@openpond/evals 0.4.2 → 0.6.0

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 (64) hide show
  1. package/CONTRACT.md +27 -1
  2. package/README.md +31 -2
  3. package/conformance/telemetry/v1/invalid-batch.json +17 -0
  4. package/conformance/telemetry/v1/valid-batch.json +58 -0
  5. package/dist/artifact-verification.js +172 -0
  6. package/dist/builtin-benchmarks/harness-refiner.js +469 -24
  7. package/dist/execution-contracts.js +153 -0
  8. package/dist/execution-receipts.js +185 -0
  9. package/dist/graders.js +8 -2
  10. package/dist/index.js +10 -0
  11. package/dist/learned-preference.js +334 -0
  12. package/dist/preferences.js +803 -0
  13. package/dist/rollouts.js +241 -0
  14. package/dist/tasksets.js +24 -0
  15. package/dist/telemetry/index.js +4 -0
  16. package/dist/telemetry-analysis.js +183 -0
  17. package/dist/telemetry-bundle.js +60 -0
  18. package/dist/telemetry-catalog.js +50 -0
  19. package/dist/telemetry.js +112 -0
  20. package/dist/types/artifact-verification.d.ts +31 -0
  21. package/dist/types/artifact-verification.d.ts.map +1 -0
  22. package/dist/types/builtin-benchmarks/harness-refiner.d.ts +22 -0
  23. package/dist/types/builtin-benchmarks/harness-refiner.d.ts.map +1 -1
  24. package/dist/types/conformance.d.ts +44 -0
  25. package/dist/types/conformance.d.ts.map +1 -1
  26. package/dist/types/evidence/conformance.d.ts +24 -24
  27. package/dist/types/evidence/contracts.d.ts +42 -42
  28. package/dist/types/execution-contracts.d.ts +806 -0
  29. package/dist/types/execution-contracts.d.ts.map +1 -0
  30. package/dist/types/execution-receipts.d.ts +55 -0
  31. package/dist/types/execution-receipts.d.ts.map +1 -0
  32. package/dist/types/graders.d.ts.map +1 -1
  33. package/dist/types/index.d.ts +10 -0
  34. package/dist/types/index.d.ts.map +1 -1
  35. package/dist/types/learned-preference.d.ts +282 -0
  36. package/dist/types/learned-preference.d.ts.map +1 -0
  37. package/dist/types/preferences.d.ts +583 -0
  38. package/dist/types/preferences.d.ts.map +1 -0
  39. package/dist/types/review-conformance.d.ts +20 -15
  40. package/dist/types/review-conformance.d.ts.map +1 -1
  41. package/dist/types/rollouts.d.ts +221 -0
  42. package/dist/types/rollouts.d.ts.map +1 -0
  43. package/dist/types/tasksets.d.ts +89 -0
  44. package/dist/types/tasksets.d.ts.map +1 -1
  45. package/dist/types/telemetry/index.d.ts +5 -0
  46. package/dist/types/telemetry/index.d.ts.map +1 -0
  47. package/dist/types/telemetry-analysis.d.ts +116 -0
  48. package/dist/types/telemetry-analysis.d.ts.map +1 -0
  49. package/dist/types/telemetry-bundle.d.ts +309 -0
  50. package/dist/types/telemetry-bundle.d.ts.map +1 -0
  51. package/dist/types/telemetry-catalog.d.ts +269 -0
  52. package/dist/types/telemetry-catalog.d.ts.map +1 -0
  53. package/dist/types/telemetry.d.ts +266 -0
  54. package/dist/types/telemetry.d.ts.map +1 -0
  55. package/package.json +20 -2
  56. package/schemas/telemetry/v1/evidence-completeness.schema.json +82 -0
  57. package/schemas/telemetry/v1/evidence-reference.schema.json +41 -0
  58. package/schemas/telemetry/v1/metric-definition.schema.json +96 -0
  59. package/schemas/telemetry/v1/metric-observation.schema.json +182 -0
  60. package/schemas/telemetry/v1/run-metric-summary.schema.json +98 -0
  61. package/schemas/telemetry/v1/run-telemetry-batch.schema.json +405 -0
  62. package/schemas/telemetry/v1/run-telemetry-event.schema.json +201 -0
