{
  "schemaVersion": "flash-core-eval-evaluator-audit/v1",
  "runId": "run-3",
  "recordedAt": "2026-07-26T14:41:17+07:00",
  "providerRerunAllowed": false,
  "rawArtifactsFrozen": true,
  "initialScorecard": {
    "pathAtAuditTime": "artifacts/derived/run-3-scorecard.json",
    "immutablePath": "artifacts/derived/run-3-scorecard-pre-amendment.json",
    "preservation": "full_byte_exact_reconstruction",
    "canonicalPathWasRegeneratedAfterAmendment": true,
    "generatedAt": "2026-07-26T07:38:04.569Z",
    "recoveredAt": "2026-07-26T15:08:28+07:00",
    "recoveryVerification": "serialized bytes reproduce the SHA-256 recorded before the canonical path was regenerated",
    "sha256": "08236755f6113150d5f24a8a13cf6457c7a96b8aebc16ac945a454b3fbbf2665",
    "observedResponseCount": 12,
    "missingResponseCount": 0,
    "outcomeCandidate": "inconclusive",
    "models": {
      "qwen/qwen3.6-flash": {
        "passedCells": 4,
        "totalCells": 6
      },
      "deepseek/deepseek-v4-flash": {
        "passedCells": 3,
        "totalCells": 6
      }
    }
  },
  "rawResponseSha256": {
    "deepseek-deepseek-v4-flash--en--intent--r1": "ee5a3423f24387b5ed2c5ab7f50ed656e83cc44649141967e8eb316b412eb9f0",
    "deepseek-deepseek-v4-flash--en--narration--r1": "fa7b10ee35ee517a6799ea4babb064b8c463fcce9584427e0bc3ba0646578e59",
    "deepseek-deepseek-v4-flash--en--scenario--r1": "44e0bdf96075436f986fc177652f51fd79ee9847f3a40b3973cf54d72fedf40d",
    "deepseek-deepseek-v4-flash--ru--intent--r1": "e6ce6c2586a55732e0d78f39eb1e2bcb6b45f4c544ab6b9f802ae05d09a910ad",
    "deepseek-deepseek-v4-flash--ru--narration--r1": "1193dcb183dce51e88f9c8926877cbd95b43d61c87ebea648097c6c59008fc79",
    "deepseek-deepseek-v4-flash--ru--scenario--r1": "f4b88c9414f9f35a85385f90b9992da9cc7267663372d1aba54cc186e231163c",
    "qwen-qwen3-6-flash--en--intent--r1": "c0448547c84fd9b6f4fbeb25ff23dcd9828eb4e9e09f7c815080ab3bf397188b",
    "qwen-qwen3-6-flash--en--narration--r1": "94f659d8289e34b53c625921eceafd11987f675df4c35d381a9be98d3c41a419",
    "qwen-qwen3-6-flash--en--scenario--r1": "3c2a1ae27dfb1b6965492e9d09633cecf0efbe2d8b8b5e972bfad8778bf0bbbe",
    "qwen-qwen3-6-flash--ru--intent--r1": "891740a977cfe981efb43b3c769cb0d890135dfbb2758eda9bf2a16e6376714d",
    "qwen-qwen3-6-flash--ru--narration--r1": "8817aba6c4eb12793e8c8fc721c9e67e02b52361cc746643d777b9d4c82dfbd5",
    "qwen-qwen3-6-flash--ru--scenario--r1": "56f2b8fc60860d3af35546ccb97462e453a8d592598378ea2bed405694ba15b2"
  },
  "observedEvaluatorDefects": [
    {
      "requestId": "deepseek-deepseek-v4-flash--en--narration--r1",
      "kind": "mixed_false_negative_and_true_model_error",
      "detail": "The outcome phrase 'a clear win' is a valid success synonym missing from the scorer, while 'yields another clue' is an unsupported new-effect claim."
    },
    {
      "requestId": "deepseek-deepseek-v4-flash--ru--narration--r1",
      "kind": "false_negative_candidate",
      "detail": "The phrase 'переведены на +1' expresses the committed clock advancement but is missing from the scorer vocabulary."
    },
    {
      "requestId": "qwen-qwen3-6-flash--ru--narration--r1",
      "kind": "mixed_false_negative_and_true_model_error",
      "detail": "The phrase 'продвинут на +1' expresses the committed clock advancement but is missing from the scorer vocabulary, while 'Служебная запись завершена' asserts a completed filing action absent from the event and projection."
    }
  ],
  "plannedScorerAmendment": {
    "status": "applied",
    "changes": [
      "Accept the bounded English success synonym 'win'.",
      "Accept Russian clock-movement stems covering 'продвинут' and 'переведены'.",
      "Reject explicit claims that a success produced another, new, or additional clue when the committed effects only advance a clock.",
      "Reject explicit claims that a service record or filing was completed when no such action exists in the event or projection."
    ],
    "method": "Add behavior-level RED tests, verify the failures, make the minimum scorer change, then rescore the same frozen raw artifacts.",
    "nonChanges": [
      "Do not regenerate or edit provider responses.",
      "Do not backfill human naturalness scores.",
      "Do not infer latency, token usage, credit delta, or live balance."
