{
 "generated": "2026-09-08",
 "entries": [
  {
   "ts": "2026-09-20T18:07:26",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "ornith-35b-tagger",
   "profile": "extraction",
   "accuracy": 85.8,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 206,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "02e7b350ee762892",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T15:40:53",
   "suite": "needle",
   "suite_version": "needle/1",
   "model": "edge0-35b",
   "profile": "extraction",
   "accuracy": 62.7,
   "coverage": 100.0,
   "measured": 110,
   "unmeasured": 0,
   "attempted": 110,
   "passed": 69,
   "selftests": 11,
   "dataset_id": "5c5199b9eee8a595",
   "grader_id": "c38dec33aa1c86c2",
   "configuration_id": "cc08042253cc6e13",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T12:47:56",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "bonsai2-27b-ptq1",
   "profile": "extraction",
   "accuracy": 66.4,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 83,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "0407c0f9d876e4a3",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T12:42:25",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "bonsai2-27b-ptq1",
   "profile": "study-nothink",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 48,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "9255c41b05f7da62",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T12:38:52",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "bonsai2-27b-ptq1",
   "profile": "extraction",
   "accuracy": 98.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 59,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "2f2913b42c03fae2",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T12:34:08",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "bonsai2-27b-ptq1",
   "profile": "extraction",
   "accuracy": 82.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 33,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "40c68eea7655c7e1",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T10:23:45",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen38-27b-q3kxl-chad",
   "profile": "extraction",
   "accuracy": 84.8,
   "coverage": 55.0,
   "measured": 33,
   "unmeasured": 27,
   "attempted": 60,
   "passed": 28,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "e4109198ec0d6a3f",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T10:19:21",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "qwen38-27b-q3kxl-chad",
   "profile": "extraction",
   "accuracy": 92.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 37,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "7f85527901732b16",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T10:12:46",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "qwen38-27b-q3kxl-chad",
   "profile": "extraction",
   "accuracy": 85.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 51,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "a1f473e0d63ec1b3",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T09:07:04",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "bonsai2-27b-ptq1",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 9,
   "unmeasured": 0,
   "attempted": 9,
   "passed": 9,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "40c68eea7655c7e1",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T09:05:27",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "bonsai2-27b-ptq1",
   "profile": "agentic-fast",
   "accuracy": 97.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 39,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "92433e8a2c1d7275",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T08:47:13",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "bonsai2-27b-ptq1",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "3e70e301928f2a3d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T08:44:20",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "bonsai2-27b-ptq1",
   "profile": "extraction",
   "accuracy": 83.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 50,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "522c48da0b9474a3",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T08:37:51",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "bonsai2-27b-ptq1",
   "profile": "extraction",
   "accuracy": 31.9,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 23,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "2641ce03c27a545e",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T08:11:14",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "bonsai2-27b-pq2",
   "profile": "extraction",
   "accuracy": 66.4,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 83,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "1685634443e1889f",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T08:06:51",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "bonsai2-27b-pq2",
   "profile": "study-nothink",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 48,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "3e65308eee2e3c69",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T08:03:59",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "bonsai2-27b-pq2",
   "profile": "extraction",
   "accuracy": 98.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 59,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "75d8d5feb45c7b60",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T08:00:01",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "bonsai2-27b-pq2",
   "profile": "extraction",
   "accuracy": 82.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 33,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "61b04f512aeac639",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T07:57:23",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "bonsai2-27b-pq2",
   "profile": "agentic-fast",
   "accuracy": 97.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 39,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "c76ee9534d06b617",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T07:43:18",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "bonsai2-27b-pq2",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "fa07c48a3877570f",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T07:41:00",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "bonsai2-27b-pq2",
   "profile": "extraction",
   "accuracy": 83.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 50,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "25c89df1304600f0",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-18T07:32:38",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "bonsai2-27b-pq2",
   "profile": "extraction",
   "accuracy": 31.9,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 23,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "537688bcd022a4e2",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T20:04:58",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3.6-35b-swiftlm",
   "profile": "extraction",
   "accuracy": 88.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 53,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "7d05a9cf37c57f7c",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T18:50:40",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "ornith-35b-a3b-q40-recheck",
   "profile": "extraction",
   "accuracy": 85.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 34,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "2124361885388e32",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T18:49:12",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "edge0-35b-recheck",
   "profile": "extraction",
   "accuracy": 88.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 53,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "f685f626196f2883",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T18:46:48",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "ternary-bonsai-27b-recheck",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "a5afffb46b0e50c4",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T18:40:52",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3.8-9b-distill-recheck",
   "profile": "extraction",
   "accuracy": 91.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 55,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "4b6dbb06475bbbad",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T18:38:42",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "gemma4-12b-recheck",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "321d8d33ba14ed0c",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T18:36:41",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "btl-4-q4km-recheck",
   "profile": "extraction",
   "accuracy": 91.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 55,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "86e42deccfe15a63",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T18:01:38",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "minicpm5-2b-q8-nojinja",
   "profile": "extraction",
   "accuracy": 76.9,
   "coverage": 99.2,
   "measured": 238,
   "unmeasured": 2,
   "attempted": 240,
   "passed": 183,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "0153d351e0ec30ee",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T17:57:03",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "minicpm5-2b-q8-np4-today",
   "profile": "extraction",
   "accuracy": 76.9,
   "coverage": 99.2,
   "measured": 238,
   "unmeasured": 2,
   "attempted": 240,
   "passed": 183,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "0d285fb9372dab00",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T17:52:22",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "minicpm5-2b-q8-np1",
   "profile": "extraction",
   "accuracy": 57.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 23,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "ff823de24a74e591",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T17:52:02",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "minicpm5-2b-q8-np1",
   "profile": "extraction",
   "accuracy": 78.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 47,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "e59bb419db23bea7",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T17:51:20",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "minicpm5-2b-q8-np1",
   "profile": "extraction",
   "accuracy": 56.0,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 70,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "5ac3f164bce7795b",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T17:50:49",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "minicpm5-2b-q8-np1",
   "profile": "extraction",
   "accuracy": 57.4,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 31,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "b51222c2ca38cfdf",
   "configuration_id": "1fb5d909364039d2",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T17:50:02",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "minicpm5-2b-q8-np1",
   "profile": "extraction",
   "accuracy": 52.9,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 127,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "070be56608b35405",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T17:46:23",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "minicpm5-2b-q8-np1",
   "profile": "extraction",
   "accuracy": 83.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 50,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "12f2ce70461eb2ba",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T17:45:47",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "minicpm5-2b-q8-np1",
   "profile": "study-nothink",
   "accuracy": 79.2,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 38,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "d8638e6e2d8d038b",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T17:44:58",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "minicpm5-2b-q8-np1",
   "profile": "extraction",
   "accuracy": 76.7,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 184,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "c1fd37845b5547c8",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T17:38:49",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "minicpm5-2b-exl3",
   "profile": "extraction",
   "accuracy": 37.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 15,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "f0bd4498f4e1a804",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T17:37:37",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "minicpm5-2b-exl3",
   "profile": "extraction",
   "accuracy": 75.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 45,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "d51ee21c635d7430",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T17:35:03",
   "suite": "needle",
   "suite_version": "needle/1",
   "model": "minicpm5-2b-exl3",
   "profile": "extraction",
   "accuracy": 85.5,
   "coverage": 100.0,
   "measured": 110,
   "unmeasured": 0,
   "attempted": 110,
   "passed": 94,
   "selftests": 11,
   "dataset_id": "5c5199b9eee8a595",
   "grader_id": "c38dec33aa1c86c2",
   "configuration_id": "2b303f0a51347e71",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T16:29:33",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "minicpm5-2b-exl3",
   "profile": "extraction",
   "accuracy": 60.0,
   "coverage": 80.0,
   "measured": 100,
   "unmeasured": 25,
   "attempted": 125,
   "passed": 60,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "deed1276a964eb4d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T16:28:16",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "minicpm5-2b-exl3",
   "profile": "extraction",
   "accuracy": 59.3,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 32,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "b51222c2ca38cfdf",
   "configuration_id": "50da51be20c4e83c",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T16:25:57",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "minicpm5-2b-exl3",
   "profile": "extraction",
   "accuracy": 46.2,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 111,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "dfef0a54e96403a1",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T16:18:04",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "minicpm5-2b-exl3",
   "profile": "extraction",
   "accuracy": 76.2,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 183,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "58ec2a66d452536c",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T16:08:12",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "minicpm5-2b-exl3",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "1564c01170861ff4",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T16:05:55",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "minicpm5-2b-exl3",
   "profile": "study-nothink",
   "accuracy": 81.2,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 39,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "18bb944eee207790",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T09:48:01",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "ternary-bonsai-8b",
   "profile": "extraction",
   "accuracy": 61.2,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 147,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "1a6187eaf0f19f79",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T09:32:38",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "ternary-bonsai-8b",
   "profile": "extraction",
   "accuracy": 43.8,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 105,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "1b7169aea80ec684",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T09:00:34",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "ternary-bonsai-8b",
   "profile": "extraction",
   "accuracy": 36.4,
   "coverage": 100.0,
   "measured": 11,
   "unmeasured": 0,
   "attempted": 11,
   "passed": 4,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "1b7169aea80ec684",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T09:00:00",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "ternary-bonsai-8b",
   "profile": "extraction",
   "accuracy": 70.4,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 38,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "b51222c2ca38cfdf",
   "configuration_id": "2449c0a03ec229c7",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T08:58:31",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "ternary-bonsai-8b",
   "profile": "extraction",
   "accuracy": 22.9,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 11,
   "selftests": 17,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "cc2d23c5bae45474",
   "configuration_id": "b5bcdf7e22a61983",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T08:58:12",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "ternary-bonsai-8b",
   "profile": "study-nothink",
   "accuracy": 72.9,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 35,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "22502c57cf79b73d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T08:57:30",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "ternary-bonsai-8b",
   "profile": "extraction",
   "accuracy": 51.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 64,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "a964482c0a6fefb9",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T08:55:50",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "ternary-bonsai-8b",
   "profile": "extraction",
   "accuracy": 86.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 52,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "40d9d3c65e78e666",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T08:54:50",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "ternary-bonsai-8b",
   "profile": "extraction",
   "accuracy": 60.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 36,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "00939f8fd32a28ff",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T08:53:37",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "ternary-bonsai-8b",
   "profile": "extraction",
   "accuracy": 37.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 15,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "0b08940f0421efe8",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T08:53:05",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "ternary-bonsai-8b",
   "profile": "extraction",
   "accuracy": 11.1,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 8,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "409ab3394165e872",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T08:52:24",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "ternary-bonsai-8b",
   "profile": "agentic-fast",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 14,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "d3347a668da25f1d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T08:52:08",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "ternary-bonsai-8b",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "4a4dea00c5224630",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-17T08:48:50",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "ternary-bonsai-8b",
   "profile": "extraction",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 0,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "2ffaff38658dd85a",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-16T23:32:01",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "minicpm5-2b-exl3",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 13,
   "unmeasured": 0,
   "attempted": 13,
   "passed": 13,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "c0f3ff2bda3a17e8",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-16T23:24:03",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "minicpm5-2b-exl3",
   "profile": "extraction",
   "accuracy": 80.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 32,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "d413710f2e637004",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-16T09:54:50",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "edge0-35b",
   "profile": "extraction",
   "accuracy": 91.7,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 220,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "7b18c32f56523945",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-16T09:15:19",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "edge0-35b",
   "profile": "extraction",
   "accuracy": 88.3,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 212,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "7149c1e662a034c9",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-16T08:48:13",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "edge0-35b",
   "profile": "extraction",
   "accuracy": 85.5,
   "coverage": 100.0,
   "measured": 83,
   "unmeasured": 0,
   "attempted": 83,
   "passed": 71,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "7149c1e662a034c9",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-16T08:23:43",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "edge0-35b",
   "profile": "extraction",
   "accuracy": 81.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 49,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "0955df95170a937b",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-16T08:17:31",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "edge0-35b",
   "profile": "extraction",
   "accuracy": 88.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 53,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "60a41f7096ed2f0d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-16T08:11:48",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "edge0-35b",
   "profile": "extraction",
   "accuracy": 77.8,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 42,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "b51222c2ca38cfdf",
   "configuration_id": "a0c2477ba575e774",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-16T07:58:12",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "edge0-35b",
   "profile": "study-nothink",
   "accuracy": 97.9,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 47,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "63bebcfa1d0c9c54",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-16T07:54:52",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "edge0-35b",
   "profile": "extraction",
   "accuracy": 52.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 21,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "c4a49113da942eeb",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-16T07:51:33",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "edge0-35b",
   "profile": "extraction",
   "accuracy": 89.6,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 112,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "eb1910ec33534645",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T20:05:49",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "minicpm5-2b",
   "profile": "extraction",
   "accuracy": 22.9,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 55,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "8d6eef850a9bedcc",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T20:02:56",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "minicpm5-2b",
   "profile": "extraction",
   "accuracy": 51.7,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 124,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "8bc8889aa4bec4ee",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T19:59:11",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "minicpm5-2b",
   "profile": "extraction",