  63. package/schemas/telemetry/v1/telemetry-cohort.schema.json +100 -0
  64. package/schemas/telemetry/v1/telemetry-export-bundle.schema.json +655 -0
@@ -0,0 +1 @@
1
+ {"version":3,"file":"execution-contracts.d.ts","sourceRoot":"","sources":["../../../../src/execution-contracts.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAcxB,eAAO,MAAM,mBAAmB;;;EAAmC,CAAC;AACpE,eAAO,MAAM,kBAAkB;;;;;;;;;EAS7B,CAAC;AACH,eAAO,MAAM,yBAAyB;;;;;;;;;;;;EAYpC,CAAC;AAEH,eAAO,MAAM,+BAA+B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAcjC,CAAC;AACZ,eAAO,MAAM,wBAAwB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAE1B,CAAC;AAEZ,eAAO,MAAM,8BAA8B;;;;;EAKzC,CAAC;AACH,eAAO,MAAM,8BAA8B;;;;EAIzC,CAAC;AACH,eAAO,MAAM,2BAA2B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAa7B,CAAC;AACZ,eAAO,MAAM,6BAA6B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAO/B,CAAC;AACZ,eAAO,MAAM,sBAAsB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAExB,CAAC;AAEZ,eAAO,MAAM,+BAA+B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAYjC,CAAC;AACZ,eAAO,MAAM,wBAAwB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAE1B,CAAC;AAEZ,eAAO,MAAM,4BAA4B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAuBvC,CAAC;AAkCH,eAAO,MAAM,0BAA0B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAA6D,CAAC;AACrG,eAAO,MAAM,mBAAmB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAGK,CAAC;AAEtC,MAAM,MAAM,aAAa,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,mBAAmB,CAAC,CAAC;AAChE,MAAM,MAAM,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC;AAC9D,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,kBAAkB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,wBAAwB,CAAC,CAAC;AAC1E,MAAM,MAAM,qBAAqB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,2BAA2B,CAAC,CAAC;AAChF,MAAM,MAAM,gBAAgB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,sBAAsB,CAAC,CAAC;AACtE,MAAM,MAAM,kBAAkB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,wBAAwB,CAAC,CAAC;AAC1E,MAAM,MAAM,sBAAsB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,4BAA4B,CAAC,CAAC;AAClF,MAAM,MAAM,aAAa,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,mBAAmB,CAAC,CAAC"}
@@ -0,0 +1,55 @@
1
+ import type { FailureClass, ImmutableArtifactRef } from "@openpond/harness";
2
+ import { ArtifactManifestContentSchema, EnvironmentReleaseContentSchema, VerifierSetReleaseContentSchema, type ArtifactManifest, type AttemptOutcomeClass, type EnvironmentRelease, type FailureOwner, type RewardComponentReceipt, type RewardReceipt, type VerifierSetRelease } from "./execution-contracts.js";
3
+ import { type TasksetRelease } from "./tasksets.js";
4
+ export declare function createEnvironmentRelease(input: Parameters<typeof EnvironmentReleaseContentSchema.parse>[0]): EnvironmentRelease;
5
+ export declare function createArtifactManifest(input: Parameters<typeof ArtifactManifestContentSchema.parse>[0]): ArtifactManifest;
6
+ export declare function createVerifierSetRelease(input: Parameters<typeof VerifierSetReleaseContentSchema.parse>[0]): VerifierSetRelease;
7
+ export declare function bindTasksetExecutionReleases(input: {
8
+ taskset: TasksetRelease;
9
+ environment: EnvironmentRelease;
10
+ verifierSet: VerifierSetRelease;
11
+ }): TasksetRelease;
12
+ export declare function classifyAttemptOutcome(input: {
13
+ failureClass: FailureClass | null;
14
+ timeoutKind?: "task_deadline" | "infrastructure_timeout";
15
+ }): {
16
