sam-jung commited on
Commit
9864488
·
1 Parent(s): b5ef2ff

Replace capped search results with uncapped runs

Browse files

Remove 1,269 superseded capped search runs. Preserve all 4,650 current runs, including byte-identical retained records in 461 mixed shards. Add original multi-domain labels to the task metadata and update the domain summary. Keep private visibility and Git history.

This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. README.md +7 -9
  2. UPLOAD-STATUS.json +7 -9
  3. analysis-summary-20261104.json +688 -199
  4. analysis-summary-legacy-4269.json +0 -1
  5. analysis-summary.json +688 -199
  6. release-manifest-legacy-4269.json +0 -0
  7. release-manifest.json +0 -0
  8. rollouts-00003.jsonl.gz +0 -3
  9. rollouts-00005.jsonl.gz +2 -2
  10. rollouts-00007.jsonl.gz +0 -3
  11. rollouts-00008.jsonl.gz +2 -2
  12. rollouts-00010.jsonl.gz +0 -3
  13. rollouts-00013.jsonl.gz +2 -2
  14. rollouts-00016.jsonl.gz +2 -2
  15. rollouts-00020.jsonl.gz +2 -2
  16. rollouts-00021.jsonl.gz +2 -2
  17. rollouts-00023.jsonl.gz +2 -2
  18. rollouts-00029.jsonl.gz +2 -2
  19. rollouts-00030.jsonl.gz +2 -2
  20. rollouts-00031.jsonl.gz +2 -2
  21. rollouts-00032.jsonl.gz +2 -2
  22. rollouts-00037.jsonl.gz +2 -2
  23. rollouts-00038.jsonl.gz +0 -3
  24. rollouts-00039.jsonl.gz +0 -3
  25. rollouts-00043.jsonl.gz +2 -2
  26. rollouts-00044.jsonl.gz +2 -2
  27. rollouts-00045.jsonl.gz +2 -2
  28. rollouts-00046.jsonl.gz +2 -2
  29. rollouts-00047.jsonl.gz +2 -2
  30. rollouts-00048.jsonl.gz +2 -2
  31. rollouts-00052.jsonl.gz +2 -2
  32. rollouts-00053.jsonl.gz +2 -2
  33. rollouts-00055.jsonl.gz +2 -2
  34. rollouts-00058.jsonl.gz +2 -2
  35. rollouts-00063.jsonl.gz +2 -2
  36. rollouts-00065.jsonl.gz +2 -2
  37. rollouts-00070.jsonl.gz +2 -2
  38. rollouts-00073.jsonl.gz +0 -3
  39. rollouts-00074.jsonl.gz +2 -2
  40. rollouts-00076.jsonl.gz +2 -2
  41. rollouts-00078.jsonl.gz +2 -2
  42. rollouts-00080.jsonl.gz +2 -2
  43. rollouts-00082.jsonl.gz +0 -3
  44. rollouts-00083.jsonl.gz +2 -2
  45. rollouts-00085.jsonl.gz +2 -2
  46. rollouts-00101.jsonl.gz +2 -2
  47. rollouts-00102.jsonl.gz +2 -2
  48. rollouts-00105.jsonl.gz +0 -3
  49. rollouts-00106.jsonl.gz +2 -2
  50. rollouts-00111.jsonl.gz +2 -2
README.md CHANGED
@@ -8,29 +8,27 @@ configs:
8
  ---
9
  # CrossSearch: 100-task paper set
10
 
11
- This resource contains **100 tasks and 5,919 retained, completed evaluation runs**. The current evaluation comprises **3,300 native-search runs** (11 model configurations, three final attempts per task) and **1,350 uncapped search-backend runs** (three models, 50 tasks, three alternative backends, three attempts). The previous **1,269 capped search-backend runs** remain available as a separate historical cohort.
12
 
13
  ## November 2026 update
14
 
15
- This update adds **1,650 completed runs**: 300 Gemini 4 Argon native-search runs and 1,350 uncapped Exa, Parallel and Firecrawl runs for GPT-5.6 Sol, Kimi K3 and Gemini 3.8 Flash. Each original task/model/condition/replicate cell is represented once in this new cohort. Earlier incomplete attempts are not substituted for complete results, and valid low and zero scores are retained.
16
 
17
- The added files begin with `rollouts-20261104-`. They contain the original final answers, rubric judgments, available model reasoning, saved request/response captures and trace files, with operational credentials and local author/account paths removed. The source archives remain unchanged. Search/fetch call-count limits are disabled in the new search-backend cohort; task time, token, context and response-size limits are unchanged. Keep the uncapped and historical capped cohorts separate when comparing results.
18
 
19
 
20
  ## Files
21
 
22
- - tasks.jsonl.gz: expert questions, frozen rubrics and evidence maps.
23
  - tasks.tar.gz: starting workspace assets, separated from grader-only reference material.
24
  - rollouts-*.jsonl.gz: saved agent trace files, API request/response captures, final answers, original rewards and criterion judgments. The 669 earlier encrypted source runs and all 1,650 added protected archives are exported as readable derivatives. Source archives remain unchanged; no private key is distributed.
25
- - run-index.jsonl.gz: lookup of all 5,919 retained run IDs, tasks, models, conditions, rewards and trajectory shards.
26
  - run-index-20261104.jsonl.gz: just the 1,650 runs added in this update, including `release_cohort` and `search_call_limit`.
27
- - run-index-legacy-4269.jsonl.gz and release-manifest-legacy-4269.json: the previous release indexes, preserved intact.
28
  - judge-models.csv: one row for each of the 415 audited encrypted extension records, with evidence references and recorded judge labels.
29
  - CrossSearch-Judge-Prompt.txt and CrossSearch-Agent-Prompt.txt: appendix prompt templates, message roles and source notes.
30
  - grading-protocol.md: scoring protocol and judge-version uncertainty.
31
  - judge-model-audit.json: recorded judge fields for the 415 encrypted extension records.
32
- - analysis-summary.json and analysis-summary-20261104.json: updated 11-model native results, uncapped search comparisons, efficiency values and reporting definitions. Every metric group is labeled by cohort.
33
- - analysis-summary-legacy-4269.json: the earlier frozen paper/blog results, retained unchanged.
34
  - release-manifest.json: final counts, shard checksums, redaction totals and verification status. Check this manifest for completeness.
35
 
36
  ## Reading the data
@@ -80,4 +78,4 @@ Some exceptionally large traces use `large_record_parts`. Their normal rollout r
80
  argon = [r for r in added if r['condition'] == 'native'] # 300
81
  uncapped = [r for r in added if r['condition'] != 'native'] # 1,350
82
 
83
- The current leaderboard combines the 3,000 earlier native runs with the 300 Argon runs. The current search comparison uses the 1,350 added uncapped runs; it does not pool them with the historical capped runs.
 
8
  ---
9
  # CrossSearch: 100-task paper set
10
 
11
+ This resource contains **100 tasks and 4,650 completed evaluation runs**. The current evaluation comprises **3,300 native-search runs** (11 model configurations, three final attempts per task) and **1,350 uncapped search-backend runs** (three models, 50 tasks, three alternative backends, three attempts).
12
 
13
  ## November 2026 update
14
 
15
+ This release includes **1,650 updated runs**: 300 Gemini 4 Argon native-search runs and 1,350 uncapped Exa, Parallel and Firecrawl runs for GPT-5.6 Sol, Kimi K3 and Gemini 3.8 Flash. Each original task/model/condition/replicate cell is represented once in this new cohort. Earlier incomplete attempts are not substituted for complete results, and valid low and zero scores are retained.
16
 
17
+ The added files begin with `rollouts-20261104-`. They contain the original final answers, rubric judgments, available model reasoning, saved request/response captures and trace files, with operational credentials and local author/account paths removed. The source archives remain unchanged. Search/fetch call-count limits are disabled in the new search-backend cohort; task time, token, context and response-size limits are unchanged. The uncapped search-backend results replace the earlier capped comparison.
18
 
19
 
20
  ## Files
21
 
22
+ - tasks.jsonl.gz: expert questions, domain labels, frozen rubrics and evidence maps.
23
  - tasks.tar.gz: starting workspace assets, separated from grader-only reference material.
24
  - rollouts-*.jsonl.gz: saved agent trace files, API request/response captures, final answers, original rewards and criterion judgments. The 669 earlier encrypted source runs and all 1,650 added protected archives are exported as readable derivatives. Source archives remain unchanged; no private key is distributed.
25
+ - run-index.jsonl.gz: lookup of all 4,650 run IDs, tasks, models, conditions, rewards and trajectory shards.
26
  - run-index-20261104.jsonl.gz: just the 1,650 runs added in this update, including `release_cohort` and `search_call_limit`.
 
27
  - judge-models.csv: one row for each of the 415 audited encrypted extension records, with evidence references and recorded judge labels.
28
  - CrossSearch-Judge-Prompt.txt and CrossSearch-Agent-Prompt.txt: appendix prompt templates, message roles and source notes.
29
  - grading-protocol.md: scoring protocol and judge-version uncertainty.
30
  - judge-model-audit.json: recorded judge fields for the 415 encrypted extension records.
31
+ - analysis-summary.json and analysis-summary-20261104.json: updated 11-model native results, uncapped search comparisons, efficiency values and reporting definitions.
 
32
  - release-manifest.json: final counts, shard checksums, redaction totals and verification status. Check this manifest for completeness.
33
 
34
  ## Reading the data
 
78
  argon = [r for r in added if r['condition'] == 'native'] # 300
79
  uncapped = [r for r in added if r['condition'] != 'native'] # 1,350
80
 
81
+ The current leaderboard combines the 3,000 earlier native runs with the 300 Argon runs. The current search comparison uses the 1,350 uncapped runs.
UPLOAD-STATUS.json CHANGED
@@ -1,13 +1,11 @@
1
  {
2
  "status": "complete_private_release",
3
  "tasks": 100,
4
- "retained_runs": 5919,
5
- "latest_update_runs": 1650,
6
- "current_native_runs": 3300,
7
- "current_uncapped_ablation_runs": 1350,
8
- "retained_legacy_capped_runs": 1269,
9
- "data_commit": "fd44de24985e49be626dc9e7483e14494326fe7e",
10
  "visibility": "private",
11
- "manifest_sha256": "11ebaa1816f3803bc371bb991f50e1adcd59102b2625f0531ce84423ff612b97",
12
- "original_data_files_preserved": true
13
- }
 
1
  {
2
  "status": "complete_private_release",
3
  "tasks": 100,
4
+ "runs": 4650,
5
+ "native_runs": 3300,
6
+ "uncapped_ablation_runs": 1350,
7
+ "superseded_capped_runs_removed": 1269,
 