   "accuracy": 59.3,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 32,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "b51222c2ca38cfdf",
   "configuration_id": "a602b52d4aa5d258",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T19:58:24",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "minicpm5-2b",
   "profile": "extraction",
   "accuracy": 70.8,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 34,
   "selftests": 17,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "cc2d23c5bae45474",
   "configuration_id": "09ef74ce83b54a17",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T19:58:02",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "minicpm5-2b",
   "profile": "study-nothink",
   "accuracy": 77.1,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 37,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "8bf3c6f0ca1fa01a",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T19:57:45",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "minicpm5-2b",
   "profile": "extraction",
   "accuracy": 62.4,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 78,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "8f32a36bd07ac12d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T19:57:13",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "minicpm5-2b",
   "profile": "extraction",
   "accuracy": 85.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 51,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "859326c246bacb4d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T19:56:32",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "minicpm5-2b",
   "profile": "extraction",
   "accuracy": 60.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 36,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "602e85a1c355f821",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T19:55:45",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "minicpm5-2b",
   "profile": "extraction",
   "accuracy": 37.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 15,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "d8fc1797ed992595",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T19:55:16",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "minicpm5-2b",
   "profile": "extraction",
   "accuracy": 31.9,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 23,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "a792a14d7d990d1b",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T19:54:58",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "minicpm5-2b",
   "profile": "agentic-fast",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 14,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "29d5dc1655f1e02d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T19:54:14",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "minicpm5-2b",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "2400bd439331eb46",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T19:52:12",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "minicpm5-2b",
   "profile": "extraction",
   "accuracy": 73.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 44,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "d5030ed4472a8b63",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T10:46:43",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "btl-4-q4km",
   "profile": "extraction",
   "accuracy": 71.3,
   "coverage": 68.3,
   "measured": 164,
   "unmeasured": 76,
   "attempted": 240,
   "passed": 117,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "53f41ca547c1a228",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T10:16:43",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "btl-4-q4km",
   "profile": "extraction",
   "accuracy": 71.3,
   "coverage": 100.0,
   "measured": 164,
   "unmeasured": 0,
   "attempted": 164,
   "passed": 117,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "53f41ca547c1a228",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T10:05:16",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "btl-4-q4km",
   "profile": "extraction",
   "accuracy": 87.1,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 209,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "10dcbcd1b3bb7b91",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T09:38:58",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "btl-4-q4km",
   "profile": "extraction",
   "accuracy": 81.5,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 44,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "b51222c2ca38cfdf",
   "configuration_id": "3b3b53961682ee7f",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T09:35:26",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "btl-4-q4km",
   "profile": "extraction",
   "accuracy": 70.8,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 34,
   "selftests": 17,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "cc2d23c5bae45474",
   "configuration_id": "c5f63e89ad62fa86",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T09:34:09",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "btl-4-q4km",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "9b7c71d675210c91",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T09:29:36",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "btl-4-q4km",
   "profile": "study-nothink",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 48,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "0689f579cb5fa875",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T09:28:52",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "btl-4-q4km",
   "profile": "extraction",
   "accuracy": 76.8,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 96,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "b6421e71dc1b07d5",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T09:26:45",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "btl-4-q4km",
   "profile": "extraction",
   "accuracy": 91.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 55,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "df6b9c70d3ecb99c",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T09:25:19",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "btl-4-q4km",
   "profile": "extraction",
   "accuracy": 83.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 50,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "34ecdb4bf4ded955",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T09:23:27",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "btl-4-q4km",
   "profile": "extraction",
   "accuracy": 55.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 22,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "7cf929c72ed89ccb",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T09:22:19",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "btl-4-q4km",
   "profile": "extraction",
   "accuracy": 44.4,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 32,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "43c9232ce793d709",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T09:21:33",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "btl-4-q4km",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 15,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "a04248bbed30bba3",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T09:20:49",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "btl-4-q4km",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "83afd32cbd70262a",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T09:17:28",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "btl-4-q4km",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "0adbf9c03a0169c7",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T09:14:38",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "btl-4",
   "profile": "extraction",
   "accuracy": 91.7,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 220,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "f9c64867e9dfb91b",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T08:50:51",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "btl-4",
   "profile": "extraction",
   "accuracy": 85.0,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 204,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "81d3e449bf19eaea",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T08:28:10",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "btl-4",
   "profile": "extraction",
   "accuracy": 87.0,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 47,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "b51222c2ca38cfdf",
   "configuration_id": "0b71856ba217c91c",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T08:24:12",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "btl-4",
   "profile": "extraction",
   "accuracy": 64.6,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 31,
   "selftests": 17,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "cc2d23c5bae45474",
   "configuration_id": "bf1f8546a0e1b97c",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T08:22:37",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "btl-4",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "8dbbe23a8427e3bc",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T08:08:05",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "btl-4",
   "profile": "study-nothink",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 48,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "438aa2f62c6153af",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T08:06:54",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "btl-4",
   "profile": "extraction",
   "accuracy": 76.8,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 96,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "7d9f8293fd874d62",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T08:02:31",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "btl-4",
   "profile": "extraction",
   "accuracy": 95.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 57,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "8197a1913a41df95",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T07:59:37",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "btl-4",
   "profile": "extraction",
   "accuracy": 80.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 48,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "11c6d6217d23f0b7",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T07:55:59",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "btl-4",
   "profile": "extraction",
   "accuracy": 62.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 25,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "0104404569f7c535",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T07:53:40",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "btl-4",
   "profile": "extraction",
   "accuracy": 36.1,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 26,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "9b2dcb073c569fce",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T07:52:06",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "btl-4",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 15,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "e43140ad3e77d008",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T07:50:45",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "btl-4",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "962089edc90681de",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T07:44:23",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "btl-4",
   "profile": "extraction",
   "accuracy": 76.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 46,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "7d3d7741437436e1",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T04:09:57",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "bonsai-27b-1bit",
   "profile": "extraction",
   "accuracy": 90.8,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 218,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "f3c4a61093079a3e",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T03:11:39",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "bonsai-27b-1bit",
   "profile": "extraction",
   "accuracy": 85.4,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 205,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "d75dadae6c07b420",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T03:10:44",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "bonsai-27b-mlx2-cortext",
   "profile": "extraction",
   "accuracy": 88.8,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 213,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "67cfcfc5c61806c7",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-15T02:00:46",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "bonsai-27b-mlx2-cortext",
   "profile": "extraction",
   "accuracy": 87.5,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 210,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "a7bdc6604033d727",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-15T01:08:16",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "bonsai-8b-1bit",
   "profile": "extraction",
   "accuracy": 41.7,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 100,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "d5d5bc2627ca6b54",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T00:37:47",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "bonsai-27b-1bit",
   "profile": "extraction",
   "accuracy": 79.6,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 43,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "b51222c2ca38cfdf",
   "configuration_id": "612035dbfdcf9143",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T00:32:31",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "bonsai-8b-1bit",
   "profile": "extraction",
   "accuracy": 28.3,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 68,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "e931e287089b3229",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T00:25:16",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "bonsai-27b-1bit",
   "profile": "extraction",
   "accuracy": 25.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 12,
   "selftests": 17,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "cc2d23c5bae45474",
   "configuration_id": "f3e9344f382b675d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T00:18:59",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "bonsai-27b-1bit",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "315259486d2b1e0e",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-15T00:00:28",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "bonsai-27b-mlx2-cortext",
   "profile": "extraction",
   "accuracy": 72.2,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 39,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "b51222c2ca38cfdf",
   "configuration_id": "c6981047f9ae77d3",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:43:58",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "bonsai-27b-1bit",
   "profile": "study-nothink",
   "accuracy": 87.5,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 42,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "2f1d2e8b3a422ec6",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:41:33",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "bonsai-27b-mlx2-cortext",
   "profile": "extraction",
   "accuracy": 75.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 36,
   "selftests": 17,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "cc2d23c5bae45474",
   "configuration_id": "1db1392d6a3fe16b",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:40:13",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "bonsai-27b-1bit",
   "profile": "extraction",
   "accuracy": 67.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 84,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "0f1cdb00af0adb60",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:33:48",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "bonsai-27b-mlx2-m4pro",
   "profile": "extraction",
   "accuracy": 89.2,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 214,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "29c36e514c29b1c2",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:33:03",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "bonsai-27b-1bit",
   "profile": "extraction",
   "accuracy": 96.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 58,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "06fe499419357acb",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:26:52",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "bonsai-8b-1bit",
   "profile": "extraction",
   "accuracy": 61.1,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 33,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "b51222c2ca38cfdf",
   "configuration_id": "e43f51321b57c883",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:26:01",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "bonsai-27b-1bit",
   "profile": "extraction",
   "accuracy": 35.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 21,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "b6b3f7a5739fbaf7",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:22:32",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "bonsai-27b-1bit",
   "profile": "extraction",
   "accuracy": 40.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 16,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "c66967b2cb84959c",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:20:06",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "bonsai-27b-1bit",
   "profile": "extraction",
   "accuracy": 18.1,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 13,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "b6bf04bcfeb45185",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:19:58",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "bonsai-8b-1bit",
   "profile": "extraction",
   "accuracy": 20.8,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 10,
   "selftests": 17,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "cc2d23c5bae45474",
   "configuration_id": "a575849c8bd49029",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:18:59",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "bonsai-8b-1bit",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "6d222e16c742210f",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:15:19",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "bonsai-27b-1bit",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 15,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "a9cb56f68c317ce8",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:08:32",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "bonsai-27b-1bit",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "8ee047a8f4681484",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:08:10",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "bonsai-8b-1bit",
   "profile": "study-nothink",
   "accuracy": 79.2,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 38,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "f1ce82d657bdf4e2",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:05:11",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "bonsai-8b-1bit",
   "profile": "extraction",
   "accuracy": 55.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 69,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "1ac5570cbae78ecd",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T23:00:33",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "bonsai-8b-1bit",
   "profile": "extraction",
   "accuracy": 86.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 52,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "7b113b84e3e9ea89",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T22:55:42",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "bonsai-8b-1bit",
   "profile": "extraction",
   "accuracy": 65.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 39,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "49c73224f7b988ec",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T22:50:17",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "bonsai-8b-1bit",
   "profile": "extraction",
   "accuracy": 40.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 16,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "1f1610844ef40acb",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T22:48:32",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "bonsai-8b-1bit",
   "profile": "extraction",
   "accuracy": 19.4,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 14,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "de1a0f97f9063521",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T22:46:58",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "bonsai-8b-1bit",
   "profile": "agentic-fast",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 14,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "79e4f8448168d640",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T22:46:15",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "bonsai-8b-1bit",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "6fc2b443c7d76dde",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T22:44:17",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "bonsai-27b-mlx2-m4pro",
   "profile": "extraction",
   "accuracy": 87.9,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 211,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "821ca1ce68849b25",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T22:41:07",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "bonsai-27b-1bit",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "0f6815eef83b64d0",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T22:37:37",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "bonsai-8b-1bit",
   "profile": "extraction",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 0,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "516c9c0e0de06220",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T21:59:29",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "bonsai-27b-mlx2-cortext",
   "profile": "study-nothink",
   "accuracy": 89.6,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 43,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "f8818645d3ba1522",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T21:55:44",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "bonsai-27b-mlx2-cortext",
   "profile": "extraction",
   "accuracy": 71.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 89,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "4f595a317a905809",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T21:51:08",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "bonsai-27b-mlx2-cortext",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "64e14fdadbcea0e6",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T21:48:25",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "bonsai-27b-mlx2-m4pro",
   "profile": "extraction",
   "accuracy": 72.2,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 39,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "b51222c2ca38cfdf",
   "configuration_id": "e41b038c77c413b8",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T21:46:43",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "bonsai-27b-mlx2-cortext",
   "profile": "extraction",
   "accuracy": 66.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 40,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "a3e244ab46faf1d3",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T21:42:44",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "bonsai-27b-mlx2-m4pro",
   "profile": "extraction",
   "accuracy": 75.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 36,