+ outcomeClass: AttemptOutcomeClass;
17
+ failureOwner: FailureOwner | null;
18
+ };
19
+ export declare function createRewardReceipt(input: {
20
+ id: string;
21
+ attemptRef: {
22
+ id: string;
23
+ contentHash: string;
24
+ };
25
+ verifierSet: VerifierSetRelease;
26
+ artifactManifest: ArtifactManifest;
27
+ outcomeClass: AttemptOutcomeClass;
28
+ failureOwner: FailureOwner | null;
29
+ components: RewardComponentReceipt[];
30
+ visibleEvidenceRefs?: ImmutableArtifactRef[];
31
+ privilegedEvidenceRefs?: ImmutableArtifactRef[];
32
+ supersedes?: {
33
+ id: string;
34
+ contentHash: string;
35
+ } | null;
36
+ createdAt: string;
37
+ metadata?: Record<string, unknown>;
38
+ }): RewardReceipt;
39
+ export declare function regradeRewardReceipt(input: {
40
+ original: RewardReceipt;
41
+ id: string;
42
+ verifierSet: VerifierSetRelease;
43
+ artifactManifest: ArtifactManifest;
44
+ components: RewardComponentReceipt[];
45
+ outcomeClass?: AttemptOutcomeClass;
46
+ failureOwner?: FailureOwner | null;
47
+ createdAt: string;
48
+ metadata?: Record<string, unknown>;
49
+ }): RewardReceipt;
50
+ export declare function verifyEnvironmentRelease(release: EnvironmentRelease): boolean;
51
+ export declare function verifyArtifactManifest(manifest: ArtifactManifest): boolean;
52
+ export declare function verifyVerifierSetRelease(release: VerifierSetRelease): boolean;
53
+ export declare function verifyRewardReceipt(receipt: RewardReceipt): boolean;
54
+ export declare function isScorableOutcome(outcome: AttemptOutcomeClass): boolean;
55
+ //# sourceMappingURL=execution-receipts.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"execution-receipts.d.ts","sourceRoot":"","sources":["../../../../src/execution-receipts.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,YAAY,EAAE,oBAAoB,EAAE,MAAM,mBAAmB,CAAC;AAE5E,OAAO,EACL,6BAA6B,EAE7B,+BAA+B,EAK/B,+BAA+B,EAE/B,KAAK,gBAAgB,EACrB,KAAK,mBAAmB,EACxB,KAAK,kBAAkB,EACvB,KAAK,YAAY,EACjB,KAAK,sBAAsB,EAC3B,KAAK,aAAa,EAClB,KAAK,kBAAkB,EACxB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAGL,KAAK,cAAc,EACpB,MAAM,eAAe,CAAC;AAEvB,wBAAgB,wBAAwB,CACtC,KAAK,EAAE,UAAU,CAAC,OAAO,+BAA+B,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,GACjE,kBAAkB,CAGpB;AAED,wBAAgB,sBAAsB,CACpC,KAAK,EAAE,UAAU,CAAC,OAAO,6BAA6B,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,GAC/D,gBAAgB,CAGlB;AAED,wBAAgB,wBAAwB,CACtC,KAAK,EAAE,UAAU,CAAC,OAAO,+BAA+B,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,GACjE,kBAAkB,CAGpB;AAED,wBAAgB,4BAA4B,CAAC,KAAK,EAAE;IAClD,OAAO,EAAE,cAAc,CAAC;IACxB,WAAW,EAAE,kBAAkB,CAAC;IAChC,WAAW,EAAE,kBAAkB,CAAC;CACjC,GAAG,cAAc,CAsBjB;AAED,wBAAgB,sBAAsB,CAAC,KAAK,EAAE;IAC5C,YAAY,EAAE,YAAY,GAAG,IAAI,CAAC;IAClC,WAAW,CAAC,EAAE,eAAe,GAAG,wBAAwB,CAAC;CAC1D,GAAG;IAAE,YAAY,EAAE,mBAAmB,CAAC;IAAC,YAAY,EAAE,YAAY,GAAG,IAAI,CAAA;CAAE,CAmB3E;AAED,wBAAgB,mBAAmB,CAAC,KAAK,EAAE;IACzC,EAAE,EAAE,MAAM,CAAC;IACX,UAAU,EAAE;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC;IAChD,WAAW,EAAE,kBAAkB,CAAC;IAChC,gBAAgB,EAAE,gBAAgB,CAAC;IACnC,YAAY,EAAE,mBAAmB,CAAC;IAClC,YAAY,EAAE,YAAY,GAAG,IAAI,CAAC;IAClC,UAAU,EAAE,sBAAsB,EAAE,CAAC;IACrC,mBAAmB,CAAC,EAAE,oBAAoB,EAAE,CAAC;IAC7C,sBAAsB,CAAC,EAAE,oBAAoB,EAAE,CAAC;IAChD,UAAU,CAAC,EAAE;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAC;IACxD,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC,GAAG,aAAa,CAoEhB;AAED,wBAAgB,oBAAoB,CAAC,KAAK,EAAE;IAC1C,QAAQ,EAAE,aAAa,CAAC;IACxB,EAAE,EAAE,MAAM,CAAC;IACX,WAAW,EAAE,kBAAkB,CAAC;IAChC,gBAAgB,EAAE,gBAAgB,CAAC;IACnC,UAAU,EAAE,sBAAsB,EAAE,CAAC;IACrC,YAAY,CAAC,EAAE,mBAAmB,CAAC;IACnC,YAAY,CAAC,EAAE,YAAY,GAAG,IAAI,CAAC;IACnC,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC,GAAG,aAAa,CAwBhB;AAED,wBAAgB,wBAAwB,CAAC,OAAO,EAAE,kBAAkB,GAAG,OAAO,CAE7E;AAED,wBAAgB,sBAAsB,CAAC,QAAQ,EAAE,gBAAgB,GAAG,OAAO,CAE1E;AAED,wBAAgB,wBAAwB,CAAC,OAAO,EAAE,kBAAkB,GAAG,OAAO,CAE7E;AAED,wBAAgB,mBAAmB,CAAC,OAAO,EAAE,aAAa,GAAG,OAAO,CAKnE;AAED,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,mBAAmB,GAAG,OAAO,CAKvE"}
@@ -1 +1 @@
1
- {"version":3,"file":"graders.d.ts","sourceRoot":"","sources":["../../../../src/graders.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAGxB,OAAO,KAAK,EAA2B,UAAU,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAErF,eAAO,MAAM,2BAA2B;;;;;;;;;;;;;;;;;;kBAW7B,CAAC;AACZ,eAAO,MAAM,oBAAoB;;;;;;;;;;;;;;;;;;;kBAAkF,CAAC;AAEpH,MAAM,MAAM,eAAe,GAAG;IAC5B,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAChC,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,mBAAmB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CACrC,CAAC;AACF,MAAM,MAAM,gBAAgB,GAAG,CAAC,KAAK,EAAE;IAAE,MAAM,EAAE,OAAO,CAAC,UAAU,EAAE;QAAE,IAAI,EAAE,aAAa,CAAA;KAAE,CAAC,CAAC;IAAC,IAAI,EAAE,UAAU,CAAC;IAAC,QAAQ,EAAE,eAAe,CAAA;CAAE,KAAK,OAAO,CAAC,IAAI,CAAC,cAAc,EAAE,eAAe,GAAG,UAAU,GAAG,eAAe,GAAG,aAAa,CAAC,CAAC,CAAC;AAC/O,MAAM,MAAM,oBAAoB,GAAG,CAAC,KAAK,EAAE;IAAE,MAAM,EAAE,OAAO,CAAC,UAAU,EAAE;QAAE,IAAI,EAAE,iBAAiB,CAAA;KAAE,CAAC,CAAC;IAAC,IAAI,EAAE,UAAU,CAAC;IAAC,QAAQ,EAAE,eAAe,CAAA;CAAE,KAAK,OAAO,CAAC,IAAI,CAAC,cAAc,EAAE,eAAe,GAAG,UAAU,GAAG,eAAe,GAAG,aAAa,CAAC,CAAC,CAAC;AAEvP,wBAAsB,aAAa,CAAC,KAAK,EAAE;IACzC,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE,eAAe,CAAC;IAC1B,OAAO,EAAE,UAAU,EAAE,CAAC;IACtB,UAAU,CAAC,EAAE,gBAAgB,CAAC;IAC9B,cAAc,CAAC,EAAE,oBAAoB,CAAC;CACvC,GAAG,OAAO,CAAC,cAAc,EAAE,CAAC,CAwB5B;AA4CD,MAAM,MAAM,cAAc,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,oBAAoB,CAAC,CAAC"}
1
+ {"version":3,"file":"graders.d.ts","sourceRoot":"","sources":["../../../../src/graders.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAGxB,OAAO,KAAK,EAA2B,UAAU,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAErF,eAAO,MAAM,2BAA2B;;;;;;;;;;;;;;;;;;kBAW7B,CAAC;AACZ,eAAO,MAAM,oBAAoB;;;;;;;;;;;;;;;;;;;kBAAkF,CAAC;AAEpH,MAAM,MAAM,eAAe,GAAG;IAC5B,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAChC,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,mBAAmB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CACrC,CAAC;AACF,MAAM,MAAM,gBAAgB,GAAG,CAAC,KAAK,EAAE;IAAE,MAAM,EAAE,OAAO,CAAC,UAAU,EAAE;QAAE,IAAI,EAAE,aAAa,CAAA;KAAE,CAAC,CAAC;IAAC,IAAI,EAAE,UAAU,CAAC;IAAC,QAAQ,EAAE,eAAe,CAAA;CAAE,KAAK,OAAO,CAAC,IAAI,CAAC,cAAc,EAAE,eAAe,GAAG,UAAU,GAAG,eAAe,GAAG,aAAa,CAAC,CAAC,CAAC;AAC/O,MAAM,MAAM,oBAAoB,GAAG,CAAC,KAAK,EAAE;IAAE,MAAM,EAAE,OAAO,CAAC,UAAU,EAAE;QAAE,IAAI,EAAE,iBAAiB,CAAA;KAAE,CAAC,CAAC;IAAC,IAAI,EAAE,UAAU,CAAC;IAAC,QAAQ,EAAE,eAAe,CAAA;CAAE,KAAK,OAAO,CAAC,IAAI,CAAC,cAAc,EAAE,eAAe,GAAG,UAAU,GAAG,eAAe,GAAG,aAAa,CAAC,CAAC,CAAC;AAEvP,wBAAsB,aAAa,CAAC,KAAK,EAAE;IACzC,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE,eAAe,CAAC;IAC1B,OAAO,EAAE,UAAU,EAAE,CAAC;IACtB,UAAU,CAAC,EAAE,gBAAgB,CAAC;IAC9B,cAAc,CAAC,EAAE,oBAAoB,CAAC;CACvC,GAAG,OAAO,CAAC,cAAc,EAAE,CAAC,CAwB5B;AAkDD,MAAM,MAAM,cAAc,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,oBAAoB,CAAC,CAAC"}
@@ -1,10 +1,20 @@
1
1
  export * from "./compatibility.js";
2
2
  export * from "./benchmarks.js";
