean-baseline", "warm-1000", "warm-32000", "warm-8000", "warmup-discard" ], "blocks_repeated": { "warmup-discard": 2, "clean-baseline": 2, "warm-1000": 2, "warm-8000": 2 }, "block_occurrences": [ { "block_id": "warmup-discard", "block": "warmup-discard", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 0, "n_discarded": 1, "first_at": "2026-08-12T12:53:02-06:00", "last_at": "2026-08-12T12:53:02-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/0", "response_failures": 0, "response_failure_counts_of_n": "0/0", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "clean-baseline", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": "un-strengthened 1000-token block of 3 calls — the clean baseline by PREREG-C5's Population table; the warm set at the same tier is 10", "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:53:03-06:00", "last_at": "2026-08-12T12:53:05-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 3, "min": 0.832, "median": 0.848, "max": 0.854, "values": [ 0.832, 0.848, 0.854 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 3, "null_counter_split": { "total": 3, "with_response_text": 3, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-1000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:53:06-06:00", "last_at": "2026-08-12T12:53:13-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 10, "min": 0.823, "median": 0.835, "max": 0.845, "values": [ 0.823, 0.825, 0.827, 0.834, 0.835, 0.836, 0.837, 0.842, 0.843, 0.845 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 10, "null_counter_split": { "total": 10, "with_response_text": 10, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-8000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 8000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:53:14-06:00", "last_at": "2026-08-12T12:53:23-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 10, "min": 0.877, "median": 0.889, "max": 1.374, "values": [ 0.877, 0.884, 0.885, 0.888, 0.888, 0.889, 0.89, 0.896, 0.898, 1.374 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 10, "null_counter_split": { "total": 10, "with_response_text": 10, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-32000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 32000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:53:24-06:00", "last_at": "2026-08-12T12:53:36-06:00", "decode_tok_s": { "n": 10, "min": 103.18, "median": 124.975, "max": 125.4, "values": [ 103.18, 124.19, 124.28, 124.51, 124.88, 125.07, 125.28, 125.33, 125.34, 125.4 ] }, "ttft_proxy_ms": { "n": 10, "min": 959.06, "median": 982.395, "max": 2660.38, "values": [ 959.06, 964.0, 968.47, 979.71, 981.81, 982.98, 984.57, 989.95, 1022.04, 2660.38 ] }, "prompt_tok_s": { "n": 10, "min": 18470.14, "median": 556973.31, "max": 559368.55, "values": [ 18470.14, 551938.87, 555252.84, 555879.85, 556808.49, 557138.13, 558587.34, 558723.89, 558860.52, 559368.55 ] }, "wall_s": { "n": 10, "min": 1.171, "median": 1.196, "max": 2.9, "values": [ 1.171, 1.172, 1.18, 1.188, 1.194, 1.199, 1.2, 1.21, 1.238, 2.9 ] }, "load_s": { "n": 10, "min": 0.461, "median": 0.469, "max": 0.473, "values": [ 0.461, 0.464, 0.465, 0.466, 0.468, 0.469, 0.47, 0.47, 0.472, 0.473 ] }, "prompt_eval_count": { "n": 10, "min": 31997, "median": 31997.0, "max": 31997, "values": [ 31997, 31997, 31997, 31997, 31997, 31997, 31997, 31997, 31997, 31997 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warmup-discard", "block": "warmup-discard", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 0, "n_discarded": 1, "first_at": "2026-08-12T18:58:59-06:00", "last_at": "2026-08-12T18:58:59-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/0", "response_failures": 0, "response_failure_counts_of_n": "0/0", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "clean-baseline", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T18:58:59-06:00", "last_at": "2026-08-12T18:59:01-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 3, "min": 0.806, "median": 0.811, "max": 0.818, "values": [ 0.806, 0.811, 0.818 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 3, "null_counter_split": { "total": 3, "with_response_text": 3, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-1000", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T18:59:02-06:00", "last_at": "2026-08-12T18:59:09-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 10, "min": 0.813, "median": 0.819, "max": 0.822, "values": [ 0.813, 0.816, 0.818, 0.818, 0.819, 0.819, 0.82, 0.821, 0.821, 0.822 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 10, "null_counter_split": { "total": 10, "with_response_text": 10, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-8000", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 8000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T18:59:10-06:00", "last_at": "2026-08-12T18:59:18-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 10, "min": 0.87, "median": 0.879, "max": 1.346, "values": [ 0.87, 0.872, 0.875, 0.877, 0.878, 0.88, 0.881, 0.882, 0.883, 1.346 ] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 10, "null_counter_split": { "total": 10, "with_response_text": 10, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" } ], "anomalies": { "null_counter_calls": 46, "null_counter_split": { "total": 46, "with_response_text": 46, "empty_response_text": 0 }, "null_counter_note": "HTTP 200, every counter and done_reason null. Excluded from every statistic by counting rule 4; cause unexplained — transport_error is null and nothing in the manifest or the logs accounts for it", "empty_length_calls": 0, "transport_errors": 0, "response_failures": "0/56", "ceiling": 0.1, "threshold_arithmetic": "floor(0.1 x 56) + 1 = 6 — the smallest count that exceeds the ceiling at this denominator" }, "contention": { "fence_status": "UNMEASURED", "fence_note": "UNMEASURED by construction: the GIN-overlap receipt is not in the raw JSONL, so a rebuild cannot evaluate the registered contention fence. It is not evidence the blocks were clean.", "gin_note": "PREREG-C5 makes confirming gin_overlap_regex against a real guide journal line a precondition for PUBLISHING the overlap column; the raw windows are stored either way so a corrected pattern can be re-applied offline", "overlapping_blocks": [] }, "not_recoverable_from_raw": [ "gin_overlap (per block) — queried at summary time, never written to raw", "runner_argv / batch_size — captured into harness/manifest.json, not raw" ], "failure_ledger": "response failures 0/56 · 46/56 null-counter (46 with text, 0 without) · 3 blocks re-run in the completion pass · cold-load NOT-RUN — prod resident; unloading the live seat is out of contract." }, { "arm": "gemma4:12b", "registered": { "tag": "gemma4:12b", "warm": true, "cold_load": true, "strength_sweep": false, "dflash": null, "expected_calls": 37 }, "outcome": "RANKED", "outcome_why": "carried under protocol", "primary_tier": { "block_id": "warm-1000", "occurrences": 1, "usable": [ "10/10" ], "median_decode_tok_s": [ 105.665 ], "measurable": true, "note": "the primary metric is decode tok/s at the 1k tier; an arm with 0 usable counters here has no primary figure whatever its outcome state says about its other tiers" }, "population": { "recorded_calls": 36, "counted_calls": 34, "discarded_warmups": 2, "registered_total_including_warmup": 37, "reconciles": false, "note": "the registered per-arm total INCLUDES that arm's one discarded warmup (golden/c5-roster.json). A completion re-block adds calls on purpose, so a re-blocked arm reconciles ABOVE its registered total and the occurrence list shows where." }, "blocks_present": [ "clean-baseline", "cold-load", "warm-1000", "warm-32000", "warm-8000", "warmup-discard" ], "blocks_repeated": { "warmup-discard": 2 }, "block_occurrences": [ { "block_id": "warmup-discard", "block": "warmup-discard", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 0, "n_discarded": 1, "first_at": "2026-08-12T12:54:48-06:00", "last_at": "2026-08-12T12:54:48-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/0", "response_failures": 0, "response_failure_counts_of_n": "0/0", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "clean-baseline", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": "un-strengthened 1000-token block of 3 calls — the clean baseline by PREREG-C5's Population table; the warm set at the same tier is 10", "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:54:51-06:00", "last_at": "2026-08-12T12:54:53-06:00", "decode_tok_s": { "n": 3, "min": 105.51, "median": 105.65, "max": 105.79, "values": [ 105.51, 105.65, 105.79 ] }, "ttft_proxy_ms": { "n": 3, "min": 630.29, "median": 637.04, "max": 638.34, "values": [ 630.29, 637.04, 638.34 ] }, "prompt_tok_s": { "n": 3, "min": 5705.95, "median": 5756.28, "max": 6038.22, "values": [ 5705.95, 5756.28, 6038.22 ] }, "wall_s": { "n": 3, "min": 0.704, "median": 0.711, "max": 0.712, "values": [ 0.704, 0.711, 0.712 ] }, "load_s": { "n": 3, "min": 0.453, "median": 0.455, "max": 0.455, "values": [ 0.453, 0.455, 0.455 ] }, "prompt_eval_count": { "n": 3, "min": 997, "median": 997, "max": 997, "values": [ 997, 997, 997 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-1000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:54:53-06:00", "last_at": "2026-08-12T12:55:00-06:00", "decode_tok_s": { "n": 10, "min": 104.88, "median": 105.665, "max": 105.98, "values": [ 104.88, 105.31, 105.41, 105.45, 105.6, 105.73, 105.83, 105.83, 105.91, 105.98 ] }, "ttft_proxy_ms": { "n": 10, "min": 631.31, "median": 637.35, "max": 641.18, "values": [ 631.31, 633.6, 635.37, 636.83, 636.87, 637.83, 638.4, 639.07, 639.4, 641.18 ] }, "prompt_tok_s": { "n": 10, "min": 5713.89, "median": 5735.575, "max": 5765.91, "values": [ 5713.89, 5716.97, 5719.5, 5724.19, 5733.25, 5737.9, 5739.85, 5755.25, 5759.78, 5765.91 ] }, "wall_s": { "n": 10, "min": 0.704, "median": 0.71, "max": 0.715, "values": [ 0.704, 0.707, 0.708, 0.71, 0.71, 0.711, 0.711, 0.712, 0.713, 0.715 ] }, "load_s": { "n": 10, "min": 0.449, "median": 0.455, "max": 0.458, "values": [ 0.449, 0.452, 0.452, 0.454, 0.454, 0.455, 0.455, 0.455, 0.456, 0.458 ] }, "prompt_eval_count": { "n": 10, "min": 997, "median": 997.0, "max": 997, "values": [ 997, 997, 997, 997, 997, 997, 997, 997, 997, 997 