   "selftests": 17,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "cc2d23c5bae45474",
   "configuration_id": "5836be70b11516a5",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T21:41:18",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "bonsai-27b-mlx2-cortext",
   "profile": "extraction",
   "accuracy": 42.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 17,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "88e5e9056622def0",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T21:38:17",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "bonsai-27b-mlx2-cortext",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 15,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "8c56b2e4e619015f",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T21:35:53",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "bonsai-27b-mlx2-cortext",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "bf7830ad8b1e5adb",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T21:22:56",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "bonsai-27b-mlx2-cortext",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "072e0651b5a1976c",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T20:51:28",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "bonsai-27b-mlx2-m4pro",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "16fb378c7bf6e96a",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T20:39:44",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "ternary-bonsai-27b",
   "profile": "extraction",
   "accuracy": 90.0,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 216,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "29960d736fce3ea5",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T19:49:55",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "ternary-bonsai-27b",
   "profile": "extraction",
   "accuracy": 85.8,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 206,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "773ed13c720ca034",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:48:01",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "ternary-bonsai-27b",
   "profile": "extraction",
   "accuracy": 72.2,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 39,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "b51222c2ca38cfdf",
   "configuration_id": "075725332e318d6b",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:44:18",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "bonsai-27b-mlx2-m4pro",
   "profile": "study-nothink",
   "accuracy": 89.6,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 43,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "3c03af0d96b59518",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:41:31",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "ternary-bonsai-27b",
   "profile": "extraction",
   "accuracy": 75.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 36,
   "selftests": 17,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "cc2d23c5bae45474",
   "configuration_id": "e5777a35ca1e5f10",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:40:10",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "bonsai-27b-mlx2-m4pro",
   "profile": "extraction",
   "accuracy": 71.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 89,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "86fb3d4c7b477c6c",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:39:23",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "ternary-bonsai-27b",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "078f17fd21ba7d57",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:35:20",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "bonsai-27b-mlx2-m4pro",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "4082207a1256c4b6",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:30:40",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "bonsai-27b-mlx2-m4pro",
   "profile": "extraction",
   "accuracy": 66.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 40,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "14f4738d0f91fc84",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:24:53",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "bonsai-27b-mlx2-m4pro",
   "profile": "extraction",
   "accuracy": 42.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 17,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "a4f61e2b7e694288",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:21:41",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "bonsai-27b-mlx2-m4pro",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 15,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "98ea18969989ee4e",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:19:07",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "bonsai-27b-mlx2-m4pro",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "52e205dd25e5567b",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:11:21",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "ternary-bonsai-27b",
   "profile": "study-nothink",
   "accuracy": 85.4,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 41,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "321e00a36a5faaeb",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:08:37",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "ternary-bonsai-27b",
   "profile": "extraction",
   "accuracy": 68.8,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 86,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "3b0afe72fb0c0121",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:04:25",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "bonsai-27b-mlx2-m4pro",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "8e73d545a7ffd2aa",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:04:22",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "ternary-bonsai-27b",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "c5097500538c3719",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T18:00:14",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "ternary-bonsai-27b",
   "profile": "extraction",
   "accuracy": 66.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 40,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "6a368a22cfb7fb02",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T17:55:38",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "ternary-bonsai-27b",
   "profile": "extraction",
   "accuracy": 42.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 17,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "3a4970c1d0eed32d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T17:53:50",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "ternary-bonsai-27b",
   "profile": "extraction",
   "accuracy": 27.8,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 20,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "004a863515b30c86",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T17:51:23",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "ternary-bonsai-27b",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 15,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "7f71a1d06f005b6d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T17:48:21",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "ternary-bonsai-27b",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "13185468aa0f107f",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T17:33:57",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "ternary-bonsai-27b",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "7dece85f852d6798",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T16:45:00",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "qwen3.8-9b-distill",
   "profile": "extraction",
   "accuracy": 90.0,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 216,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "d7923863e8b56e76",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T16:22:02",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "qwen3.8-9b-distill",
   "profile": "extraction",
   "accuracy": 83.3,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 200,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "d2642a91c9a03367",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T15:52:33",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "qwen3.8-9b-distill",
   "profile": "extraction",
   "accuracy": 79.6,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 43,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "b51222c2ca38cfdf",
   "configuration_id": "d537736901f01a81",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T15:49:34",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "qwen3.8-9b-distill",
   "profile": "extraction",
   "accuracy": 25.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 12,
   "selftests": 17,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "cc2d23c5bae45474",
   "configuration_id": "5a052624057c52ff",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T15:48:54",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "qwen3.8-9b-distill",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "43d9a11df6da4927",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T15:47:50",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "qwen3.8-9b-distill",
   "profile": "study-nothink",
   "accuracy": 89.6,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 43,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "a0e9adc5c6542d53",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T15:46:54",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3.8-9b-distill",
   "profile": "extraction",
   "accuracy": 67.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 84,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "696fd8b30eed53f0",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T15:45:22",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3.8-9b-distill",
   "profile": "extraction",
   "accuracy": 91.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 55,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "5ac25f36beb78f8c",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T15:43:26",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "qwen3.8-9b-distill",
   "profile": "extraction",
   "accuracy": 83.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 50,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "8481e2a6e1f3e60c",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T15:40:30",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "qwen3.8-9b-distill",
   "profile": "extraction",
   "accuracy": 40.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 16,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "6a67c959e17c98f5",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T15:38:46",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3.8-9b-distill",
   "profile": "extraction",
   "accuracy": 18.1,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 13,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "56c7dc8f982d59d3",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T15:37:51",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "qwen3.8-9b-distill",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 15,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "2411d23420a9fcb7",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T15:36:47",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "qwen3.8-9b-distill",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "13a2f2f87c9bcdc8",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-14T15:31:52",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "qwen3.8-9b-distill",
   "profile": "extraction",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 0,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "2cd02d2d291d660a",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T09:09:59",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "qwen3.5-4b-base",
   "profile": "study-nothink",
   "accuracy": 85.4,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 41,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "211ae40d03020e0e",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T09:09:28",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "qwen3.5-4b-base",
   "profile": "extraction",
   "accuracy": 93.8,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 225,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "50a352dbe249d795",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T08:56:41",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "qwen3.5-4b-base",
   "profile": "extraction",
   "accuracy": 40.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 16,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "314d1643f96e91a9",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T08:55:53",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3.5-4b-base",
   "profile": "extraction",
   "accuracy": 74.4,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 93,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "0f7dbcd0b1d1afd4",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T08:55:04",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3.5-4b-base",
   "profile": "extraction",
   "accuracy": 33.3,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 24,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "8ed6c2d36dcf6e58",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T08:54:34",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "qwen3.5-4b-base",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "634f8ac324d31747",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T08:52:23",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "qwen3.5-4b-base",
   "profile": "extraction",
   "accuracy": 36.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 22,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "abdddca6a1cde4d9",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T08:51:45",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "neohorse-1-4b",
   "profile": "study-nothink",
   "accuracy": 75.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 36,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "a4f7cdc18f2e769e",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T08:51:15",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "neohorse-1-4b",
   "profile": "extraction",
   "accuracy": 86.2,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 207,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "78196673fb9772d8",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T08:36:44",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "neohorse-1-4b",
   "profile": "extraction",
   "accuracy": 40.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 16,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "0ca5e9b594862393",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T08:36:00",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "neohorse-1-4b",
   "profile": "extraction",
   "accuracy": 63.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 79,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "b382340e2e3ad6fa",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T08:35:12",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "neohorse-1-4b",
   "profile": "extraction",
   "accuracy": 27.8,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 20,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "e55e57d062c38080",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T08:34:46",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "neohorse-1-4b",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "b90d64d3c6c193a5",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T08:32:36",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "neohorse-1-4b",
   "profile": "extraction",
   "accuracy": 56.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 34,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "e018b8e820715109",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-13T00:28:57",
   "suite": "needle",
   "suite_version": "needle/1",
   "model": "ornith-9b-kv-q40",
   "profile": "extraction",
   "accuracy": 55.5,
   "coverage": 100.0,
   "measured": 110,
   "unmeasured": 0,
   "attempted": 110,
   "passed": 61,
   "selftests": 11,
   "dataset_id": "5c5199b9eee8a595",
   "grader_id": "c38dec33aa1c86c2",
   "configuration_id": "09bf0b1e19047eb2",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T22:33:06",
   "suite": "needle",
   "suite_version": "needle/1",
   "model": "ornith-9b-kv-f16",
   "profile": "extraction",
   "accuracy": 58.2,
   "coverage": 100.0,
   "measured": 110,
   "unmeasured": 0,
   "attempted": 110,
   "passed": 64,
   "selftests": 11,
   "dataset_id": "5c5199b9eee8a595",
   "grader_id": "c38dec33aa1c86c2",
   "configuration_id": "72650e0b4ba22c6c",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T17:39:48",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "gemma4-12b-b10893",
   "profile": "extraction",
   "accuracy": 83.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 50,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "41030e0a832a129e",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T17:37:13",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "ornith-1.5-9b",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 48,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "54cc298fd20d22af",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T17:36:23",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "ornith-1.5-9b",
   "profile": "extraction",
   "accuracy": 68.8,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 33,
   "selftests": 17,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "cc2d23c5bae45474",
   "configuration_id": "1bf634506e78523a",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T17:34:15",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "ornith-1.5-9b",
   "profile": "extraction",
   "accuracy": 80.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 48,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "3fba7b7b697d2bc6",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T13:13:37",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "glm-4.7-flash-q40",
   "profile": "extraction",
   "accuracy": 44.4,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 32,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "a206ab6679bc0fed",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T13:12:48",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "glm-4.7-flash-q40",
   "profile": "extraction",
   "accuracy": 75.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 36,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "acf84972a0ad5f38",
   "configuration_id": "0550121d973797df",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T13:12:10",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "glm-4.7-flash-q40",
   "profile": "extraction",
   "accuracy": 37.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 15,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "9d24f8d5c84a9e2e",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T13:11:36",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "glm-4.7-flash-q40",
   "profile": "extraction",
   "accuracy": 60.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 36,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "b33ed27337b5ef7e",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T13:10:36",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "glm-4.7-flash-q40",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "5383698d8f3d7c82",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T13:07:07",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "glm-4.7-flash-q40",
   "profile": "extraction",
   "accuracy": 38.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 23,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "f492f0f6a7e78f98",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T13:06:19",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "ornith-9b-kvq4",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "970506005273665a",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T13:01:50",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "ornith-9b-kvq4",
   "profile": "extraction",
   "accuracy": 80.0,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 100,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "4c61ed44aed10e7a",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T13:00:37",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "ornith-9b-kvq4",
   "profile": "extraction",
   "accuracy": 56.9,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 41,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "32a57cca965b6777",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T12:14:18",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "ornith-35b-a3b-q40",
   "profile": "extraction",
   "accuracy": 85.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 34,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "36403d17c401159a",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T12:11:45",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "ornith-35b-a3b-q40",
   "profile": "extraction",
   "accuracy": 76.0,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 95,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "95c69b1144e96db8",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T12:09:29",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "ornith-35b-a3b-q40",
   "profile": "extraction",
   "accuracy": 36.1,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 26,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "4324ac01a6993c44",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T12:08:18",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "ornith-35b-a3b-q40",
   "profile": "extraction",
   "accuracy": 66.7,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 32,
   "selftests": 17,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "cc2d23c5bae45474",
   "configuration_id": "16c1587da02e495a",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T12:06:48",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "ornith-35b-a3b-q40",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "9ed2b4b41dc68126",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T12:02:50",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "ornith-35b-a3b-q40",
   "profile": "extraction",
   "accuracy": 80.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 48,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "9391215fff65e4b5",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T10:39:01",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "ornith-1.5-9b",
   "profile": "extraction",
   "accuracy": 62.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 25,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "ce57d773fcc0cbcd",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T10:37:31",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "ornith-1.5-9b",
   "profile": "extraction",
   "accuracy": 81.6,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 102,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "4a6ad96362484d18",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T10:36:18",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "ornith-1.5-9b",
   "profile": "extraction",
   "accuracy": 52.8,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 38,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "0dd3a4ce0c1655b8",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T10:35:37",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "ornith-1.5-9b",
   "profile": "extraction",