 
8
  "visibility": "private",
9
+ "manifest_sha256": "35c9f28f02f0c6f1ae37a81b590f43183bdb5e952f7361a033f7fa202e0b1380",
10
+ "all_4650_original_record_contents_preserved": true
11
+ }
analysis-summary-20261104.json CHANGED
@@ -2,8 +2,7 @@
2
  "task_count": 100,
3
  "current_native_runs": 3300,
4
  "current_uncapped_search_runs": 1350,
5
- "retained_legacy_capped_runs": 1269,
6
- "retained_run_count": 5919,
7
  "added_run_count": 1650,
8
  "native_models": [
9
  {
@@ -931,16 +930,6 @@
931
  "image_audio_video_union_tasks": 33,
932
  "nontext_evidence_tasks": 30,
933
  "nontext_evidence_rubric_items": 90,
934
- "primary_domains": {
935
- "engineering": 46,
936
- "medicine": 13,
937
- "science": 7,
938
- "aviation": 1,
939
- "law": 7,
940
- "public_records": 14,
941
- "finance": 9,
942
- "other": 3
943
- },
944
  "rubric_categories": [
945
  {
946
  "category": "Fact / rule identification",
@@ -1259,19 +1248,699 @@
1259
  "Joined application / comparison",
1260
  "Fact / rule identification"
1261
  ],
1262
- "provenance": "Primary domains are original author-provided submission labels. Rubric criterionType is empty for all 663 items. Four reporting categories are exploratory post-hoc coding of primary requirement and original source-support graph, not original labels, a new grading rubric, or a causal failure taxonomy. Calculation and grounding/limits labels were reviewed against criterion text. Remainder splits on whether the linked source graph uses both workspace and web sources.",
1263
  "rate_definition": "Pooled pass rate of saved per-item boolean judgments across all 11 native-search models, all 100 tasks and all 3 trials per model/task. Each of 663 items has 33 judgments. Categories have different difficulty and composition; rates are descriptive and do not establish causes.",
1264
  "scoring_basis": "Saved per-item outcomes for non-perfect attempts; validated full-credit records imply every unit-weight item passed. No new grading.",
1265
- "inventory_definition": "Delivered environment/artifacts files excluding the standard answer_format.md. Original submission URLs without delivered local files are excluded. Image-only PDF means no extractable text and at least one embedded raster image; PDFs with an OCR layer are not identified as image-only."
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1266
  },
1267
  "cost_note": "Eight model costs are recorded. Gemini 4 Argon, Gemini 3.8 Flash and Kimi K3 costs use saved input, cached-input and output token counts at the November 3, 2026 rate snapshot. Agent cost excludes failed reruns, grading, tools and infrastructure.",
1268
- "cohort_note": "The 1,269 earlier capped search runs are retained separately and are not pooled with the 1,350 uncapped runs.",
1269
  "tasks": 100,
1270
- "runs": 5919,
1271
  "native_runs": 3300,
1272
- "ablation_runs": 2619,
1273
  "uncapped_ablation_runs": 1350,
1274
- "legacy_capped_ablation_runs": 1269,
1275
  "current_evaluation_runs": 4650,
1276
  "metrics": [
1277
  {
@@ -1334,26 +2003,6 @@
1334
  "3": 11
1335
  }
1336
  },
1337
- {
1338
- "model": "google/gemini-3.8-flash",
1339
- "condition": "exa",
1340
- "cohort": "legacy_capped",
1341
- "tasks": 47,
1342
- "runs": 141,
1343
- "mean_reward": 0.8852179432624113,
1344
- "full_credit_pass_at_1": 0.5460992907801419,
1345
- "pass_power_k": {
1346
- "1": 0.5460992907801419,
1347
- "2": 0.475177304964539,
1348
- "3": 0.425531914893617
1349
- },
1350
- "perfect_attempts_per_task_histogram": {
1351
- "0": 17,
1352
- "1": 3,
1353
- "2": 7,
1354
- "3": 20
1355
- }
1356
- },
1357
  {
1358
  "model": "google/gemini-3.8-flash",
1359
  "condition": "exa",
@@ -1374,26 +2023,6 @@
1374
  "3": 23
1375
  }
1376
  },
1377
- {
1378
- "model": "google/gemini-3.8-flash",
1379
- "condition": "firecrawl",
1380
- "cohort": "legacy_capped",
1381
- "tasks": 47,
1382
- "runs": 141,
1383
- "mean_reward": 0.8536526382978723,
1384
- "full_credit_pass_at_1": 0.44680851063829785,
1385
- "pass_power_k": {
1386
- "1": 0.44680851063829785,
1387
- "2": 0.3262411347517731,
1388
- "3": 0.2765957446808511
1389
- },
1390
- "perfect_attempts_per_task_histogram": {
1391
- "0": 17,
1392
- "1": 10,
1393
- "2": 7,
1394
- "3": 13
1395
- }
1396
- },
1397
  {
1398
  "model": "google/gemini-3.8-flash",
1399
  "condition": "firecrawl",
@@ -1434,26 +2063,6 @@
1434
  "3": 38
1435
  }
1436
  },
1437
- {
1438
- "model": "google/gemini-3.8-flash",
1439
- "condition": "parallel",
1440
- "cohort": "legacy_capped",
1441
- "tasks": 47,
1442
- "runs": 141,
1443
- "mean_reward": 0.8851616595744681,
1444
- "full_credit_pass_at_1": 0.5673758865248227,
1445
- "pass_power_k": {
1446
- "1": 0.5673758865248227,
1447
- "2": 0.475177304964539,
1448
- "3": 0.44680851063829785
1449
- },
1450
- "perfect_attempts_per_task_histogram": {
1451
- "0": 13,
1452
- "1": 9,
1453
- "2": 4,
1454
- "3": 21
1455
- }
1456
- },
1457
  {
1458
  "model": "google/gemini-3.8-flash",
1459
  "condition": "parallel",
@@ -1494,26 +2103,6 @@
1494
  "3": 81
1495
  }
1496
  },
1497
- {
1498
- "model": "gpt-5.6-sol",
1499
- "condition": "exa",
1500
- "cohort": "legacy_capped",
1501
- "tasks": 47,
1502
- "runs": 141,
1503
- "mean_reward": 0.9034142765957447,
1504
- "full_credit_pass_at_1": 0.6382978723404256,
1505
- "pass_power_k": {
1506
- "1": 0.6382978723404256,
1507
- "2": 0.5035460992907801,
1508
- "3": 0.40425531914893614
1509
- },
1510
- "perfect_attempts_per_task_histogram": {
1511
- "0": 9,
1512
- "1": 5,
1513
- "2": 14,
1514
- "3": 19
1515
- }
1516
- },
1517
  {
1518
  "model": "gpt-5.6-sol",
1519
  "condition": "exa",
@@ -1534,26 +2123,6 @@
1534
  "3": 30
1535
  }
1536
  },
1537
- {
1538
- "model": "gpt-5.6-sol",
1539
- "condition": "firecrawl",
1540
- "cohort": "legacy_capped",
1541
- "tasks": 47,
1542
- "runs": 141,
1543
- "mean_reward": 0.8824806028368795,
1544
- "full_credit_pass_at_1": 0.5957446808510638,
1545
- "pass_power_k": {
1546
- "1": 0.5957446808510638,
1547
- "2": 0.48936170212765956,
1548
- "3": 0.425531914893617
1549
- },
1550
- "perfect_attempts_per_task_histogram": {
1551
- "0": 12,
1552
- "1": 6,
1553
- "2": 9,
1554
- "3": 20
1555
- }
1556
- },
1557
  {
1558
  "model": "gpt-5.6-sol",
1559
  "condition": "firecrawl",
@@ -1594,26 +2163,6 @@
1594
  "3": 47
1595
  }
1596
  },
1597
- {
1598
- "model": "gpt-5.6-sol",
1599
- "condition": "parallel",
1600
- "cohort": "legacy_capped",
1601
- "tasks": 47,
1602
- "runs": 141,
1603
- "mean_reward": 0.8948098156028369,
1604
- "full_credit_pass_at_1": 0.6099290780141844,
1605
- "pass_power_k": {
1606
- "1": 0.6099290780141844,
1607
- "2": 0.5106382978723404,
1608
- "3": 0.44680851063829785
1609
- },
1610
- "perfect_attempts_per_task_histogram": {
1611
- "0": 12,
1612
- "1": 5,
1613
- "2": 9,
1614
- "3": 21
1615
- }
1616
- },
1617
  {
1618
  "model": "gpt-5.6-sol",
1619
  "condition": "parallel",
@@ -1674,26 +2223,6 @@
1674
  "3": 73
1675
  }
1676
  },
1677
- {
1678
- "model": "moonshotai/kimi-k3",
1679
- "condition": "exa",
1680
- "cohort": "legacy_capped",
1681
- "tasks": 47,
1682
- "runs": 141,
1683
- "mean_reward": 0.7981263191489362,
1684
- "full_credit_pass_at_1": 0.375886524822695,
1685
- "pass_power_k": {
1686
- "1": 0.375886524822695,
1687
- "2": 0.23404255319148937,
1688
- "3": 0.19148936170212766
1689
- },
1690
- "perfect_attempts_per_task_histogram": {
1691
- "0": 18,
1692
- "1": 14,
1693
- "2": 6,
1694
- "3": 9
1695
- }
1696
- },
1697
  {
1698
  "model": "moonshotai/kimi-k3",
1699
  "condition": "exa",
@@ -1714,26 +2243,6 @@
1714
  "3": 17
1715
  }
1716
  },
1717
- {
1718
- "model": "moonshotai/kimi-k3",
1719
- "condition": "firecrawl",
1720
- "cohort": "legacy_capped",
1721
- "tasks": 47,
1722
- "runs": 141,
1723
- "mean_reward": 0.7683709078014185,
1724
- "full_credit_pass_at_1": 0.36879432624113473,
1725
- "pass_power_k": {
1726
- "1": 0.36879432624113473,
1727
- "2": 0.2553191489361702,
1728
- "3": 0.19148936170212766
1729
- },
1730
- "perfect_attempts_per_task_histogram": {
1731
- "0": 22,
1732
- "1": 7,
1733
- "2": 9,
1734
- "3": 9
1735
- }
1736
- },
1737
  {
1738
  "model": "moonshotai/kimi-k3",
1739
  "condition": "firecrawl",
@@ -1774,26 +2283,6 @@
1774
  "3": 23
1775
  }
1776
  },
1777
- {
1778
- "model": "moonshotai/kimi-k3",
1779
- "condition": "parallel",
1780
- "cohort": "legacy_capped",
1781
- "tasks": 47,
1782
- "runs": 141,
1783
- "mean_reward": 0.8110883475177305,
1784
- "full_credit_pass_at_1": 0.375886524822695,
1785
- "pass_power_k": {
1786
- "1": 0.375886524822695,
1787
- "2": 0.2553191489361702,
1788
- "3": 0.19148936170212766
1789
- },
1790
- "perfect_attempts_per_task_histogram": {
1791
- "0": 21,
1792
- "1": 8,
1793
- "2": 9,
1794
- "3": 9
1795
- }
1796
- },
1797
  {
1798
  "model": "moonshotai/kimi-k3",
1799
  "condition": "parallel",
@@ -1854,5 +2343,5 @@
1854
  }
1855
  ],
1856
  "no_new_grading_or_model_calls": true,
1857
- "method": "Metrics are aggregated from retained completed run indexes. Cohort labels distinguish native, current uncapped and historical capped runs; capped and uncapped results are never pooled."
1858
  }
 
2
  "task_count": 100,
3
  "current_native_runs": 3300,
4
  "current_uncapped_search_runs": 1350,
5
+ "retained_run_count": 4650,
 
6
  "added_run_count": 1650,
7
  "native_models": [
8
  {
 
930
  "image_audio_video_union_tasks": 33,
931
  "nontext_evidence_tasks": 30,
932
  "nontext_evidence_rubric_items": 90,
 
 
 
 
 
 
 
 
 
 
933
  "rubric_categories": [
934
  {
935
  "category": "Fact / rule identification",
 