3
+ export * from "./artifact-verification.js";
3
4
  export * from "./evidence/index.js";
5
+ export * from "./execution-contracts.js";
6
+ export * from "./execution-receipts.js";
4
7
  export * from "./graders.js";
5
8
  export * from "./harness.js";
9
+ export * from "./learned-preference.js";
6
10
  export * from "./runs.js";
7
11
  export * from "./model-improvement-qualification.js";
12
+ export * from "./preferences.js";
8
13
  export * from "./review-conformance.js";
14
+ export * from "./rollouts.js";
15
+ export * from "./telemetry.js";
16
+ export * from "./telemetry-catalog.js";
17
+ export * from "./telemetry-analysis.js";
18
+ export * from "./telemetry-bundle.js";
9
19
  export * from "./tasksets.js";
10
20
  //# sourceMappingURL=index.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,oBAAoB,CAAC;AACnC,cAAc,iBAAiB,CAAC;AAChC,cAAc,qBAAqB,CAAC;AACpC,cAAc,cAAc,CAAC;AAC7B,cAAc,cAAc,CAAC;AAC7B,cAAc,WAAW,CAAC;AAC1B,cAAc,sCAAsC,CAAC;AACrD,cAAc,yBAAyB,CAAC;AACxC,cAAc,eAAe,CAAC"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,oBAAoB,CAAC;AACnC,cAAc,iBAAiB,CAAC;AAChC,cAAc,4BAA4B,CAAC;AAC3C,cAAc,qBAAqB,CAAC;AACpC,cAAc,0BAA0B,CAAC;AACzC,cAAc,yBAAyB,CAAC;AACxC,cAAc,cAAc,CAAC;AAC7B,cAAc,cAAc,CAAC;AAC7B,cAAc,yBAAyB,CAAC;AACxC,cAAc,WAAW,CAAC;AAC1B,cAAc,sCAAsC,CAAC;AACrD,cAAc,kBAAkB,CAAC;AACjC,cAAc,yBAAyB,CAAC;AACxC,cAAc,eAAe,CAAC;AAC9B,cAAc,gBAAgB,CAAC;AAC/B,cAAc,wBAAwB,CAAC;AACvC,cAAc,yBAAyB,CAAC;AACxC,cAAc,uBAAuB,CAAC;AACtC,cAAc,eAAe,CAAC"}
@@ -0,0 +1,282 @@
1
+ import { z } from "zod";
2
+ import { type ComparisonAssignment, type PreferenceComparisonRelease, type PreferenceReceipt } from "./preferences.js";
3
+ import type { TasksetRelease } from "./tasksets.js";
4
+ declare const PreferenceDatasetPartitionSchema: z.ZodEnum<{
5
+ reward_train: "reward_train";
6
+ reward_validation: "reward_validation";
7
+ reward_qualification: "reward_qualification";
8
+ }>;
9
+ export declare const PreferenceEvidenceAuthoritySchema: z.ZodEnum<{
10
+ human: "human";
11
+ synthetic_fixture: "synthetic_fixture";
12
+ }>;
13
+ export declare const RewardModelQualificationKindSchema: z.ZodEnum<{
14
+ synthetic_smoke: "synthetic_smoke";
15
+ human_heldout: "human_heldout";
16
+ }>;
17
+ export declare const PreferenceDatasetReleaseContentSchema: z.ZodObject<{
18
+ schemaVersion: z.ZodLiteral<"openpond.preferenceDatasetRelease.v1">;
19
+ id: z.ZodString;
20
+ revision: z.ZodNumber;
21
+ tasksetRelease: z.ZodObject<{
22
+ id: z.ZodString;
23
+ contentHash: z.ZodString;
24
+ }, z.core.$strict>;
25
+ comparisonRelease: z.ZodObject<{
26
+ id: z.ZodString;
27
+ contentHash: z.ZodString;
28
+ }, z.core.$strict>;
29
+ authority: z.ZodEnum<{
30
+ human: "human";
31
+ synthetic_fixture: "synthetic_fixture";
32
+ }>;
33
+ qualificationEligibility: z.ZodEnum<{
34
+ smoke_only: "smoke_only";
35
+ human_heldout: "human_heldout";
36
+ }>;
37
+ fixtureRelease: z.ZodNullable<z.ZodObject<{
38
+ id: z.ZodString;
39
+ contentHash: z.ZodString;
40
+ }, z.core.$strict>>;
41
+ groups: z.ZodArray<z.ZodObject<{
42
+ id: z.ZodString;
43
+ assignmentRef: z.ZodObject<{
44
+ id: z.ZodString;
45
+ contentHash: z.ZodString;
46
+ }, z.core.$strict>;
47
+ scenarioRef: z.ZodObject<{
48
+ id: z.ZodString;
49
+ contentHash: z.ZodString;
50
+ }, z.core.$strict>;
51
+ scenarioSplit: z.ZodEnum<{
52
+ train: "train";
53
+ validation: "validation";
54
+ }>;
55
+ preferenceResultRef: z.ZodObject<{
56
