@openpond/evals 0.3.0 → 0.4.0

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 (47) hide show
  1. package/CONTRACT.md +11 -3
  2. package/README.md +27 -8
  3. package/RELEASING.md +13 -46
  4. package/dist/compatibility.js +2 -2
  5. package/dist/conformance.js +1 -2
  6. package/dist/evidence/authoring.js +1 -1
  7. package/dist/evidence/conformance.js +1 -1
  8. package/dist/evidence/contracts.js +1 -2
  9. package/dist/evidence/eligibility.js +1 -1
  10. package/dist/graders.js +1 -1
  11. package/dist/harness.js +33 -101
  12. package/dist/index.js +2 -3
  13. package/dist/model-improvement-qualification.js +125 -0
  14. package/dist/review-conformance.js +209 -0
  15. package/dist/runs.js +1 -9
  16. package/dist/tasksets.js +2 -15
  17. package/dist/types/compatibility.d.ts +1 -1
  18. package/dist/types/compatibility.d.ts.map +1 -1
  19. package/dist/types/conformance.d.ts +2 -0
  20. package/dist/types/conformance.d.ts.map +1 -1
  21. package/dist/types/evidence/conformance.d.ts +20 -20
  22. package/dist/types/evidence/contracts.d.ts +18 -18
  23. package/dist/types/harness.d.ts +3 -409
  24. package/dist/types/harness.d.ts.map +1 -1
  25. package/dist/types/index.d.ts +2 -3
  26. package/dist/types/index.d.ts.map +1 -1
  27. package/dist/types/model-improvement-qualification.d.ts +222 -0
  28. package/dist/types/model-improvement-qualification.d.ts.map +1 -0
  29. package/dist/types/review-conformance.d.ts +607 -0
  30. package/dist/types/review-conformance.d.ts.map +1 -0
  31. package/dist/types/runs.d.ts +1 -9
  32. package/dist/types/runs.d.ts.map +1 -1
  33. package/dist/types/tasksets.d.ts +2 -25
  34. package/dist/types/tasksets.d.ts.map +1 -1
  35. package/package.json +14 -10
  36. package/dist/common.js +0 -67
  37. package/dist/harness-improvements.js +0 -329
  38. package/dist/harness-workspaces.js +0 -368
  39. package/dist/sha256.js +0 -91
  40. package/dist/types/common.d.ts +0 -51
  41. package/dist/types/common.d.ts.map +0 -1
  42. package/dist/types/harness-improvements.d.ts +0 -523
  43. package/dist/types/harness-improvements.d.ts.map +0 -1
  44. package/dist/types/harness-workspaces.d.ts +0 -802
  45. package/dist/types/harness-workspaces.d.ts.map +0 -1
  46. package/dist/types/sha256.d.ts +0 -2
  47. package/dist/types/sha256.d.ts.map +0 -1
@@ -1,332 +1,7 @@
1
1
  import { z } from "zod";
2
- import { ImmutableArtifactRefSchema } from "./common.js";
2
+ import { HarnessTraceSchema, ImmutableArtifactRefSchema, type ModelAction, type ToolObservation } from "@openpond/harness";
3
3
  import { type AttemptReceipt, type RunManifest } from "./runs.js";
4
- export declare const PortabilityReportSchema: z.ZodObject<{
5
- portable: z.ZodBoolean;
6
- blockers: z.ZodArray<z.ZodString>;
7
- localOnlyAssetRefs: z.ZodArray<z.ZodString>;
8
- hostPrivateAssetRefs: z.ZodArray<z.ZodString>;
9
- }, z.core.$strict>;
10
- export declare const AgentSnapshotContentSchema: z.ZodObject<{
11
- schemaVersion: z.ZodLiteral<"openpond.agentSnapshot.v2">;
12
- id: z.ZodString;
13
- sourceRelease: z.ZodNullable<z.ZodObject<{
14
- id: z.ZodString;
15
- contentHash: z.ZodString;
16
- }, z.core.$strict>>;
17
- instructions: z.ZodArray<z.ZodObject<{
18
- id: z.ZodString;
19
- path: z.ZodString;
20
- contentHash: z.ZodString;
21
- sizeBytes: z.ZodNumber;
22
- mediaType: z.ZodString;
23
- visibility: z.ZodEnum<{
24
- policy: "policy";
25
- verifier: "verifier";
26
- host_private: "host_private";
27
- }>;
28
- }, z.core.$strict>>;
29
- skills: z.ZodArray<z.ZodObject<{
30
- id: z.ZodString;
31
- path: z.ZodString;
32
- contentHash: z.ZodString;
33
- sizeBytes: z.ZodNumber;
34
- mediaType: z.ZodString;
35
- visibility: z.ZodEnum<{
36
- policy: "policy";
37
- verifier: "verifier";
38
- host_private: "host_private";
39
- }>;
40
- }, z.core.$strict>>;
41
- agents: z.ZodArray<z.ZodObject<{
42
- id: z.ZodString;
43
- path: z.ZodString;
44
- contentHash: z.ZodString;
45