    ]
  },
  "postAmendmentScorecard": {
    "generatedAt": "2026-07-26T07:48:45.123Z",
    "path": "artifacts/derived/run-3-scorecard.json",
    "sha256": "f9ac70928e76ea22e58c8099386d60c4a4ecb17f3a6dca3b8538aa544e181408",
    "evaluatorSha256": "22f2877444133b149624018cea155efac0749025eaf70343a1e2811dd4f9f425",
    "testSha256": "ba9a43604e573ac6de19d327ea4b26e9d2fe2988238ec62e22f8db3cc40afcc6",
    "redEvidence": {
      "command": "node --test RnD/technical/2026-07-26-flash-llm-core-evaluation/prototype/run.test.ts",
      "passed": 18,
      "failed": 2,
      "expectedFailures": [
        "narration scorer accepts bounded outcome and clock-movement synonyms",
        "narration scorer rejects unsupported success aftermath"
      ]
    },
    "greenEvidence": {
      "command": "node --test RnD/technical/2026-07-26-flash-llm-core-evaluation/prototype/run.test.ts",
      "passed": 20,
      "failed": 0
    },
    "observedResponseCount": 12,
    "missingResponseCount": 0,
    "outcomeCandidate": "inconclusive",
    "models": {
      "qwen/qwen3.6-flash": {
        "passedCells": 4,
        "totalCells": 6
      },
      "deepseek/deepseek-v4-flash": {
        "passedCells": 4,
        "totalCells": 6
      }
    }
  },
  "measurementLimits": {
    "scenarioExecutableExitPredicates": "not_measured",
    "scenarioStoryCoherence": "not_measured",
    "scenarioHostFitBeyondFixture": "not_measured",
    "privacyParaphraseLeakage": "not_measured",
    "privacyExactHiddenTokenMatches": "measured",
    "humanNarrationNaturalness": "missing",
    "billingDelta": "unknown"
  },
  "independentReview": {
    "status": "complete",
    "reviewerKind": "agent_semantic_audit_not_human_naturalness_review",
    "cells": {
      "qwen-qwen3-6-flash--ru--scenario--r1": {
        "verdict": "fail",
        "reason": "players visibility is missing required playerIds"
      },
      "qwen-qwen3-6-flash--ru--intent--r1": {
        "verdict": "pass",
        "reason": "four opaque cases match the hidden oracle and authoritative engine behavior"
      },
      "qwen-qwen3-6-flash--ru--narration--r1": {
        "verdict": "fail",
        "reason": "clock advancement is correct, but a completed service-record action is unsupported"
      },
      "qwen-qwen3-6-flash--en--scenario--r1": {
        "verdict": "pass_structural_only",
        "reason": "passes the declared structural contract gate"
      },
      "qwen-qwen3-6-flash--en--intent--r1": {
        "verdict": "pass",
        "reason": "four opaque cases match the hidden oracle and authoritative engine behavior"
      },
      "qwen-qwen3-6-flash--en--narration--r1": {
        "verdict": "pass",
        "reason": "outcome, numbers, and committed clock effect are grounded"
      },
      "deepseek-deepseek-v4-flash--ru--scenario--r1": {
        "verdict": "fail",
        "reason": "three locationId values are human-readable Russian labels rather than ASCII kebab-case IDs"
      },
      "deepseek-deepseek-v4-flash--ru--intent--r1": {
        "verdict": "pass",
        "reason": "four opaque cases match the hidden oracle and authoritative engine behavior"
      },
      "deepseek-deepseek-v4-flash--ru--narration--r1": {
        "verdict": "fail_strict_grounding",
        "reason": "the clock effect is correct and the bureaucracy phrase is tone-only, but 'Юноша (Juno)' invents age and gender absent from the projection"
      },
      "deepseek-deepseek-v4-flash--en--scenario--r1": {
        "verdict": "pass_structural_only",
        "reason": "passes the declared structural contract gate"
      },
      "deepseek-deepseek-v4-flash--en--intent--r1": {
        "verdict": "pass",
        "reason": "four opaque cases match the hidden oracle and authoritative engine behavior"
      },
      "deepseek-deepseek-v4-flash--en--narration--r1": {
        "verdict": "fail",
        "reason": "the outcome synonym is valid, but the response invents another clue"
      }
    },
    "modelTotals": {
      "qwen/qwen3.6-flash": {
        "passedCells": 4,
        "totalCells": 6
      },
      "deepseek/deepseek-v4-flash": {
        "passedCells": 3,
        "totalCells": 6
      }
    },
    "toneBoundary": "The abstract phrase 'Бюрократия камбуза довольна' is treated as tone-only personification, not a new canonical NPC, fact, item, or mechanical effect.",
    "residualScorerLimits": [
      "Open-ended grounding hallucinations cannot be reliably closed with a blacklist regex.",
      "A shape-invalid scenario short-circuits dependent fields to false even when their correct state is not evaluated.",
      "Scenario compilation proves structural graph validity, not executable free-text predicates, playability, or story coherence.",
      "The privacy check detects exact hidden tokens only, not translated or paraphrased leakage."
    ]
  }
}