+ id: z.ZodString;
57
+ contentHash: z.ZodString;
58
+ }, z.core.$strict>;
59
+ receiptRefs: z.ZodArray<z.ZodObject<{
60
+ id: z.ZodString;
61
+ contentHash: z.ZodString;
62
+ }, z.core.$strict>>;
63
+ attemptRefs: z.ZodArray<z.ZodObject<{
64
+ id: z.ZodString;
65
+ contentHash: z.ZodString;
66
+ }, z.core.$strict>>;
67
+ artifactManifestRefs: z.ZodArray<z.ZodObject<{
68
+ id: z.ZodString;
69
+ contentHash: z.ZodString;
70
+ }, z.core.$strict>>;
71
+ orderedBuckets: z.ZodArray<z.ZodArray<z.ZodString>>;
72
+ rejectAll: z.ZodBoolean;
73
+ partition: z.ZodEnum<{
74
+ reward_train: "reward_train";
75
+ reward_validation: "reward_validation";
76
+ reward_qualification: "reward_qualification";
77
+ }>;
78
+ metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
79
+ }, z.core.$strict>>;
80
+ derivedPairs: z.ZodArray<z.ZodObject<{
81
+ groupId: z.ZodString;
82
+ preferredAttemptRef: z.ZodObject<{
83
+ id: z.ZodString;
84
+ contentHash: z.ZodString;
85
+ }, z.core.$strict>;
86
+ dispreferredAttemptRef: z.ZodObject<{
87
+ id: z.ZodString;
88
+ contentHash: z.ZodString;
89
+ }, z.core.$strict>;
90
+ relation: z.ZodEnum<{
91
+ preferred: "preferred";
92
+ tie: "tie";
93
+ }>;
94
+ }, z.core.$strict>>;
95
+ createdAt: z.ZodString;
96
+ metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
97
+ }, z.core.$strict>;
98
+ export declare const PreferenceDatasetReleaseSchema: z.ZodObject<{
99
+ schemaVersion: z.ZodLiteral<"openpond.preferenceDatasetRelease.v1">;
100
+ id: z.ZodString;
101
+ revision: z.ZodNumber;
102
+ tasksetRelease: z.ZodObject<{
103
+ id: z.ZodString;
104
+ contentHash: z.ZodString;
105
+ }, z.core.$strict>;
106
+ comparisonRelease: z.ZodObject<{
107
+ id: z.ZodString;
108
+ contentHash: z.ZodString;
109
+ }, z.core.$strict>;
110
+ authority: z.ZodEnum<{
111
+ human: "human";
112
+ synthetic_fixture: "synthetic_fixture";
113
+ }>;
114
+ qualificationEligibility: z.ZodEnum<{
115
+ smoke_only: "smoke_only";
116
+ human_heldout: "human_heldout";
117
+ }>;
118
+ fixtureRelease: z.ZodNullable<z.ZodObject<{
119
+ id: z.ZodString;
120
+ contentHash: z.ZodString;
121
+ }, z.core.$strict>>;
122
+ groups: z.ZodArray<z.ZodObject<{
123
+ id: z.ZodString;
124
+ assignmentRef: z.ZodObject<{
125
+ id: z.ZodString;
126
+ contentHash: z.ZodString;
127
+ }, z.core.$strict>;
128
+ scenarioRef: z.ZodObject<{
129
+ id: z.ZodString;
130
+ contentHash: z.ZodString;
131
+ }, z.core.$strict>;
132
+ scenarioSplit: z.ZodEnum<{
133
+ train: "train";
134
+ validation: "validation";
135
+ }>;
136
+ preferenceResultRef: z.ZodObject<{
137
+ id: z.ZodString;
138
+ contentHash: z.ZodString;
139
+ }, z.core.$strict>;
140
+ receiptRefs: z.ZodArray<z.ZodObject<{
141
+ id: z.ZodString;
142
+ contentHash: z.ZodString;
143
+ }, z.core.$strict>>;
144
+ attemptRefs: z.ZodArray<z.ZodObject<{
145
+ id: z.ZodString;
146
+ contentHash: z.ZodString;
147
+ }, z.core.$strict>>;
148
+ artifactManifestRefs: z.ZodArray<z.ZodObject<{
149
+ id: z.ZodString;
150
+ contentHash: z.ZodString;
151
+ }, z.core.$strict>>;
152
+ orderedBuckets: z.ZodArray<z.ZodArray<z.ZodString>>;
153
+ rejectAll: z.ZodBoolean;
154
+ partition: z.ZodEnum<{
155
+ reward_train: "reward_train";
156
+ reward_validation: "reward_validation";
157
+ reward_qualification: "reward_qualification";
158
+ }>;
159
+ metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
160
+ }, z.core.$strict>>;
161
+ derivedPairs: z.ZodArray<z.ZodObject<{
162