- sizeBytes: z.ZodNumber;
46
- mediaType: z.ZodString;
47
- visibility: z.ZodEnum<{
48
- policy: "policy";
49
- verifier: "verifier";
50
- host_private: "host_private";
51
- }>;
52
- }, z.core.$strict>>;
53
- toolDeclarations: z.ZodArray<z.ZodObject<{
54
- name: z.ZodString;
55
- description: z.ZodString;
56
- inputSchema: z.ZodRecord<z.ZodString, z.ZodUnknown>;
57
- inputSchemaHash: z.ZodString;
58
- sideEffect: z.ZodEnum<{
59
- read: "read";
60
- write: "write";
61
- }>;
62
- timeoutMs: z.ZodNumber;
63
- }, z.core.$strict>>;
64
- capabilityRequirements: z.ZodArray<z.ZodObject<{
65
- id: z.ZodString;
66
- required: z.ZodBoolean;
67
- scopes: z.ZodArray<z.ZodString>;
68
- }, z.core.$strict>>;
69
- dependencyLock: z.ZodObject<{
70
- id: z.ZodString;
71
- path: z.ZodString;
72
- contentHash: z.ZodString;
73
- sizeBytes: z.ZodNumber;
74
- mediaType: z.ZodString;
75
- visibility: z.ZodEnum<{
76
- policy: "policy";
77
- verifier: "verifier";
78
- host_private: "host_private";
79
- }>;
80
- }, z.core.$strict>;
81
- portability: z.ZodObject<{
82
- portable: z.ZodBoolean;
83
- blockers: z.ZodArray<z.ZodString>;
84
- localOnlyAssetRefs: z.ZodArray<z.ZodString>;
85
- hostPrivateAssetRefs: z.ZodArray<z.ZodString>;
86
- }, z.core.$strict>;
87
- metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
88
- }, z.core.$strict>;
89
- export declare const AgentSnapshotSchema: z.ZodObject<{
90
- schemaVersion: z.ZodLiteral<"openpond.agentSnapshot.v2">;
91
- id: z.ZodString;
92
- sourceRelease: z.ZodNullable<z.ZodObject<{
93
- id: z.ZodString;
94
- contentHash: z.ZodString;
95
- }, z.core.$strict>>;
96
- instructions: z.ZodArray<z.ZodObject<{
97
- id: z.ZodString;
98
- path: z.ZodString;
99
- contentHash: z.ZodString;
100
- sizeBytes: z.ZodNumber;
101
- mediaType: z.ZodString;
102
- visibility: z.ZodEnum<{
103
- policy: "policy";
104
- verifier: "verifier";
105
- host_private: "host_private";
106
- }>;
107
- }, z.core.$strict>>;
108
- skills: z.ZodArray<z.ZodObject<{
109
- id: z.ZodString;
110
- path: z.ZodString;
111
- contentHash: z.ZodString;
112
- sizeBytes: z.ZodNumber;
113
- mediaType: z.ZodString;
114
- visibility: z.ZodEnum<{
115
- policy: "policy";
116
- verifier: "verifier";
117
- host_private: "host_private";
118
- }>;
119
- }, z.core.$strict>>;
120
- agents: z.ZodArray<z.ZodObject<{
121
- id: z.ZodString;
122
- path: z.ZodString;
123
- contentHash: z.ZodString;
124
- sizeBytes: z.ZodNumber;
125
- mediaType: z.ZodString;
126
- visibility: z.ZodEnum<{
127
- policy: "policy";
128
- verifier: "verifier";
129
- host_private: "host_private";
130
- }>;
131
- }, z.core.$strict>>;
132
- toolDeclarations: z.ZodArray<z.ZodObject<{
133
- name: z.ZodString;
134
- description: z.ZodString;
135
- inputSchema: z.ZodRecord<z.ZodString, z.ZodUnknown>;
136
- inputSchemaHash: z.ZodString;
137
- sideEffect: z.ZodEnum<{
138
- read: "read";
139
- write: "write";
140
- }>;
141
- timeoutMs: z.ZodNumber;
142
- }, z.core.$strict>>;
143
- capabilityRequirements: z.ZodArray<z.ZodObject<{
144
- id: z.ZodString;
145
- required: z.ZodBoolean;
146
- scopes: z.ZodArray<z.ZodString>;
147
- }, z.core.$strict>>;
148
- dependencyLock: z.ZodObject<{
149
- id: z.ZodString;
150
- path: z.ZodString;
151
- contentHash: z.ZodString;
152
- sizeBytes: z.ZodNumber;
153
- mediaType: z.ZodString;
154
- visibility: z.ZodEnum<{
155
- policy: "policy";
156
- verifier: "verifier";
157
- host_private: "host_private";
158
- }>;
159
- }, z.core.$strict>;
160
- portability: z.ZodObject<{
161
- portable: z.ZodBoolean;
162
- blockers: z.ZodArray<z.ZodString>;
163
- localOnlyAssetRefs: z.ZodArray<z.ZodString>;
164
- hostPrivateAssetRefs: z.ZodArray<z.ZodString>;
165
- }, z.core.$strict>;
166
- metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
167
- contentHash: z.ZodString;
168
- }, z.core.$strict>;
169
- export declare const LifecycleContractSchema: z.ZodObject<{
170
- create: z.ZodLiteral<true>;
171
- reset: z.ZodLiteral<true>;