1248
  "Joined application / comparison",
1249
  "Fact / rule identification"
1250
  ],
1251
+ "provenance": "Domain counts use all author-provided submission labels; tasks may belong to multiple domains. Rubric criterionType is empty for all 663 items. Four reporting categories are exploratory post-hoc coding of primary requirement and original source-support graph, not original labels, a new grading rubric, or a causal failure taxonomy. Calculation and grounding/limits labels were reviewed against criterion text. Remainder splits on whether the linked source graph uses both workspace and web sources.",
1252
  "rate_definition": "Pooled pass rate of saved per-item boolean judgments across all 11 native-search models, all 100 tasks and all 3 trials per model/task. Each of 663 items has 33 judgments. Categories have different difficulty and composition; rates are descriptive and do not establish causes.",
1253
  "scoring_basis": "Saved per-item outcomes for non-perfect attempts; validated full-credit records imply every unit-weight item passed. No new grading.",
1254
+ "inventory_definition": "Delivered environment/artifacts files excluding the standard answer_format.md. Original submission URLs without delivered local files are excluded. Image-only PDF means no extractable text and at least one embedded raster image; PDFs with an OCR layer are not identified as image-only.",
1255
+ "domain_counts": {
1256
+ "engineering": 54,
1257
+ "public_records": 41,
1258
+ "finance": 15,
1259
+ "law": 14,
1260
+ "other": 14,
1261
+ "medicine": 13,
1262
+ "science": 12,
1263
+ "aviation": 4
1264
+ },
1265
+ "multi_domain_task_count": 55,
1266
+ "domain_assignment_count": 167,
1267
+ "task_domain_labels": [
1268
+ {
1269
+ "task": "6-ghz-canopy-deployment-screening-for-a-distribution-centre-wi-fi-refresh",
1270
+ "domains": [
1271
+ "engineering",
1272
+ "public_records"
1273
+ ]
1274
+ },
1275
+ {
1276
+ "task": "acetaminophen-overdose-treatment-and-stopping-decision-294",
1277
+ "domains": [
1278
+ "medicine"
1279
+ ]
1280
+ },
1281
+ {
1282
+ "task": "adult-laryngoscope-set-assessment",
1283
+ "domains": [
1284
+ "medicine",
1285
+ "engineering"
1286
+ ]
1287
+ },
1288
+ {
1289
+ "task": "aggrow-program-farmer-eligibility-check",
1290
+ "domains": [
1291
+ "science",
1292
+ "finance",
1293
+ "public_records"
1294
+ ]
1295
+ },
1296
+ {
1297
+ "task": "agroforestry-meal-pipeline-data-integrity-and-survival-bound-audit",
1298
+ "domains": [
1299
+ "science",
1300
+ "other"
1301
+ ]
1302
+ },
1303
+ {
1304
+ "task": "aha-bls-certification-verification",
1305
+ "domains": [
1306
+ "medicine"
1307
+ ]
1308
+ },
1309
+ {
1310
+ "task": "ai-recommendation-service-integrity-and-sre-coverage-audit",
1311
+ "domains": [
1312
+ "engineering"
1313
+ ]
1314
+ },
1315
+ {
1316
+ "task": "aviation-cargo-lithium-battery-exemption-compliance-check",
1317
+ "domains": [
1318
+ "aviation",
1319
+ "public_records",
1320
+ "other"
1321
+ ]
1322
+ },
1323
+ {
1324
+ "task": "aviation-engine-mount-torque-audit-ac-43-13-1b-131",
1325
+ "domains": [
1326
+ "engineering",
1327
+ "aviation",
1328
+ "public_records"
1329
+ ]
1330
+ },
1331
+ {
1332
+ "task": "ax417-lithium-battery-baggage-release-audit",
1333
+ "domains": [
1334
+ "law",
1335
+ "engineering",
1336
+ "aviation"
1337
+ ]
1338
+ },
1339
+ {
1340
+ "task": "beamex-fb660-calibration-audit",
1341
+ "domains": [
1342
+ "engineering"
1343
+ ]
1344
+ },
1345
+ {
1346
+ "task": "belkin-charging-accessory-recall-screening-across-two-overlapping-cpsc-recalls",
1347
+ "domains": [
1348
+ "engineering",
1349
+ "public_records"
1350
+ ]
1351
+ },
1352
+ {
1353
+ "task": "beryllium-sampling-records-checked-against-current-exposure-limits",
1354
+ "domains": [
1355
+ "science",
1356
+ "public_records"
1357
+ ]
1358
+ },
1359
+ {
1360
+ "task": "brickwork-final-bill-audit-against-cpwd-dsr-2023",
1361
+ "domains": [
1362
+ "engineering",
1363
+ "public_records"
1364
+ ]
1365
+ },
1366
+ {
1367
+ "task": "building-c-ryobi-electric-pressure-washer-recall-screen",
1368
+ "domains": [
1369
+ "engineering",
1370
+ "public_records"
1371
+ ]
1372
+ },
1373
+ {
1374
+ "task": "bulk-email-sender-compliance-screening",
1375
+ "domains": [
1376
+ "engineering",
1377
+ "public_records"
1378
+ ]
1379
+ },
1380
+ {
1381
+ "task": "caips-delivery-status-vs-2025-target-kenya",
1382
+ "domains": [
1383
+ "public_records",
1384
+ "finance"
1385
+ ]
1386
+ },
1387
+ {
1388
+ "task": "campaign-24v-744-scope-and-repair-version-audit",
1389
+ "domains": [
1390
+ "public_records"
1391
+ ]
1392
+ },
1393
+ {
1394
+ "task": "candidate-neem-oil-fungicide-formulations-vs-a-historic-european-patent-s-claim-",
1395
+ "domains": [
1396
+ "law",
1397
+ "science",
1398
+ "other"
1399
+ ]
1400
+ },
1401
+ {
1402
+ "task": "champions-league-suspension-audit-carvajal-vs-mendy-2023-24-semifinal",
1403
+ "domains": [
1404
+ "public_records",
1405
+ "other"
1406
+ ]
1407
+ },
1408
+ {
1409
+ "task": "cisco-ios-xe-exposure-and-compromise-screening",
1410
+ "domains": [
1411
+ "public_records",
1412
+ "engineering"
1413
+ ]
1414
+ },
1415
+ {
1416
+ "task": "commercial-partnership-zakat-compliance-and-audit-assessment",
1417
+ "domains": [
1418
+ "finance"
1419
+ ]
1420
+ },
1421
+ {
1422
+ "task": "comps-model-revenue-integrity-check-against-a-public-restatement",
1423
+ "domains": [
1424
+ "finance",
1425
+ "public_records"
1426
+ ]
1427
+ },
1428
+ {
1429
+ "task": "computing-landed-cost-for-used-clothing-consignment-into-mombasa-march-2025",
1430
+ "domains": [
1431
+ "law",
1432
+ "finance",
1433
+ "public_records"
1434
+ ]
1435
+ },
1436
+ {
1437
+ "task": "covered-call-coverage-check-after-a-listed-contract-adjustment",
1438
+ "domains": [
1439
+ "finance",
1440
+ "public_records"
1441
+ ]
1442
+ },
1443
+ {
1444
+ "task": "designing-a-flow-cyotometry-panel",
1445
+ "domains": [
1446
+ "science"
1447
+ ]
1448
+ },
1449
+ {
1450
+ "task": "determine-which-dogs-are-eligible-for-international-travel-based-on-their-rabies",
1451
+ "domains": [
1452
+ "medicine",
1453
+ "public_records",
1454
+ "science"
1455
+ ]
1456
+ },
1457
+ {
1458
+ "task": "diagnosing-bright-sky-background-in-milky-way-photography-session",
1459
+ "domains": [
1460
+ "science",
1461
+ "other"
1462
+ ]
1463
+ },
1464
+ {
1465
+ "task": "didion-milling-dust-explosion-from-the-plant-floor-to-the-courtroom",
1466
+ "domains": [
1467
+ "public_records",
1468
+ "science",
1469
+ "law"
1470
+ ]
1471
+ },
1472
+ {
1473
+ "task": "dio-http-client-version-security-check",
1474
+ "domains": [
1475
+ "engineering"
1476
+ ]
1477
+ },
1478
+ {
1479
+ "task": "dns-resolution-incident",
1480
+ "domains": [
1481
+ "engineering"
1482
+ ]
1483
+ },
1484
+ {
1485
+ "task": "dnssec-chain-of-trust-rollover-failure-and-recovery-audit",
1486
+ "domains": [
1487
+ "engineering"
1488
+ ]
1489
+ },
1490
+ {
1491
+ "task": "docker-base-image-cve-exposure-check",
1492
+ "domains": [
1493
+ "engineering"
1494
+ ]
1495
+ },
1496
+ {
1497
+ "task": "emergency-poc-test-kit-and-iv-fluid-shelf-life-extension-audit",
1498
+ "domains": [
1499
+ "medicine"
1500
+ ]
1501
+ },
1502
+ {
1503
+ "task": "extraction-photocatalyst-data-of-ta2o5",
1504
+ "domains": [
1505
+ "science",
1506
+ "other"
1507
+ ]
1508
+ },
1509
+ {
1510
+ "task": "fact-check-a-map-s-color-scale-against-census-data",
1511
+ "domains": [
1512
+ "public_records"
1513
+ ]
1514
+ },
1515
+ {
1516
+ "task": "fact-checking-a-literature-review-product-description-against-the-manufacturer-s",
1517
+ "domains": [
1518
+ "engineering"
1519
+ ]
1520
+ },
1521
+ {
1522
+ "task": "fairfax-bridge-posting-compliance-across-restriction-history",
1523
+ "domains": [
1524
+ "public_records",
1525
+ "engineering"
1526
+ ]
1527
+ },
1528
+ {
1529
+ "task": "fern-hollow-collapse-geometry-and-initiating-member-findings",
1530
+ "domains": [
1531
+ "engineering",
1532
+ "public_records"
1533
+ ]
1534
+ },
1535
+ {
1536
+ "task": "fieldkit-native-sign-in-migration-audit",
1537
+ "domains": [
1538
+ "engineering"
1539
+ ]
1540
+ },
1541
+ {
1542
+ "task": "fintech-transaction-record-handling-and-retention-compliance-audit",
1543
+ "domains": [
1544
+ "engineering",
1545
+ "law"
1546
+ ]
1547
+ },
1548
+ {
1549
+ "task": "frozen-waffle-recall-inventory-check",
1550
+ "domains": [
1551
+ "public_records",
1552
+ "other"
1553
+ ]
1554
+ },
1555
+ {
1556
+ "task": "github-actions-runner-migration-across-multiple-ci-cd-workflows",
1557
+ "domains": [
1558
+ "engineering"
1559
+ ]
1560
+ },
1561
+ {
1562
+ "task": "github-api-secondary-rate-limit-ticket-dismissal-check-334",
1563
+ "domains": [
1564
+ "engineering"
1565
+ ]
1566
+ },
1567
+ {
1568
+ "task": "github-rest-api-compatibility-review",
1569
+ "domains": [
1570
+ "engineering"
1571
+ ]
1572
+ },
1573
+ {
1574
+ "task": "grandmother-s-old-egfr-calculator-vs-current-kidney-function-guidelines",
1575
+ "domains": [
1576
+ "medicine",
1577
+ "public_records"
1578
+ ]
1579
+ },
1580
+ {
1581
+ "task": "handlebars-patch-triage-across-two-pdf-generation-features",
1582
+ "domains": [
1583
+ "engineering",
1584
+ "public_records"
1585
+ ]
1586
+ },
1587
+ {
1588
+ "task": "heavy-haul-permit-limit-on-a-us-1-route-from-the-national-bridge-inventory",
1589
+ "domains": [
1590
+ "engineering",
1591
+ "public_records",
1592
+ "law"
1593
+ ]
1594
+ },
1595
+ {
1596
+ "task": "hemodialysis-patient-protein-intake-adequacy-vs-kdoqi-2020",
1597
+ "domains": [
1598
+ "medicine"
1599
+ ]
1600
+ },
1601
+ {
1602
+ "task": "hospital-pharmacy-medication-safety-audit",
1603
+ "domains": [
1604
+ "medicine"
1605
+ ]
1606
+ },
1607
+ {
1608
+ "task": "hsg-imaging",
1609
+ "domains": [
1610
+ "medicine"
1611
+ ]
1612
+ },
1613
+ {
1614
+ "task": "insider-sale-red-flag-for-advisory-client-of-gild",
1615
+ "domains": [
1616
+ "public_records",
1617
+ "finance"
1618
+ ]
1619
+ },
1620
+ {
1621
+ "task": "inspection-and-compliance-assessment-of-sets-of-q690-steel-plate-214",
1622
+ "domains": [
1623
+ "engineering"
1624
+ ]
1625
+ },
1626
+ {
1627
+ "task": "iphone-11-battery-replacement-warranty-investigation",
1628
+ "domains": [
1629
+ "other",
1630
+ "engineering"
1631
+ ]
1632
+ },
1633
+ {
1634
+ "task": "jwt-library-vulnerability-exposure-check-for-finance-tracker-auth",
1635
+ "domains": [
1636
+ "engineering",
1637
+ "public_records"
1638
+ ]
1639
+ },
1640
+ {
1641
+ "task": "kenya-poultry-feed-cost-report-update-280",
1642
+ "domains": [
1643
+ "public_records",
1644
+ "other"
1645
+ ]
1646
+ },
1647
+ {
1648
+ "task": "kenya-power-lighting-company-commercial-and-industrial-c-i-tariff-period-verific",
1649
+ "domains": [
1650
+ "engineering"
1651
+ ]
1652
+ },
1653
+ {
1654
+ "task": "kenyan-electronic-goods-import-tax-liability-calculation",
1655
+ "domains": [
1656
+ "law",
1657
+ "finance"
1658
+ ]
1659
+ },
1660
+ {
1661
+ "task": "kidde-fire-extinguisher-recall-and-replacement-audit",
1662
+ "domains": [
1663
+ "public_records",
1664
+ "law"
1665
+ ]
1666
+ },
1667
+ {
1668
+ "task": "kilifi-ecvr-tracker-reconciliation-against-iebc-records",
1669
+ "domains": [
1670
+ "public_records"
1671
+ ]
1672
+ },
1673
+ {
1674
+ "task": "kitchen-range-recall-screening-for-five-property-assets",
1675
+ "domains": [
1676
+ "engineering",
1677
+ "public_records"
1678
+ ]
1679
+ },
1680
+ {
1681
+ "task": "landed-cost-calculation-for-an-imitation-jewelry-shipment-from-china-to-ethiopia",
1682
+ "domains": [
1683
+ "law",
1684
+ "finance"
1685
+ ]
1686
+ },
1687
+ {
1688
+ "task": "leaked-ci-credential-breach-notification-scope-deadline",
1689
+ "domains": [
1690
+ "engineering",
1691
+ "law"
1692
+ ]
1693
+ },
1694
+ {
1695
+ "task": "lekki-toll-gate-article-fact-check-casualty-claim-and-victim-list-misattribution",
1696
+ "domains": [
1697
+ "public_records",
1698
+ "other"
1699
+ ]
1700
+ },
1701
+ {
1702
+ "task": "lodash-prototype-pollution-exposure-check-for-a-node-api-project",
1703
+ "domains": [
1704
+ "engineering"
1705
+ ]
1706
+ },
1707
+ {
1708
+ "task": "management-of-severe-plasmodium-falciparum-malaria-with-anaemia-in-a-quinine-sen",
1709
+ "domains": [
1710
+ "medicine"
1711
+ ]
1712
+ },
1713
+ {
1714
+ "task": "mimedx-2015-revenue-the-growth-that-was-partly-fake",
1715
+ "domains": [
1716
+ "finance",
1717
+ "public_records",
1718
+ "law"
1719
+ ]
1720
+ },
1721
+ {
1722
+ "task": "nimbus-lqa-redis-licensing-exposure",
1723
+ "domains": [
1724
+ "law",
1725
+ "engineering"
1726
+ ]
1727
+ },
1728
+ {
1729
+ "task": "overdue-lycoming-connecting-rod-ad-inspection-cessna-172s-n832xx",
1730
+ "domains": [
1731
+ "other",
1732
+ "aviation"
1733
+ ]
1734
+ },
1735
+ {
1736
+ "task": "patient-metformin-dispensing-log-screening-against-2020-fda-ndma-recall",
1737
+ "domains": [
1738
+ "medicine",
1739
+ "public_records"
1740
+ ]
1741
+ },
1742
+ {
1743
+ "task": "pavement-structure-determination-for-a-rural-access-road-in-ethiopia-north-shoa-",
1744
+ "domains": [
1745
+ "engineering"
1746
+ ]
1747
+ },
1748
+ {
1749
+ "task": "paystack-account-disablement-mapping-a-cited-reason-to-the-actual-policy-text",
1750
+ "domains": [
1751
+ "finance",
1752
+ "law",
1753
+ "engineering"
1754
+ ]
1755
+ },
1756
+ {
1757
+ "task": "pharmaceutical-wastewater-safety-test",
1758
+ "domains": [
1759
+ "engineering"
1760
+ ]
1761
+ },
1762
+ {
1763
+ "task": "pharmacy-chlorthalidone-lot-screening-against-fda-recall",
1764
+ "domains": [
1765
+ "medicine",
1766
+ "public_records"
1767
+ ]
1768
+ },
1769
+ {
1770
+ "task": "pipeline-restart-decision-and-release-audit",
1771
+ "domains": [
1772
+ "engineering",
1773
+ "science",
1774
+ "public_records"
1775
+ ]
1776
+ },
1777
+ {
1778
+ "task": "piper-pa-28-fleet-wing-spar-airworthiness-directive-ad-2020-26-16-compliance-scr",
1779
+ "domains": [
1780
+ "engineering",
1781
+ "public_records"
1782
+ ]
1783
+ },
1784
+ {
1785
+ "task": "planar-diffuser-cfd-validation-review",
1786
+ "domains": [
1787
+ "science",
1788
+ "engineering"
1789
+ ]
1790
+ },
1791
+ {
1792
+ "task": "portfolio-memo-verification-for-holdings-h-01-fy2025",
1793
+ "domains": [
1794
+ "finance"
1795
+ ]
1796
+ },
1797
+ {
1798
+ "task": "post-maintenance-fiber-uplink-compliance-review-111",
1799
+ "domains": [
1800
+ "engineering"
1801
+ ]
1802
+ },
1803
+ {
1804
+ "task": "post-maintenance-memory-population-review-for-three-server-hosts",
1805
+ "domains": [
1806
+ "engineering"
1807
+ ]
1808
+ },
1809
+ {
1810
+ "task": "print-and-epub-distribution-readiness-check-for-a-fantasy-short-story",
1811
+ "domains": [
1812
+ "other"
1813
+ ]
1814
+ },
1815
+ {
1816
+ "task": "q3-2025-mileage-reimbursement-tax-compliance-check",
1817
+ "domains": [
1818
+ "finance"
1819
+ ]
1820
+ },
1821
+ {
1822
+ "task": "re-vlab-production-hosting-go-live-audit",
1823
+ "domains": [
1824
+ "engineering"
1825
+ ]
1826
+ },
1827
+ {
1828
+ "task": "recall-reconciliation-log-from-the-national-register-for-a-hospital-in-2025",
1829
+ "domains": [
1830
+ "medicine"
1831
+ ]
1832
+ },
1833
+ {
1834
+ "task": "reconciling-an-investor-deck-against-sec-filings-for-capstone-companies-inc",
1835
+ "domains": [
1836
+ "finance",
1837
+ "public_records"
1838
+ ]
1839
+ },
1840
+ {
1841
+ "task": "redistribution-license-clearance-for-the-4-2-on-prem-appliance-image",
1842
+ "domains": [
1843
+ "engineering"
1844
+ ]
1845
+ },
1846
+ {
1847
+ "task": "regresshion-exposure-across-five-firmware-builds",
1848
+ "domains": [
1849
+ "engineering"
1850
+ ]
1851
+ },
1852
+ {
1853
+ "task": "remaining-fatigue-life-review-of-three-welded-hot-spots-on-an-offshore-topside-f",
1854
+ "domains": [
1855
+ "engineering"
1856
+ ]
1857
+ },
1858
+ {
1859
+ "task": "sec-non-gaap-comment-letter-reconciliation-for-a-filed-earnings-release",
1860
+ "domains": [
1861
+ "finance",
1862
+ "public_records"
1863
+ ]
1864
+ },
1865
+ {
1866
+ "task": "signal-warrant-evaluation-for-a-small-city-intersection-mutcd-chapter-4c",
1867
+ "domains": [
1868
+ "engineering"
1869
+ ]
1870
+ },
1871
+ {
1872
+ "task": "sitop-ups1600-firmware-advisory-screen-for-building-b-assets-222",
1873
+ "domains": [
1874
+ "engineering",
1875
+ "public_records"
1876
+ ]
1877
+ },
1878
+ {
1879
+ "task": "thermography-measurement-validity",
1880
+ "domains": [
1881
+ "engineering"
1882
+ ]
1883
+ },
1884
+ {
1885
+ "task": "tracing-a-transitively-vulnerable-package-across-two-monorepo-workspaces-346",
1886
+ "domains": [
1887
+ "engineering"
1888
+ ]
1889
+ },
1890
+ {
1891
+ "task": "troubleshoot-why-my-esp32-cannot-initialize-a-mpu6050-chip-despite-correct-wirin",
1892
+ "domains": [
1893
+ "engineering"
1894
+ ]
1895
+ },
1896
+ {
1897
+ "task": "ufrj-sisu-2026-and-enem-comparison",
1898
+ "domains": [
1899
+ "public_records",
1900
+ "other"
1901
+ ]
1902
+ },
1903
+ {
1904
+ "task": "unknown-schematic-identification",
1905
+ "domains": [
1906
+ "engineering"
1907
+ ]
1908
+ },
1909
+ {
1910
+ "task": "v204-relief-valve-accumulation-check",
1911
+ "domains": [
1912
+ "engineering",
1913
+ "other"
1914
+ ]
1915
+ },
1916
+ {
1917
+ "task": "vfd-continuous-current-suitability-assessment",
1918
+ "domains": [
1919
+ "engineering"
1920
+ ]
1921
+ },
1922
+ {
1923
+ "task": "workstation-ssd-firmware-failure-screening-388",
1924
+ "domains": [
1925
+ "engineering",
1926
+ "science"
1927
+ ]
1928
+ },
1929
+ {
1930
+ "task": "wrongful-termination-of-employment-and-available-legal-remedies-232",
1931
+ "domains": [
1932
+ "law"
1933
+ ]
1934
+ }
1935
+ ]
1936
  },
1937
  "cost_note": "Eight model costs are recorded. Gemini 4 Argon, Gemini 3.8 Flash and Kimi K3 costs use saved input, cached-input and output token counts at the November 3, 2026 rate snapshot. Agent cost excludes failed reruns, grading, tools and infrastructure.",
1938
+ "cohort_note": "The uncapped search comparison replaces the earlier capped runs. This release contains 3,300 native-search runs and 1,350 uncapped search runs.",
1939
  "tasks": 100,
1940
+ "runs": 4650,
1941
  "native_runs": 3300,
1942
+ "ablation_runs": 1350,
1943
  "uncapped_ablation_runs": 1350,
 
