@openpond/evals 0.11.18 → 0.11.20

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (140) hide show
  1. package/dist/connected-evidence/contracts.js +15 -5
  2. package/dist/connected-evidence/imports.js +78 -17
  3. package/dist/connected-evidence/normalize.js +18 -7
  4. package/dist/connected-evidence/sources/claude-branches.js +52 -0
  5. package/dist/connected-evidence/sources/claude-code.js +7 -29
  6. package/dist/connected-evidence/sources/codex-authorization.js +40 -0
  7. package/dist/connected-evidence/sources/codex.js +7 -3
  8. package/dist/connected-evidence/sources/grok-build.js +137 -0
  9. package/dist/connected-evidence/sources/opencode.js +101 -0
  10. package/dist/connected-evidence/sources/pi.js +187 -0
  11. package/dist/graders.js +11 -1
  12. package/dist/learning/budgeted-reward-fixtures.js +1 -0
  13. package/dist/learning/contracts.js +10 -0
  14. package/dist/learning/index.js +1 -0
  15. package/dist/learning/model-judge.js +22 -5
  16. package/dist/learning/reward-calibration-closure.js +110 -0
  17. package/dist/learning/reward-calibration.js +20 -10
  18. package/dist/learning/reward-checks.js +1 -0
  19. package/dist/native-conversations/appimage-runtime.js +131 -0
  20. package/dist/native-conversations/collector-branches.js +79 -0
  21. package/dist/native-conversations/collector-contracts.js +9 -0
  22. package/dist/native-conversations/collector-destinations.js +27 -0
  23. package/dist/native-conversations/collector-errors.js +42 -0
  24. package/dist/native-conversations/collector-progress.js +114 -0
  25. package/dist/native-conversations/collector-store.js +251 -0
  26. package/dist/native-conversations/collector-sync-requests.js +52 -0
  27. package/dist/native-conversations/collector.js +329 -0
  28. package/dist/native-conversations/contracts.js +11 -0
  29. package/dist/native-conversations/database.js +149 -0
  30. package/dist/native-conversations/discovery.js +135 -0
  31. package/dist/native-conversations/history.js +327 -0
  32. package/dist/native-conversations/index.js +11 -0
  33. package/dist/native-conversations/location.js +29 -0
  34. package/dist/native-conversations/openclaw-cold-archive.js +91 -0
  35. package/dist/native-conversations/openclaw-database.js +130 -0
  36. package/dist/native-conversations/readiness.js +4 -0
  37. package/dist/native-conversations/source-locations.js +112 -0
  38. package/dist/native-conversations/supervisor.js +208 -0
  39. package/dist/native-conversations/titles.js +38 -0
  40. package/dist/rewards.js +1 -1
  41. package/dist/types/connected-evidence/contracts.d.ts +15 -0
  42. package/dist/types/connected-evidence/contracts.d.ts.map +1 -1
  43. package/dist/types/connected-evidence/imports.d.ts +7 -0
  44. package/dist/types/connected-evidence/imports.d.ts.map +1 -1
  45. package/dist/types/connected-evidence/normalize.d.ts +7 -0
  46. package/dist/types/connected-evidence/normalize.d.ts.map +1 -1
  47. package/dist/types/connected-evidence/sources/claude-branches.d.ts +19 -0
  48. package/dist/types/connected-evidence/sources/claude-branches.d.ts.map +1 -0
  49. package/dist/types/connected-evidence/sources/claude-code.d.ts +5 -1
  50. package/dist/types/connected-evidence/sources/claude-code.d.ts.map +1 -1
  51. package/dist/types/connected-evidence/sources/codex-authorization.d.ts +18 -0
  52. package/dist/types/connected-evidence/sources/codex-authorization.d.ts.map +1 -0
  53. package/dist/types/connected-evidence/sources/codex.d.ts +5 -1
  54. package/dist/types/connected-evidence/sources/codex.d.ts.map +1 -1
  55. package/dist/types/connected-evidence/sources/grok-build.d.ts +79 -0
  56. package/dist/types/connected-evidence/sources/grok-build.d.ts.map +1 -0
  57. package/dist/types/connected-evidence/sources/history.d.ts +10 -2