172
- step: z.ZodLiteral<true>;
173
- collect: z.ZodLiteral<true>;
174
- destroy: z.ZodLiteral<true>;
175
- resetScope: z.ZodEnum<{
176
- task: "task";
177
- attempt: "attempt";
178
- }>;
179
- }, z.core.$strict>;
180
- export declare const GraderInterfaceContractSchema: z.ZodObject<{
181
- visibleEvidence: z.ZodArray<z.ZodString>;
182
- privilegedEvidence: z.ZodArray<z.ZodString>;
183
- privateVerifierIsolation: z.ZodBoolean;
184
- }, z.core.$strict>;
185
- export declare const HarnessReleaseContentSchema: z.ZodObject<{
186
- schemaVersion: z.ZodLiteral<"openpond.harnessRelease.v2">;
187
- id: z.ZodString;
188
- agentSnapshot: z.ZodObject<{
189
- id: z.ZodString;
190
- contentHash: z.ZodString;
191
- }, z.core.$strict>;
192
- program: z.ZodObject<{
193
- id: z.ZodString;
194
- path: z.ZodString;
195
- contentHash: z.ZodString;
196
- sizeBytes: z.ZodNumber;
197
- mediaType: z.ZodString;
198
- visibility: z.ZodEnum<{
199
- policy: "policy";
200
- verifier: "verifier";
201
- host_private: "host_private";
202
- }>;
203
- }, z.core.$strict>;
204
- tools: z.ZodArray<z.ZodObject<{
205
- name: z.ZodString;
206
- description: z.ZodString;
207
- inputSchema: z.ZodRecord<z.ZodString, z.ZodUnknown>;
208
- inputSchemaHash: z.ZodString;
209
- sideEffect: z.ZodEnum<{
210
- read: "read";
211
- write: "write";
212
- }>;
213
- timeoutMs: z.ZodNumber;
214
- }, z.core.$strict>>;
215
- lifecycle: z.ZodObject<{
216
- create: z.ZodLiteral<true>;
217
- reset: z.ZodLiteral<true>;
218
- step: z.ZodLiteral<true>;
219
- collect: z.ZodLiteral<true>;
220
- destroy: z.ZodLiteral<true>;
221
- resetScope: z.ZodEnum<{
222
- task: "task";
223
- attempt: "attempt";
224
- }>;
225
- }, z.core.$strict>;
226
- graderInterface: z.ZodObject<{
227
- visibleEvidence: z.ZodArray<z.ZodString>;
228
- privilegedEvidence: z.ZodArray<z.ZodString>;
229
- privateVerifierIsolation: z.ZodBoolean;
230
- }, z.core.$strict>;
231
- files: z.ZodArray<z.ZodObject<{
232
- id: z.ZodString;
233
- path: z.ZodString;
234
- contentHash: z.ZodString;
235
- sizeBytes: z.ZodNumber;
236
- mediaType: z.ZodString;
237
- visibility: z.ZodEnum<{
238
- policy: "policy";
239
- verifier: "verifier";
240
- host_private: "host_private";
241
- }>;
242
- }, z.core.$strict>>;
243
- metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
244
- }, z.core.$strict>;
245
- export declare const HarnessReleaseSchema: z.ZodObject<{
246
- schemaVersion: z.ZodLiteral<"openpond.harnessRelease.v2">;
247
- id: z.ZodString;
248
- agentSnapshot: z.ZodObject<{
249
- id: z.ZodString;
250
- contentHash: z.ZodString;
251
- }, z.core.$strict>;
252
- program: z.ZodObject<{
253
- id: z.ZodString;
254
- path: z.ZodString;
255
- contentHash: z.ZodString;
256
- sizeBytes: z.ZodNumber;
257
- mediaType: z.ZodString;
258
- visibility: z.ZodEnum<{
259
- policy: "policy";
260
- verifier: "verifier";
261
- host_private: "host_private";
262
- }>;
263
- }, z.core.$strict>;
264
- tools: z.ZodArray<z.ZodObject<{
265
- name: z.ZodString;
266
- description: z.ZodString;
267
- inputSchema: z.ZodRecord<z.ZodString, z.ZodUnknown>;
268
- inputSchemaHash: z.ZodString;
269
- sideEffect: z.ZodEnum<{
270
- read: "read";
271
- write: "write";
272
- }>;
273
- timeoutMs: z.ZodNumber;
274
- }, z.core.$strict>>;
275
- lifecycle: z.ZodObject<{
276
- create: z.ZodLiteral<true>;
277
- reset: z.ZodLiteral<true>;
278
- step: z.ZodLiteral<true>;
279
- collect: z.ZodLiteral<true>;
280
- destroy: z.ZodLiteral<true>;
281
- resetScope: z.ZodEnum<{
282
- task: "task";
283
- attempt: "attempt";
284
- }>;
285
- }, z.core.$strict>;
286
- graderInterface: z.ZodObject<{
287
- visibleEvidence: z.ZodArray<z.ZodString>;
288
- privilegedEvidence: z.ZodArray<z.ZodString>;
289
- privateVerifierIsolation: z.ZodBoolean;
290
- }, z.core.$strict>;
291
- files: z.ZodArray<z.ZodObject<{
292
- id: z.ZodString;
293
- path: z.ZodString;
294
- contentHash: z.ZodString;
295
- sizeBytes: z.ZodNumber;
296