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-8000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 8000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:55:00-06:00", "last_at": "2026-08-12T12:55:10-06:00", "decode_tok_s": { "n": 10, "min": 95.29, "median": 101.91, "max": 102.21, "values": [ 95.29, 101.43, 101.49, 101.75, 101.87, 101.95, 101.97, 102.02, 102.15, 102.21 ] }, "ttft_proxy_ms": { "n": 10, "min": 636.16, "median": 643.865, "max": 1506.23, "values": [ 636.16, 639.87, 640.2, 640.76, 642.03, 645.7, 650.26, 650.58, 652.49, 1506.23 ] }, "prompt_tok_s": { "n": 10, "min": 8436.34, "median": 97507.16, "max": 97918.45, "values": [ 8436.34, 96992.12, 97078.07, 97429.34, 97470.9, 97543.42, 97591.04, 97641.08, 97801.1, 97918.45 ] }, "wall_s": { "n": 10, "min": 0.87, "median": 0.879, "max": 2.795, "values": [ 0.87, 0.872, 0.874, 0.875, 0.875, 0.882, 0.885, 0.887, 0.888, 2.795 ] }, "load_s": { "n": 10, "min": 0.452, "median": 0.457, "max": 0.465, "values": [ 0.452, 0.454, 0.455, 0.455, 0.457, 0.457, 0.457, 0.462, 0.464, 0.465 ] }, "prompt_eval_count": { "n": 10, "min": 7997, "median": 7997.0, "max": 7997, "values": [ 7997, 7997, 7997, 7997, 7997, 7997, 7997, 7997, 7997, 7997 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-32000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 32000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:55:11-06:00", "last_at": "2026-08-12T12:55:27-06:00", "decode_tok_s": { "n": 10, "min": 92.1, "median": 95.195, "max": 95.54, "values": [ 92.1, 94.49, 95.04, 95.13, 95.19, 95.2, 95.2, 95.21, 95.22, 95.54 ] }, "ttft_proxy_ms": { "n": 10, "min": 988.09, "median": 1000.21, "max": 4114.89, "values": [ 988.09, 996.05, 996.95, 999.6, 999.94, 1000.48, 1000.99, 1011.49, 1014.46, 4114.89 ] }, "prompt_tok_s": { "n": 10, "min": 10001.12, "median": 372348.235, "max": 373574.16, "values": [ 10001.12, 370267.08, 371535.4, 371975.96, 372283.24, 372413.23, 372708.21, 373121.1, 373377.99, 373574.16 ] }, "wall_s": { "n": 10, "min": 1.32, "median": 1.333, "max": 5.09, "values": [ 1.32, 1.326, 1.332, 1.333, 1.333, 1.334, 1.336, 1.349, 1.35, 5.09 ] }, "load_s": { "n": 10, "min": 0.457, "median": 0.461, "max": 0.466, "values": [ 0.457, 0.459, 0.459, 0.46, 0.461, 0.461, 0.463, 0.465, 0.466, 0.466 ] }, "prompt_eval_count": { "n": 10, "min": 31997, "median": 31997.0, "max": 31997, "values": [ 31997, 31997, 31997, 31997, 31997, 31997, 31997, 31997, 31997, 31997 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "cold-load", "block": "cold-load", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": null, "reasoning_strength": null, "n_calls": 1, "n_discarded": 0, "first_at": "2026-08-12T12:55:28-06:00", "last_at": "2026-08-12T12:55:28-06:00", "decode_tok_s": { "n": 1, "min": 90.58, "median": 90.58, "max": 90.58, "values": [ 90.58 ] }, "ttft_proxy_ms": { "n": 1, "min": 3086.3, "median": 3086.3, "max": 3086.3, "values": [ 3086.3 ] }, "prompt_tok_s": { "n": 1, "min": 5041.54, "median": 5041.54, "max": 5041.54, "values": [ 5041.54 ] }, "wall_s": { "n": 1, "min": 3.179, "median": 3.179, "max": 3.179, "values": [ 3.179 ] }, "load_s": { "n": 1, "min": 2.879, "median": 2.879, "max": 2.879, "values": [ 2.879 ] }, "prompt_eval_count": { "n": 1, "min": 997, "median": 997, "max": 997, "values": [ 997 ] }, "usable_counter_calls": 1, "usable_counts_of_n": "1/1", "response_failures": 0, "response_failure_counts_of_n": "0/1", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warmup-discard", "block": "warmup-discard", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 0, "n_discarded": 1, "first_at": "2026-08-12T18:58:54-06:00", "last_at": "2026-08-12T18:58:54-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/0", "response_failures": 0, "response_failure_counts_of_n": "0/0", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" } ], "anomalies": { "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "null_counter_note": "HTTP 200, every counter and done_reason null. Excluded from every statistic by counting rule 4; cause unexplained — transport_error is null and nothing in the manifest or the logs accounts for it", "empty_length_calls": 0, "transport_errors": 0, "response_failures": "0/34", "ceiling": 0.1, "threshold_arithmetic": "floor(0.1 x 34) + 1 = 4 — the smallest count that exceeds the ceiling at this denominator" }, "contention": { "fence_status": "UNMEASURED", "fence_note": "UNMEASURED by construction: the GIN-overlap receipt is not in the raw JSONL, so a rebuild cannot evaluate the registered contention fence. It is not evidence the blocks were clean.", "gin_note": "PREREG-C5 makes confirming gin_overlap_regex against a real guide journal line a precondition for PUBLISHING the overlap column; the raw windows are stored either way so a corrected pattern can be re-applied offline", "overlapping_blocks": [] }, "not_recoverable_from_raw": [ "gin_overlap (per block) — queried at summary time, never written to raw", "runner_argv / batch_size — captured into harness/manifest.json, not raw" ], "failure_ledger": "response failures 0/34" }, { "arm": "qwen3.5:27b", "registered": { "tag": "qwen3.5:27b", "warm": true, "cold_load": true, "strength_sweep": false, "dflash": null, "expected_calls": 