  58. package/dist/types/connected-evidence/sources/history.d.ts.map +1 -1
  59. package/dist/types/connected-evidence/sources/opencode.d.ts +78 -0
  60. package/dist/types/connected-evidence/sources/opencode.d.ts.map +1 -0
  61. package/dist/types/connected-evidence/sources/pi.d.ts +79 -0
  62. package/dist/types/connected-evidence/sources/pi.d.ts.map +1 -0
  63. package/dist/types/graders.d.ts +81 -0
  64. package/dist/types/graders.d.ts.map +1 -1
  65. package/dist/types/learning/budgeted-reward-fixtures.d.ts +2 -1
  66. package/dist/types/learning/budgeted-reward-fixtures.d.ts.map +1 -1
  67. package/dist/types/learning/contracts.d.ts +167 -0
  68. package/dist/types/learning/contracts.d.ts.map +1 -1
  69. package/dist/types/learning/index.d.ts +1 -0
  70. package/dist/types/learning/index.d.ts.map +1 -1
  71. package/dist/types/learning/judge-budget.d.ts +3 -0
  72. package/dist/types/learning/judge-budget.d.ts.map +1 -1
  73. package/dist/types/learning/model-judge.d.ts +6 -1
  74. package/dist/types/learning/model-judge.d.ts.map +1 -1
  75. package/dist/types/learning/operations.d.ts +44 -0
  76. package/dist/types/learning/operations.d.ts.map +1 -1
  77. package/dist/types/learning/repository.d.ts +81 -0
  78. package/dist/types/learning/repository.d.ts.map +1 -1
  79. package/dist/types/learning/reward-calibration-closure.d.ts +2323 -0
  80. package/dist/types/learning/reward-calibration-closure.d.ts.map +1 -0
  81. package/dist/types/learning/reward-calibration.d.ts +1 -1
  82. package/dist/types/learning/reward-calibration.d.ts.map +1 -1
  83. package/dist/types/learning/reward-checks.d.ts +35 -0
  84. package/dist/types/learning/reward-checks.d.ts.map +1 -1
  85. package/dist/types/learning/transport.d.ts +103 -0
  86. package/dist/types/learning/transport.d.ts.map +1 -1
  87. package/dist/types/native-conversations/appimage-runtime.d.ts +12 -0
  88. package/dist/types/native-conversations/appimage-runtime.d.ts.map +1 -0
  89. package/dist/types/native-conversations/collector-branches.d.ts +22 -0
  90. package/dist/types/native-conversations/collector-branches.d.ts.map +1 -0
  91. package/dist/types/native-conversations/collector-contracts.d.ts +92 -0
  92. package/dist/types/native-conversations/collector-contracts.d.ts.map +1 -0
  93. package/dist/types/native-conversations/collector-destinations.d.ts +18 -0
  94. package/dist/types/native-conversations/collector-destinations.d.ts.map +1 -0
  95. package/dist/types/native-conversations/collector-errors.d.ts +13 -0
  96. package/dist/types/native-conversations/collector-errors.d.ts.map +1 -0
  97. package/dist/types/native-conversations/collector-progress.d.ts +24 -0
  98. package/dist/types/native-conversations/collector-progress.d.ts.map +1 -0
  99. package/dist/types/native-conversations/collector-store.d.ts +31 -0
  100. package/dist/types/native-conversations/collector-store.d.ts.map +1 -0
  101. package/dist/types/native-conversations/collector-sync-requests.d.ts +25 -0
  102. package/dist/types/native-conversations/collector-sync-requests.d.ts.map +1 -0
  103. package/dist/types/native-conversations/collector.d.ts +8 -0
  104. package/dist/types/native-conversations/collector.d.ts.map +1 -0
  105. package/dist/types/native-conversations/contracts.d.ts +47 -0
  106. package/dist/types/native-conversations/contracts.d.ts.map +1 -0
  107. package/dist/types/native-conversations/database.d.ts +14 -0
  108. package/dist/types/native-conversations/database.d.ts.map +1 -0
  109. package/dist/types/native-conversations/discovery.d.ts +9 -0
  110. package/dist/types/native-conversations/discovery.d.ts.map +1 -0
  111. package/dist/types/native-conversations/history.d.ts +23 -0
  112. package/dist/types/native-conversations/history.d.ts.map +1 -0
  113. package/dist/types/native-conversations/index.d.ts +12 -0
