{
  "checks": {
    "active_trial_marker_cleared": true,
    "all_receipts_http_200_and_pinned_jev_1_14_0": true,
    "all_request_hashes_and_payloads_reconstructed_from_frozen_inputs": true,
    "exact_ordered_trial_inventory_matches_frozen_plan": true,
    "fixture_and_core_source_hashes_match_frozen_plan": true,
    "four_rgb_alias_pairs_have_identical_requests_and_unknown_verdicts": true,
    "independent_counts_brier_and_latency_quantiles_match_published_metrics": true,
    "input_token_cost_arithmetic_checked": true,
    "measured_summary_matches_original_metrics_and_hashes": true,
    "occluded_and_unreleased_observability_references_reviewed": true,
    "presence_threshold_application_checked": true,
    "publication_regressions_passed": 13,
    "results_equal_append_only_attempt_ledger": true,
    "rgb_requests_have_no_measured_state": true,
    "stored_probabilities_verdicts_usage_match_provider_receipts": true
  },
  "evaluation_plan_sha256": "5f02006190d97bdaca43f139f5bf221cece23bba868a5e9ee0f424618ea87d8f",
  "findings": {
    "estimated_input_cost_usd": 0.007644756,
    "input_tokens": 182018,
    "output_tokens": 4776,
    "overall_latency_seconds": {
      "max": 4.091210667043924,
      "p50": 1.7764145623659715,
      "p95": 3.287287432909943
    },
    "rgb": {
      "ambiguous_abstention": [
        8,
        16
      ],
      "observability_reference_agreement": [
        24,
        32
      ],
      "physical_false_completion": [
        0,
        24
      ],
      "physical_positives_called_complete": [
        0,
        8
      ]
    },
    "rgb_telemetry": {
      "observability_reference_agreement": [
        25,
        32
      ],
      "physical_false_completion": [
        0,
        24
      ],
      "physical_positives_called_complete": [
        1,
        8
      ]
    }
  },
  "frozen_core_source": "19acf0734d63ecebc07140c8d84302a54baee4c4",
  "inventory": {
    "attempts_per_trial": 1,
    "completed": 64,
    "conditions": {
      "rgb": 32,
      "rgb_telemetry": 32
    },
    "errors": 0,
    "planned": 64,
    "unrun": 0,
    "valid": 64
  },
  "limitations": [
    "This review confirms internal provenance, arithmetic and stated evidence semantics; it is not an independently rerun provider experiment.",
    "Four correlated layouts with authored diagnostic states; 64 calls are not 64 independent real-world tasks.",
    "RGB and telemetry use different observability reference labels; their accuracy difference is not a general accuracy-gain estimate.",
    "No false complete verdicts were observed, but telemetry recognized only one of eight physical positives; this does not establish a reliable completion detector.",
    "Telemetry is privileged simulator state, not a validated robot sensor or camera estimator.",
    "Latency includes request preparation, upload, network, hosted response and parsing from preloaded images; rendering, camera capture and actuation are excluded.",
    "Cost is derived from returned input usage and the frozen documented price; it is not a billing receipt.",
    "No learned controller, direct actuator loop, local Jev execution or real-robot transfer is demonstrated."
  ],
  "original_results_sha256": "18c71ad02cd50763b04a90028b46740b98bd91ffa9d6071f5a179d4c22537ae4",
  "review_scope": "Independent read-only review of the completed frozen provider run; no requests rerun and no protocol or scorer edits.",
  "reviewed_at": "2026-10-19T04:14:14.525991+00:00",
  "schema": "dvidia.image-goal-independent-run-review/v1",
  "status": "passed"
}