37 }, "outcome": "RANKED", "outcome_why": "carried under protocol", "primary_tier": { "block_id": "warm-1000", "occurrences": 1, "usable": [ "10/10" ], "median_decode_tok_s": [ 66.45 ], "measurable": true, "note": "the primary metric is decode tok/s at the 1k tier; an arm with 0 usable counters here has no primary figure whatever its outcome state says about its other tiers" }, "population": { "recorded_calls": 37, "counted_calls": 36, "discarded_warmups": 1, "registered_total_including_warmup": 37, "reconciles": true, "note": "the registered per-arm total INCLUDES that arm's one discarded warmup (golden/c5-roster.json). A completion re-block adds calls on purpose, so a re-blocked arm reconciles ABOVE its registered total and the occurrence list shows where." }, "blocks_present": [ "clean-baseline", "cold-load", "warm-1000", "warm-32000", "warm-8000", "warmup-discard" ], "blocks_repeated": {}, "block_occurrences": [ { "block_id": "warmup-discard", "block": "warmup-discard", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 0, "n_discarded": 1, "first_at": "2026-08-12T18:45:06-06:00", "last_at": "2026-08-12T18:45:06-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/0", "response_failures": 0, "response_failure_counts_of_n": "0/0", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "clean-baseline", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T18:45:16-06:00", "last_at": "2026-08-12T18:45:20-06:00", "decode_tok_s": { "n": 3, "min": 66.48, "median": 66.48, "max": 66.5, "values": [ 66.48, 66.48, 66.5 ] }, "ttft_proxy_ms": { "n": 3, "min": 435.47, "median": 439.41, "max": 441.57, "values": [ 435.47, 439.41, 441.57 ] }, "prompt_tok_s": { "n": 3, "min": 10272.91, "median": 10277.9, "max": 10675.13, "values": [ 10272.91, 10277.9, 10675.13 ] }, "wall_s": { "n": 3, "min": 1.585, "median": 1.589, "max": 1.59, "values": [ 1.585, 1.589, 1.59 ] }, "load_s": { "n": 3, "min": 0.333, "median": 0.336, "max": 0.342, "values": [ 0.333, 0.336, 0.342 ] }, "prompt_eval_count": { "n": 3, "min": 996, "median": 996, "max": 996, "values": [ 996, 996, 996 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-1000", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T18:45:21-06:00", "last_at": "2026-08-12T18:45:35-06:00", "decode_tok_s": { "n": 10, "min": 66.29, "median": 66.45, "max": 66.49, "values": [ 66.29, 66.4, 66.4, 66.45, 66.45, 66.45, 66.46, 66.46, 66.48, 66.49 ] }, "ttft_proxy_ms": { "n": 10, "min": 434.58, "median": 435.87, "max": 439.82, "values": [ 434.58, 434.77, 435.17, 435.43, 435.59, 436.15, 437.91, 438.5, 438.55, 439.82 ] }, "prompt_tok_s": { "n": 10, "min": 10054.21, "median": 10203.195, "max": 10252.19, "values": [ 10054.21, 10162.64, 10186.55, 10191.03, 10200.53, 10205.86, 10223.98, 10233.65, 10233.86, 10252.19 ] }, "wall_s": { "n": 10, "min": 1.584, "median": 1.586, "max": 1.594, "values": [ 1.584, 1.584, 1.584, 1.585, 1.585, 1.587, 1.588, 1.589, 1.591, 1.594 ] }, "load_s": { "n": 10, "min": 0.331, "median": 0.333, "max": 0.335, "values": [ 0.331, 0.331, 0.331, 0.332, 0.332, 0.333, 0.334, 0.335, 0.335, 0.335 ] }, "prompt_eval_count": { "n": 10, "min": 996, "median": 996.0, "max": 996, "values": [ 996, 996, 996, 996, 996, 996, 996, 996, 996, 996 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-8000", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 8000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T18:45:37-06:00", "last_at": "2026-08-12T18:45:55-06:00", "decode_tok_s": { "n": 10, "min": 49.93, "median": 65.36, "max": 65.53, "values": [ 49.93, 54.1, 58.98, 65.31, 65.36, 65.36, 65.43, 65.43, 65.49, 65.53 ] }, "ttft_proxy_ms": { "n": 10, "min": 493.31, "median": 499.62, "max": 2390.73, "values": [ 493.31, 494.07, 494.38, 495.25, 496.2, 503.04, 503.29, 505.23, 526.34, 2390.73 ] }, "prompt_tok_s": { "n": 10, "min": 4026.69, "median": 80538.7, "max": 80984.45, "values": [ 4026.69, 75835.32, 77253.05, 80403.02, 80490.43, 80586.97, 80658.5, 80682.92, 80745.66, 80984.45 ] }, "wall_s": { "n": 10, "min": 1.679, "median": 1.684, "max": 3.938, "values": [ 1.679, 1.68, 1.68, 1.681, 1.682, 1.687, 1.687, 1.839, 1.947, 3.938 ] }, "load_s": { "n": 10, "min": 0.335, "median": 0.338, "max": 0.361, "values": [ 0.335, 0.336, 0.336, 0.337, 0.337, 0.338, 0.341, 0.345, 0.346, 0.361 ] }, "prompt_eval_count": { "n": 10, "min": 7996, "median": 7996.0, "max": 7996, "values": [ 7996, 7996, 7996, 7996, 7996, 7996, 7996, 7996, 7996, 7996 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-32000", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 32000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T18:45:57-06:00", "last_at": "2026-08-12T18:46:21-06:00", "decode_tok_s": { "n": 10, "min": 60.97, "median": 61.045, "max": 61.12, "values": [ 60.97, 61.01, 61.01, 61.03, 61.04, 61.05, 61.1, 61.11, 61.12, 61.12 ] }, "ttft_proxy_ms": { "n": 10, "min": 670.08, "median": 674.515, "max": 7532.68, "values": [ 670.08, 671.13, 672.15, 673.87, 674.37, 674.66, 676.89, 686.73, 686.94, 7532.68 ] }, "prompt_tok_s": { "n": 10, "min": 4597.05, "median": 312518.965, "max": 314024.93, "values": [ 4597.05, 311296.62, 311988.69, 312277.96, 