{
  "schema": "../RESULT-SCHEMA.json",
  "results": [
    {
      "id": "chess-reference-v0.1-2026-08-11",
      "path": "chess-reference-result-v0.1.json",
      "game": "chess",
      "protocol": "chess-v0.1",
      "model": "reference-baseline",
      "evidence_status": "retired",
      "retired_reason": "Six of twelve fixtures carry a wrong answer key: no fixture position contains a king, none is a legal chess position, and the hand-rolled move/legality logic used in place of an engine encoded two mistaken beliefs (en passant treated as mandatory; side-to-move and full-move counter not updated after a move). This 36/36 means the fixture set agrees with itself, not that the key is correct. See docs/CHESS-BENCHMARK-V0.1.md#wrong-key-the-six-broken-fixtures.",
      "trials": 3,
      "task_fixtures": 12,
      "passed": 36,
      "failed": 0,
      "incomplete": 0,
      "note": "Deterministic control fixture, not a provider-backed model result.",
      "evaluator_version": "chess-evaluator@0.2.0"
    },
    {
      "id": "chess-gemini-antigravity-v0.1-2026-08-12",
      "path": "submissions/gemini-antigravity-chess-v0.1/evaluation.json",
      "raw_path": "submissions/gemini-antigravity-chess-v0.1/raw/",
      "game": "chess",
      "protocol": "chess-v0.1",
      "model": "Gemini 3.6 Flash",
      "provider": "Google Gemini via Antigravity",
      "trials": 3,
      "task_fixtures": 12,
      "passed": 18,
      "failed": 18,
      "incomplete": 0,
      "pass_rate": 0.5,
      "consistency": 0.5,
      "evidence_status": "retired",
      "note": "WITHDRAWN 2026-08-12 for scoring against field names the evaluated model was never given; see evaluation.json. RETIRED 2026-08-12 on top of that, independently: chess-v0.1 itself carries a wrong answer key on 6/12 fixtures, so any score against it is void regardless of the field-name defect. Kept for the record, not as a result. Raw trajectories unchanged.",
      "retired_reason": "Doubly invalid: (1) evaluator 0.1.0 read field names the model was never given, a defect independent of the answer key; (2) chess-v0.1's answer key is wrong on 6 of 12 fixtures (no fixture position has a king; hand-rolled logic treated en passant as mandatory and failed to update side-to-move/full-move counter). Reason (2) is why the whole protocol, not just this row, is retired. See docs/CHESS-BENCHMARK-V0.1.md#wrong-key-the-six-broken-fixtures.",
      "evaluator_version": "chess-evaluator@0.1.0"
    },
    {
      "id": "chess-gemini-antigravity-v0.1-eval-0.2.0-2026-08-12",
      "path": "submissions/gemini-antigravity-chess-v0.1-eval-0.2.0/evaluation.json",
      "raw_path": "submissions/gemini-antigravity-chess-v0.1/raw/",
      "game": "chess",
      "protocol": "chess-v0.1",
      "model": "Gemini 3.6 Flash",
      "provider": "Google Gemini via Antigravity",
      "trials": 3,
      "task_fixtures": 12,
      "passed": 36,
      "failed": 0,
      "contract_violations": 0,
      "incomplete": 0,
      "pass_rate": 1,
      "consistency": 1,
      "evaluator_version": "chess-evaluator@0.2.0",
      "response_contract": "chess-response-v0.1",
      "evidence_status": "retired",
      "retired_reason": "Six of twelve fixtures carry a wrong answer key (no fixture position has a king; hand-rolled move/legality logic treated en passant as mandatory and failed to update side-to-move/full-move counter after a move). This 36/36 means the submission agreed with the key's mistakes, not that it played correctly. See docs/CHESS-BENCHMARK-V0.1.md#wrong-key-the-six-broken-fixtures.",
      "note": "Same raw trajectories, rescored against the contract the model was actually given. Supersedes the withdrawn 0.1.0 verdict on the field-name defect. RETIRED 2026-08-12: chess-v0.1's answer key itself is wrong on 6/12 fixtures, so this perfect profile means agreement with the key, not a capability result."
    },
    {
      "id": "claude-opus-4-6-chess-v0.1-2026-08-12",
      "path": "submissions/claude-opus-4-6-chess-v0.1/",
      "raw_path": "submissions/claude-opus-4-6-chess-v0.1/raw/",
      "game": "chess",
      "protocol": "chess-v0.1",
      "model": "Claude Opus 4.6 (Thinking)",
      "provider": "Anthropic via Antigravity",
      "trials": 2,
      "task_fixtures": 12,
      "evidence_status": "finding_not_a_score",
      "note": "Two of three protocol-required trials completed (trial 1 hit RESOURCE_EXHAUSTED / HTTP 429 before producing output; not substituted or reconstructed). Both completed trials disagree with the chess-v0.1 answer key on the same 6 of 12 fixtures, 100% consistent across trials, and agree with the key on the other 6. No score is published: trial count is below the protocol's minimum of three, and scoring against a confirmed-wrong key would not mean anything regardless. See FINDING.md."
    }
  ]
}