- mediaType: z.ZodString;
297
- visibility: z.ZodEnum<{
298
- policy: "policy";
299
- verifier: "verifier";
300
- host_private: "host_private";
301
- }>;
302
- }, z.core.$strict>>;
303
- metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
304
- contentHash: z.ZodString;
305
- }, z.core.$strict>;
306
- export declare const ModelActionSchema: z.ZodObject<{
307
- id: z.ZodString;
308
- turn: z.ZodNumber;
309
- kind: z.ZodEnum<{
310
- message: "message";
311
- terminal: "terminal";
312
- tool_call: "tool_call";
313
- }>;
314
- name: z.ZodNullable<z.ZodString>;
315
- arguments: z.ZodRecord<z.ZodString, z.ZodUnknown>;
316
- content: z.ZodNullable<z.ZodString>;
317
- }, z.core.$strict>;
318
- export declare const ToolObservationSchema: z.ZodObject<{
319
- actionId: z.ZodString;
320
- turn: z.ZodNumber;
321
- terminal: z.ZodBoolean;
322
- output: z.ZodRecord<z.ZodString, z.ZodUnknown>;
323
- artifactRefs: z.ZodArray<z.ZodObject<{
324
- id: z.ZodString;
325
- contentHash: z.ZodString;
326
- mediaType: z.ZodDefault<z.ZodNullable<z.ZodString>>;
327
- sizeBytes: z.ZodDefault<z.ZodNullable<z.ZodNumber>>;
328
- }, z.core.$strict>>;
329
- }, z.core.$strict>;
4
+ export * from "@openpond/harness/harness";
330
5
  export type HarnessLease = {
331
6
  id: string;
332
7
  metadata: Record<string, unknown>;
@@ -357,7 +32,7 @@ export type HarnessExecutionInput = {
357
32
  };
358
33
  export type HarnessExecutionResult = {
359
34
  receipt: AttemptReceipt;
360
- trace: unknown;
35
+ trace: z.infer<typeof HarnessTraceSchema>;
361
36
  output: Record<string, unknown>;
362
37
  };
363
38
  export interface HarnessExecutor {
@@ -381,85 +56,4 @@ export declare function executeRuntimeProtocol(input: {
381
56
  signal?: AbortSignal;
382
57
  now?: () => string;
383
58
  }): Promise<HarnessExecutionResult>;
384
- export declare function createAgentSnapshot(input: z.input<typeof AgentSnapshotContentSchema>): AgentSnapshot;
385
- export declare function createHarnessRelease(input: z.input<typeof HarnessReleaseContentSchema>): HarnessRelease;
386
- export declare const HarnessLifecycleEventSchema: z.ZodObject<{
387
- sequence: z.ZodNumber;
388
- type: z.ZodEnum<{
389
- reset: "reset";
390
- terminal: "terminal";
391
- created: "created";
392
- action: "action";
393
- observation: "observation";
394
- failure: "failure";
395
- collected: "collected";
396
- destroyed: "destroyed";
397
- }>;
398
- payloadHash: z.ZodString;
399
- metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
400
- }, z.core.$strict>;
401
- export declare const HarnessTraceSchema: z.ZodObject<{
402
- schemaVersion: z.ZodLiteral<"openpond.harnessTrace.v1">;
403
- manifest: z.ZodObject<{
404
- id: z.ZodString;
405
- contentHash: z.ZodString;
406
- }, z.core.$strict>;
407
- taskId: z.ZodString;
408
- seed: z.ZodString;
409
- events: z.ZodArray<z.ZodObject<{
410
- sequence: z.ZodNumber;
411
- type: z.ZodEnum<{
412
- reset: "reset";
413
- terminal: "terminal";
414
- created: "created";
415
- action: "action";
416
- observation: "observation";
417
- failure: "failure";
418
- collected: "collected";
419
- destroyed: "destroyed";
420
- }>;
421
- payloadHash: z.ZodString;
422
- metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
423
- }, z.core.$strict>>;
424
- actions: z.ZodArray<z.ZodObject<{
425
- id: z.ZodString;
426
- turn: z.ZodNumber;
427
- kind: z.ZodEnum<{
428
- message: "message";
429
- terminal: "terminal";
430
- tool_call: "tool_call";
431
- }>;
432
- name: z.ZodNullable<z.ZodString>;
433
- arguments: z.ZodRecord<z.ZodString, z.ZodUnknown>;
434
- content: z.ZodNullable<z.ZodString>;
435
- }, z.core.$strict>>;
436
- observations: z.ZodArray<z.ZodObject<{
437
- actionId: z.ZodString;
438
- turn: z.ZodNumber;
439
- terminal: z.ZodBoolean;
440
- output: z.ZodRecord<z.ZodString, z.ZodUnknown>;
441
- artifactRefs: z.ZodArray<z.ZodObject<{
442
- id: z.ZodString;
443
- contentHash: z.ZodString;
444