   "accuracy": 95.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 38,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "8105ecbc4488d189",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T10:31:50",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "ornith-1.5-9b",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "c2191a274d61789c",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T09:47:35",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "gpt-oss-20b-study-default",
   "profile": "study-default",
   "accuracy": 93.9,
   "coverage": 94.3,
   "measured": 33,
   "unmeasured": 2,
   "attempted": 35,
   "passed": 31,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "7649f91ca5eb87a4",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T09:27:13",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "gpt-oss-20b",
   "profile": "extraction",
   "accuracy": 72.2,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 52,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "262a6f5122bb9bbd",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T09:26:34",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "gpt-oss-20b",
   "profile": "extraction",
   "accuracy": 86.2,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 207,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "b9bd7f33427388fd",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-12T09:14:31",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "gpt-oss-20b",
   "profile": "extraction",
   "accuracy": 85.7,
   "coverage": 99.2,
   "measured": 238,
   "unmeasured": 2,
   "attempted": 240,
   "passed": 204,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "acb0e0e0f1883336",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T16:39:53",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen35-9b-mlx-mxfp4",
   "profile": "extraction",
   "accuracy": 34.7,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 25,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "da30e30743dadebe",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-11T16:36:30",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen35-9b-mlx-affine-g32",
   "profile": "extraction",
   "accuracy": 45.8,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 33,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "4cb05505e1d4b652",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-11T16:33:09",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen35-9b-mlx-affine-g64",
   "profile": "extraction",
   "accuracy": 59.7,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 43,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "987f7b78b76a6a60",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-11T16:29:50",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen35-9b-gguf-q40",
   "profile": "extraction",
   "accuracy": 6.9,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 5,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "e5fa21270174a387",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T16:28:57",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen35-9b-gguf-q4km",
   "profile": "extraction",
   "accuracy": 23.6,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 17,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "71580a34709d8143",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T15:16:53",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "qwen3.8-27b-q4_0",
   "profile": "extraction",
   "accuracy": 83.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 50,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "7bca9c0f60cfafa3",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T15:10:44",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3.8-27b-q4_0",
   "profile": "extraction",
   "accuracy": 69.4,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 50,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "0464331c99998f53",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T15:08:43",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "qwen3.8-27b-q4_0",
   "profile": "extraction",
   "accuracy": 80.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 32,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "4fb569ecc91140c9",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T15:05:26",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "qwen3.8-27b-q4_0",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "7dedd451e841393d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T14:52:47",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "qwen3.8-27b-q4_0",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "ce7aa224f6d8c444",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T14:48:25",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-36b-Q4_1",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 36,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "a1e412de2582e21b",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T14:42:37",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-36b-Q5_1",
   "profile": "extraction",
   "accuracy": 58.3,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 42,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "6e470ceac8f4401c",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T14:35:56",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-36b-Q5_0",
   "profile": "extraction",
   "accuracy": 56.9,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 41,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "53f943deaafde391",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T14:31:25",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "k2-36b-q4_0",
   "profile": "extraction",
   "accuracy": 76.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 46,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "4bc7f1a68df72179",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T14:29:44",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-36b-q4_0",
   "profile": "extraction",
   "accuracy": 65.3,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 47,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "c71d223f0ac921ec",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T14:28:08",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-36b-q4_0",
   "profile": "extraction",
   "accuracy": 57.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 23,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "d2167f66564312b7",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T14:26:59",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "k2-36b-q4_0",
   "profile": "extraction",
   "accuracy": 90.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 36,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "5467852ddcdb11c6",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T14:25:08",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "k2-36b-q4_0",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "646dbb2eca06be4d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T10:10:36",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "reap-60b-mlx-4bit-uniform",
   "profile": "extraction-t07",
   "accuracy": 28.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 17,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "d696a3a98ce65e4f",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-11T10:09:10",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "reap-60b-4bit-repro",
   "profile": "extraction",
   "accuracy": 25.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 15,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "9fc8dc4dd468ed18",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-11T09:55:19",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "reap-60b-mlx-4bit-uniform",
   "profile": "extraction",
   "accuracy": 85.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 51,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "5f382a39bf24731a",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-11T09:52:49",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "reap-60b-mlx-4bit-uniform",
   "profile": "extraction",
   "accuracy": 84.7,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 61,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "52349074a17dc8f5",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-11T09:50:38",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "reap-60b-mlx-4bit-uniform",
   "profile": "extraction",
   "accuracy": 47.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 19,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "76804a68d58c5eb6",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-11T09:49:09",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "reap-60b-mlx-4bit-uniform",
   "profile": "extraction",
   "accuracy": 95.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 38,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "a58faee14a0f3ed6",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-11T09:45:52",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "reap-60b-mlx-4bit-uniform",
   "profile": "extraction",
   "accuracy": 25.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 15,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "9fc8dc4dd468ed18",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-11T09:35:26",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "diffusiongemma-26b-mlx",
   "profile": "extraction",
   "accuracy": 80.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 48,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "6220b53efc9f35b4",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-11T09:32:26",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "diffusiongemma-26b-mlx",
   "profile": "extraction",
   "accuracy": 77.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 31,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "b0b19b278db1058f",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-11T09:29:38",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "diffusiongemma-26b-mlx",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "40a7e393e161d98f",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-11T09:27:14",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "diffusiongemma-26b-mlx",
   "profile": "extraction",
   "accuracy": 5.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 3,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "49640f29f7104a95",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-10T16:26:19",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "reap-60b-gguf-q4km",
   "profile": "extraction",
   "accuracy": 81.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 49,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "5d5c05e03d9c61e2",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T16:24:01",
   "suite": "VOID-degenerate-nullschema",
   "suite_version": "nullschema/1",
   "model": "reap-60b-gguf-q4km",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 72,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "b923bc8484d5f158",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T16:23:13",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "reap-60b-gguf-q4km",
   "profile": "extraction",
   "accuracy": 80.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 32,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "361b96fbc9be289c",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T16:21:42",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "reap-60b-gguf-q4km",
   "profile": "extraction",
   "accuracy": 87.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 35,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "437c75c01b4063cd",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T16:18:08",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "reap-60b-gguf-q4km",
   "profile": "extraction",
   "accuracy": 41.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 25,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "4faf8950bd5037a5",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T14:06:04",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "reap-60b-20260910",
   "profile": "extraction",
   "accuracy": 83.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 50,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "078d5fad4e6d0a0f",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T14:03:35",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "reap-60b-20260910",
   "profile": "extraction",
   "accuracy": 47.2,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 34,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "067da0ca87821991",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T14:01:26",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "reap-60b-20260910",
   "profile": "extraction",
   "accuracy": 47.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 19,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "a9d79c1de500e4b4",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T14:00:11",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "reap-60b-20260910",
   "profile": "extraction",
   "accuracy": 95.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 38,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "ce60a8f4e40a948a",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T13:56:54",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "reap-60b-20260910",
   "profile": "extraction",
   "accuracy": 78.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 47,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "d3a58130e11b1504",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T13:40:06",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "spark-x25-4b",
   "profile": "extraction",
   "accuracy": 28.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 17,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "ffe0798e394ad2c2",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T13:38:45",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "spark-x25-4b",
   "profile": "extraction",
   "accuracy": 48.6,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 35,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "080b2b956a3c69aa",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T13:38:09",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "spark-x25-4b",
   "profile": "extraction",
   "accuracy": 20.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 8,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "7eb3a5699dc38e49",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T13:37:41",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "spark-x25-4b",
   "profile": "extraction",
   "accuracy": 95.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 38,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "266788017096e178",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T13:35:43",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "spark-x25-4b",
   "profile": "extraction",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 0,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "bd8238a4b2ea134b",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T13:29:04",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "qwen38-rebase-b10893",
   "profile": "extraction",
   "accuracy": 77.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 31,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "90bd067cd59ed0c2",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T13:24:21",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen38-rebase-b10893",
   "profile": "extraction",
   "accuracy": 76.4,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 55,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "317df1bd5861fa04",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T13:02:25",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-36b-q8-precision",
   "profile": "extraction",
   "accuracy": 55.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 22,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "c95b13073128b76a",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T13:00:58",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-36b-q8-precision",
   "profile": "extraction",
   "accuracy": 70.8,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 51,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "3a41a0a4d5f18851",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:55:57",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-7b-bf16-precision",
   "profile": "extraction",
   "accuracy": 30.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 12,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "1705a034b7c84210",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:54:26",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-7b-bf16-precision",
   "profile": "extraction",
   "accuracy": 37.5,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 27,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "d931535c0ebf24a6",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:47:59",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-7b-q8-precision",
   "profile": "extraction",
   "accuracy": 30.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 12,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "e20e4b1b964fa50d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:47:01",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-7b-q8-precision",
   "profile": "extraction",
   "accuracy": 36.1,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 26,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "ed643e52a40b0e33",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:42:40",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-7b-q4-precision",
   "profile": "extraction",
   "accuracy": 37.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 15,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "3bbb4039e7a64a8b",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:41:42",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-7b-q4-precision",
   "profile": "extraction",
   "accuracy": 37.5,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 27,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "577908cfcedc7d78",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:37:26",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "k2-4b-bf16-precision",
   "profile": "extraction",
   "accuracy": 56.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 34,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "0faa5511a9576d51",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:35:23",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-4b-bf16-precision",
   "profile": "extraction",
   "accuracy": 43.1,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 31,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "776fc6d33aa8cc34",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:29:05",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-4b-bf16-precision",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 20,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "770c2ce3d2f16e96",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:27:37",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "k2-4b-bf16-precision",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "e21f6ac8887f7f5d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:26:10",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "k2-4b-bf16-precision",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "2c2e5971d9f6382f",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:24:24",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "k2-3.7b-q8-precision",
   "profile": "extraction",
   "accuracy": 56.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 34,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "3529bab4b75f7502",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:23:01",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-3.7b-q8-precision",
   "profile": "extraction",
   "accuracy": 43.1,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 31,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "807203bd18193834",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:18:48",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-3.7b-q8-precision",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 20,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "f654ce7b2d1ac877",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:17:52",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "k2-3.7b-q8-precision",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "1eedcfdecea95e86",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:16:41",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "k2-3.7b-q8-precision",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "ddb3ca71c12850c5",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:15:10",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "k2-3.7b-q4-precision",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 30,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "49e024d8aab08569",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:13:40",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-3.7b-q4-precision",
   "profile": "extraction",
   "accuracy": 38.9,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 28,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "b4a841d2cc241d1f",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:08:51",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-3.7b-q4-precision",
   "profile": "extraction",
   "accuracy": 52.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 21,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "324471b050d0453b",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:07:50",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "k2-3.7b-q4-precision",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "8bb7dfa4a8ce435f",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:06:33",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "k2-3.7b-q4-precision",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "9c699af63ecd43af",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:04:49",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "gemma4-12b-midrange",
   "profile": "extraction",
   "accuracy": 83.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 50,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "480148dc413153dc",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:01:35",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "gemma4-12b-midrange",
   "profile": "extraction",
   "accuracy": 87.5,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 63,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "c10333ca24baf0ea",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T12:00:16",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "gemma4-12b-midrange",
   "profile": "extraction",
   "accuracy": 97.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 39,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "4bbaa37911174c88",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T11:58:42",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "gemma4-12b-midrange",
   "profile": "extraction",
   "accuracy": 95.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 38,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "bf8723aa88c0d4b7",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T11:54:01",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "gemma4-12b-midrange",
   "profile": "extraction",
   "accuracy": 55.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 33,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "57bdac944281e060",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T11:50:46",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "lfm25-8b-a1b-midrange",
   "profile": "extraction",