1944
  "current_evaluation_runs": 4650,
1945
  "metrics": [
1946
  {
 
2003
  "3": 11
2004
  }
2005
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2006
  {
2007
  "model": "google/gemini-3.8-flash",
2008
  "condition": "exa",
 
2023
  "3": 23
2024
  }
2025
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2026
  {
2027
  "model": "google/gemini-3.8-flash",
2028
  "condition": "firecrawl",
 
2063
  "3": 38
2064
  }
2065
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2066
  {
2067
  "model": "google/gemini-3.8-flash",
2068
  "condition": "parallel",
 
2103
  "3": 81
2104
  }
2105
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2106
  {
2107
  "model": "gpt-5.6-sol",
2108
  "condition": "exa",
 
2123
  "3": 30
2124
  }
2125
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2126
  {
2127
  "model": "gpt-5.6-sol",
2128
  "condition": "firecrawl",
 
2163
  "3": 47
2164
  }
2165
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2166
  {
2167
  "model": "gpt-5.6-sol",
2168
  "condition": "parallel",
 
2223
  "3": 73
2224
  }
2225
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2226
  {
2227
  "model": "moonshotai/kimi-k3",
2228
  "condition": "exa",
 
2243
  "3": 17
2244
  }
2245
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2246
  {
2247
  "model": "moonshotai/kimi-k3",
2248
  "condition": "firecrawl",
 
2283
  "3": 23
2284
  }
2285
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2286
  {
2287
  "model": "moonshotai/kimi-k3",
2288
  "condition": "parallel",
 
2343
  }
2344
  ],
2345
  "no_new_grading_or_model_calls": true,
2346
+ "method": "Metrics are aggregated from the completed runs in this release. Native-search and uncapped search-backend conditions are reported separately."
2347
  }
analysis-summary-legacy-4269.json DELETED
@@ -1 +0,0 @@
1
- {"tasks":100,"runs":4269,"native_runs":3000,"ablation_runs":1269,"metrics":[{"model":"anthropic/claude-opus-5","condition":"native","tasks":100,"runs":300,"mean_reward":0.9224380066666666,"full_credit_pass_at_1":0.6966666666666667,"pass_power_k":{"1":0.6966666666666668,"2":0.59,"3":0.51},"perfect_attempts_per_task_histogram":{"2":24,"3":51,"0":17,"1":8}},{"model":"anthropic/claude-opus-5.5","condition":"native","tasks":100,"runs":300,"mean_reward":0.96980951,"full_credit_pass_at_1":0.86,"pass_power_k":{"1":0.86,"2":0.8,"3":0.76},"perfect_attempts_per_task_histogram":{"0":6,"2":12,"3":76,"1":6}},{"model":"anthropic/claude-sonnet-5","condition":"native","tasks":100,"runs":300,"mean_reward":0.7452871033333334,"full_credit_pass_at_1":0.3433333333333333,"pass_power_k":{"1":0.34333333333333327,"2":0.17666666666666667,"3":0.11},"perfect_attempts_per_task_histogram":{"0":39,"1":30,"3":11,"2":20}},{"model":"google/gemini-3.8-flash","condition":"exa","tasks":47,"runs":141,"mean_reward":0.8852179432624113,"full_credit_pass_at_1":0.5460992907801419,"pass_power_k":{"1":0.5460992907801419,"2":0.47517730496453897,"3":0.425531914893617},"perfect_attempts_per_task_histogram":{"0":17,"3":20,"1":3,"2":7}},{"model":"google/gemini-3.8-flash","condition":"firecrawl","tasks":47,"runs":141,"mean_reward":0.8536526382978723,"full_credit_pass_at_1":0.44680851063829785,"pass_power_k":{"1":0.44680851063829785,"2":0.3262411347517731,"3":0.2765957446808511},"perfect_attempts_per_task_histogram":{"1":10,"3":13,"0":17,"2":7}},{"model":"google/gemini-3.8-flash","condition":"native","tasks":100,"runs":300,"mean_reward":0.8724971966666667,"full_credit_pass_at_1":0.58,"pass_power_k":{"1":0.58,"2":0.46,"3":0.38},"perfect_attempts_per_task_histogram":{"3":38,"2":24,"0":26,"1":12}},{"model":"google/gemini-3.8-flash","condition":"parallel","tasks":47,"runs":141,"mean_reward":0.8851616595744681,"full_credit_pass_at_1":0.5673758865248227,"pass_power_k":{"1":0.5673758865248227,"2":0.47517730496453897,"3":0.44680851063829785},"perfect_attempts_per_task_histogram":{"3":21,"1":9,"2":4,"0":13}},{"model":"gpt-5.6-sol","condition":"exa","tasks":47,"runs":141,"mean_reward":0.9034142765957447,"full_credit_pass_at_1":0.6382978723404256,"pass_power_k":{"1":0.6382978723404256,"2":0.5035460992907802,"3":0.40425531914893614},"perfect_attempts_per_task_histogram":{"0":9,"2":14,"1":5,"3":19}},{"model":"gpt-5.6-sol","condition":"firecrawl","tasks":47,"runs":141,"mean_reward":0.8824806028368795,"full_credit_pass_at_1":0.5957446808510638,"pass_power_k":{"1":0.5957446808510638,"2":0.48936170212765956,"3":0.425531914893617},"perfect_attempts_per_task_histogram":{"2":9,"3":20,"0":12,"1":6}},{"model":"gpt-5.6-sol","condition":"native","tasks":100,"runs":300,"mean_reward":0.9074536733333334,"full_credit_pass_at_1":0.68,"pass_power_k":{"1":0.68,"2":0.5533333333333333,"3":0.47},"perfect_attempts_per_task_histogram":{"1":13,"0":15,"2":25,"3":47}},{"model":"gpt-5.6-sol","condition":"parallel","tasks":47,"runs":141,"mean_reward":0.8948098156028369,"full_credit_pass_at_1":0.6099290780141844,"pass_power_k":{"1":0.6099290780141844,"2":0.5106382978723404,"3":0.44680851063829785},"perfect_attempts_per_task_histogram":{"2":9,"0":12,"3":21,"1":5}},{"model":"gpt-6-astra","condition":"native","tasks":100,"runs":300,"mean_reward":0.9581732766666666,"full_credit_pass_at_1":0.82,"pass_power_k":{"1":0.82,"2":0.7833333333333333,"3":0.76},"perfect_attempts_per_task_histogram":{"3":76,"0":13,"2":7,"1":4}},{"model":"gpt-6.1-sol","condition":"native","tasks":100,"runs":300,"mean_reward":0.94528174,"full_credit_pass_at_1":0.7933333333333333,"pass_power_k":{"1":0.7933333333333333,"2":0.75,"3":0.73},"perfect_attempts_per_task_histogram":{"3":73,"0":14,"1":7,"2":6}},{"model":"moonshotai/kimi-k3","condition":"exa","tasks":47,"runs":141,"mean_reward":0.7981263191489362,"full_credit_pass_at_1":0.375886524822695,"pass_power_k":{"1":0.3758865248226951,"2":0.23404255319148937,"3":0.19148936170212766},"perfect_attempts_per_task_histogram":{"3":9,"0":18,"1":14,"2":6}},{"model":"moonshotai/kimi-k3","condition":"firecrawl","tasks":47,"runs":141,"mean_reward":0.7683709078014185,"full_credit_pass_at_1":0.36879432624113473,"pass_power_k":{"1":0.36879432624113473,"2":0.2553191489361702,"3":0.19148936170212766},"perfect_attempts_per_task_histogram":{"0":22,"1":7,"2":9,"3":9}},{"model":"moonshotai/kimi-k3","condition":"native","tasks":100,"runs":300,"mean_reward":0.8092198866666667,"full_credit_pass_at_1":0.41,"pass_power_k":{"1":0.41,"2":0.2733333333333333,"3":0.23},"perfect_attempts_per_task_histogram":{"1":28,"0":36,"3":23,"2":13}},{"model":"moonshotai/kimi-k3","condition":"parallel","tasks":47,"runs":141,"mean_reward":0.8110883475177305,"full_credit_pass_at_1":0.375886524822695,"pass_power_k":{"1":0.3758865248226951,"2":0.2553191489361702,"3":0.19148936170212766},"perfect_attempts_per_task_histogram":{"1":8,"0":21,"2":9,"3":9}},{"model":"thinkingmachines/inkling","condition":"native","tasks":100,"runs":300,"mean_reward":0.41612396,"full_credit_pass_at_1":0.03,"pass_power_k":{"1":0.03,"2":0.0,"3":0.0},"perfect_attempts_per_task_histogram":{"0":91,"1":9}},{"model":"x-ai/grok-4.6","condition":"native","tasks":100,"runs":300,"mean_reward":0.9147341299999999,"full_credit_pass_at_1":0.69,"pass_power_k":{"1":0.69,"2":0.5933333333333334,"3":0.54},"perfect_attempts_per_task_histogram":{"3":54,"1":13,"2":16,"0":17}}],"no_new_grading_or_model_calls":true,"method":"Mean reward and full-credit probability from three selected final attempts per task/configuration; pass^k is the mean probability that all k attempts pass, comb(c,k)/comb(3,k)."}
 