+ groupId: z.ZodString;
163
+ preferredAttemptRef: z.ZodObject<{
164
+ id: z.ZodString;
165
+ contentHash: z.ZodString;
166
+ }, z.core.$strict>;
167
+ dispreferredAttemptRef: z.ZodObject<{
168
+ id: z.ZodString;
169
+ contentHash: z.ZodString;
170
+ }, z.core.$strict>;
171
+ relation: z.ZodEnum<{
172
+ preferred: "preferred";
173
+ tie: "tie";
174
+ }>;
175
+ }, z.core.$strict>>;
176
+ createdAt: z.ZodString;
177
+ metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
178
+ contentHash: z.ZodString;
179
+ }, z.core.$strict>;
180
+ export declare const RewardModelQualificationReportContentSchema: z.ZodObject<{
181
+ schemaVersion: z.ZodLiteral<"openpond.rewardModelQualificationReport.v1">;
182
+ id: z.ZodString;
183
+ kind: z.ZodEnum<{
184
+ synthetic_smoke: "synthetic_smoke";
185
+ human_heldout: "human_heldout";
186
+ }>;
187
+ rewardModelVersion: z.ZodObject<{
188
+ id: z.ZodString;
189
+ contentHash: z.ZodString;
190
+ }, z.core.$strict>;
191
+ preferenceDatasetRelease: z.ZodObject<{
192
+ id: z.ZodString;
193
+ contentHash: z.ZodString;
194
+ }, z.core.$strict>;
195
+ tasksetRelease: z.ZodObject<{
196
+ id: z.ZodString;
197
+ contentHash: z.ZodString;
198
+ }, z.core.$strict>;
199
+ processorRelease: z.ZodObject<{
200
+ id: z.ZodString;
201
+ contentHash: z.ZodString;
202
+ }, z.core.$strict>;
203
+ metrics: z.ZodObject<{
204
+ sampleCount: z.ZodNumber;
205
+ finiteScoreRate: z.ZodNumber;
206
+ scoreVariance: z.ZodNumber;
207
+ checkpointReloadPassed: z.ZodBoolean;
208
+ processorCompatibilityPassed: z.ZodBoolean;
209
+ invalidAttemptExclusionPassed: z.ZodBoolean;
210
+ orderedPairAccuracy: z.ZodNullable<z.ZodNumber>;
211
+ bucketAccuracy: z.ZodNullable<z.ZodNumber>;
212
+ tieAgreement: z.ZodNullable<z.ZodNumber>;
213
+ }, z.core.$strict>;
214
+ passed: z.ZodBoolean;
215
+ productionRewardEligible: z.ZodBoolean;
216
+ createdAt: z.ZodString;
217
+ metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
218
+ }, z.core.$strict>;
219
+ export declare const RewardModelQualificationReportSchema: z.ZodObject<{
220
+ schemaVersion: z.ZodLiteral<"openpond.rewardModelQualificationReport.v1">;
221
+ id: z.ZodString;
222
+ kind: z.ZodEnum<{
223
+ synthetic_smoke: "synthetic_smoke";
224
+ human_heldout: "human_heldout";
225
+ }>;
226
+ rewardModelVersion: z.ZodObject<{
227
+ id: z.ZodString;
228
+ contentHash: z.ZodString;
229
+ }, z.core.$strict>;
230
+ preferenceDatasetRelease: z.ZodObject<{
231
+ id: z.ZodString;
232
+ contentHash: z.ZodString;
233
+ }, z.core.$strict>;
234
+ tasksetRelease: z.ZodObject<{
235
+ id: z.ZodString;
236
+ contentHash: z.ZodString;
237
+ }, z.core.$strict>;
238
+ processorRelease: z.ZodObject<{
239
+ id: z.ZodString;
240
+ contentHash: z.ZodString;
241
+ }, z.core.$strict>;
242
+ metrics: z.ZodObject<{
243
+ sampleCount: z.ZodNumber;
244
+ finiteScoreRate: z.ZodNumber;
245
+ scoreVariance: z.ZodNumber;
246
+ checkpointReloadPassed: z.ZodBoolean;
247
+ processorCompatibilityPassed: z.ZodBoolean;
248
+ invalidAttemptExclusionPassed: z.ZodBoolean;
249
+ orderedPairAccuracy: z.ZodNullable<z.ZodNumber>;
250
+ bucketAccuracy: z.ZodNullable<z.ZodNumber>;
251
+ tieAgreement: z.ZodNullable<z.ZodNumber>;
252
+ }, z.core.$strict>;
253
+ passed: z.ZodBoolean;
254
+ productionRewardEligible: z.ZodBoolean;
255
+ createdAt: z.ZodString;
256
+ metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
257
+ contentHash: z.ZodString;
258
+ }, z.core.$strict>;
259