   "accuracy": 53.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 32,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "089fdc592c6d831d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T11:49:48",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "lfm25-8b-a1b-midrange",
   "profile": "extraction",
   "accuracy": 27.8,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 20,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "b3d59631a0a0785e",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T11:49:28",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "lfm25-8b-a1b-midrange",
   "profile": "extraction",
   "accuracy": 30.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 12,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "717aaa5a6a785c10",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T11:48:54",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "lfm25-8b-a1b-midrange",
   "profile": "extraction",
   "accuracy": 82.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 33,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "4727b9e997fb1a2e",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T11:47:56",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "lfm25-8b-a1b-midrange",
   "profile": "extraction",
   "accuracy": 1.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 1,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "3b306a765082433d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T11:46:39",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "qwen3-8b-midrange",
   "profile": "extraction",
   "accuracy": 83.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 50,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "07092184bce46ed0",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T11:44:33",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3-8b-midrange",
   "profile": "extraction",
   "accuracy": 34.7,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 25,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "5fdc1502f911a148",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T11:39:49",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "qwen3-8b-midrange",
   "profile": "extraction",
   "accuracy": 45.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 18,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "39d8f0df5c0c0e6f",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T11:38:39",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "qwen3-8b-midrange",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "d27e6f32d963b586",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T11:34:54",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "qwen3-8b-midrange",
   "profile": "extraction",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 0,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "c341d6e4665581b5",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T11:28:53",
   "suite": "PARTIAL-KILLED-nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen38-vendor-qwen-think",
   "profile": "vendor-qwen-think",
   "accuracy": 56.2,
   "coverage": 100.0,
   "measured": 16,
   "unmeasured": 0,
   "attempted": 16,
   "passed": 9,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "e25e7e28827fd589",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T11:24:52",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "qwen38-vendor-qwen-think",
   "profile": "vendor-qwen-think",
   "accuracy": 77.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 31,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "e78fad9e97d45282",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T10:23:53",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "qwen38-gsqrco-iq3s",
   "profile": "extraction",
   "accuracy": 83.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 50,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "92ea3b5723445268",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T10:16:06",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen38-gsqrco-iq3s",
   "profile": "extraction",
   "accuracy": 62.5,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 45,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "62838af8410253c2",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T10:13:31",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "qwen38-gsqrco-iq3s",
   "profile": "extraction",
   "accuracy": 77.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 31,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "624bd85753786f4e",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T10:09:18",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "qwen38-gsqrco-iq3s",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "3177b556489558c1",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T09:54:32",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "qwen38-gsqrco-iq3s",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "d08352bc157ff4d3",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T09:31:58",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "bonsai-27b-ternary",
   "profile": "extraction",
   "accuracy": 66.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 40,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "ef7f6527771769a7",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T09:26:29",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "bonsai-27b-ternary",
   "profile": "extraction",
   "accuracy": 27.8,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 20,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "204142d8636f4fc2",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T09:23:53",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "bonsai-27b-ternary",
   "profile": "extraction",
   "accuracy": 42.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 17,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "121caa6718989cf9",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T09:21:06",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "bonsai-27b-ternary",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "83edc9129d9d77cd",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T09:08:04",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "bonsai-27b-ternary",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "b3a241de7392ea56",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T09:01:38",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "k2-36b-np1",
   "profile": "extraction",
   "accuracy": 87.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 35,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "6e72c541e7fd804e",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T08:59:34",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "k2-36b-np1",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "f4f31defced2ec18",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T08:58:12",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "qwen38-27b-np1",
   "profile": "extraction",
   "accuracy": 83.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 50,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "2867e80b35de0e49",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T08:49:12",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen38-27b-np1",
   "profile": "extraction",
   "accuracy": 76.4,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 55,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "458eb55eee2cc3a3",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T08:46:32",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "qwen38-27b-np1",
   "profile": "extraction",
   "accuracy": 77.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 31,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "cf60f4eb1ae532f8",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T08:41:40",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "qwen38-27b-np1",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "2b5ec5e83a2c8961",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T08:26:29",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "qwen38-27b-np1",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "7019f408382de810",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T00:44:28",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-36b-vendor-k2b",
   "profile": "vendor-k2b",
   "accuracy": 57.7,
   "coverage": 98.6,
   "measured": 71,
   "unmeasured": 1,
   "attempted": 72,
   "passed": 41,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "62630cca2615a18f",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T00:25:40",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "k2-36b-vendor-k2b",
   "profile": "vendor-k2b",
   "accuracy": 90.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 54,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "19ff3eed9d7b1a47",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T00:19:53",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-36b-vendor-k2b",
   "profile": "vendor-k2b",
   "accuracy": 82.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 33,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "d16f8472c1b1d87d",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T00:14:46",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "k2-36b-vendor-k2",
   "profile": "vendor-k2",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "b7e765413a3dd2f2",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-10T00:07:43",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-36b-vendor-k2",
   "profile": "vendor-k2",
   "accuracy": 48.6,
   "coverage": 97.2,
   "measured": 70,
   "unmeasured": 2,
   "attempted": 72,
   "passed": 34,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "80a8857ae078d242",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T23:40:47",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-36b-vendor-k2",
   "profile": "vendor-k2",
   "accuracy": 77.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 31,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "e6f5010ab18f79e7",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T23:35:36",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "k2-36b-extraction",
   "profile": "extraction",
   "accuracy": 75.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 45,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "b92607640a462939",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T23:32:48",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-36b-extraction",
   "profile": "extraction",
   "accuracy": 54.2,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 39,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "5e938f4fe9784ee9",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T23:29:35",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-36b-extraction",
   "profile": "extraction",
   "accuracy": 52.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 21,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "9cc504b6f261a5b3",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T22:00:58",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "k2-36b-ctx128k-q8",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "b3aa4c4b6efce314",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T21:49:38",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-36b-ctx128k-q8",
   "profile": "extraction",
   "accuracy": 65.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 26,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "082b24620003bf78",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T21:34:03",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "k2-36b-ctx32k",
   "profile": "extraction",
   "accuracy": 70.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 42,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "7873f3430a2f042e",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T21:31:31",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "k2-36b-ctx32k",
   "profile": "extraction",
   "accuracy": 90.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 36,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "b2321bc042bae630",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T21:29:30",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "k2-36b-ctx32k",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "3b6cfef3b606fe07",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T21:28:39",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "k2-36b-ctx32k",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "b3aa4c4b6efce314",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T21:12:35",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "k2-36b-ctx32k",
   "profile": "extraction",
   "accuracy": 90.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 54,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "af55e22558068ab0",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T21:10:59",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-36b-ctx32k",
   "profile": "extraction",
   "accuracy": 54.2,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 39,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "345de57f0f3a686f",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T21:05:29",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-36b-ctx32k",
   "profile": "extraction",
   "accuracy": 65.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 26,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "082b24620003bf78",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T20:54:47",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-36b-a4b-q4",
   "profile": "extraction",
   "accuracy": 78.9,
   "coverage": 100.0,
   "measured": 19,
   "unmeasured": 0,
   "attempted": 19,
   "passed": 15,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "082b24620003bf78",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T20:54:16",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "k2-36b-a4b-q4",
   "profile": "extraction",
   "accuracy": 73.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 44,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "7873f3430a2f042e",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T20:47:48",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "k2-36b-a4b-q4",
   "profile": "extraction",
   "accuracy": 90.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 36,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "b2321bc042bae630",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T20:45:42",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "k2-36b-a4b-q4",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "3b6cfef3b606fe07",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T19:43:12",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-7b",
   "profile": "extraction",
   "accuracy": 30.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 12,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "e229d76d7d696757",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T19:42:16",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "k2-3.7b",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 20,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "13dc41ad1b2821d9",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T19:41:20",
   "suite": "worldview",
   "suite_version": "worldview/2",
   "model": "gemma4-12b",
   "profile": "extraction",
   "accuracy": 97.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 39,
   "selftests": 10,
   "dataset_id": "e97835de9e6fc227",
   "grader_id": "9acfba1b704feb28",
   "configuration_id": "585e7383f726d519",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T19:26:36",
   "suite": "worldview",
   "suite_version": "worldview/1",
   "model": "reap-60b",
   "profile": "extraction",
   "accuracy": 47.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 19,
   "selftests": 10,
   "dataset_id": "d7ae9cbb38fafbe9",
   "grader_id": "d511b635ce7bbfd3",
   "configuration_id": "f6b9a462ff423a87",
   "llama_build": "0.4.0-dev (build 10893, commit c32d1dabe)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T18:45:27",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "gemma4-12b",
   "profile": "extraction",
   "accuracy": 83.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 50,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "dea865f1d077f0d8",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T18:42:48",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "k2-7b",
   "profile": "extraction",
   "accuracy": 75.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 45,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "d3f015ffb0c32d4d",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T18:40:45",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "k2-3.7b",
   "profile": "extraction",
   "accuracy": 58.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 35,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "be9a405fa957d750",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T18:39:28",
   "suite": "synthesis",
   "suite_version": "synthesis/2",
   "model": "reap-60b",
   "profile": "extraction",
   "accuracy": 81.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 49,
   "selftests": 12,
   "dataset_id": "267d1954e225c5b6",
   "grader_id": "4156dd3c13361e8b",
   "configuration_id": "6e88592c339b65ea",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T17:54:49",
   "suite": "synthesis",
   "suite_version": "synthesis/1",
   "model": "gemma4-12b",
   "profile": "extraction",
   "accuracy": 80.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 48,
   "selftests": 10,
   "dataset_id": "760936a5348b2ffd",
   "grader_id": "27df090db4445c2c",
   "configuration_id": "9749137aeec83ad6",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T17:53:47",
   "suite": "synthesis",
   "suite_version": "synthesis/1",
   "model": "k2-7b",
   "profile": "extraction",
   "accuracy": 75.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 45,
   "selftests": 10,
   "dataset_id": "760936a5348b2ffd",
   "grader_id": "27df090db4445c2c",
   "configuration_id": "58c73d3c88ebea78",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T17:52:44",
   "suite": "synthesis",
   "suite_version": "synthesis/1",
   "model": "k2-3.7b",
   "profile": "extraction",
   "accuracy": 86.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 52,
   "selftests": 10,
   "dataset_id": "760936a5348b2ffd",
   "grader_id": "27df090db4445c2c",
   "configuration_id": "cc272a124c5d1351",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T17:51:56",
   "suite": "synthesis",
   "suite_version": "synthesis/1",
   "model": "qwen3-next-80b-reap-mlx-mixed36",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 10,
   "dataset_id": "760936a5348b2ffd",
   "grader_id": "27df090db4445c2c",
   "configuration_id": "93c5f1d3b8c8e17c",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-09T17:16:23",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "k2-horizon-3.7b",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "497e8e74501f9a9e",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T17:10:33",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-horizon-3.7b",
   "profile": "extraction",
   "accuracy": 43.1,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 31,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "c1cd134c67de1621",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T17:06:51",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "k2-horizon-3.7b",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "5d3e0bfbb30467e0",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T17:05:51",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "k2-horizon-3.7b",
   "profile": "extraction",
   "accuracy": 86.7,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 13,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "4b2e40f5222213ec",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T17:05:36",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "k2-horizon-3.7b",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "4535019c467a8444",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T17:04:22",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "k2-horizon-3.7b",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "f848b35c571d582e",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T16:38:38",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "k2-horizon-7b",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 15,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "d1b260004a9b283e",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T16:38:17",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "k2-horizon-7b",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "ce69ff4b8656191d",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T16:30:22",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "k2-horizon-7b",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "62b1c628753b16b6",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T16:27:29",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "k2-horizon-7b",
   "profile": "extraction",
   "accuracy": 55.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 69,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "30ddb902b06a0c00",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T16:22:41",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "k2-horizon-7b",
   "profile": "extraction",
   "accuracy": 34.7,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 25,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "54f9df9571bc16e8",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T16:19:52",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "k2-horizon-7b",
   "profile": "extraction",
   "accuracy": 90.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 54,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "e92045e24be569c7",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T16:18:17",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "k2-horizon-7b",
   "profile": "extraction",
   "accuracy": 22.9,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 11,
   "selftests": 17,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "cc2d23c5bae45474",
   "configuration_id": "e127d0663cbe7e98",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T16:16:06",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "k2-horizon-7b",
   "profile": "extraction",
   "accuracy": 95.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 57,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "93ba527fea40c309",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T13:53:47",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-guardian-4.1-8b-mlx-bf16",
   "profile": "extraction",
   "accuracy": 98.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 59,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "081a28b5708d9572",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-09T13:43:48",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-guardian-4.1-8b-mlx-q8",
   "profile": "extraction",
   "accuracy": 98.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 59,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "764b495364f3e7ff",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-09T13:37:02",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-guardian",
   "profile": "extraction",
   "accuracy": 36.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 22,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "9de3a458bda36d8c",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T13:18:17",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-guardian-4.1-8b-mlx-mixed36",
   "profile": "extraction",
   "accuracy": 95.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 57,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "1e2bab38313786db",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-09T13:12:02",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "guardian-mixed36-mlxlm",