 
analysis-summary.json CHANGED
@@ -2,8 +2,7 @@
2
  "task_count": 100,
3
  "current_native_runs": 3300,
4
  "current_uncapped_search_runs": 1350,
5
- "retained_legacy_capped_runs": 1269,
6
- "retained_run_count": 5919,
7
  "added_run_count": 1650,
8
  "native_models": [
9
  {
@@ -931,16 +930,6 @@
931
  "image_audio_video_union_tasks": 33,
932
  "nontext_evidence_tasks": 30,
933
  "nontext_evidence_rubric_items": 90,
934
- "primary_domains": {
935
- "engineering": 46,
936
- "medicine": 13,
937
- "science": 7,
938
- "aviation": 1,
939
- "law": 7,
940
- "public_records": 14,
941
- "finance": 9,
942
- "other": 3
943
- },
944
  "rubric_categories": [
945
  {
946
  "category": "Fact / rule identification",
@@ -1259,19 +1248,699 @@
1259
  "Joined application / comparison",
1260
  "Fact / rule identification"
1261
  ],
1262
- "provenance": "Primary domains are original author-provided submission labels. Rubric criterionType is empty for all 663 items. Four reporting categories are exploratory post-hoc coding of primary requirement and original source-support graph, not original labels, a new grading rubric, or a causal failure taxonomy. Calculation and grounding/limits labels were reviewed against criterion text. Remainder splits on whether the linked source graph uses both workspace and web sources.",
1263
  "rate_definition": "Pooled pass rate of saved per-item boolean judgments across all 11 native-search models, all 100 tasks and all 3 trials per model/task. Each of 663 items has 33 judgments. Categories have different difficulty and composition; rates are descriptive and do not establish causes.",
1264
  "scoring_basis": "Saved per-item outcomes for non-perfect attempts; validated full-credit records imply every unit-weight item passed. No new grading.",
1265
- "inventory_definition": "Delivered environment/artifacts files excluding the standard answer_format.md. Original submission URLs without delivered local files are excluded. Image-only PDF means no extractable text and at least one embedded raster image; PDFs with an OCR layer are not identified as image-only."
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1266
  },
1267
  "cost_note": "Eight model costs are recorded. Gemini 4 Argon, Gemini 3.8 Flash and Kimi K3 costs use saved input, cached-input and output token counts at the November 3, 2026 rate snapshot. Agent cost excludes failed reruns, grading, tools and infrastructure.",
1268
- "cohort_note": "The 1,269 earlier capped search runs are retained separately and are not pooled with the 1,350 uncapped runs.",
1269
  "tasks": 100,
1270
- "runs": 5919,
1271
  "native_runs": 3300,
1272
- "ablation_runs": 2619,
1273
  "uncapped_ablation_runs": 1350,
1274
- "legacy_capped_ablation_runs": 1269,
1275
  "current_evaluation_runs": 4650,
1276
  "metrics": [
1277
  {
@@ -1334,26 +2003,6 @@
1334
  "3": 11
1335
  }
1336
  },
1337
- {
1338
- "model": "google/gemini-3.8-flash",
1339
- "condition": "exa",
1340
- "cohort": "legacy_capped",
1341
- "tasks": 47,
1342
- "runs": 141,
1343
- "mean_reward": 0.8852179432624113,
1344
- "full_credit_pass_at_1": 0.5460992907801419,
1345
- "pass_power_k": {
1346
- "1": 0.5460992907801419,
1347
- "2": 0.475177304964539,
1348
- "3": 0.425531914893617
1349
- },
1350
- "perfect_attempts_per_task_histogram": {
1351
- "0": 17,
1352
- "1": 3,
1353
- "2": 7,
1354
- "3": 20
1355
- }
1356
- },
1357
  {
1358
  "model": "google/gemini-3.8-flash",
1359
  "condition": "exa",
@@ -1374,26 +2023,6 @@
1374
  "3": 23
1375
  }
1376
  },
1377
- {
1378
- "model": "google/gemini-3.8-flash",
1379
- "condition": "firecrawl",
1380
- "cohort": "legacy_capped",
1381
- "tasks": 47,
1382
- "runs": 141,
1383
- "mean_reward": 0.8536526382978723,
1384
- "full_credit_pass_at_1": 0.44680851063829785,
1385
- "pass_power_k": {
1386
- "1": 0.44680851063829785,
1387
- "2": 0.3262411347517731,
1388
- "3": 0.2765957446808511
1389
- },
1390
- "perfect_attempts_per_task_histogram": {
1391
- "0": 17,
1392
- "1": 10,
1393
- "2": 7,
1394
- "3": 13
1395
- }
1396
- },
1397
  {
1398
  "model": "google/gemini-3.8-flash",
1399
  "condition": "firecrawl",
@@ -1434,26 +2063,6 @@
1434
  "3": 38
1435
  }
1436
  },
1437
- {
1438
- "model": "google/gemini-3.8-flash",
1439
- "condition": "parallel",
1440
- "cohort": "legacy_capped",
1441
- "tasks": 47,
1442
- "runs": 141,
1443
- "mean_reward": 0.8851616595744681,
1444
- "full_credit_pass_at_1": 0.5673758865248227,
1445
- "pass_power_k": {
1446
- "1": 0.5673758865248227,
1447
- "2": 0.475177304964539,
1448
- "3": 0.44680851063829785
1449
- },
1450
- "perfect_attempts_per_task_histogram": {
1451
- "0": 13,
1452
- "1": 9,
1453
- "2": 4,
1454
- "3": 21
1455
- }
1456
- },
1457
  {
1458
  "model": "google/gemini-3.8-flash",
1459
  "condition": "parallel",
@@ -1494,26 +2103,6 @@
1494
  "3": 81
1495
  }
1496
  },
1497
- {
1498
- "model": "gpt-5.6-sol",
1499
- "condition": "exa",
1500
- "cohort": "legacy_capped",
1501
- "tasks": 47,
1502
- "runs": 141,
1503
- "mean_reward": 0.9034142765957447,
1504
- "full_credit_pass_at_1": 0.6382978723404256,
1505
- "pass_power_k": {
1506
- "1": 0.6382978723404256,
1507
- "2": 0.5035460992907801,
1508
- "3": 0.40425531914893614
1509
- },
1510
- "perfect_attempts_per_task_histogram": {
1511
- "0": 9,
1512
- "1": 5,
1513
- "2": 14,
1514
- "3": 19
1515
- }
1516
- },
1517
  {
1518
  "model": "gpt-5.6-sol",
1519
  "condition": "exa",
@@ -1534,26 +2123,6 @@
1534
  "3": 30
1535
  }
1536
  },
1537
- {
1538
- "model": "gpt-5.6-sol",
1539
- "condition": "firecrawl",
1540
- "cohort": "legacy_capped",
1541
- "tasks": 47,
1542
- "runs": 141,
1543
- "mean_reward": 0.8824806028368795,
1544
- "full_credit_pass_at_1": 0.5957446808510638,
1545
- "pass_power_k": {
1546
- "1": 0.5957446808510638,
1547
- "2": 0.48936170212765956,
1548
- "3": 0.425531914893617
1549
- },
1550
- "perfect_attempts_per_task_histogram": {
1551
- "0": 12,
1552
- "1": 6,
1553
- "2": 9,
1554
- "3": 20
1555
- }
1556
- },
1557
  {
1558
  "model": "gpt-5.6-sol",
1559
  "condition": "firecrawl",
@@ -1594,26 +2163,6 @@
1594
  "3": 47
1595
  }
1596
  },
1597
- {
1598
- "model": "gpt-5.6-sol",
1599
- "condition": "parallel",
1600
- "cohort": "legacy_capped",
1601
- "tasks": 47,
1602
- "runs": 141,
1603
- "mean_reward": 0.8948098156028369,
1604
- "full_credit_pass_at_1": 0.6099290780141844,
1605
- "pass_power_k": {
1606
- "1": 0.6099290780141844,
1607
- "2": 0.5106382978723404,
1608
- "3": 0.44680851063829785
1609
- },
1610
- "perfect_attempts_per_task_histogram": {
1611
- "0": 12,
1612
- "1": 5,
1613
- "2": 9,
1614
- "3": 21
1615
- }
1616
- },
1617
  {
1618
  "model": "gpt-5.6-sol",
1619
  "condition": "parallel",
@@ -1674,26 +2223,6 @@
1674
  "3": 73
1675
  }
1676
  },
1677
- {
1678
- "model": "moonshotai/kimi-k3",
1679
- "condition": "exa",
1680
- "cohort": "legacy_capped",
1681
- "tasks": 47,
1682
- "runs": 141,
1683
- "mean_reward": 0.7981263191489362,
1684
- "full_credit_pass_at_1": 0.375886524822695,
1685
- "pass_power_k": {
1686
- "1": 0.375886524822695,
1687
- "2": 0.23404255319148937,
1688
- "3": 0.19148936170212766
1689
- },
1690
- "perfect_attempts_per_task_histogram": {
1691
- "0": 18,
1692
- "1": 14,
1693
- "2": 6,
1694
- "3": 9
1695
- }
1696
- },
1697
  {
1698
  "model": "moonshotai/kimi-k3",
1699
  "condition": "exa",
@@ -1714,26 +2243,6 @@
1714
  "3": 17
1715
  }
1716
  },
1717
- {
1718
- "model": "moonshotai/kimi-k3",
1719
- "condition": "firecrawl",
1720
- "cohort": "legacy_capped",
1721
- "tasks": 47,
1722
- "runs": 141,
1723
- "mean_reward": 0.7683709078014185,
1724
- "full_credit_pass_at_1": 0.36879432624113473,
1725
- "pass_power_k": {
1726
- "1": 0.36879432624113473,
1727
- "2": 0.2553191489361702,
1728
- "3": 0.19148936170212766
1729
- },
1730
- "perfect_attempts_per_task_histogram": {
1731
- "0": 22,
1732
- "1": 7,
1733
- "2": 9,
1734
- "3": 9
1735
- }
1736
- },
1737
  {
1738
  "model": "moonshotai/kimi-k3",
1739
  "condition": "firecrawl",
@@ -1774,26 +2283,6 @@
1774
  "3": 23
1775
  }
1776
  },
1777
- {
1778
- "model": "moonshotai/kimi-k3",
1779
- "condition": "parallel",
1780
- "cohort": "legacy_capped",
1781
- "tasks": 47,
1782
- "runs": 141,
1783
- "mean_reward": 0.8110883475177305,
1784
- "full_credit_pass_at_1": 0.375886524822695,
1785
- "pass_power_k": {
1786
- "1": 0.375886524822695,
1787
- "2": 0.2553191489361702,
1788
- "3": 0.19148936170212766
1789
- },
1790
- "perfect_attempts_per_task_histogram": {
1791
- "0": 21,
1792
- "1": 8,
1793
- "2": 9,
1794
- "3": 9
1795
- }
1796
- },
1797
  {
1798
  "model": "moonshotai/kimi-k3",
1799
  "condition": "parallel",
@@ -1854,5 +2343,5 @@
1854
  }
1855
  ],
1856
  "no_new_grading_or_model_calls": true,
1857
- "method": "Metrics are aggregated from retained completed run indexes. Cohort labels distinguish native, current uncapped and historical capped runs; capped and uncapped results are never pooled."
1858
  }
 