- mediaType: z.ZodDefault<z.ZodNullable<z.ZodString>>;
445
- sizeBytes: z.ZodDefault<z.ZodNullable<z.ZodNumber>>;
446
- }, z.core.$strict>>;
447
- }, z.core.$strict>>;
448
- terminal: z.ZodBoolean;
449
- failureClass: z.ZodNullable<z.ZodEnum<{
450
- policy_failure: "policy_failure";
451
- grader_failure: "grader_failure";
452
- environment_failure: "environment_failure";
453
- infrastructure_failure: "infrastructure_failure";
454
- timeout: "timeout";
455
- cancelled: "cancelled";
456
- }>>;
457
- output: z.ZodRecord<z.ZodString, z.ZodUnknown>;
458
- contentHash: z.ZodString;
459
- }, z.core.$strict>;
460
- export type AgentSnapshot = z.infer<typeof AgentSnapshotSchema>;
461
- export type HarnessRelease = z.infer<typeof HarnessReleaseSchema>;
462
- export type ModelAction = z.infer<typeof ModelActionSchema>;
463
- export type ToolObservation = z.infer<typeof ToolObservationSchema>;
464
- export type HarnessTrace = z.infer<typeof HarnessTraceSchema>;
465
59
  //# sourceMappingURL=harness.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"harness.d.ts","sourceRoot":"","sources":["../../../../src/harness.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAEL,0BAA0B,EAO3B,MAAM,aAAa,CAAC;AAErB,OAAO,EAAwB,KAAK,cAAc,EAAE,KAAK,WAAW,EAAE,MAAM,WAAW,CAAC;AAExF,eAAO,MAAM,uBAAuB;;;;;kBAKzB,CAAC;AAEZ,eAAO,MAAM,0BAA0B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAgB5B,CAAC;AACZ,eAAO,MAAM,mBAAmB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAAiF,CAAC;AAElH,eAAO,MAAM,uBAAuB;;;;;;;;;;kBAOzB,CAAC;AACZ,eAAO,MAAM,6BAA6B;;;;kBAI/B,CAAC;AAEZ,eAAO,MAAM,2BAA2B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAU7B,CAAC;AACZ,eAAO,MAAM,oBAAoB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAAkF,CAAC;AAEpH,eAAO,MAAM,iBAAiB;;;;;;;;;;;kBAOnB,CAAC;AACZ,eAAO,MAAM,qBAAqB;;;;;;;;;;;kBAMvB,CAAC;AAEZ,MAAM,MAAM,YAAY,GAAG;IAAE,EAAE,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAAE,CAAC;AAC7E,MAAM,WAAW,cAAc;IAC7B,MAAM,CAAC,KAAK,EAAE;QAAE,QAAQ,EAAE,WAAW,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,WAAW,CAAA;KAAE,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC;IACnH,KAAK,CAAC,KAAK,EAAE,YAAY,EAAE,KAAK,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,WAAW,CAAA;KAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IACxF,IAAI,CAAC,KAAK,EAAE,YAAY,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,WAAW,GAAG,OAAO,CAAC,eAAe,CAAC,CAAC;IAC9F,OAAO,CAAC,KAAK,EAAE,YAAY,EAAE,MAAM,EAAE,WAAW,GAAG,OAAO,CAAC;QAAE,YAAY,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,0BAA0B,CAAC,EAAE,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;KAAE,CAAC,CAAC;IAC9J,OAAO,CAAC,KAAK,EAAE,YAAY,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CAC7C;AAED,MAAM,MAAM,qBAAqB,GAAG;IAClC,QAAQ,EAAE,WAAW,CAAC;IACtB,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,CAAC;AACF,MAAM,MAAM,sBAAsB,GAAG;IACnC,OAAO,EAAE,cAAc,CAAC;IACxB,KAAK,EAAE,OAAO,CAAC;IACf,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACjC,CAAC;AACF,MAAM,WAAW,eAAe;IAC9B,OAAO,CAAC,KAAK,EAAE,qBAAqB,GAAG,OAAO,CAAC,sBAAsB,CAAC,CAAC;CACxE;AACD,MAAM,WAAW,gBAAgB;IAC/B,GAAG,CAAC,KAAK,EAAE;QACT,QAAQ,EAAE,WAAW,CAAC;QACtB,OAAO,EAAE,MAAM,EAAE,CAAC;QAClB,KAAK,EAAE,MAAM,EAAE,CAAC;QAChB,WAAW,EAAE,MAAM,CAAC;QACpB,MAAM,CAAC,EAAE,WAAW,CAAC;KACtB,GAAG,OAAO,CAAC,sBAAsB,EAAE,CAAC,CAAC;CACvC;AAED,wBAAsB,sBAAsB,CAAC,KAAK,EAAE;IAClD,QAAQ,EAAE,WAAW,CAAC;IACtB,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,WAAW,EAAE,CAAC;IACvB,OAAO,EAAE,cAAc,CAAC;IACxB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,GAAG,CAAC,EAAE,MAAM,MAAM,CAAC;CACpB,GAAG,OAAO,CAAC,sBAAsB,CAAC,CAuGlC;AAED,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,0BAA0B,CAAC,GAAG,aAAa,CAGpG;AACD,wBAAgB,oBAAoB,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,2BAA2B,CAAC,GAAG,cAAc,CAGvG;AAED,eAAO,MAAM,2BAA2B;;;;;;;;;;;;;;kBAK7B,CAAC;AACZ,eAAO,MAAM,kBAAkB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAYpB,CAAC;AAEZ,MAAM,MAAM,aAAa,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,mBAAmB,CAAC,CAAC;AAChE,MAAM,MAAM,cAAc,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,oBAAoB,CAAC,CAAC;AAClE,MAAM,MAAM,WAAW,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,iBAAiB,CAAC,CAAC;AAC5D,MAAM,MAAM,eAAe,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,qBAAqB,CAAC,CAAC;AACpE,MAAM,MAAM,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC"}