312402.97, 312634.96, 312864.24, 312925.44, 313047.9, 314024.93 ] }, "wall_s": { "n": 10, "min": 1.998, "median": 2.004, "max": 8.865, "values": [ 1.998, 2.001, 2.001, 2.001, 2.004, 2.004, 2.007, 2.015, 2.018, 8.865 ] }, "load_s": { "n": 10, "min": 0.338, "median": 0.342, "max": 0.351, "values": [ 0.338, 0.34, 0.34, 0.34, 0.342, 0.342, 0.343, 0.343, 0.344, 0.351 ] }, "prompt_eval_count": { "n": 10, "min": 31996, "median": 31996.0, "max": 31996, "values": [ 31996, 31996, 31996, 31996, 31996, 31996, 31996, 31996, 31996, 31996 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "cold-load", "block": "cold-load", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": null, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T18:46:24-06:00", "last_at": "2026-08-12T18:46:35-06:00", "decode_tok_s": { "n": 3, "min": 65.93, "median": 65.99, "max": 66.05, "values": [ 65.93, 65.99, 66.05 ] }, "ttft_proxy_ms": { "n": 3, "min": 4289.8, "median": 4289.8, "max": 4541.18, "values": [ 4289.8, 4289.8, 4541.18 ] }, "prompt_tok_s": { "n": 3, "min": 3159.03, "median": 3160.47, "max": 3166.73, "values": [ 3159.03, 3160.47, 3166.73 ] }, "wall_s": { "n": 3, "min": 5.449, "median": 5.451, "max": 5.699, "values": [ 5.449, 5.451, 5.699 ] }, "load_s": { "n": 3, "min": 3.968, "median": 3.969, "max": 4.22, "values": [ 3.968, 3.969, 4.22 ] }, "prompt_eval_count": { "n": 3, "min": 996, "median": 996, "max": 996, "values": [ 996, 996, 996 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" } ], "anomalies": { "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "null_counter_note": "HTTP 200, every counter and done_reason null. Excluded from every statistic by counting rule 4; cause unexplained — transport_error is null and nothing in the manifest or the logs accounts for it", "empty_length_calls": 0, "transport_errors": 0, "response_failures": "0/36", "ceiling": 0.1, "threshold_arithmetic": "floor(0.1 x 36) + 1 = 4 — the smallest count that exceeds the ceiling at this denominator" }, "contention": { "fence_status": "UNMEASURED", "fence_note": "UNMEASURED by construction: the GIN-overlap receipt is not in the raw JSONL, so a rebuild cannot evaluate the registered contention fence. It is not evidence the blocks were clean.", "gin_note": "PREREG-C5 makes confirming gin_overlap_regex against a real guide journal line a precondition for PUBLISHING the overlap column; the raw windows are stored either way so a corrected pattern can be re-applied offline", "overlapping_blocks": [] }, "not_recoverable_from_raw": [ "gin_overlap (per block) — queried at summary time, never written to raw", "runner_argv / batch_size — captured into harness/manifest.json, not raw" ], "failure_ledger": "response failures 0/36" }, { "arm": "qwen3.6:27b", "registered": { "tag": "qwen3.6:27b", "warm": true, "cold_load": true, "strength_sweep": false, "dflash": null, "expected_calls": 37 }, "outcome": "RANKED", "outcome_why": "carried under protocol", "primary_tier": { "block_id": "warm-1000", "occurrences": 1, "usable": [ "10/10" ], "median_decode_tok_s": [ 66.485 ], "measurable": true, "note": "the primary metric is decode tok/s at the 1k tier; an arm with 0 usable counters here has no primary figure whatever its outcome state says about its other tiers" }, "population": { "recorded_calls": 37, "counted_calls": 36, "discarded_warmups": 1, "registered_total_including_warmup": 37, "reconciles": true, "note": "the registered per-arm total INCLUDES that arm's one discarded warmup (golden/c5-roster.json). A completion re-block adds calls on purpose, so a re-blocked arm reconciles ABOVE its registered total and the occurrence list shows where." }, "blocks_present": [ "clean-baseline", "cold-load", "warm-1000", "warm-32000", "warm-8000", "warmup-discard" ], "blocks_repeated": {}, "block_occurrences": [ { "block_id": "warmup-discard", "block": "warmup-discard", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 0, "n_discarded": 1, "first_at": "2026-08-12T18:46:46-06:00", "last_at": "2026-08-12T18:46:46-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/0", "response_failures": 0, "response_failure_counts_of_n": "0/0", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "clean-baseline", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T18:46:57-06:00", "last_at": "2026-08-12T18:47:00-06:00", "decode_tok_s": { "n": 3, "min": 66.44, "median": 66.48, "max": 66.51, "values": [ 66.44, 66.48, 66.51 ] }, "ttft_proxy_ms": { "n": 3, "min": 431.46, "median": 437.2, "max": 437.51, "values": [ 431.46, 437.2, 437.51 ] }, "prompt_tok_s": { "n": 3, "min": 10184.78, "median": 10221.67, "max": 10550.62, "values": [ 10184.78, 10221.67, 10550.62 ] }, "wall_s": { "n": 3, "min": 1.431, "median": 1.436, "max": 1.436, "values": [ 1.431, 1.436, 1.436 ] }, "load_s": { "n": 3, "min": 0.331, "median": 0.333, "max": 0.334, "values": [ 0.331, 0.333, 0.334 ] }, "prompt_eval_count": { "n": 3, "min": 996, "median": 996, "max": 996, "values": [ 996, 996, 996 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-1000", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T18:47:01-06:00", "last_at": "2026-08-12T18:47:14-06:00", "decode_tok_s": { "n": 