2
  "task_count": 100,
3
  "current_native_runs": 3300,
4
  "current_uncapped_search_runs": 1350,
5
+ "retained_run_count": 4650,
 
6
  "added_run_count": 1650,
7
  "native_models": [
8
  {
 
930
  "image_audio_video_union_tasks": 33,
931
  "nontext_evidence_tasks": 30,
932
  "nontext_evidence_rubric_items": 90,
 
 
 
 
 
 
 
 
 
 
933
  "rubric_categories": [
934
  {
935
  "category": "Fact / rule identification",
 
1248
  "Joined application / comparison",
1249
  "Fact / rule identification"
1250
  ],
1251
+ "provenance": "Domain counts use all author-provided submission labels; tasks may belong to multiple domains. Rubric criterionType is empty for all 663 items. Four reporting categories are exploratory post-hoc coding of primary requirement and original source-support graph, not original labels, a new grading rubric, or a causal failure taxonomy. Calculation and grounding/limits labels were reviewed against criterion text. Remainder splits on whether the linked source graph uses both workspace and web sources.",
1252
  "rate_definition": "Pooled pass rate of saved per-item boolean judgments across all 11 native-search models, all 100 tasks and all 3 trials per model/task. Each of 663 items has 33 judgments. Categories have different difficulty and composition; rates are descriptive and do not establish causes.",
1253
  "scoring_basis": "Saved per-item outcomes for non-perfect attempts; validated full-credit records imply every unit-weight item passed. No new grading.",
1254
+ "inventory_definition": "Delivered environment/artifacts files excluding the standard answer_format.md. Original submission URLs without delivered local files are excluded. Image-only PDF means no extractable text and at least one embedded raster image; PDFs with an OCR layer are not identified as image-only.",
1255
+ "domain_counts": {
1256
+ "engineering": 54,
1257
+ "public_records": 41,
1258
+ "finance": 15,
1259
+ "law": 14,
1260
+ "other": 14,
1261
+ "medicine": 13,
1262
+ "science": 12,
1263
+ "aviation": 4
1264
+ },
1265
+ "multi_domain_task_count": 55,
1266
+ "domain_assignment_count": 167,
1267
+ "task_domain_labels": [
1268
+ {
1269
+ "task": "6-ghz-canopy-deployment-screening-for-a-distribution-centre-wi-fi-refresh",
1270
+ "domains": [
1271
+ "engineering",
1272
+ "public_records"
1273
+ ]
1274
+ },
1275
+ {
1276
+ "task": "acetaminophen-overdose-treatment-and-stopping-decision-294",
1277
+ "domains": [
1278
+ "medicine"
1279
+ ]
1280
+ },
1281
+ {
1282
+ "task": "adult-laryngoscope-set-assessment",
1283
+ "domains": [
1284
+ "medicine",
1285
+ "engineering"
1286
+ ]
1287
+ },
1288
+ {
1289
+ "task": "aggrow-program-farmer-eligibility-check",
1290
+ "domains": [
1291
+ "science",
1292
+ "finance",
1293
+ "public_records"
1294
+ ]
1295
+ },
1296
+ {
1297
+ "task": "agroforestry-meal-pipeline-data-integrity-and-survival-bound-audit",
1298
+ "domains": [
1299
+ "science",
1300
+ "other"
1301
+ ]
1302
+ },
1303
+ {
1304
+ "task": "aha-bls-certification-verification",
1305
+ "domains": [
1306
+ "medicine"
1307
+ ]
1308
+ },
1309
+ {
1310
+ "task": "ai-recommendation-service-integrity-and-sre-coverage-audit",
1311
+ "domains": [
1312
+ "engineering"
1313
+ ]
1314
+ },
1315
+ {
1316
+ "task": "aviation-cargo-lithium-battery-exemption-compliance-check",
1317
+ "domains": [
1318
+ "aviation",
1319
+ "public_records",
1320
+ "other"
1321
+ ]
1322
+ },
1323
+ {
1324
+ "task": "aviation-engine-mount-torque-audit-ac-43-13-1b-131",
1325
+ "domains": [
1326
+ "engineering",
1327
+ "aviation",
1328
+ "public_records"
1329
+ ]
1330
+ },
1331
+ {
1332
+ "task": "ax417-lithium-battery-baggage-release-audit",
1333
+ "domains": [
1334
+ "law",
1335
+ "engineering",
1336
+ "aviation"
1337
+ ]
1338
+ },
1339
+ {
1340
+ "task": "beamex-fb660-calibration-audit",
1341
+ "domains": [
1342
+ "engineering"
1343
+ ]
1344
+ },
1345
+ {
1346
+ "task": "belkin-charging-accessory-recall-screening-across-two-overlapping-cpsc-recalls",
1347
+ "domains": [
1348
+ "engineering",
1349
+ "public_records"
1350
+ ]
1351
+ },
1352
+ {
1353
+ "task": "beryllium-sampling-records-checked-against-current-exposure-limits",
1354
+ "domains": [
1355
+ "science",
1356
+ "public_records"
1357
+ ]
1358
+ },
1359
+ {
1360
+ "task": "brickwork-final-bill-audit-against-cpwd-dsr-2023",
1361
+ "domains": [
1362
+ "engineering",
1363
+ "public_records"
1364
+ ]
1365
+ },
1366
+ {
1367
+ "task": "building-c-ryobi-electric-pressure-washer-recall-screen",
1368
+ "domains": [
1369
+ "engineering",
1370
+ "public_records"
1371
+ ]
1372
+ },
1373
+ {
1374
+ "task": "bulk-email-sender-compliance-screening",
1375
+ "domains": [
1376
+ "engineering",
1377
+ "public_records"
1378
+ ]
1379
+ },
1380
+ {
1381
+ "task": "caips-delivery-status-vs-2025-target-kenya",
1382
+ "domains": [
1383
+ "public_records",
1384
+ "finance"
1385
+ ]
1386
+ },
1387
+ {
1388
+ "task": "campaign-24v-744-scope-and-repair-version-audit",
1389
+ "domains": [
1390
+ "public_records"
1391
+ ]
1392
+ },
1393
+ {
1394
+ "task": "candidate-neem-oil-fungicide-formulations-vs-a-historic-european-patent-s-claim-",
1395
+ "domains": [
1396
+ "law",
1397
+ "science",
1398
+ "other"
1399
+ ]
1400
+ },
1401
+ {
1402
+ "task": "champions-league-suspension-audit-carvajal-vs-mendy-2023-24-semifinal",
1403
+ "domains": [
1404
+ "public_records",
1405
+ "other"
1406
+ ]
1407
+ },
1408
+ {
1409
+ "task": "cisco-ios-xe-exposure-and-compromise-screening",
1410
+ "domains": [
1411
+ "public_records",
1412
+ "engineering"
1413
+ ]
1414
+ },
1415
+ {
1416
+ "task": "commercial-partnership-zakat-compliance-and-audit-assessment",
1417
+ "domains": [
1418
+ "finance"
1419
+ ]
1420
+ },
1421
+ {
1422
+ "task": "comps-model-revenue-integrity-check-against-a-public-restatement",
1423
+ "domains": [
1424
+ "finance",
1425
+ "public_records"
1426
+ ]
1427
+ },
1428
+ {
1429
+ "task": "computing-landed-cost-for-used-clothing-consignment-into-mombasa-march-2025",
1430
+ "domains": [
1431
+ "law",
1432
+ "finance",
1433
+ "public_records"
1434
+ ]
1435
+ },
1436
+ {
1437
+ "task": "covered-call-coverage-check-after-a-listed-contract-adjustment",
1438
+ "domains": [
1439
+ "finance",
1440
+ "public_records"
1441
+ ]
1442
+ },
1443
+ {
1444
+ "task": "designing-a-flow-cyotometry-panel",
1445
+ "domains": [
1446
+ "science"
1447
+ ]
1448
+ },
1449
+ {
1450
+ "task": "determine-which-dogs-are-eligible-for-international-travel-based-on-their-rabies",
1451
+ "domains": [
1452
+ "medicine",
1453
+ "public_records",
1454
+ "science"
1455
+ ]
1456
+ },
1457
+ {
1458
+ "task": "diagnosing-bright-sky-background-in-milky-way-photography-session",
1459
+ "domains": [
1460
+ "science",
1461
+ "other"
1462
+ ]
1463
+ },
1464
+ {
1465
+ "task": "didion-milling-dust-explosion-from-the-plant-floor-to-the-courtroom",
1466
+ "domains": [
1467
+ "public_records",
1468
+ "science",
1469
+ "law"
1470
+ ]
1471
+ },
1472
+ {
1473
+ "task": "dio-http-client-version-security-check",
1474
+ "domains": [
1475
+ "engineering"
1476
+ ]
1477
+ },
1478
+ {
1479
+ "task": "dns-resolution-incident",
1480
+ "domains": [
1481
+ "engineering"
1482
+ ]
1483
+ },
1484
+ {
1485
+ "task": "dnssec-chain-of-trust-rollover-failure-and-recovery-audit",
1486
+ "domains": [
1487
+ "engineering"
1488
+ ]
1489
+ },
1490
+ {
1491
+ "task": "docker-base-image-cve-exposure-check",
1492
+ "domains": [
1493
+ "engineering"
1494
+ ]
1495
+ },
1496
+ {
1497
+ "task": "emergency-poc-test-kit-and-iv-fluid-shelf-life-extension-audit",
1498
+ "domains": [
1499
+ "medicine"
1500
+ ]
1501
+ },
1502
+ {
1503
+ "task": "extraction-photocatalyst-data-of-ta2o5",
1504
+ "domains": [
1505
+ "science",
1506
+ "other"
1507
+ ]
1508
+ },
1509
+ {
1510
+ "task": "fact-check-a-map-s-color-scale-against-census-data",
1511
+ "domains": [
1512
+ "public_records"
1513
+ ]
1514
+ },
1515
+ {
1516
+ "task": "fact-checking-a-literature-review-product-description-against-the-manufacturer-s",
1517
+ "domains": [
1518
+ "engineering"
1519
+ ]
1520
+ },
1521
+ {
1522
+ "task": "fairfax-bridge-posting-compliance-across-restriction-history",
1523
+ "domains": [
1524
+ "public_records",
1525
+ "engineering"
1526
+ ]
1527
+ },
1528
+ {
1529
+ "task": "fern-hollow-collapse-geometry-and-initiating-member-findings",
1530
+ "domains": [
1531
+ "engineering",
1532
+ "public_records"
1533
+ ]
1534
+ },
1535
+ {
1536
+ "task": "fieldkit-native-sign-in-migration-audit",
1537
+ "domains": [
1538
+ "engineering"
1539
+ ]
1540
+ },
1541
+ {
1542
+ "task": "fintech-transaction-record-handling-and-retention-compliance-audit",
1543
+ "domains": [
1544
+ "engineering",
1545
+ "law"
1546
+ ]
1547
+ },
1548
+ {
1549
+ "task": "frozen-waffle-recall-inventory-check",
1550
+ "domains": [
1551
+ "public_records",
1552
+ "other"
1553
+ ]
1554
+ },
1555
+ {
1556
+ "task": "github-actions-runner-migration-across-multiple-ci-cd-workflows",
1557
+ "domains": [
1558
+ "engineering"
1559
+ ]
1560
+ },
1561
+ {
1562
+ "task": "github-api-secondary-rate-limit-ticket-dismissal-check-334",
1563
+ "domains": [
1564
+ "engineering"
1565
+ ]
1566
+ },
1567
+ {
1568
+ "task": "github-rest-api-compatibility-review",
1569
+ "domains": [
1570
+ "engineering"
1571
+ ]
1572
+ },
1573
+ {
1574
+ "task": "grandmother-s-old-egfr-calculator-vs-current-kidney-function-guidelines",
1575
+ "domains": [
1576
+ "medicine",
1577
+ "public_records"
1578
+ ]
1579
+ },
1580
+ {
1581