1
+ {"version":3,"file":"harness.d.ts","sourceRoot":"","sources":["../../../../src/harness.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAGL,kBAAkB,EAClB,0BAA0B,EAI1B,KAAK,WAAW,EAChB,KAAK,eAAe,EACrB,MAAM,mBAAmB,CAAC;AAE3B,OAAO,EAEL,KAAK,cAAc,EACnB,KAAK,WAAW,EACjB,MAAM,WAAW,CAAC;AAEnB,cAAc,2BAA2B,CAAC;AAE1C,MAAM,MAAM,YAAY,GAAG;IAAE,EAAE,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAAE,CAAC;AAE7E,MAAM,WAAW,cAAc;IAC7B,MAAM,CAAC,KAAK,EAAE;QACZ,QAAQ,EAAE,WAAW,CAAC;QACtB,MAAM,EAAE,MAAM,CAAC;QACf,IAAI,EAAE,MAAM,CAAC;QACb,MAAM,EAAE,WAAW,CAAC;KACrB,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC;IAC1B,KAAK,CACH,KAAK,EAAE,YAAY,EACnB,KAAK,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,WAAW,CAAA;KAAE,GAC3C,OAAO,CAAC,IAAI,CAAC,CAAC;IACjB,IAAI,CACF,KAAK,EAAE,YAAY,EACnB,MAAM,EAAE,WAAW,EACnB,MAAM,EAAE,WAAW,GAClB,OAAO,CAAC,eAAe,CAAC,CAAC;IAC5B,OAAO,CACL,KAAK,EAAE,YAAY,EACnB,MAAM,EAAE,WAAW,GAClB,OAAO,CAAC;QACT,YAAY,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,0BAA0B,CAAC,EAAE,CAAC;QAC3D,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;KACnC,CAAC,CAAC;IACH,OAAO,CAAC,KAAK,EAAE,YAAY,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CAC7C;AAED,MAAM,MAAM,qBAAqB,GAAG;IAClC,QAAQ,EAAE,WAAW,CAAC;IACtB,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,CAAC;AAEF,MAAM,MAAM,sBAAsB,GAAG;IACnC,OAAO,EAAE,cAAc,CAAC;IACxB,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC;IAC1C,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACjC,CAAC;AAEF,MAAM,WAAW,eAAe;IAC9B,OAAO,CAAC,KAAK,EAAE,qBAAqB,GAAG,OAAO,CAAC,sBAAsB,CAAC,CAAC;CACxE;AAED,MAAM,WAAW,gBAAgB;IAC/B,GAAG,CAAC,KAAK,EAAE;QACT,QAAQ,EAAE,WAAW,CAAC;QACtB,OAAO,EAAE,MAAM,EAAE,CAAC;QAClB,KAAK,EAAE,MAAM,EAAE,CAAC;QAChB,WAAW,EAAE,MAAM,CAAC;QACpB,MAAM,CAAC,EAAE,WAAW,CAAC;KACtB,GAAG,OAAO,CAAC,sBAAsB,EAAE,CAAC,CAAC;CACvC;AAED,wBAAsB,sBAAsB,CAAC,KAAK,EAAE;IAClD,QAAQ,EAAE,WAAW,CAAC;IACtB,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,WAAW,EAAE,CAAC;IACvB,OAAO,EAAE,cAAc,CAAC;IACxB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,GAAG,CAAC,EAAE,MAAM,MAAM,CAAC;CACpB,GAAG,OAAO,CAAC,sBAAsB,CAAC,CA8JlC"}
@@ -1,10 +1,9 @@
1
- export * from "./common.js";
2
1
  export * from "./compatibility.js";
3
2
  export * from "./evidence/index.js";
4
3
  export * from "./graders.js";
5
4
  export * from "./harness.js";
6
- export * from "./harness-improvements.js";
7
- export * from "./harness-workspaces.js";
8
5
  export * from "./runs.js";
6
+ export * from "./model-improvement-qualification.js";
7
+ export * from "./review-conformance.js";
9
8
  export * from "./tasksets.js";
10
9
  //# sourceMappingURL=index.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,aAAa,CAAC;AAC5B,cAAc,oBAAoB,CAAC;AACnC,cAAc,qBAAqB,CAAC;AACpC,cAAc,cAAc,CAAC;AAC7B,cAAc,cAAc,CAAC;AAC7B,cAAc,2BAA2B,CAAC;AAC1C,cAAc,yBAAyB,CAAC;AACxC,cAAc,WAAW,CAAC;AAC1B,cAAc,eAAe,CAAC"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,oBAAoB,CAAC;AACnC,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"}
@@ -0,0 +1,222 @@
1
+ import { z } from "zod";
2
+ export declare const ModelImprovementDecisionSchema: z.ZodEnum<{
3
+ no_training: "no_training";
4
+ sft: "sft";
5
+ preference: "preference";
6
+ rl: "rl";
7
+ }>;
8
+ export declare const ModelImprovementSignalSchema: z.ZodObject<{
9
+ kind: z.ZodEnum<{
10
+ none: "none";
11
+ demonstrations: "demonstrations";
12
+ chosen_rejected: "chosen_rejected";
13
+ scalar_reward: "scalar_reward";
14
+ }>;
15
+ strength: z.ZodEnum<{
16
+ absent: "absent";
17
+ weak: "weak";
18
+ usable: "usable";
19
+ }>;
20
+ calibrated: z.ZodBoolean;