  114. package/dist/types/native-conversations/index.d.ts.map +1 -0
  115. package/dist/types/native-conversations/location.d.ts +3 -0
  116. package/dist/types/native-conversations/location.d.ts.map +1 -0
  117. package/dist/types/native-conversations/openclaw-cold-archive.d.ts +8 -0
  118. package/dist/types/native-conversations/openclaw-cold-archive.d.ts.map +1 -0
  119. package/dist/types/native-conversations/openclaw-database.d.ts +14 -0
  120. package/dist/types/native-conversations/openclaw-database.d.ts.map +1 -0
  121. package/dist/types/native-conversations/readiness.d.ts +5 -0
  122. package/dist/types/native-conversations/readiness.d.ts.map +1 -0
  123. package/dist/types/native-conversations/source-locations.d.ts +8 -0
  124. package/dist/types/native-conversations/source-locations.d.ts.map +1 -0
  125. package/dist/types/native-conversations/supervisor.d.ts +21 -0
  126. package/dist/types/native-conversations/supervisor.d.ts.map +1 -0
  127. package/dist/types/native-conversations/titles.d.ts +4 -0
  128. package/dist/types/native-conversations/titles.d.ts.map +1 -0
  129. package/dist/types/rewards.d.ts +50 -0
  130. package/dist/types/rewards.d.ts.map +1 -1
  131. package/dist/types/telemetry-analysis.d.ts +1 -1
  132. package/dist/types/telemetry-bundle.d.ts +2 -2
  133. package/dist/types/telemetry-catalog.d.ts +21 -21
  134. package/dist/types/telemetry.d.ts +2 -2
  135. package/package.json +11 -3
  136. package/schemas/learning/v1/command-request.schema.json +122 -91
  137. package/schemas/learning/v1/decision.schema.json +63 -0
  138. package/schemas/learning/v1/grade.schema.json +91 -0
  139. package/schemas/learning/v1/reward_check.schema.json +91 -0
  140. package/schemas/learning/v1/source.schema.json +28 -0
@@ -572,6 +572,69 @@
572
572
  "type": "null"
573
573
  }
574
574
  ]
575
+ },
576
+ "transportPolicy": {
577
+ "type": "object",
578
+ "properties": {
579
+ "schemaVersion": {
580
+ "type": "string",
581
+ "const": "openpond.modelJudgeTransportPolicy.v1"
582
+ },
583
+ "responseFormat": {
584
+ "type": "string",
585
+ "const": "judgment_json_schema"
586
+ },
587
+ "reasoningEffort": {
588
+ "type": "string",
589
+ "enum": [
590
+ "off",
591
+ "low",
592
+ "medium",
593
+ "high",
594
+ "xhigh",
595
+ "max"
596
+ ]
597
+ },
598
+ "maximumOutputTokens": {
599
+ "type": "integer",
600
+ "exclusiveMinimum": 0,
601
+ "maximum": 1000000
602
+ }
603
+ },
604
+ "required": [
605
+ "schemaVersion",
606
+ "responseFormat",
607
+ "reasoningEffort",
608
+ "maximumOutputTokens"
609
+ ],
610
+ "additionalProperties": false
611
+ },
612
+ "effectiveRequestHash": {
613
+ "$ref": "#/$defs/__schema2"
614
+ },
615
+ "finishReason": {
616
+ "anyOf": [
617
+ {
618
+ "type": "string",
619
+ "minLength": 1,
620
+ "maxLength": 200
621
+ },
622
+ {
623
+ "type": "null"
624
+ }
625
+ ]
626
+ },
627
+ "reasoningTokens": {
628
+ "anyOf": [
629
+ {
630
+ "type": "integer",
631
+ "minimum": 0,
632
+ "maximum": 9007199254740991
633
+ },
634
+ {
635
+ "type": "null"
636
+ }
637
+ ]
575
638
  }
576
639
  },
577
640
  "required": [
@@ -549,6 +549,69 @@
549
549
  "type": "null"
550
550
  }
551
551
  ]
552
+ },
553
+ "transportPolicy": {
554
+ "type": "object",
555
+ "properties": {
556
+ "schemaVersion": {
557
+ "type": "string",
558
+ "const": "openpond.modelJudgeTransportPolicy.v1"
559
+ },
560
+ "responseFormat": {
561
+ "type": "string",
562
+ "const": "judgment_json_schema"
563
+ },
564
+ "reasoningEffort": {
565
+ "type": "string",
566
+ "enum": [
567
+ "off",
568
+ "low",
569
+ "medium",
570
+ "high",
571
+ "xhigh",
572
+ "max"
573
+ ]
574
+ },
575
+ "maximumOutputTokens": {
576
+ "type": "integer",
577
+ "exclusiveMinimum": 0,
578
+ "maximum": 1000000
579
+ }
580
+ },
581
+ "required": [
582
+ "schemaVersion",
583
+ "responseFormat",
584