   "profile": "extraction",
   "accuracy": null,
   "coverage": 0.0,
   "measured": 0,
   "unmeasured": 60,
   "attempted": 60,
   "passed": 0,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "b56b3c67a8112f7f",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-09T12:50:23",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-guardian",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "fe46fd4c1301cd33",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T12:47:47",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-guardian",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 41,
   "unmeasured": 0,
   "attempted": 41,
   "passed": 41,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "fe46fd4c1301cd33",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T12:38:47",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-guardian-gguf-q4km",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "6af9b4aa83361ef8",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T12:28:45",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-guardian",
   "profile": "extraction",
   "accuracy": 37.3,
   "coverage": 100.0,
   "measured": 59,
   "unmeasured": 0,
   "attempted": 59,
   "passed": 22,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "9de3a458bda36d8c",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T12:27:20",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-guardian-gguf-q4km",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 36,
   "unmeasured": 0,
   "attempted": 36,
   "passed": 36,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "6af9b4aa83361ef8",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T12:22:24",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-guardian",
   "profile": "extraction",
   "accuracy": 40.0,
   "coverage": 100.0,
   "measured": 5,
   "unmeasured": 0,
   "attempted": 5,
   "passed": 2,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "9de3a458bda36d8c",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T12:22:08",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-guardian-gguf-q4km",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 33,
   "unmeasured": 0,
   "attempted": 33,
   "passed": 33,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "6af9b4aa83361ef8",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T12:12:38",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-guardian-mlx-mixed36",
   "profile": "extraction",
   "accuracy": null,
   "coverage": 0.0,
   "measured": 0,
   "unmeasured": 60,
   "attempted": 60,
   "passed": 0,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "d9be4d5a49d0f4d5",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-09T12:05:48",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "qwen3-next-80b-reap",
   "profile": "extraction",
   "accuracy": 35.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 21,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "e9925436d57909a1",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T12:04:26",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "gemma4-e2b",
   "profile": "extraction",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 0,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "022b5819dd82ff96",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T12:03:35",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "granite-4.2-30b",
   "profile": "extraction",
   "accuracy": 20.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 12,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "bf1109eb4e492a3d",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T11:57:59",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "qwen3.6-35b",
   "profile": "extraction",
   "accuracy": 65.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 39,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "8e5c69081ab588b0",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T11:56:35",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "qwen3.8-27b",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "ac125d630f2b70bc",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-09T00:03:46",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "mlx-community-Qwen3.6-35B-A3B-4bit",
   "profile": "extraction",
   "accuracy": 60.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 36,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "4a82e7591c905bf0",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-09T00:01:59",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "qwen3-next-80b-reap-mlx-mixed36",
   "profile": "extraction",
   "accuracy": 78.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 47,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "8352ab9b62086e37",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-09T00:00:03",
   "suite": "toolcall",
   "suite_version": "toolcall/1",
   "model": "qwen3.8-27b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "10af88c170513155",
   "grader_id": "804e8fc9457649d8",
   "configuration_id": "8886d10bc276895c",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T23:06:07",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "mlx-community-Qwen3.6-35B-A3B-4bit",
   "profile": "extraction",
   "accuracy": 27.1,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 13,
   "selftests": 16,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "486601ba23034dd6",
   "configuration_id": "4fc6c3fdb9c9b02c",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T23:03:45",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "q36-toolparser",
   "profile": "extraction",
   "accuracy": null,
   "coverage": 0.0,
   "measured": 0,
   "unmeasured": 48,
   "attempted": 48,
   "passed": 0,
   "selftests": 16,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "486601ba23034dd6",
   "configuration_id": "2225a59526198b68",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T23:02:41",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "mlx-community-Qwen3.6-35B-A3B-4bit",
   "profile": "extraction",
   "accuracy": 90.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 54,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "4ed45f89d9eca30b",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T22:21:01",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3.8-27b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "1b2a598c6ef11638",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T22:15:06",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "qwen3.8-27b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 75.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 36,
   "selftests": 16,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "486601ba23034dd6",
   "configuration_id": "2750a525d9018a88",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T21:51:17",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "flashnext-4.27bpw-m64",
   "profile": "extraction",
   "accuracy": null,
   "coverage": 0.0,
   "measured": 0,
   "unmeasured": 48,
   "attempted": 48,
   "passed": 0,
   "selftests": 16,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "486601ba23034dd6",
   "configuration_id": "92b28699e44d08e7",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T21:50:54",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "qwen3-next-80b-reap",
   "profile": "extraction",
   "accuracy": null,
   "coverage": 0.0,
   "measured": 0,
   "unmeasured": 48,
   "attempted": 48,
   "passed": 0,
   "selftests": 16,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "486601ba23034dd6",
   "configuration_id": "cbb517440b51fc70",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T21:50:38",
   "suite": "contract",
   "suite_version": "contract/2",
   "model": "qwen3-next-80b-reap-mlx-mixed36",
   "profile": "extraction",
   "accuracy": 54.2,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 26,
   "selftests": 16,
   "dataset_id": "1fe53d30cb5480f6",
   "grader_id": "486601ba23034dd6",
   "configuration_id": "e059879c3195b24c",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T21:42:32",
   "suite": "contract",
   "suite_version": "contract/1",
   "model": "qwen3-next-80b-reap-mlx-q4",
   "profile": "extraction",
   "accuracy": 47.2,
   "coverage": 100.0,
   "measured": 36,
   "unmeasured": 0,
   "attempted": 36,
   "passed": 17,
   "selftests": 9,
   "dataset_id": "ab4bb33923bd0ed6",
   "grader_id": "2dd146852463f032",
   "configuration_id": "14fac1620eb065e0",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T21:41:43",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3-next-80b-reap-mlx-q4",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "72e18625ddbdaf10",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T21:38:11",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3-next-80b-reap-mlx-mixed36",
   "profile": "extraction",
   "accuracy": 98.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 59,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "c37510db6b5ff9cf",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T21:09:36",
   "suite": "contract",
   "suite_version": "contract/1",
   "model": "reapq4-mlx_lm",
   "profile": "extraction",
   "accuracy": null,
   "coverage": 0.0,
   "measured": 0,
   "unmeasured": 36,
   "attempted": 36,
   "passed": 0,
   "selftests": 9,
   "dataset_id": "ab4bb33923bd0ed6",
   "grader_id": "2dd146852463f032",
   "configuration_id": "224bdfd4a5d2e672",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T21:09:32",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "reapq4-mlx_lm",
   "profile": "extraction",
   "accuracy": null,
   "coverage": 0.0,
   "measured": 0,
   "unmeasured": 60,
   "attempted": 60,
   "passed": 0,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "2f64770a4290f344",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T21:08:15",
   "suite": "contract",
   "suite_version": "contract/1",
   "model": "reapmix36-rapid-mlx",
   "profile": "extraction",
   "accuracy": null,
   "coverage": 0.0,
   "measured": 0,
   "unmeasured": 36,
   "attempted": 36,
   "passed": 0,
   "selftests": 9,
   "dataset_id": "ab4bb33923bd0ed6",
   "grader_id": "2dd146852463f032",
   "configuration_id": "46d4a279b80fca4d",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T21:08:15",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "reapmix36-rapid-mlx",
   "profile": "extraction",
   "accuracy": null,
   "coverage": 0.0,
   "measured": 0,
   "unmeasured": 60,
   "attempted": 60,
   "passed": 0,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "8077358e07c7af36",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T21:07:11",
   "suite": "contract",
   "suite_version": "contract/1",
   "model": "reapq4-rapid-mlx",
   "profile": "extraction",
   "accuracy": null,
   "coverage": 0.0,
   "measured": 0,
   "unmeasured": 36,
   "attempted": 36,
   "passed": 0,
   "selftests": 9,
   "dataset_id": "ab4bb33923bd0ed6",
   "grader_id": "2dd146852463f032",
   "configuration_id": "788a11f836a3345b",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T21:07:11",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "reapq4-rapid-mlx",
   "profile": "extraction",
   "accuracy": null,
   "coverage": 0.0,
   "measured": 0,
   "unmeasured": 60,
   "attempted": 60,
   "passed": 0,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "ba21b624ebdf49bd",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T20:31:49",
   "suite": "contract",
   "suite_version": "contract/1",
   "model": "granite-4.2-30b",
   "profile": "extraction",
   "accuracy": 66.7,
   "coverage": 100.0,
   "measured": 36,
   "unmeasured": 0,
   "attempted": 36,
   "passed": 24,
   "selftests": 9,
   "dataset_id": "ab4bb33923bd0ed6",
   "grader_id": "2dd146852463f032",
   "configuration_id": "cc8ca71e6e299aee",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T20:29:55",
   "suite": "contract",
   "suite_version": "contract/1",
   "model": "gemma4-12b",
   "profile": "extraction",
   "accuracy": 66.7,
   "coverage": 100.0,
   "measured": 36,
   "unmeasured": 0,
   "attempted": 36,
   "passed": 24,
   "selftests": 9,
   "dataset_id": "ab4bb33923bd0ed6",
   "grader_id": "2dd146852463f032",
   "configuration_id": "0d1de531e8b00c72",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T19:59:29",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "flashnext-4.27bpw-m64",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "17c29ba75eeadc76",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T19:55:08",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "flashnext-4.27bpw-m64",
   "profile": "extraction",
   "accuracy": 95.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 57,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "028bac48a65abdd9",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T19:43:15",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "flashnext-4.27bpw-m64",
   "profile": "extraction",
   "accuracy": 58.3,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 42,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "a1f73dd74c73dbc6",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T19:35:27",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "flashnext-4.27bpw-m64",
   "profile": "extraction",
   "accuracy": 81.6,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 102,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "2867d56b04f8bbb4",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T19:10:55",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "qwen3-next-80b-reap-ag",
   "profile": "agentic-fast",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "ccae884465959abc",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T19:03:01",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "qwen3-next-80b-reap-ag",
   "profile": "agentic-fast",
   "accuracy": 90.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 36,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "2b03adebd9317650",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T18:59:30",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "qwen3-next-80b-reap-ag",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 15,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "3907525a897ced43",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T16:52:36",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3-next-80b-reap",
   "profile": "extraction",
   "accuracy": 97.2,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 70,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "cc55008d40e60573",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T16:51:48",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3-next-80b-reap",
   "profile": "extraction",
   "accuracy": 87.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 109,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "7e1bbf58551d87ef",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T16:50:05",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3-next-80b-reap",
   "profile": "extraction",
   "accuracy": 98.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 59,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "81efb9eaae394534",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": null,
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T11:56:22",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "glm-4.7-flash-mlx-mixed_4_6",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "c1b73af4ac52ca1f",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.32.0"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T11:54:29",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "glm-4.7-flash-mlx-mixed_3_6",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "99894ae40745ad56",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.32.0"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T11:12:06",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "muse-glimmer-30b",
   "profile": "extraction",
   "accuracy": 85.6,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 107,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "d9d25da365e04cc8",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T08:46:40",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "muse-glimmer-30b",
   "profile": "extraction",
   "accuracy": 86.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 52,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "63338a6fdcbdcddd",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T06:39:37",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "laguna-xs2-mlx-q4",
   "profile": "extraction",
   "accuracy": null,
   "coverage": 0.0,
   "measured": 0,
   "unmeasured": 11,
   "attempted": 11,
   "passed": 0,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "3c17fa93e654c0b4",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:50:22",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "glm-4.7-flash-mlx-q4",
   "profile": "extraction",
   "accuracy": 40.3,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 29,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "b669c85bfa68b0b5",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:49:18",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "glm-4.7-flash-mlx-q4",
   "profile": "extraction",
   "accuracy": 53.6,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 67,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "2432689cc6751eea",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:47:19",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "glm-4.7-flash-mlx-q4",
   "profile": "extraction",
   "accuracy": 96.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 58,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "c80ceb72756eafac",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:45:08",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "glm-4.7-flash-mlx-q8",
   "profile": "extraction",
   "accuracy": 59.7,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 43,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "97ef09a912feb8be",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:44:03",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "glm-4.7-flash-mlx-q8",
   "profile": "extraction",
   "accuracy": 76.8,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 96,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "6d0a345ab4ecfa54",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:41:51",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "glm-4.7-flash-mlx-q8",
   "profile": "extraction",
   "accuracy": 96.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 58,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "b4008e66dcadf553",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:38:48",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3-8b-mlx-8bit",
   "profile": "extraction",
   "accuracy": 56.9,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 41,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "b9aaca6eee78b218",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:37:54",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3-8b-mlx-8bit",
   "profile": "extraction",
   "accuracy": 74.4,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 93,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "e34a89daf5181d58",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:36:05",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3-8b-mlx-8bit",
   "profile": "extraction",
   "accuracy": 96.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 58,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "02039a8c6b78a23b",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:34:14",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "nemotron-lightning-30b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 9.7,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 7,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "023ab1ea82fce0de",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:33:05",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "nemotron-lightning-30b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 56.8,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 71,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "15a43d7b2024f68c",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:30:54",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "nemotron-lightning-30b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 90.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 54,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "813491de8103980c",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:28:43",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "granite-4.2-30b-gguf",
   "profile": "extraction",
   "accuracy": 90.3,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 65,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "ad5177ffa2d55fa5",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:26:42",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "granite-4.2-30b-gguf",
   "profile": "extraction",
   "accuracy": 84.0,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 105,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "e353f603986dcdcd",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:22:10",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-4.2-30b-gguf",
   "profile": "extraction",
   "accuracy": 98.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 59,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "001987450ecbaa8a",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:15:08",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "ornith-1.5-35b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 62.5,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 45,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "1fb1a5ad2ee784e3",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:14:00",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "ornith-1.5-35b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 67.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 84,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "a438aff9707ec8a1",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:11:47",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "ornith-1.5-35b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 88.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 53,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "4f3b9812298e963e",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:09:30",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "granite-4.2-30b-mlx-q4",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 72,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "48b3d526e6d6e6a3",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:09:30",
   "suite": "VOID-degenerate-nullschema",
   "suite_version": "nullschema/1",
   "model": "granite-4.2-30b-mlx-q4",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 72,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "48b3d526e6d6e6a3",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx@dflash-venv": "0.32.2",
    "mlx@mlx-main-venv": "0.32.2",
    "mlx@mlx-venv": "0.32.2",
    "mlx@rapid-mlx": "0.32.2",
    "mlx_lm@dflash-venv": "0.31.3",
    "mlx_lm@mlx-main-venv": "0.32.0",
    "mlx_lm@mlx-venv": "0.31.3",
    "mlx_lm@rapid-mlx": "0.31.3",
    "rapid-mlx": "0.13.4",
    "vllm_mlx@dflash-venv": "0.13.4",
    "vllm_mlx@rapid-mlx": "0.13.4"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:06:51",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "granite-4.2-30b-mlx-q4",
   "profile": "extraction",
   "accuracy": 86.4,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 108,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "db0a9174aa469c68",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-08T00:01:22",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-4.2-30b-mlx-q4",