21
+ confounded: z.ZodBoolean;
22
+ variance: z.ZodNullable<z.ZodNumber>;
23
+ evidenceRefs: z.ZodArray<z.ZodObject<{
24
+ id: z.ZodString;
25
+ contentHash: z.ZodString;
26
+ }, z.core.$strict>>;
27
+ }, z.core.$strict>;
28
+ export declare const ModelImprovementQualificationReceiptContentSchema: z.ZodObject<{
29
+ schemaVersion: z.ZodLiteral<"openpond.modelImprovementQualificationReceipt.v1">;
30
+ id: z.ZodString;
31
+ review: z.ZodObject<{
32
+ id: z.ZodString;
33
+ contentHash: z.ZodString;
34
+ }, z.core.$strict>;
35
+ harnessRelease: z.ZodObject<{
36
+ id: z.ZodString;
37
+ contentHash: z.ZodString;
38
+ }, z.core.$strict>;
39
+ tasksetRelease: z.ZodNullable<z.ZodObject<{
40
+ id: z.ZodString;
41
+ contentHash: z.ZodString;
42
+ }, z.core.$strict>>;
43
+ baselineEvaluation: z.ZodNullable<z.ZodObject<{
44
+ id: z.ZodString;
45
+ contentHash: z.ZodString;
46
+ }, z.core.$strict>>;
47
+ model: z.ZodObject<{
48
+ provider: z.ZodString;
49
+ model: z.ZodString;
50
+ revision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
51
+ artifactHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
52
+ tokenizerRevision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
53
+ chatTemplateHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
54
+ }, z.core.$strict>;
55
+ environmentHash: z.ZodNullable<z.ZodString>;
56
+ toolContractHash: z.ZodNullable<z.ZodString>;
57
+ permissionContractHash: z.ZodNullable<z.ZodString>;
58
+ policyHash: z.ZodNullable<z.ZodString>;
59
+ verifierRef: z.ZodNullable<z.ZodObject<{
60
+ id: z.ZodString;
61
+ contentHash: z.ZodString;
62
+ }, z.core.$strict>>;
63
+ sourcePolicies: z.ZodArray<z.ZodObject<{
64
+ policy: z.ZodObject<{
65
+ id: z.ZodString;
66
+ contentHash: z.ZodString;
67
+ }, z.core.$strict>;
68
+ state: z.ZodEnum<{
69
+ authorized: "authorized";
70
+ revoked: "revoked";
71
+ deleted: "deleted";
72
+ expired: "expired";
73
+ }>;
74
+ checkedAt: z.ZodString;
75
+ }, z.core.$strict>>;
76
+ trainingEvidenceRefs: z.ZodArray<z.ZodObject<{
77
+ id: z.ZodString;
78
+ contentHash: z.ZodString;
79
+ }, z.core.$strict>>;
80
+ frozenEvaluationEvidenceRefs: z.ZodArray<z.ZodObject<{
81
+ id: z.ZodString;
82
+ contentHash: z.ZodString;
83
+ }, z.core.$strict>>;
84
+ privacyApproval: z.ZodNullable<z.ZodObject<{
85
+ id: z.ZodString;
86
+ contentHash: z.ZodString;
87
+ }, z.core.$strict>>;
88
+ budgetApproval: z.ZodNullable<z.ZodObject<{
89
+ id: z.ZodString;
90
+ contentHash: z.ZodString;
91
+ }, z.core.$strict>>;
92
+ maximumCostUsd: z.ZodNumber;
93
+ signal: z.ZodObject<{
94
+ kind: z.ZodEnum<{
95
+ none: "none";
96
+ demonstrations: "demonstrations";
97
+ chosen_rejected: "chosen_rejected";
98
+ scalar_reward: "scalar_reward";
99
+ }>;
100
+ strength: z.ZodEnum<{
101
+ absent: "absent";
102
+ weak: "weak";
103
+ usable: "usable";
104
+ }>;
105
+ calibrated: z.ZodBoolean;
106
+ confounded: z.ZodBoolean;
107
+ variance: z.ZodNullable<z.ZodNumber>;
108
+ evidenceRefs: z.ZodArray<z.ZodObject<{
109
+ id: z.ZodString;
110
+ contentHash: z.ZodString;
111
+ }, z.core.$strict>>;
112
+ }, z.core.$strict>;
113
+ decision: z.ZodEnum<{
114
+ no_training: "no_training";
115
+ sft: "sft";
116
+ preference: "preference";
117
+ rl: "rl";
118
+ }>;
119
+ reasons: z.ZodArray<z.ZodString>;
120
+ createdAt: z.ZodString;
121
+ metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
122
+ }, z.core.$strict>;
123
+ export declare const ModelImprovementQualificationReceiptSchema: z.ZodObject<{
124
+ schemaVersion: z.ZodLiteral<"openpond.modelImprovementQualificationReceipt.v1">;
125
+ id: z.ZodString;
126
+ review: z.ZodObject<{
127
+ id: z.ZodString;
128
+ contentHash: z.ZodString;
129
+ }, z.core.$strict>;
130
+ harnessRelease: z.ZodObject<{
131
+ id: z.ZodString;
132
+ contentHash: z.ZodString;
133
+ }, z.core.$strict>;
134
+ tasksetRelease: z.ZodNullable<z.ZodObject<{
135
+ id: z.ZodString;
136
+ contentHash: z.ZodString;
137
+ }, z.core.$strict>>;