10, "min": 66.35, "median": 66.485, "max": 66.55, "values": [ 66.35, 66.38, 66.44, 66.47, 66.48, 66.49, 66.49, 66.51, 66.51, 66.55 ] }, "ttft_proxy_ms": { "n": 10, "min": 435.43, "median": 437.56, "max": 441.02, "values": [ 435.43, 436.49, 436.59, 436.69, 437.49, 437.63, 438.67, 439.21, 439.71, 441.02 ] }, "prompt_tok_s": { "n": 10, "min": 10155.49, "median": 10201.885, "max": 10230.49, "values": [ 10155.49, 10157.46, 10171.05, 10183.11, 10201.47, 10202.3, 10216.85, 10220.21, 10227.55, 10230.49 ] }, "wall_s": { "n": 10, "min": 1.434, "median": 1.436, "max": 1.44, "values": [ 1.434, 1.435, 1.435, 1.435, 1.436, 1.436, 1.437, 1.438, 1.439, 1.44 ] }, "load_s": { "n": 10, "min": 0.332, "median": 0.334, "max": 0.338, "values": [ 0.332, 0.332, 0.332, 0.333, 0.334, 0.334, 0.335, 0.335, 0.336, 0.338 ] }, "prompt_eval_count": { "n": 10, "min": 996, "median": 996.0, "max": 996, "values": [ 996, 996, 996, 996, 996, 996, 996, 996, 996, 996 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-8000", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 8000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T18:47:15-06:00", "last_at": "2026-08-12T18:47:31-06:00", "decode_tok_s": { "n": 10, "min": 65.47, "median": 65.54, "max": 65.58, "values": [ 65.47, 65.49, 65.5, 65.51, 65.52, 65.56, 65.56, 65.56, 65.57, 65.58 ] }, "ttft_proxy_ms": { "n": 10, "min": 491.1, "median": 494.45, "max": 2405.26, "values": [ 491.1, 491.31, 493.49, 493.52, 493.89, 495.01, 496.2, 496.76, 496.91, 2405.26 ] }, "prompt_tok_s": { "n": 10, "min": 3978.68, "median": 81529.445, "max": 82071.7, "values": [ 3978.68, 81294.86, 81416.54, 81482.09, 81513.65, 81545.24, 81699.38, 81786.29, 81798.84, 82071.7 ] }, "wall_s": { "n": 10, "min": 1.491, "median": 1.496, "max": 3.406, "values": [ 1.491, 1.493, 1.494, 1.494, 1.496, 1.496, 1.497, 1.497, 1.5, 3.406 ] }, "load_s": { "n": 10, "min": 0.335, "median": 0.337, "max": 0.339, "values": [ 0.335, 0.335, 0.336, 0.336, 0.337, 0.337, 0.338, 0.338, 0.338, 0.339 ] }, "prompt_eval_count": { "n": 10, "min": 7996, "median": 7996.0, "max": 7996, "values": [ 7996, 7996, 7996, 7996, 7996, 7996, 7996, 7996, 7996, 7996 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-32000", "block": "warm", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": 32000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T18:47:32-06:00", "last_at": "2026-08-12T18:47:53-06:00", "decode_tok_s": { "n": 10, "min": 61.28, "median": 61.305, "max": 61.39, "values": [ 61.28, 61.28, 61.29, 61.3, 61.3, 61.31, 61.31, 61.32, 61.33, 61.39 ] }, "ttft_proxy_ms": { "n": 10, "min": 669.8, "median": 675.625, "max": 7623.77, "values": [ 669.8, 672.25, 672.26, 674.46, 675.47, 675.78, 675.88, 676.94, 677.73, 7623.77 ] }, "prompt_tok_s": { "n": 10, "min": 4535.88, "median": 307271.385, "max": 312457.89, "values": [ 4535.88, 306574.94, 306645.45, 306804.23, 307057.45, 307485.32, 307514.87, 308856.61, 309540.85, 312457.89 ] }, "wall_s": { "n": 10, "min": 1.536, "median": 1.542, "max": 8.491, "values": [ 1.536, 1.537, 1.539, 1.541, 1.541, 1.543, 1.543, 1.546, 1.552, 8.491 ] }, "load_s": { "n": 10, "min": 0.339, "median": 0.341, "max": 0.343, "values": [ 0.339, 0.339, 0.339, 0.339, 0.34, 0.341, 0.342, 0.342, 0.343, 0.343 ] }, "prompt_eval_count": { "n": 10, "min": 31996, "median": 31996.0, "max": 31996, "values": [ 31996, 31996, 31996, 31996, 31996, 31996, 31996, 31996, 31996, 31996 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "cold-load", "block": "cold-load", "block_id_source": "record", "run_label": "c5b", "inference_note": null, "tier_tokens": null, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T18:47:55-06:00", "last_at": "2026-08-12T18:48:06-06:00", "decode_tok_s": { "n": 3, "min": 65.75, "median": 65.93, "max": 65.99, "values": [ 65.75, 65.93, 65.99 ] }, "ttft_proxy_ms": { "n": 3, "min": 4287.95, "median": 4292.1, "max": 4293.21, "values": [ 4287.95, 4292.1, 4293.21 ] }, "prompt_tok_s": { "n": 3, "min": 3143.18, "median": 3150.53, "max": 3159.97, "values": [ 3143.18, 3150.53, 3159.97 ] }, "wall_s": { "n": 3, "min": 5.295, "median": 5.299, "max": 5.301, "values": [ 5.295, 5.299, 5.301 ] }, "load_s": { "n": 3, "min": 3.965, "median": 3.969, "max": 3.972, "values": [ 3.965, 3.969, 3.972 ] }, "prompt_eval_count": { "n": 3, "min": 996, "median": 996, "max": 996, "values": [ 996, 996, 996 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" } ], "anomalies": { "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "null_counter_note": "HTTP 200, every counter and done_reason null. Excluded from every statistic by counting rule 4; cause unexplained — transport_error is null and nothing in the manifest or the logs accounts for it", "empty_length_calls": 0, "transport_errors": 0, "response_failures": "0/36", "ceiling": 0.1, "threshold_arithmetic": "floor(0.1 x 36) + 1 = 4 — the smallest count that exceeds the ceiling at this denominator" }, "contention": { "fence_status": "UNMEASURED", "fence_note": "UNMEASURED by construction: the GIN-overlap receipt is not in the raw JSONL, so a rebuild cannot evaluate the registered contention fence. It is not evidence the blocks were