+ export declare function createPreferenceDatasetRelease(input: z.input<typeof PreferenceDatasetReleaseContentSchema>): PreferenceDatasetRelease;
260
+ export declare function materializePreferenceDatasetRelease(input: {
261
+ id: string;
262
+ revision: number;
263
+ tasksetRelease: TasksetRelease;
264
+ comparisonRelease: PreferenceComparisonRelease;
265
+ authority: "human" | "synthetic_fixture";
266
+ groups: ReadonlyArray<{
267
+ assignment: ComparisonAssignment;
268
+ receipt: PreferenceReceipt;
269
+ partition: z.infer<typeof PreferenceDatasetPartitionSchema>;
270
+ }>;
271
+ createdAt: string;
272
+ metadata?: Record<string, unknown>;
273
+ }): PreferenceDatasetRelease;
274
+ export declare function verifyPreferenceDatasetRelease(value: unknown): value is PreferenceDatasetRelease;
275
+ export declare function createRewardModelQualificationReport(input: z.input<typeof RewardModelQualificationReportContentSchema>): RewardModelQualificationReport;
276
+ export declare function verifyRewardModelQualificationReport(value: unknown): value is RewardModelQualificationReport;
277
+ export type PreferenceEvidenceAuthority = z.infer<typeof PreferenceEvidenceAuthoritySchema>;
278
+ export type PreferenceDatasetRelease = z.infer<typeof PreferenceDatasetReleaseSchema>;
279
+ export type RewardModelQualificationKind = z.infer<typeof RewardModelQualificationKindSchema>;
280
+ export type RewardModelQualificationReport = z.infer<typeof RewardModelQualificationReportSchema>;
281
+ export {};
282
+ //# sourceMappingURL=learned-preference.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"learned-preference.d.ts","sourceRoot":"","sources":["../../../../src/learned-preference.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAUxB,OAAO,EAGL,KAAK,oBAAoB,EACzB,KAAK,2BAA2B,EAChC,KAAK,iBAAiB,EACvB,MAAM,kBAAkB,CAAC;AAC1B,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,eAAe,CAAC;AAEpD,QAAA,MAAM,gCAAgC;;;;EAIpC,CAAC;AAEH,eAAO,MAAM,iCAAiC;;;EAG5C,CAAC;AAEH,eAAO,MAAM,kCAAkC;;;EAG7C,CAAC;AAwEH,eAAO,MAAM,qCAAqC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAuEhD,CAAC;AAEH,eAAO,MAAM,8BAA8B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAEhC,CAAC;AAcZ,eAAO,MAAM,2CAA2C;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBA+BtD,CAAC;AAEH,eAAO,MAAM,oCAAoC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAGpC,CAAC;AAEd,wBAAgB,8BAA8B,CAC5C,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,qCAAqC,CAAC,GAC3D,wBAAwB,CAM1B;AAED,wBAAgB,mCAAmC,CAAC,KAAK,EAAE;IACzD,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,cAAc,EAAE,cAAc,CAAC;IAC/B,iBAAiB,EAAE,2BAA2B,CAAC;IAC/C,SAAS,EAAE,OAAO,GAAG,mBAAmB,CAAC;IACzC,MAAM,EAAE,aAAa,CAAC;QACpB,UAAU,EAAE,oBAAoB,CAAC;QACjC,OAAO,EAAE,iBAAiB,CAAC;QAC3B,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,gCAAgC,CAAC,CAAC;KAC7D,CAAC,CAAC;IACH,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC,GAAG,wBAAwB,CA4E3B;AAED,wBAAgB,8BAA8B,CAC5C,KAAK,EAAE,OAAO,GACb,KAAK,IAAI,wBAAwB,CAMnC;AAED,wBAAgB,oCAAoC,CAClD,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,2CAA2C,CAAC,GACjE,8BAA8B,CAMhC;AAED,wBAAgB,oCAAoC,CAClD,KAAK,EAAE,OAAO,GACb,KAAK,IAAI,8BAA8B,CAMzC;AAgED,MAAM,MAAM,2BAA2B,GAAG,CAAC,CAAC,KAAK,CAC/C,OAAO,iCAAiC,CACzC,CAAC;AACF,MAAM,MAAM,wBAAwB,GAAG,CAAC,CAAC,KAAK,CAC5C,OAAO,8BAA8B,CACtC,CAAC;AACF,MAAM,MAAM,4BAA4B,GAAG,CAAC,CAAC,KAAK,CAChD,OAAO,kCAAkC,CAC1C,CAAC;AACF,MAAM,MAAM,8BAA8B,GAAG,CAAC,CAAC,KAAK,CAClD,OAAO,oCAAoC,CAC5C,CAAC"}