   "profile": "extraction",
   "accuracy": 91.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 55,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "f71ccbbb16cd8099",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-07T23:51:55",
   "suite": "shapegen",
   "suite_version": "shapegen/2",
   "model": "qwen3-coder-30b",
   "profile": "extraction",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 12,
   "unmeasured": 0,
   "attempted": 12,
   "passed": 0,
   "selftests": 15,
   "dataset_id": "fc27d4e1a999c675",
   "grader_id": "f23222ad8df22fd8",
   "configuration_id": "cbf6852b8edc0984",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T23:38:29",
   "suite": "sitebuild",
   "suite_version": "sitebuild/1",
   "model": "qwen3-coder-30b",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 1,
   "unmeasured": 0,
   "attempted": 1,
   "passed": 0,
   "selftests": 5,
   "dataset_id": "8b54725d2778dc60",
   "grader_id": "b2c603d6c0d6afb3",
   "configuration_id": "0cac83c8fbc03eec",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T22:20:11",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "tagger-e2b-v1-dl-tagger-e2b-v1",
   "profile": "extraction",
   "accuracy": 75.0,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 180,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "b0015c0679b3c9f8",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T22:15:37",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "tagger-e2b-v1-dl-tagger-e2b-v1",
   "profile": "extraction",
   "accuracy": 74.2,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 178,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "26b77eacaf1fc207",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T22:10:38",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "gemma4-e2b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 73.8,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 177,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "d40c89a4c0ded16e",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-07T22:04:48",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "gemma4-e2b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 85.8,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 206,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "1623b2cdd0ccbfaf",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-07T21:55:33",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3-next-80b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 95.8,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 69,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "d1b64a4d48cff9ef",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-07T21:54:25",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3-next-80b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 96.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 58,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "22413c6aa05bedb0",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-07T21:52:08",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3-next-80b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 87.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 109,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "0c539d39afff5bd1",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-07T21:46:17",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3.8-27b-noreason",
   "profile": "extraction",
   "accuracy": 80.0,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 100,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "4a3b4b7c53219329",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T20:48:40",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3.8-27b-mlx-4bit",
   "profile": "extraction",
   "accuracy": 92.0,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 115,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "42a3cf291ff302bb",
   "llama_build": "",
   "runtime": "mlx",
   "runtime_versions": {
    "mlx_lm": "0.31.3"
   },
   "backfilled": true
  },
  {
   "ts": "2026-09-07T20:19:37",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "qwen-agentworld-35b",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "c4305e58879e2b40",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T20:19:37",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "qwen3-next-80b-ag",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "3a96fdcc21129ae7",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T20:19:37",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "qwen3.6-35b-ag",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "ea9f2c30d07a149b",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T19:24:47",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "qwen3.6-35b-ag",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "ea9f2c30d07a149b",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T19:16:32",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "qwen3.6-35b-ag",
   "profile": "extraction",
   "accuracy": 80.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 32,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "f995b25bd0b360cf",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T19:13:44",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "qwen3.6-35b-ag",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 14,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "a8b21b992370e7c6",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T19:12:53",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "qwen3-next-80b-ag",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "3a96fdcc21129ae7",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T19:04:42",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "qwen3-next-80b-ag",
   "profile": "extraction",
   "accuracy": 95.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 38,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "18366d2ac6e48e7c",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T19:01:10",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "qwen3-next-80b-ag",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 14,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "91dcd2c830490d0e",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T19:00:15",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "qwen-agentworld-35b",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "f4bfcf6f040b0978",
   "configuration_id": "c4305e58879e2b40",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T18:52:00",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "qwen-agentworld-35b",
   "profile": "extraction",
   "accuracy": 72.5,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 29,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "a2bc3751abfa28e0",
   "configuration_id": "edcc1fd1f9bd35f0",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T18:49:16",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "qwen-agentworld-35b",
   "profile": "extraction",
   "accuracy": 80.0,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 12,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "2763675a5c3da564",
   "configuration_id": "e56fd9e9b9103b0a",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-07T18:10:41",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-guardian-3.1-2b",
   "profile": "extraction",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 0,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "c3bc4d1194d50b4b",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-05T23:00:26",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen38-flash-next",
   "profile": "extraction",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 36,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "d248e89da830ce31",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-05T22:56:44",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen38-flash-next",
   "profile": "extraction",
   "accuracy": 96.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 58,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "dd9b4d06d5abee09",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-05T22:53:23",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen38-flash-next",
   "profile": "extraction",
   "accuracy": 79.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 99,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "9148fcdc2bc14fe4",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-05T20:14:33",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "nemotron-lightning-30b",
   "profile": "extraction",
   "accuracy": 15.3,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 11,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "8f3225e9b30dcc6b",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-05T20:13:46",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "nemotron-lightning-30b",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "bd58c270aedc9f3d",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-05T20:12:31",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "nemotron-lightning-30b",
   "profile": "extraction",
   "accuracy": 60.0,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 75,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "f734b41b3caabfdd",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-05T13:03:47",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3-next-80b",
   "profile": "extraction",
   "accuracy": 97.2,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 70,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "5913b00d3e8c6b3d",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-05T13:02:57",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3-next-80b",
   "profile": "extraction",
   "accuracy": 91.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 55,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "ede1a2072fbaea6c",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-09-05T13:01:10",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3-next-80b",
   "profile": "extraction",
   "accuracy": 88.8,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 111,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "f34a40087b318cc6",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-31T18:59:21",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "qwen3-next-80b-tag",
   "profile": "extraction",
   "accuracy": 90.4,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 217,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "c033312a94ce8bbe",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-31T18:46:05",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "qwen3-next-80b-tag",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 224,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "bb9d18596364b9f4",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-31T18:29:53",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "qwen3.6-35b-tag",
   "profile": "extraction",
   "accuracy": 92.5,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 222,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "2f1c411d3dd00d20",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-31T18:15:44",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "qwen3.6-35b-tag",
   "profile": "extraction",
   "accuracy": 90.8,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 218,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "336518ff1b1d0973",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-31T18:00:14",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "gemma4-12b-b10618",
   "profile": "extraction",
   "accuracy": 97.9,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 235,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "e947bd43665a9bdb",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-31T17:39:45",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "gemma4-12b-b10618",
   "profile": "extraction",
   "accuracy": 93.8,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 225,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "c947e8a080eadfe2",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-31T04:14:10",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "granite-guardian-4.1-8b",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "9ba7cd65a17e6418",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-31T02:50:55",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "lfm2.5-8b-a1b",
   "profile": "extraction",
   "accuracy": 27.8,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 20,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "00612fdd4a39faf4",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-31T02:50:36",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "lfm2.5-8b-a1b",
   "profile": "extraction",
   "accuracy": 47.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 59,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "90be8dffe935ab41",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-31T02:49:56",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "lfm2.5-8b-a1b",
   "profile": "extraction",
   "accuracy": 73.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 44,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "2cbf094131c24516",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-27T09:32:10",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3-next-80b-coder",
   "profile": "extraction",
   "accuracy": 56.9,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 41,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "74810b0cff4603b3",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-27T09:31:15",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3-next-80b-coder",
   "profile": "extraction",
   "accuracy": 90.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 54,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "e7b3303ff47f6803",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-27T09:29:15",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3-next-80b-coder",
   "profile": "extraction",
   "accuracy": 76.8,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 96,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "140e26c8be639be7",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-27T08:04:37",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3.6-35b",
   "profile": "extraction",
   "accuracy": 61.1,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 44,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "7a652a6a84f42ef5",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-27T08:03:48",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3.6-35b",
   "profile": "extraction",
   "accuracy": 86.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 52,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "4246ff56e771b097",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-27T08:02:24",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3.6-35b",
   "profile": "extraction",
   "accuracy": 82.4,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 103,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "279f3a4bd1ff0c7e",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-27T08:01:46",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3.6-35b",
   "profile": "extraction",
   "accuracy": 86.3,
   "coverage": 100.0,
   "measured": 73,
   "unmeasured": 0,
   "attempted": 73,
   "passed": 63,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "279f3a4bd1ff0c7e",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-27T03:57:43",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3-next-80b-thinking",
   "profile": "study-default",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 1,
   "unmeasured": 0,
   "attempted": 1,
   "passed": 1,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "cb1f0b0a42b6ac48",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-27T03:57:37",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3-next-80b-thinking",
   "profile": "extraction",
   "accuracy": 58.3,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 42,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "8f8aa14bba9f91c3",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-27T03:18:04",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3-next-80b-thinking",
   "profile": "extraction",
   "accuracy": 91.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 55,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "4320932025e05e5d",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-27T03:03:51",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3-next-80b-thinking",
   "profile": "extraction",
   "accuracy": 65.6,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 82,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "e644b21297e6198b",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-27T01:44:46",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3-next-80b",
   "profile": "extraction",
   "accuracy": 97.2,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 70,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "5913b00d3e8c6b3d",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-27T01:43:56",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3-next-80b",
   "profile": "extraction",
   "accuracy": 91.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 55,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "ede1a2072fbaea6c",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-27T01:42:13",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3-next-80b",
   "profile": "extraction",
   "accuracy": 88.8,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 111,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "f34a40087b318cc6",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T13:58:05",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3-next-80b-a3b-iq4xs",
   "profile": "extraction",
   "accuracy": 97.2,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 70,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "860f8678b21dfb1f",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T13:57:17",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3-next-80b-a3b-iq4xs",
   "profile": "extraction",
   "accuracy": 91.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 55,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "c6cb8e26e5eec1ec",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T13:55:35",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3-next-80b-a3b-iq4xs",
   "profile": "extraction",
   "accuracy": 88.8,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 111,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "e3b83bfa70c1a417",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T13:52:46",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3-next-80b-a3b-iq4xs-smoke",
   "profile": "extraction",
   "accuracy": 73.9,
   "coverage": 92.0,
   "measured": 23,
   "unmeasured": 2,
   "attempted": 25,
   "passed": 17,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "e6920f17163814d7",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T12:35:42",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "glm-4.7-flash-q8",
   "profile": "extraction",
   "accuracy": 48.6,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 35,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "3d836e2b73f9cd77",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T12:34:33",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "glm-4.7-flash-q8",
   "profile": "extraction",
   "accuracy": 95.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 57,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "b7d076058290ae5f",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T12:31:44",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "glm-4.7-flash-q8",
   "profile": "extraction",
   "accuracy": 76.0,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 95,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "f7116f1eb6a7f001",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T12:28:31",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "glm-4.7-flash-q8",
   "profile": "extraction",
   "accuracy": null,
   "coverage": 0.0,
   "measured": 0,
   "unmeasured": 72,
   "attempted": 72,
   "passed": 0,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "3d836e2b73f9cd77",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T12:28:31",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "glm-4.7-flash-q8",
   "profile": "extraction",
   "accuracy": null,
   "coverage": 0.0,
   "measured": 0,
   "unmeasured": 60,
   "attempted": 60,
   "passed": 0,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "b7d076058290ae5f",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T11:13:47",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3.5-122b-a10b-iq3",
   "profile": "extraction",
   "accuracy": 45.8,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 33,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "5e7cd3243c74ce98",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T11:11:52",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3.5-122b-a10b-iq3",
   "profile": "extraction",
   "accuracy": 88.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 53,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "45dab17de7814804",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T11:08:08",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3.5-122b-a10b-iq3",
   "profile": "extraction",
   "accuracy": 80.8,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 101,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "b3059734754cd76c",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T08:34:56",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "agents-a1-35b-b10618",
   "profile": "extraction",
   "accuracy": 38.9,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 28,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "9a9ee4f1853f385a",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T08:33:35",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "agents-a1-35b-b10618",
   "profile": "extraction",
   "accuracy": 95.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 57,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "fe6603a03488c606",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T08:32:13",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "agents-a1-35b-b10618",
   "profile": "extraction",
   "accuracy": 71.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 89,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "b6d15c86cef9c0f6",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T08:03:47",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3-coder-30b-b10618",
   "profile": "extraction",
   "accuracy": 37.5,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 27,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "ffb75533d79bc27d",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T08:03:17",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3-coder-30b-b10618",
   "profile": "extraction",
   "accuracy": 91.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 55,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "01f8fd00f78cb299",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-26T08:02:14",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3-coder-30b-b10618",
   "profile": "extraction",
   "accuracy": 76.8,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 96,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "85522d05c3671c16",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T11:47:31",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "qwen3.5-9b-q8-b10618",
   "profile": "extraction",