clean.", "gin_note": "PREREG-C5 makes confirming gin_overlap_regex against a real guide journal line a precondition for PUBLISHING the overlap column; the raw windows are stored either way so a corrected pattern can be re-applied offline", "overlapping_blocks": [] }, "not_recoverable_from_raw": [ "gin_overlap (per block) — queried at summary time, never written to raw", "runner_argv / batch_size — captured into harness/manifest.json, not raw" ], "failure_ledger": "response failures 0/36" }, { "arm": "nemotron-3.5-lightning:30b-a3b", "registered": { "tag": "nemotron-3.5-lightning:30b-a3b", "warm": true, "cold_load": true, "strength_sweep": false, "dflash": null, "expected_calls": 37 }, "outcome": "RANKED", "outcome_why": "carried under protocol", "primary_tier": { "block_id": "warm-1000", "occurrences": 1, "usable": [ "10/10" ], "median_decode_tok_s": [ 112.52 ], "measurable": true, "note": "the primary metric is decode tok/s at the 1k tier; an arm with 0 usable counters here has no primary figure whatever its outcome state says about its other tiers" }, "population": { "recorded_calls": 37, "counted_calls": 36, "discarded_warmups": 1, "registered_total_including_warmup": 37, "reconciles": true, "note": "the registered per-arm total INCLUDES that arm's one discarded warmup (golden/c5-roster.json). A completion re-block adds calls on purpose, so a re-blocked arm reconciles ABOVE its registered total and the occurrence list shows where." }, "blocks_present": [ "clean-baseline", "cold-load", "warm-1000", "warm-32000", "warm-8000", "warmup-discard" ], "blocks_repeated": {}, "block_occurrences": [ { "block_id": "warmup-discard", "block": "warmup-discard", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 0, "n_discarded": 1, "first_at": "2026-08-12T12:55:35-06:00", "last_at": "2026-08-12T12:55:35-06:00", "decode_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "ttft_proxy_ms": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_tok_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "wall_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "load_s": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "prompt_eval_count": { "n": 0, "min": null, "median": null, "max": null, "values": [] }, "usable_counter_calls": 0, "usable_counts_of_n": "0/0", "response_failures": 0, "response_failure_counts_of_n": "0/0", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "clean-baseline", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": "un-strengthened 1000-token block of 3 calls — the clean baseline by PREREG-C5's Population table; the warm set at the same tier is 10", "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:55:47-06:00", "last_at": "2026-08-12T12:55:50-06:00", "decode_tok_s": { "n": 3, "min": 112.16, "median": 112.27, "max": 112.4, "values": [ 112.16, 112.27, 112.4 ] }, "ttft_proxy_ms": { "n": 3, "min": 248.02, "median": 248.06, "max": 248.94, "values": [ 248.02, 248.06, 248.94 ] }, "prompt_tok_s": { "n": 3, "min": 29102.76, "median": 29259.44, "max": 30239.8, "values": [ 29102.76, 29259.44, 30239.8 ] }, "wall_s": { "n": 3, "min": 1.089, "median": 1.09, "max": 1.097, "values": [ 1.089, 1.09, 1.097 ] }, "load_s": { "n": 3, "min": 0.211, "median": 0.212, "max": 0.212, "values": [ 0.211, 0.212, 0.212 ] }, "prompt_eval_count": { "n": 3, "min": 1000, "median": 1000, "max": 1000, "values": [ 1000, 1000, 1000 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-1000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 1000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:55:51-06:00", "last_at": "2026-08-12T12:56:00-06:00", "decode_tok_s": { "n": 10, "min": 112.43, "median": 112.52, "max": 112.63, "values": [ 112.43, 112.48, 112.48, 112.49, 112.52, 112.52, 112.53, 112.59, 112.6, 112.63 ] }, "ttft_proxy_ms": { "n": 10, "min": 246.92, "median": 248.59, "max": 250.74, "values": [ 246.92, 247.34, 247.94, 248.25, 248.45, 248.73, 248.78, 248.9, 250.13, 250.74 ] }, "prompt_tok_s": { "n": 10, "min": 29049.5, "median": 29261.6, "max": 29351.34, "values": [ 29049.5, 29087.52, 29178.34, 29196.23, 29238.91, 29284.29, 29295.44, 29304.89, 29309.18, 29351.34 ] }, "wall_s": { "n": 10, "min": 1.086, "median": 1.088, "max": 1.09, "values": [ 1.086, 1.086, 1.086, 1.088, 1.088, 1.088, 1.088, 1.089, 1.089, 1.09 ] }, "load_s": { "n": 10, "min": 0.21, "median": 0.211, "max": 0.214, "values": [ 0.21, 0.211, 0.211, 0.211, 0.211, 0.212, 0.212, 0.212, 0.213, 0.214 ] }, "prompt_eval_count": { "n": 10, "min": 1000, "median": 1000.0, "max": 1000, "values": [ 1000, 1000, 1000, 1000, 1000, 1000, 1000, 1000, 1000, 1000 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-8000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 8000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:56:02-06:00", "last_at": "2026-08-12T12:56:12-06:00", "decode_tok_s": { "n": 10, "min": 114.33, "median": 120.585, "max": 120.87, "values": [ 114.33, 120.52, 120.57, 120.58, 120.58, 120.59, 120.63, 120.71, 120.72, 120.87 ] }, "ttft_proxy_ms": { "n": 10, "min": 270.0, "median": 271.94, "max": 804.89, "values": [ 270.0, 270.89, 270.91, 271.77, 271.9, 271.98, 272.79, 273.17, 274.36, 804.89 ] }, "prompt_tok_s": { "n": 10, "min": 13978.32, "median": 208809.325, "max": 209863.59, "values": [ 13978.32, 207889.4, 208599.52, 208610.39, 208610.39, 209008.26, 209013.72, 209358.32, 209687.57, 209863.59 ] }, "wall_s": { "n": 10, "min": 1.059, "median": 1.061, "max": 1.637, "values": [ 1.059, 1.059, 1.06, 1.06, 1.061, 1.061, 1.062, 1.064, 1.068, 1.637 ] }, "load_s": { "n": 10, "min": 0.214, "median": 0.217, "max": 0.219, "values": [ 0.214, 0.215, 0.215, 0.216, 0.217, 0.217, 0.218, 0.218, 0.218, 0.219 ] }, "prompt_eval_count": { "n": 10, "min": 8000, "median": 8000.0, "max": 8000, "values": [ 8000, 8000, 8000, 8000, 8000, 8000, 8000, 8000, 8000, 8000 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "warm-32000", "block": "warm", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": 32000, "reasoning_strength": null, "n_calls": 10, "n_discarded": 0, "first_at": "2026-08-12T12:56:13-06:00", "last_at": "2026-08-12T12:56:26-06:00", "decode_tok_s": { "n": 10, "min": 107.19, "median": 107.74, "max": 107.8, "values": [ 107.19, 107.64, 107.64, 107.73, 107.74, 107.74, 107.75, 107.76, 107.78, 107.8 ] }, "ttft_proxy_ms": { "n": 10, "min": 332.99, "median": 336.49, "max": 2210.85, "values": [ 332.99, 335.39, 336.2, 336.35, 336.47, 336.51, 338.51, 341.9, 343.43, 2210.85 ] }, "prompt_tok_s": { "n": 10, "min": 16657.77, "median": 615366.865, "max": 617760.62, "values": [ 16657.77, 548856.83, 610244.48, 615148.02, 615337.28, 615396.45, 615396.45, 616000.62, 617403.05, 617760.62 ] }, "wall_s": { "n": 10, "min": 1.291, "median": 1.295, "max": 3.176, "values": [ 1.291, 1.292, 1.292, 1.292, 1.295, 1.295, 1.296, 1.298, 1.302, 3.176 ] }, "load_s": { "n": 10, "min": 0.22, "median": 0.221, "max": 0.226, "values": [ 0.22, 0.22, 0.22, 0.22, 0.221, 0.221, 0.222, 0.223, 0.224, 0.226 ] }, "prompt_eval_count": { "n": 10, "min": 32000, "median": 32000.0, "max": 32000, "values": [ 32000, 32000, 32000, 32000, 32000, 32000, 32000, 32000, 32000, 32000 ] }, "usable_counter_calls": 10, "usable_counts_of_n": "10/10", "response_failures": 0, "response_failure_counts_of_n": "0/10", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" }, { "block_id": "cold-load", "block": "cold-load", "block_id_source": "inferred", "run_label": null, "inference_note": null, "tier_tokens": null, "reasoning_strength": null, "n_calls": 3, "n_discarded": 0, "first_at": "2026-08-12T12:56:28-06:00", "last_at": "2026-08-12T12:56:53-06:00", "decode_tok_s": { "n": 3, "min": 108.29, "median": 109.92, "max": 110.19, "values": [ 108.29, 109.92, 110.19 ] }, "ttft_proxy_ms": { "n": 3, "min": 11616.46, "median": 11620.41, "max": 11622.54, "values": [ 11616.46, 11620.41, 11622.54 ] }, "prompt_tok_s": { "n": 3, "min": 3421.66, "median": 3455.19, "max": 3459.72, "values": [ 3421.66, 3455.19, 3459.72 ] }, "wall_s": { "n": 3, "min": 12.475, "median": 12.48, "max": 12.493, "values": [ 12.475, 12.48, 12.493 ] }, "load_s": { "n": 3, "min": 11.324, "median": 11.325, "max": 11.33, "values": [ 11.324, 11.325, 11.33 ] }, "prompt_eval_count": { "n": 3, "min": 1000, "median": 1000, "max": 1000, "values": [ 1000, 1000, 1000 ] }, "usable_counter_calls": 3, "usable_counts_of_n": "3/3", "response_failures": 0, "response_failure_counts_of_n": "0/3", "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "empty_length_calls": 0, "transport_errors": 0, "gin_overlap": null, "gin_overlap_note": "not recoverable from raw — the journal window was queried at summary time and stored on the Block object, not on the record" } ], "anomalies": { "null_counter_calls": 0, "null_counter_split": { "total": 0, "with_response_text": 0, "empty_response_text": 0 }, "null_counter_note": "HTTP 200, every counter and done_reason null. Excluded from every statistic by counting rule 4; cause unexplained — transport_error is null and nothing in the manifest or the logs accounts for it", "empty_length_calls": 0, "transport_errors": 0, "response_failures": "0/36", "ceiling": 0.1, "threshold_arithmetic": "floor(0.1 x 36) + 1 = 4 — the smallest count that exceeds the ceiling at this denominator" }, "contention": { "fence_status": "UNMEASURED", "fence_note": "UNMEASURED by construction: the GIN-overlap receipt is not in the raw JSONL, so a rebuild cannot evaluate the registered contention fence. It is not evidence the blocks were clean.", "gin_note": "PREREG-C5 makes confirming gin_overlap_regex against a real guide journal line a precondition for PUBLISHING the overlap column; the raw windows are stored either way so a corrected pattern can be re-applied offline", "overlapping_blocks": [] }, "not_recoverable_from_raw": [ "gin_overlap (per block) — queried at summary time, never written to raw", "runner_argv / batch_size — captured into harness/manifest.json, not raw" ], "failure_ledger": "response failures 0/36" } ], "n_rows": 8, "contamination_caveat": "Published 2026-08-13. Models with a later training cutoff may have seen these sets. We author fresh sets each cycle; this one is not a standard, it is our kit, yours to reuse.", "not_a_standard": "Our own trials, our own hardware, for our own chairs. Not first independent numbers, and not a benchmark." }