+ "task": "handlebars-patch-triage-across-two-pdf-generation-features",
1582
+ "domains": [
1583
+ "engineering",
1584
+ "public_records"
1585
+ ]
1586
+ },
1587
+ {
1588
+ "task": "heavy-haul-permit-limit-on-a-us-1-route-from-the-national-bridge-inventory",
1589
+ "domains": [
1590
+ "engineering",
1591
+ "public_records",
1592
+ "law"
1593
+ ]
1594
+ },
1595
+ {
1596
+ "task": "hemodialysis-patient-protein-intake-adequacy-vs-kdoqi-2020",
1597
+ "domains": [
1598
+ "medicine"
1599
+ ]
1600
+ },
1601
+ {
1602
+ "task": "hospital-pharmacy-medication-safety-audit",
1603
+ "domains": [
1604
+ "medicine"
1605
+ ]
1606
+ },
1607
+ {
1608
+ "task": "hsg-imaging",
1609
+ "domains": [
1610
+ "medicine"
1611
+ ]
1612
+ },
1613
+ {
1614
+ "task": "insider-sale-red-flag-for-advisory-client-of-gild",
1615
+ "domains": [
1616
+ "public_records",
1617
+ "finance"
1618
+ ]
1619
+ },
1620
+ {
1621
+ "task": "inspection-and-compliance-assessment-of-sets-of-q690-steel-plate-214",
1622
+ "domains": [
1623
+ "engineering"
1624
+ ]
1625
+ },
1626
+ {
1627
+ "task": "iphone-11-battery-replacement-warranty-investigation",
1628
+ "domains": [
1629
+ "other",
1630
+ "engineering"
1631
+ ]
1632
+ },
1633
+ {
1634
+ "task": "jwt-library-vulnerability-exposure-check-for-finance-tracker-auth",
1635
+ "domains": [
1636
+ "engineering",
1637
+ "public_records"
1638
+ ]
1639
+ },
1640
+ {
1641
+ "task": "kenya-poultry-feed-cost-report-update-280",
1642
+ "domains": [
1643
+ "public_records",
1644
+ "other"
1645
+ ]
1646
+ },
1647
+ {
1648
+ "task": "kenya-power-lighting-company-commercial-and-industrial-c-i-tariff-period-verific",
1649
+ "domains": [
1650
+ "engineering"
1651
+ ]
1652
+ },
1653
+ {
1654
+ "task": "kenyan-electronic-goods-import-tax-liability-calculation",
1655
+ "domains": [
1656
+ "law",
1657
+ "finance"
1658
+ ]
1659
+ },
1660
+ {
1661
+ "task": "kidde-fire-extinguisher-recall-and-replacement-audit",
1662
+ "domains": [
1663
+ "public_records",
1664
+ "law"
1665
+ ]
1666
+ },
1667
+ {
1668
+ "task": "kilifi-ecvr-tracker-reconciliation-against-iebc-records",
1669
+ "domains": [
1670
+ "public_records"
1671
+ ]
1672
+ },
1673
+ {
1674
+ "task": "kitchen-range-recall-screening-for-five-property-assets",
1675
+ "domains": [
1676
+ "engineering",
1677
+ "public_records"
1678
+ ]
1679
+ },
1680
+ {
1681
+ "task": "landed-cost-calculation-for-an-imitation-jewelry-shipment-from-china-to-ethiopia",
1682
+ "domains": [
1683
+ "law",
1684
+ "finance"
1685
+ ]
1686
+ },
1687
+ {
1688
+ "task": "leaked-ci-credential-breach-notification-scope-deadline",
1689
+ "domains": [
1690
+ "engineering",
1691
+ "law"
1692
+ ]
1693
+ },
1694
+ {
1695
+ "task": "lekki-toll-gate-article-fact-check-casualty-claim-and-victim-list-misattribution",
1696
+ "domains": [
1697
+ "public_records",
1698
+ "other"
1699
+ ]
1700
+ },
1701
+ {
1702
+ "task": "lodash-prototype-pollution-exposure-check-for-a-node-api-project",
1703
+ "domains": [
1704
+ "engineering"
1705
+ ]
1706
+ },
1707
+ {
1708
+ "task": "management-of-severe-plasmodium-falciparum-malaria-with-anaemia-in-a-quinine-sen",
1709
+ "domains": [
1710
+ "medicine"
1711
+ ]
1712
+ },
1713
+ {
1714
+ "task": "mimedx-2015-revenue-the-growth-that-was-partly-fake",
1715
+ "domains": [
1716
+ "finance",
1717
+ "public_records",
1718
+ "law"
1719
+ ]
1720
+ },
1721
+ {
1722
+ "task": "nimbus-lqa-redis-licensing-exposure",
1723
+ "domains": [
1724
+ "law",
1725
+ "engineering"
1726
+ ]
1727
+ },
1728
+ {
1729
+ "task": "overdue-lycoming-connecting-rod-ad-inspection-cessna-172s-n832xx",
1730
+ "domains": [
1731
+ "other",
1732
+ "aviation"
1733
+ ]
1734
+ },
1735
+ {
1736
+ "task": "patient-metformin-dispensing-log-screening-against-2020-fda-ndma-recall",
1737
+ "domains": [
1738
+ "medicine",
1739
+ "public_records"
1740
+ ]
1741
+ },
1742
+ {
1743
+ "task": "pavement-structure-determination-for-a-rural-access-road-in-ethiopia-north-shoa-",
1744
+ "domains": [
1745
+ "engineering"
1746
+ ]
1747
+ },
1748
+ {
1749
+ "task": "paystack-account-disablement-mapping-a-cited-reason-to-the-actual-policy-text",
1750
+ "domains": [
1751
+ "finance",
1752
+ "law",
1753
+ "engineering"
1754
+ ]
1755
+ },
1756
+ {
1757
+ "task": "pharmaceutical-wastewater-safety-test",
1758
+ "domains": [
1759
+ "engineering"
1760
+ ]
1761
+ },
1762
+ {
1763
+ "task": "pharmacy-chlorthalidone-lot-screening-against-fda-recall",
1764
+ "domains": [
1765
+ "medicine",
1766
+ "public_records"
1767
+ ]
1768
+ },
1769
+ {
1770
+ "task": "pipeline-restart-decision-and-release-audit",
1771
+ "domains": [
1772
+ "engineering",
1773
+ "science",
1774
+ "public_records"
1775
+ ]
1776
+ },
1777
+ {
1778
+ "task": "piper-pa-28-fleet-wing-spar-airworthiness-directive-ad-2020-26-16-compliance-scr",
1779
+ "domains": [
1780
+ "engineering",
1781
+ "public_records"
1782
+ ]
1783
+ },
1784
+ {
1785
+ "task": "planar-diffuser-cfd-validation-review",
1786
+ "domains": [
1787
+ "science",
1788
+ "engineering"
1789
+ ]
1790
+ },
1791
+ {
1792
+ "task": "portfolio-memo-verification-for-holdings-h-01-fy2025",
1793
+ "domains": [
1794
+ "finance"
1795
+ ]
1796
+ },
1797
+ {
1798
+ "task": "post-maintenance-fiber-uplink-compliance-review-111",
1799
+ "domains": [
1800
+ "engineering"
1801
+ ]
1802
+ },
1803
+ {
1804
+ "task": "post-maintenance-memory-population-review-for-three-server-hosts",
1805
+ "domains": [
1806
+ "engineering"
1807
+ ]
1808
+ },
1809
+ {
1810
+ "task": "print-and-epub-distribution-readiness-check-for-a-fantasy-short-story",
1811
+ "domains": [
1812
+ "other"
1813
+ ]
1814
+ },
1815
+ {
1816
+ "task": "q3-2025-mileage-reimbursement-tax-compliance-check",
1817
+ "domains": [
1818
+ "finance"
1819
+ ]
1820
+ },
1821
+ {
1822
+ "task": "re-vlab-production-hosting-go-live-audit",
1823
+ "domains": [
1824
+ "engineering"
1825
+ ]
1826
+ },
1827
+ {
1828
+ "task": "recall-reconciliation-log-from-the-national-register-for-a-hospital-in-2025",
1829
+ "domains": [
1830
+ "medicine"
1831
+ ]
1832
+ },
1833
+ {
1834
+ "task": "reconciling-an-investor-deck-against-sec-filings-for-capstone-companies-inc",
1835
+ "domains": [
1836
+ "finance",
1837
+ "public_records"
1838
+ ]
1839
+ },
1840
+ {
1841
+ "task": "redistribution-license-clearance-for-the-4-2-on-prem-appliance-image",
1842
+ "domains": [
1843
+ "engineering"
1844
+ ]
1845
+ },
1846
+ {
1847
+ "task": "regresshion-exposure-across-five-firmware-builds",
1848
+ "domains": [
1849
+ "engineering"
1850
+ ]
1851
+ },
1852
+ {
1853
+ "task": "remaining-fatigue-life-review-of-three-welded-hot-spots-on-an-offshore-topside-f",
1854
+ "domains": [
1855
+ "engineering"
1856
+ ]
1857
+ },
1858
+ {
1859
+ "task": "sec-non-gaap-comment-letter-reconciliation-for-a-filed-earnings-release",
1860
+ "domains": [
1861
+ "finance",
1862
+ "public_records"
1863
+ ]
1864
+ },
1865
+ {
1866
+ "task": "signal-warrant-evaluation-for-a-small-city-intersection-mutcd-chapter-4c",
1867
+ "domains": [
1868
+ "engineering"
1869
+ ]
1870
+ },
1871
+ {
1872
+ "task": "sitop-ups1600-firmware-advisory-screen-for-building-b-assets-222",
1873
+ "domains": [
1874
+ "engineering",
1875
+ "public_records"
1876
+ ]
1877
+ },
1878
+ {
1879
+ "task": "thermography-measurement-validity",
1880
+ "domains": [
1881
+ "engineering"
1882
+ ]
1883
+ },
1884
+ {
1885
+ "task": "tracing-a-transitively-vulnerable-package-across-two-monorepo-workspaces-346",
1886
+ "domains": [
1887
+ "engineering"
1888
+ ]
1889
+ },
1890
+ {
1891
+ "task": "troubleshoot-why-my-esp32-cannot-initialize-a-mpu6050-chip-despite-correct-wirin",
1892
+ "domains": [
1893
+ "engineering"
1894
+ ]
1895
+ },
1896
+ {
1897
+ "task": "ufrj-sisu-2026-and-enem-comparison",
1898
+ "domains": [
1899
+ "public_records",
1900
+ "other"
1901
+ ]
1902
+ },
1903
+ {
1904
+ "task": "unknown-schematic-identification",
1905
+ "domains": [
1906
+ "engineering"
1907
+ ]
1908
+ },
1909
+ {
1910
+ "task": "v204-relief-valve-accumulation-check",
1911
+ "domains": [
1912
+ "engineering",
1913
+ "other"
1914
+ ]
1915
+ },
1916
+ {
1917
+ "task": "vfd-continuous-current-suitability-assessment",
1918
+ "domains": [
1919
+ "engineering"
1920
+ ]
1921
+ },
1922
+ {
1923
+ "task": "workstation-ssd-firmware-failure-screening-388",
1924
+ "domains": [
1925
+ "engineering",
1926
+ "science"
1927
+ ]
1928
+ },
1929
+ {
1930
+ "task": "wrongful-termination-of-employment-and-available-legal-remedies-232",
1931
+ "domains": [
1932
+ "law"
1933
+ ]
1934
+ }
1935
+ ]
1936
  },
1937
  "cost_note": "Eight model costs are recorded. Gemini 4 Argon, Gemini 3.8 Flash and Kimi K3 costs use saved input, cached-input and output token counts at the November 3, 2026 rate snapshot. Agent cost excludes failed reruns, grading, tools and infrastructure.",
1938
+ "cohort_note": "The uncapped search comparison replaces the earlier capped runs. This release contains 3,300 native-search runs and 1,350 uncapped search runs.",
1939
  "tasks": 100,
1940
+ "runs": 4650,
1941
  "native_runs": 3300,
1942
+ "ablation_runs": 1350,
1943
  "uncapped_ablation_runs": 1350,
 
1944
  "current_evaluation_runs": 4650,
1945
  "metrics": [
1946
  {
 
2003
  "3": 11
2004
  }
2005
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2006
  {
2007
  "model": "google/gemini-3.8-flash",
2008
  "condition": "exa",
 
2023
  "3": 23
2024
  }
2025
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2026
  {
2027
  "model": "google/gemini-3.8-flash",
2028
  "condition": "firecrawl",
 