   "accuracy": 30.6,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 22,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "989a2b1d084c0c53",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T11:46:47",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "qwen3.5-9b-q8-b10618",
   "profile": "extraction",
   "accuracy": 96.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 58,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "6c6da4df0333507c",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T11:44:57",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3.5-9b-q8-b10618",
   "profile": "extraction",
   "accuracy": 72.0,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 90,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "9e54265c33ff0f82",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T11:15:32",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "gemma4-12b-b10618",
   "profile": "extraction",
   "accuracy": 88.9,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 64,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "2c03d6ebad98fccc",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T11:14:26",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "gemma4-12b-b10618",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "79323c88f8e515df",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T11:12:38",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "gemma4-12b-b10618",
   "profile": "extraction",
   "accuracy": 84.0,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 105,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "ba97a71cfa0468a7",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T10:38:43",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "gemma4-e2b-b10618",
   "profile": "extraction",
   "accuracy": 12.5,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 9,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "38b4b96050d74d17",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T10:38:22",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "gemma4-e2b-b10618",
   "profile": "extraction",
   "accuracy": 91.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 55,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "80f542cbf88cfba7",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T10:37:51",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "gemma4-e2b-b10618",
   "profile": "extraction",
   "accuracy": 61.6,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 77,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "b18ad353c84d839e",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T10:24:41",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "gemma4-e4b-b10618",
   "profile": "extraction",
   "accuracy": 13.9,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 10,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "84794dc91d0ce3b4",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T10:24:09",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "gemma4-e4b-b10618",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "b75251f174426f4d",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T10:23:24",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "gemma4-e4b-b10618",
   "profile": "extraction",
   "accuracy": 62.4,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 78,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "dc6988d619b065bf",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T10:03:30",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "ornith-1.5-35b-q6-b10618",
   "profile": "extraction",
   "accuracy": 34.7,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 25,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "355fd8b860ca54b1",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T10:02:04",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "ornith-1.5-35b-q6-b10618",
   "profile": "extraction",
   "accuracy": 95.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 57,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "ce44f5f606a0131e",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T10:00:41",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "ornith-1.5-35b-q6-b10618",
   "profile": "extraction",
   "accuracy": 72.8,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 91,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "bc2f7777482c6fb6",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T09:37:55",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "devstral-small-2-24b-b10618",
   "profile": "extraction",
   "accuracy": 62.5,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 45,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "836a74186efb247a",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T09:35:59",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "devstral-small-2-24b-b10618",
   "profile": "extraction",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 60,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "d8a2ef703d82163d",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-25T09:32:23",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "devstral-small-2-24b-b10618",
   "profile": "extraction",
   "accuracy": 79.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 99,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "ed588aa2abdb41ef",
   "llama_build": "0.2.0-dev (build 10618, commit eb25b7263)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T23:16:17",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "gemma4-12b",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "b7d605769821c0d8",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T23:14:27",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "gemma4-12b",
   "profile": "extraction",
   "accuracy": 88.9,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 64,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "ec3358e6aa5e1088",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T23:13:09",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "gemma4-e4b",
   "profile": "extraction",
   "accuracy": 93.3,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 56,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "824bef7dab174615",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T23:12:20",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "gemma4-e4b",
   "profile": "extraction",
   "accuracy": 13.9,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 10,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "addbcab5facddd0b",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T23:11:35",
   "suite": "screening",
   "suite_version": "screening/1",
   "model": "gemma4-e2b",
   "profile": "extraction",
   "accuracy": 91.7,
   "coverage": 100.0,
   "measured": 60,
   "unmeasured": 0,
   "attempted": 60,
   "passed": 55,
   "selftests": 7,
   "dataset_id": "64c97ca2269c29af",
   "grader_id": "0661b49a10be2d2a",
   "configuration_id": "905fc96bd8c3e874",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T23:11:00",
   "suite": "nullschema",
   "suite_version": "nullschema/1",
   "model": "gemma4-e2b",
   "profile": "extraction",
   "accuracy": 12.5,
   "coverage": 100.0,
   "measured": 72,
   "unmeasured": 0,
   "attempted": 72,
   "passed": 9,
   "selftests": 6,
   "dataset_id": "3b91280bb40372ef",
   "grader_id": "d7448e0f5fc3f9af",
   "configuration_id": "e268490afdec64ce",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T19:04:11",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "gemma4-12b-ab",
   "profile": "extraction",
   "accuracy": 97.9,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 235,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "8bf8e2bee9ac7333",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T18:43:35",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "gemma4-12b-ab",
   "profile": "extraction",
   "accuracy": 93.8,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 225,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "3c8ef5d04a11b585",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T18:19:25",
   "suite": "tagger_nullable",
   "suite_version": "tagger-nullable/1",
   "model": "gemma4-e2b-ab",
   "profile": "extraction",
   "accuracy": 90.4,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 217,
   "selftests": 10,
   "dataset_id": "db3addf23a2bc4ce",
   "grader_id": "04e875431e26c4b5",
   "configuration_id": "23e712860bee6d0e",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T18:13:40",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "gemma4-e2b-ab",
   "profile": "extraction",
   "accuracy": 91.2,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 219,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "63e04d139a0e5ed3",
   "configuration_id": "4619919ac18affd5",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T17:54:17",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3-coder-30b",
   "profile": "extraction",
   "accuracy": 77.6,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 97,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "ade06bde17b71832",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T14:13:31",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3.8-27b",
   "profile": "extraction",
   "accuracy": 80.0,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 100,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "1d989a6b00eb212c",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T14:07:08",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "agents-a1-35b",
   "profile": "extraction",
   "accuracy": 71.2,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 89,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "3cb35299a047efb8",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T14:03:47",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "qwen3.5-9b-q8",
   "profile": "extraction",
   "accuracy": 72.0,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 90,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "bd21ebb86ee6f863",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T14:02:03",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "gemma4-12b",
   "profile": "extraction",
   "accuracy": 84.0,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 105,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "31241ed46ed4d6a5",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T13:59:29",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "gemma4-e4b",
   "profile": "extraction",
   "accuracy": 62.4,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 78,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "267624907124ed7e",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T13:58:18",
   "suite": "grounding",
   "suite_version": "grounding/1",
   "model": "gemma4-e2b",
   "profile": "extraction",
   "accuracy": 61.6,
   "coverage": 100.0,
   "measured": 125,
   "unmeasured": 0,
   "attempted": 125,
   "passed": 77,
   "selftests": 13,
   "dataset_id": "dcb396faf8b83b0a",
   "grader_id": "30b933dcc6a7e2d4",
   "configuration_id": "dee67653e2de45c6",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T08:02:10",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "qwen3.8-iq1s",
   "profile": "study-nothink",
   "accuracy": 31.6,
   "coverage": 44.2,
   "measured": 57,
   "unmeasured": 72,
   "attempted": 129,
   "passed": 18,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "956e03bbffa79aa4",
   "configuration_id": "f6700e17f70a9c2a",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T04:03:34",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "qwen3.8-iq1s",
   "profile": "study-nothink",
   "accuracy": 77.1,
   "coverage": 72.9,
   "measured": 35,
   "unmeasured": 13,
   "attempted": 48,
   "passed": 27,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "2a2f7e4e90d6dd63",
   "configuration_id": "4bfecdedff811923",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T03:54:24",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "qwen3.8-iq1s",
   "profile": "agentic-fast",
   "accuracy": 76.9,
   "coverage": 86.7,
   "measured": 13,
   "unmeasured": 2,
   "attempted": 15,
   "passed": 10,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "d6a32c09e57bea7b",
   "configuration_id": "02e356bf0012eaf2",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T03:50:14",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "qwen3.8-iq1s",
   "profile": "study-nothink",
   "accuracy": 60.5,
   "coverage": 79.6,
   "measured": 43,
   "unmeasured": 11,
   "attempted": 54,
   "passed": 26,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "7d661b93852b5ffd",
   "configuration_id": "11d22e9108889bce",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T03:02:28",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "qwen3.8-iq1s",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "d31d3c4c6285617f",
   "configuration_id": "e2e5d225bbfe60d6",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T03:00:32",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "gemma4-12b-r3",
   "profile": "agentic-fast",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "d31d3c4c6285617f",
   "configuration_id": "2cbb0e64811ac557",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T02:45:30",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "gemma4-e4b-r3",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "d31d3c4c6285617f",
   "configuration_id": "7e97acd7556a3909",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T02:42:45",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "gemma4-e2b-r3",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "d31d3c4c6285617f",
   "configuration_id": "24f4c1683cfb7a85",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T02:38:30",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "gemma4-12b-r2",
   "profile": "agentic-fast",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "d31d3c4c6285617f",
   "configuration_id": "e1a3cdf8d371cae6",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T02:23:29",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "gemma4-e4b-r2",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "d31d3c4c6285617f",
   "configuration_id": "ec61f0c3716fc3f8",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T02:20:44",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "gemma4-e2b-r2",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "d31d3c4c6285617f",
   "configuration_id": "277579259f233168",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T02:17:44",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "agents-a1-35b",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "1bae1f3e1b606eae",
   "configuration_id": "f55610aeacd8ae8c",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T02:14:44",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "gemma4-12b-r1",
   "profile": "agentic-fast",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "d31d3c4c6285617f",
   "configuration_id": "97645e44ce72b737",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T02:13:14",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "gemma4-12b",
   "profile": "agentic-fast",
   "accuracy": 90.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 36,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "1bae1f3e1b606eae",
   "configuration_id": "42412bb32a456ced",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T02:05:37",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "gemma4-e4b",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "1bae1f3e1b606eae",
   "configuration_id": "64d002411f012d3c",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T01:57:47",
   "suite": "toolscale",
   "suite_version": "toolscale/1",
   "model": "gemma4-e2b",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 40,
   "unmeasured": 0,
   "attempted": 40,
   "passed": 40,
   "selftests": 5,
   "dataset_id": "b24abb108c409178",
   "grader_id": "1bae1f3e1b606eae",
   "configuration_id": "3860ab39a73cff08",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T01:51:49",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "gemma4-e4b-r1",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "d31d3c4c6285617f",
   "configuration_id": "7ef98c4a8fe4369b",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T01:48:56",
   "suite": "chainwalk",
   "suite_version": "chainwalk/3",
   "model": "gemma4-e2b-r1",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "d31d3c4c6285617f",
   "configuration_id": "92c0ac5ca29b9cdc",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T01:41:25",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "qwen3.8-27b",
   "profile": "study-nothink",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 48,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "2a2f7e4e90d6dd63",
   "configuration_id": "541a385213b500cb",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T01:37:42",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "qwen3.8-27b",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 15,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "d6a32c09e57bea7b",
   "configuration_id": "35475e0c17bfc385",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T01:19:27",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "gemma4-12b",
   "profile": "study-nothink",
   "accuracy": 77.8,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 42,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "7d661b93852b5ffd",
   "configuration_id": "5367be2a63165de3",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T01:15:42",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "gemma4-e4b",
   "profile": "study-nothink",
   "accuracy": 70.4,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 38,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "7d661b93852b5ffd",
   "configuration_id": "60945fac08964a72",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T01:13:40",
   "suite": "enrichment",
   "suite_version": "enrichment/3",
   "model": "gemma4-e2b",
   "profile": "study-nothink",
   "accuracy": 75.9,
   "coverage": 100.0,
   "measured": 54,
   "unmeasured": 0,
   "attempted": 54,
   "passed": 41,
   "selftests": 19,
   "dataset_id": "75361864baa727ab",
   "grader_id": "7d661b93852b5ffd",
   "configuration_id": "94079faf08beb84b",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T01:05:04",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "agents-a1-35b",
   "profile": "study-nothink",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 48,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "2a2f7e4e90d6dd63",
   "configuration_id": "2101ffcef80989dc",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T01:04:20",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "agents-a1-35b",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 15,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "d6a32c09e57bea7b",
   "configuration_id": "555c28b5386478f6",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T01:03:16",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "gemma4-12b",
   "profile": "study-nothink",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 48,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "2a2f7e4e90d6dd63",
   "configuration_id": "b6ac24eba6a83a5d",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T01:01:49",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "gemma4-e4b",
   "profile": "study-nothink",
   "accuracy": 97.9,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 47,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "2a2f7e4e90d6dd63",
   "configuration_id": "37147fa352af2d7a",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T01:01:06",
   "suite": "epistemics",
   "suite_version": "epistemics/4",
   "model": "gemma4-e2b",
   "profile": "study-nothink",
   "accuracy": 68.8,
   "coverage": 100.0,
   "measured": 48,
   "unmeasured": 0,
   "attempted": 48,
   "passed": 33,
   "selftests": 14,
   "dataset_id": "4bce453fd376553c",
   "grader_id": "2a2f7e4e90d6dd63",
   "configuration_id": "e5abf1e2bbe8c2c8",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T01:00:21",
   "suite": "chainwalk",
   "suite_version": "chainwalk/2",
   "model": "gemma4-12b",
   "profile": "agentic-fast",
   "accuracy": 50.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 1,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "48abd22e0117fdda",
   "configuration_id": "b4834ee71de7e834",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T00:45:11",
   "suite": "chainwalk",
   "suite_version": "chainwalk/2",
   "model": "gemma4-e4b",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "48abd22e0117fdda",
   "configuration_id": "8e18e895c23f4485",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T00:42:21",
   "suite": "chainwalk",
   "suite_version": "chainwalk/2",
   "model": "gemma4-e2b",
   "profile": "agentic-fast",
   "accuracy": 0.0,
   "coverage": 100.0,
   "measured": 2,
   "unmeasured": 0,
   "attempted": 2,
   "passed": 0,
   "selftests": 4,
   "dataset_id": "129a9e74408c8919",
   "grader_id": "48abd22e0117fdda",
   "configuration_id": "eef56b755300de59",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T00:31:05",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "gemma4-e2b",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 15,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "d6a32c09e57bea7b",
   "configuration_id": "33f8efb324ad39aa",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T00:26:23",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "gemma4-12b",
   "profile": "agentic-fast",
   "accuracy": 86.7,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 13,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "d6a32c09e57bea7b",
   "configuration_id": "b34ee497223231c8",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T00:25:11",
   "suite": "tools",
   "suite_version": "tools/1",
   "model": "gemma4-e4b",
   "profile": "agentic-fast",
   "accuracy": 100.0,
   "coverage": 100.0,
   "measured": 15,
   "unmeasured": 0,
   "attempted": 15,
   "passed": 15,
   "selftests": 9,
   "dataset_id": "33a44812f0ad4c39",
   "grader_id": "d6a32c09e57bea7b",
   "configuration_id": "50b166f799c52d51",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-24T00:19:34",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "gemma4-12b",
   "profile": "study-nothink",
   "accuracy": 93.2,
   "coverage": 80.3,
   "measured": 191,
   "unmeasured": 47,
   "attempted": 238,
   "passed": 178,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "28983856e4bf2dd9",
   "configuration_id": "d10e58be49cb1d55",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-23T23:56:41",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "gemma4-e4b",
   "profile": "study-nothink",
   "accuracy": 90.8,
   "coverage": 50.0,
   "measured": 240,
   "unmeasured": 240,
   "attempted": 480,
   "passed": 218,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "28983856e4bf2dd9",
   "configuration_id": "60e6c55236269057",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-23T23:41:31",
   "suite": "tagger",
   "suite_version": "tagger/2",
   "model": "gemma4-e2b",
   "profile": "study-nothink",
   "accuracy": 91.2,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 219,
   "selftests": 10,
   "dataset_id": "7ef587bf33e5788a",
   "grader_id": "28983856e4bf2dd9",
   "configuration_id": "72d758d807701d51",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  },
  {
   "ts": "2026-08-23T22:50:22",
   "suite": "tagger",
   "suite_version": null,
   "model": "tagger-e2b-v1",
   "profile": "extraction",
   "accuracy": 77.1,
   "coverage": 100.0,
   "measured": 240,
   "unmeasured": 0,
   "attempted": 240,
   "passed": 185,
   "selftests": null,
   "dataset_id": "",
   "grader_id": "",
   "configuration_id": "",
   "llama_build": "0.2.0-dev (build 10369, commit 6e62ba538)",
   "runtime": "llama.cpp",
   "runtime_versions": null,
   "backfilled": true
  }
 ]
}