138
+ baselineEvaluation: z.ZodNullable<z.ZodObject<{
139
+ id: z.ZodString;
140
+ contentHash: z.ZodString;
141
+ }, z.core.$strict>>;
142
+ model: z.ZodObject<{
143
+ provider: z.ZodString;
144
+ model: z.ZodString;
145
+ revision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
146
+ artifactHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
147
+ tokenizerRevision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
148
+ chatTemplateHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
149
+ }, z.core.$strict>;
150
+ environmentHash: z.ZodNullable<z.ZodString>;
151
+ toolContractHash: z.ZodNullable<z.ZodString>;
152
+ permissionContractHash: z.ZodNullable<z.ZodString>;
153
+ policyHash: z.ZodNullable<z.ZodString>;
154
+ verifierRef: z.ZodNullable<z.ZodObject<{
155
+ id: z.ZodString;
156
+ contentHash: z.ZodString;
157
+ }, z.core.$strict>>;
158
+ sourcePolicies: z.ZodArray<z.ZodObject<{
159
+ policy: z.ZodObject<{
160
+ id: z.ZodString;
161
+ contentHash: z.ZodString;
162
+ }, z.core.$strict>;
163
+ state: z.ZodEnum<{
164
+ authorized: "authorized";
165
+ revoked: "revoked";
166
+ deleted: "deleted";
167
+ expired: "expired";
168
+ }>;
169
+ checkedAt: z.ZodString;
170
+ }, z.core.$strict>>;
171
+ trainingEvidenceRefs: z.ZodArray<z.ZodObject<{
172
+ id: z.ZodString;
173
+ contentHash: z.ZodString;
174
+ }, z.core.$strict>>;
175
+ frozenEvaluationEvidenceRefs: z.ZodArray<z.ZodObject<{
176
+ id: z.ZodString;
177
+ contentHash: z.ZodString;
178
+ }, z.core.$strict>>;
179
+ privacyApproval: z.ZodNullable<z.ZodObject<{
180
+ id: z.ZodString;
181
+ contentHash: z.ZodString;
182
+ }, z.core.$strict>>;
183
+ budgetApproval: z.ZodNullable<z.ZodObject<{
184
+ id: z.ZodString;
185
+ contentHash: z.ZodString;
186
+ }, z.core.$strict>>;
187
+ maximumCostUsd: z.ZodNumber;
188
+ signal: z.ZodObject<{
189
+ kind: z.ZodEnum<{
190
+ none: "none";
191
+ demonstrations: "demonstrations";
192
+ chosen_rejected: "chosen_rejected";
193
+ scalar_reward: "scalar_reward";
194
+ }>;
195
+ strength: z.ZodEnum<{
196
+ absent: "absent";
197
+ weak: "weak";
198
+ usable: "usable";
199
+ }>;
200
+ calibrated: z.ZodBoolean;
201
+ confounded: z.ZodBoolean;
202
+ variance: z.ZodNullable<z.ZodNumber>;
203
+ evidenceRefs: z.ZodArray<z.ZodObject<{
204
+ id: z.ZodString;
205
+ contentHash: z.ZodString;
206
+ }, z.core.$strict>>;
207
+ }, z.core.$strict>;
208
+ decision: z.ZodEnum<{
209
+ no_training: "no_training";
210
+ sft: "sft";
211
+ preference: "preference";
212
+ rl: "rl";
213
+ }>;
214
+ reasons: z.ZodArray<z.ZodString>;
215
+ createdAt: z.ZodString;
216
+ metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
217
+ contentHash: z.ZodString;
218
+ }, z.core.$strict>;
219
+ export declare function createModelImprovementQualificationReceipt(input: z.input<typeof ModelImprovementQualificationReceiptContentSchema>): ModelImprovementQualificationReceipt;
220
+ export declare function verifyModelImprovementQualificationReceipt(value: unknown): value is ModelImprovementQualificationReceipt;
221
+ export type ModelImprovementQualificationReceipt = z.infer<typeof ModelImprovementQualificationReceiptSchema>;
222
+ //# sourceMappingURL=model-improvement-qualification.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"model-improvement-qualification.d.ts","sourceRoot":"","sources":["../../../../src/model-improvement-qualification.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAexB,eAAO,MAAM,8BAA8B;;;;;EAKzC,CAAC;AAEH,eAAO,MAAM,4BAA4B;;;;;;;;;;;;;;;;;;;kBAS9B,CAAC;AAEZ,eAAO,MAAM,iDAAiD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAkG1D,CAAC;AAEL,eAAO,MAAM,0CAA0C;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAG1C,CAAC;AAEd,wBAAgB,0CAA0C,CACxD,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,iDAAiD,CAAC,GACvE,oCAAoC,CAMtC;AAED,wBAAgB,0CAA0C,CACxD,KAAK,EAAE,OAAO,GACb,KAAK,IAAI,oCAAoC,CAS/C;AAMD,MAAM,MAAM,oCAAoC,GAAG,CAAC,CAAC,KAAK,CACxD,OAAO,0CAA0C,CAClD,CAAC"}