2063
  "3": 38
2064
  }
2065
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2066
  {
2067
  "model": "google/gemini-3.8-flash",
2068
  "condition": "parallel",
 
2103
  "3": 81
2104
  }
2105
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2106
  {
2107
  "model": "gpt-5.6-sol",
2108
  "condition": "exa",
 
2123
  "3": 30
2124
  }
2125
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2126
  {
2127
  "model": "gpt-5.6-sol",
2128
  "condition": "firecrawl",
 
2163
  "3": 47
2164
  }
2165
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2166
  {
2167
  "model": "gpt-5.6-sol",
2168
  "condition": "parallel",
 
2223
  "3": 73
2224
  }
2225
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2226
  {
2227
  "model": "moonshotai/kimi-k3",
2228
  "condition": "exa",
 
2243
  "3": 17
2244
  }
2245
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2246
  {
2247
  "model": "moonshotai/kimi-k3",
2248
  "condition": "firecrawl",
 
2283
  "3": 23
2284
  }
2285
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2286
  {
2287
  "model": "moonshotai/kimi-k3",
2288
  "condition": "parallel",
 
2343
  }
2344
  ],
2345
  "no_new_grading_or_model_calls": true,
2346
+ "method": "Metrics are aggregated from the completed runs in this release. Native-search and uncapped search-backend conditions are reported separately."
2347
  }
release-manifest-legacy-4269.json DELETED
The diff for this file is too large to render. See raw diff
 
release-manifest.json CHANGED
The diff for this file is too large to render. See raw diff
 
rollouts-00003.jsonl.gz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:3f507488faf4af64585704a4aa7518b797454256f769202ccf303ff2299daf13
3
- size 18458907
 
 
 
 
rollouts-00005.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2c8155b58d8fdb29acc0f7c9734fb7598cf7444c7c7a3c6825f5c55f8a39ed0d
3
- size 32511468
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4d3a7b4555ac161c6cd9b9731b61173109d87ffe62d3a12eed468475189b1f3e
3
+ size 24200870
rollouts-00007.jsonl.gz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:79f41bdfd14bb19a9c64c823356e03d1b58ac6fe59e0d99a0fd39306ddc66fc0
3
- size 63093299
 
 
 
 
rollouts-00008.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:841e713e98071d65ff0d432e660091f99132e94ae1bfd8b839de717df70c3a37
3
- size 32777243
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:166c64dec3ab43a15402d5347ab76f324596fcbc8d8a559186e937357bd5360b
3
+ size 20598075
rollouts-00010.jsonl.gz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:7a19c5a45aea488a1c4fa879dd7c3e0ca4b48892102e39fa9afd5203a4f5ba6d
3
- size 14080857
 
 
 
 
rollouts-00013.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6395ed5b28568239956b6745a73d2a117d2f6912fcc061c7e4e9e68f85c94e0a
3
- size 25571821
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8e6abe3c2230cd1f11898509a3925c2292f04d0686817c5de48109249915fbb9
3
+ size 24737489
rollouts-00016.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9cdcc213cc6c4d5a08cf193abf8f8e16ea6393ba29b9a42eeaed50d185ebc847
3
- size 29403130
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e24ebaddfa4875f45a4e8d47f47e76318b151ff7f2d156d0d65dbd76ae8f1580
3
+ size 28444758
rollouts-00020.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6837cce48023f4da7c89f1378587e4caf2062c1eca62a316b1587199f45038db
3
- size 33057184
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be264bfecf1ac4ae9eefce9f6666f2bbc1e05d8e0ab38a7737077935d7d7cfde
3
+ size 29967613
rollouts-00021.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:30f29ca062ae4b50da6dd8643e06c1655aae6818be147e963e2858b85073268e
3
- size 31761896
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:587d5989d8bf78916b4af74652cd9148ba9bd251da68d4d41e6a958a5bf0cf77
3
+ size 22220925
rollouts-00023.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:89011e472e31f09bcb164403bb2298616a6f60541bd363e065a27269815aa351
3
- size 30772134
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:047b4a8f6610a50597c07f7d248cd8ede95933313ebb41cc71a10a5a038b2741
3
+ size 22077201
rollouts-00029.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:80c4850401e825021154992fde791f077c3a0f7d74a251c017ec4f4056715bfb
3
- size 33047894
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74656724e640ca6539373bcb5c29558b72642d8166394c2a5b6a721c1e8cefb1
3
+ size 30633502
rollouts-00030.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ebc81005b112a82369abde5cd7eb868bad0f05cfa4eb0febe7a3bce2ba7b40fb
3
- size 30115285
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d23a76df7e7e3aa5783bbf1cd9888f450ac3d6fec6a31aac15b2ae023754d7da
3
+ size 12372147
rollouts-00031.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1b51f0c80cf681fc2df1ea6a22bb77ac738dd647c16e22e1e4236bae959b1c95
3
- size 32061474
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4bb495ebe730da8e675cb92b1dd29199ece1b8a88eb618afa75020ed66bfcc08
3
+ size 30328780
rollouts-00032.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3df7b76c4149c8ed3648ca5c0f111b14d545202a461dff8f5e6e7be4dd056f1b
3
- size 31510602
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:71e69931493ab2d4728ba7612069b9894b4cb76e9cd91481213564cb74d1a661
3
+ size 26346810
rollouts-00037.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3f086733ad3376da7ecf91bad470005ce5e397154a63f0a40b25c2d0e9d1964e
3
- size 26695639
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6a9db02da2e1abd52756869d508f238fdaaabdde342e50a1f4dce92e23af912d
3
+ size 21666524
rollouts-00038.jsonl.gz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:090b51a913c1800c3a7d127aae77f5bfb2c667d1b7d322ca13a162f1c1d227c6
3
- size 10288620
 
 
 
 
rollouts-00039.jsonl.gz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:7f6f48c74d63dc4a464fb3ca5ec3cfdfc5f41e0d8a53cf05a39b7b3139dace27
3
- size 31223703
 
 
 
 
rollouts-00043.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a6d941aa01f6d70274e05e38abced91381d9de94a9a185a9dd49d284e437de0f
3
- size 29386111
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e4eb49d3de514be98ffa2cf9a875ea50cb17cf971737185f7b2886b400823ad5
3
+ size 27292610
rollouts-00044.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:dfc321726f04d1989a1e1ae1f902ccd4748ff4e448d6675646b0faab0e4d4f42
3
- size 30893136
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6bdcfa91ebd0bca2949340a86bd49a6d291b764585a96e94b527fb0b3b633b1a
3
+ size 19235597
rollouts-00045.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:61d8d3712aa688f5a417d4d855bd653f3eb3c699d60cef1680f0fcd3974bae44
3
- size 32673932
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:83aa6ebfee86838998ddcdda542102470e9ea5bc9a3047db5245a2e49f88cf91
3
+ size 627657
rollouts-00046.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:26b375a73ae0f2e3162feb8fc27c9e7fa04c21dd023d0287152eadb808313da2
3
- size 27276072
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5ef166794c8ebe1a2329deea966315695b10e43cfcc8b6280cd46588ddd9385f
3
+ size 21772341
rollouts-00047.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1ef4ce4ef1f4005066c72b597703f83a650e7ff7ca65caf82f279598aabf5e40
3
- size 29276316
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1de8a49bc059153e7b61c847ccef2e1c4663c785c6b887d764c7d995cd6bc938
3
+ size 28583640
rollouts-00048.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ab4a3e69e575947c9e0e201399b335e0a4f6c7ed8aff140ba748cef9993f2720
3
- size 16029392
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:85b95cf72ccb7e9cc6b7d2070dcd6aac2e7953b81d5edaa0a388e50ce3376748
3
+ size 8499142
rollouts-00052.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b8d3ed8156366e1c531711c8e6731a430da909d92dc2170cacede287a769ace2
3
- size 17625713
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:76fd9bfcc35a152bf1d1ed72ddd666a88b1e47ff538d04478cfa77fbbc4b663c
3
+ size 3586558
rollouts-00053.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c7a1752fbe3c4c2ffa824a9ec664f6d73b0827f10add4e97acf4c005e7193ac5
3
- size 32984171
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d8c60701938c8462059780f91187b18604c62479bd6cc4c45a701c4b78241e28
3
+ size 14884732
rollouts-00055.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:246280a4621a739366547c2d9391bebfeddbab9293dacb29b2eec7a094af5d0f
3
- size 26210634
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fcbbea9992678d25025b34947825e3421bf83341b7ed2121d02807a8412b50fd
3
+ size 18357809
rollouts-00058.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d1a4223ed101f75673280cb4e5eb38f6c8f7db6c36f2f51f60cf85323547e17f
3
- size 22650360
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:86330653ef1e21e7c06737a17b25f1aa7ae1170ea5a2f1d32483977193d9e69c
3
+ size 21573043
rollouts-00063.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:97ff30cc5ed5cd1701bf2a52db9a1540e6a205015f3cbc1e1a79340f31fe1938
3
- size 22697741
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:78493cb345dd3fcaed50023641700d4deea7930765b187c7baa2d649a7a33eac
3
+ size 16094108
rollouts-00065.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7f587eac4afeb250a3f2ffec7a7ee9e4f1aaa809db2226b6f2173a4656b77fc3
3
- size 28629718
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:02f65bd006f6d40372e6ad619c1792987ab48ddfbd9fb17dc0027c23951cc262
3
+ size 23707029
rollouts-00070.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4af759e1ac857205869973fd1ff47011127cb089a96eef503ce6289612ca3366
3
- size 4729536
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4efd6f11dcd74b2dd332696cebe869e0f105fa7afe249dcac5be19f8dc04397d
3
+ size 2190074
rollouts-00073.jsonl.gz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:0da0727cd80d2655af236576e1f8956b741a2e8977d057be3d16cee4b17aa41e
3
- size 83220570
 
 
 
 
rollouts-00074.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:35739c8bad0b7e4a063ae377f19830a1b49260c83e4c0dd21c1d90a4bc197f1c
3
- size 33380722
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:05617d50c2ec8c3ce8d147034ff4ef7a4c2317862170dbc9fad09e79d88359a2
3
+ size 6995432
rollouts-00076.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3b7fa7f0cf1adfa02144b63d73d5a5d6f72da9a18bc741e0a949d306e69615c7
3
- size 30292272
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cd9651045003a0cf162693c0a829c26986d768faf0709c5897576e9cd6e216fe
3
+ size 17999575
rollouts-00078.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5fab4cfd70584056ac67e18a7cd48464f632b0b63f979485e25d3b0b59a84960
3
- size 12888604
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a929fb04fb1df6d3c966f7b1d72c52f7c94dfed15ce74e53b853bc40aa2b70a3
3
+ size 10312140
rollouts-00080.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5ac04603dbb91364835d629d968d0b6e104da8b32df97ad8afd4dacf6baf6e03
3
- size 31800112
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a2672f6a56bf3ecd8cd75882b3a5eb2a79f5202de076287e709d13b2458fe590
3
+ size 23760723
rollouts-00082.jsonl.gz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:49173464b770bffd9c08e1af2259dc2c9b706456bc37a15c281c284425fd42d2
3
- size 96164612
 
 
 
 
rollouts-00083.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a70243a5dc5d2c88e73424d3d3f4851f3277d3f7fe41a8b5e929b3a3ba96571e
3
- size 1509936
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f42b1a0a67bc0397ddc18e1dd6363027ba73163129d67c496e13619aee9ec258
3
+ size 1055563
rollouts-00085.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9f36b6725fb96ec6ba2282746c25eb0c4f4c91e1646514e2624543b18740b9ef
3
- size 15825921
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b825f5581e166675e5a8c205df9ca5afc1f33cb604be755b22fb354cd8c06ffc
3
+ size 888560
rollouts-00101.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5e39212275e57477a925821ba44f9a8f74b9aa574059ea14196f96956fd95075
3
- size 19098222
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cf3048512a23f8460f3e45d15ab30b23a09df842dab041b111c3d3c7eddb8363
3
+ size 17386737
rollouts-00102.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:08a642b9c52b0a0118c8f0358ec1b65ba7485f1e58bfc11de83e400a45c3739b
3
- size 30943829
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:07cddb73ac72471a4810384d8b1af5510f8deb12572822bc62843c6dd68734e4
3
+ size 28075897
rollouts-00105.jsonl.gz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:4036abb14c9bf48ca5f15338b7b79f2c088686a81574b350134d550face56c2d
3
- size 40572992
 
 
 
 
rollouts-00106.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:82c0f1134a689d4bf33571a35f231c3e7870be2f337b658bbb42ce838ae348f7
3
- size 25099367
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dd8f9e5bc4b9a49611519415e08c2159cb0af170dd5823fb8c5bcc7b67c10361
3
+ size 22590522
rollouts-00111.jsonl.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:84e882e1f658f8d86a604c89f6a17423b04df1884d8ef907bf6d4bdc596bdc7e
3
- size 24735508
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ef863da2de5cc83b7eb2838805b09784923fe05ef4ae707c620187526348a35
3
+ size 4380215