+ "reasoningEffort",
585
+ "maximumOutputTokens"
586
+ ],
587
+ "additionalProperties": false
588
+ },
589
+ "effectiveRequestHash": {
590
+ "$ref": "#/$defs/__schema2"
591
+ },
592
+ "finishReason": {
593
+ "anyOf": [
594
+ {
595
+ "type": "string",
596
+ "minLength": 1,
597
+ "maxLength": 200
598
+ },
599
+ {
600
+ "type": "null"
601
+ }
602
+ ]
603
+ },
604
+ "reasoningTokens": {
605
+ "anyOf": [
606
+ {
607
+ "type": "integer",
608
+ "minimum": 0,
609
+ "maximum": 9007199254740991
610
+ },
611
+ {
612
+ "type": "null"
613
+ }
614
+ ]
552
615
  }
553
616
  },
554
617
  "required": [
@@ -730,6 +793,34 @@
730
793
  "type": "null"
731
794
  }
732
795
  ]
796
+ },
797
+ "effectiveRequestHash": {
798
+ "type": "string",
799
+ "pattern": "^[a-f0-9]{64}$"
800
+ },
801
+ "finishReason": {
802
+ "anyOf": [
803
+ {
804
+ "type": "string",
805
+ "minLength": 1,
806
+ "maxLength": 200
807
+ },
808
+ {
809
+ "type": "null"
810
+ }
811
+ ]
812
+ },
813
+ "reasoningTokens": {
814
+ "anyOf": [
815
+ {
816
+ "type": "integer",
817
+ "minimum": 0,
818
+ "maximum": 9007199254740991
819
+ },
820
+ {
821
+ "type": "null"
822
+ }
823
+ ]
733
824
  }
734
825
  },
735
826
  "required": [
@@ -468,6 +468,69 @@
468
468
  "type": "null"
469
469
  }
470
470
  ]
471
+ },
472
+ "transportPolicy": {
473
+ "type": "object",
474
+ "properties": {
475
+ "schemaVersion": {
476
+ "type": "string",
477
+ "const": "openpond.modelJudgeTransportPolicy.v1"
478
+ },
479
+ "responseFormat": {
480
+ "type": "string",
481
+ "const": "judgment_json_schema"
482
+ },
483
+ "reasoningEffort": {
484
+ "type": "string",
485
+ "enum": [
486
+ "off",
487
+ "low",
488
+ "medium",
489
+ "high",
490
+ "xhigh",
491
+ "max"
492
+ ]
493
+ },
494
+ "maximumOutputTokens": {
495
+ "type": "integer",
496
+ "exclusiveMinimum": 0,
497
+ "maximum": 1000000
498
+ }
499
+ },
500
+ "required": [
501
+ "schemaVersion",
502
+ "responseFormat",
503
+ "reasoningEffort",
504
+ "maximumOutputTokens"
505
+ ],
506
+ "additionalProperties": false
507
+ },
508
+ "effectiveRequestHash": {
509
+ "$ref": "#/$defs/__schema2"
510
+ },
511
+ "finishReason": {
512
+ "anyOf": [
513
+ {
514
+ "type": "string",
515
+ "minLength": 1,
516
+ "maxLength": 200
517
+ },
518
+ {
519
+ "type": "null"
520
+ }
521
+ ]
522
+ },
523
+ "reasoningTokens": {
524
+ "anyOf": [
525
+ {
526
+ "type": "integer",
527
+ "minimum": 0,
528
+ "maximum": 9007199254740991
529
+ },
530
+ {
531
+ "type": "null"
532
+ }
533
+ ]
471
534
  }
472
535
  },
473
536
  "required": [
@@ -655,6 +718,34 @@
655
718
  "type": "null"
656
719
  }
657
720
  ]
721
+ },
722
+ "effectiveRequestHash": {
723
+ "type": "string",
724
+ "pattern": "^[a-f0-9]{64}$"
725
+ },
726
+ "finishReason": {
727
+ "anyOf": [
728
+ {
729
+ "type": "string",
730
+ "minLength": 1,
731
+ "maxLength": 200
732
+ },
733
+ {
734
+ "type": "null"
735
+ }
736
+ ]
737
+ },
738
+ "reasoningTokens": {
739
+ "anyOf": [
740
+ {
741
+ "type": "integer",
742
+ "minimum": 0,
743
+ "maximum": 9007199254740991
744
+ },
745
+ {
746
+ "type": "null"
747
+ }
748
+ ]
658
749
  }
659
750
  },
660
751
  "required": [
@@ -160,6 +160,34 @@
160
160
  ],
161
161
  "additionalProperties": false
162
162
  },
163
+ "capturedOrigin": {
164
+ "type": "object",
165
+ "properties": {
166
+ "receiptId": {
167
+ "$ref": "#/$defs/__schema0"
168
+ },
169
+ "ownerUserId": {
170
+ "$ref": "#/$defs/__schema0"
171
+ },
172
+ "modelId": {
173
+ "$ref": "#/$defs/__schema0"
174
+ },
175
+ "packageHash": {
176
+ "$ref": "#/$defs/__schema2"
177
+ },
178
+ "taskset": {
179
+ "$ref": "#/$defs/__schema1"
180
+ }
181
+ },
182
+ "required": [
183
+ "receiptId",
184
+ "ownerUserId",
185
+ "modelId",
186
+ "packageHash",
187
+ "taskset"
188
+ ],
189
+ "additionalProperties": false
190
+ },
163
191
  "contentHash": {
164
192
  "$ref": "#/$defs/__schema2"
165
193
  }