@openpond/evals 0.1.0

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.
@@ -0,0 +1,293 @@
1
+ import { z } from "zod";
2
+ export declare const ModelRefSchema: z.ZodObject<{
3
+ provider: z.ZodString;
4
+ model: z.ZodString;
5
+ revision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
6
+ artifactHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
7
+ tokenizerRevision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
8
+ chatTemplateHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
9
+ }, z.core.$strict>;
10
+ export declare const RuntimeTargetBindingSchema: z.ZodObject<{
11
+ adapterId: z.ZodString;
12
+ placement: z.ZodEnum<{
13
+ local: "local";
14
+ remote: "remote";
15
+ colocated: "colocated";
16
+ provider_native: "provider_native";
17
+ }>;
18
+ runtimeVersion: z.ZodString;
19
+ capabilityReceipt: z.ZodString;
20
+ }, z.core.$strict>;
21
+ export declare const RunLimitsSchema: z.ZodObject<{
22
+ maxTurns: z.ZodNumber;
23
+ timeoutMs: z.ZodNumber;
24
+ maxOutputBytes: z.ZodNumber;
25
+ maximumSpendUsd: z.ZodNullable<z.ZodNumber>;
26
+ }, z.core.$strict>;
27
+ export declare const RunManifestContentSchema: z.ZodObject<{
28
+ schemaVersion: z.ZodLiteral<"openpond.runManifest.v1">;
29
+ id: z.ZodString;
30
+ harnessRelease: z.ZodObject<{
31
+ id: z.ZodString;
32
+ contentHash: z.ZodString;
33
+ }, z.core.$strict>;
34
+ tasksetRelease: z.ZodObject<{
35
+ id: z.ZodString;
36
+ contentHash: z.ZodString;
37
+ }, z.core.$strict>;
38
+ model: z.ZodObject<{
39
+ provider: z.ZodString;
40
+ model: z.ZodString;
41
+ revision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
42
+ artifactHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
43
+ tokenizerRevision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
44
+ chatTemplateHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
45
+ }, z.core.$strict>;
46
+ runtimeTarget: z.ZodObject<{
47
+ adapterId: z.ZodString;
48
+ placement: z.ZodEnum<{
49
+ local: "local";
50
+ remote: "remote";
51
+ colocated: "colocated";
52
+ provider_native: "provider_native";
53
+ }>;
54
+ runtimeVersion: z.ZodString;
55
+ capabilityReceipt: z.ZodString;
56
+ }, z.core.$strict>;
57
+ limits: z.ZodObject<{
58
+ maxTurns: z.ZodNumber;
59
+ timeoutMs: z.ZodNumber;
60
+ maxOutputBytes: z.ZodNumber;
61
+ maximumSpendUsd: z.ZodNullable<z.ZodNumber>;
62
+ }, z.core.$strict>;
63
+ approval: z.ZodNullable<z.ZodObject<{
64
+ id: z.ZodString;
65
+ contentHash: z.ZodString;
66
+ }, z.core.$strict>>;
67
+ createdAt: z.ZodString;
68
+ metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
69
+ }, z.core.$strict>;
70
+ export declare const RunManifestSchema: z.ZodObject<{
71
+ schemaVersion: z.ZodLiteral<"openpond.runManifest.v1">;
72
+ id: z.ZodString;
73
+ harnessRelease: z.ZodObject<{
74
+ id: z.ZodString;
75
+ contentHash: z.ZodString;
76
+ }, z.core.$strict>;
77
+ tasksetRelease: z.ZodObject<{
78
+ id: z.ZodString;
79
+ contentHash: z.ZodString;
80
+ }, z.core.$strict>;
81
+ model: z.ZodObject<{
82
+ provider: z.ZodString;
83
+ model: z.ZodString;
84
+ revision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
85
+ artifactHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
86
+ tokenizerRevision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
87
+ chatTemplateHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
88
+ }, z.core.$strict>;
89
+ runtimeTarget: z.ZodObject<{
90
+ adapterId: z.ZodString;
91
+ placement: z.ZodEnum<{
92
+ local: "local";
93
+ remote: "remote";
94
+ colocated: "colocated";
95
+ provider_native: "provider_native";
96
+ }>;
97
+ runtimeVersion: z.ZodString;
98
+ capabilityReceipt: z.ZodString;
99
+ }, z.core.$strict>;
100
+ limits: z.ZodObject<{
101
+ maxTurns: z.ZodNumber;
102
+ timeoutMs: z.ZodNumber;
103
+ maxOutputBytes: z.ZodNumber;
104
+ maximumSpendUsd: z.ZodNullable<z.ZodNumber>;
105
+ }, z.core.$strict>;
106
+ approval: z.ZodNullable<z.ZodObject<{
107
+ id: z.ZodString;
108
+ contentHash: z.ZodString;
109
+ }, z.core.$strict>>;
110
+ createdAt: z.ZodString;
111
+ metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
112
+ contentHash: z.ZodString;
113
+ }, z.core.$strict>;
114
+ export declare const AttemptReceiptContentSchema: z.ZodObject<{
115
+ schemaVersion: z.ZodLiteral<"openpond.attemptReceipt.v1">;
116
+ id: z.ZodString;
117
+ runManifest: z.ZodObject<{
118
+ id: z.ZodString;
119
+ contentHash: z.ZodString;
120
+ }, z.core.$strict>;
121
+ taskId: z.ZodString;
122
+ seed: z.ZodString;
123
+ terminal: z.ZodBoolean;
124
+ failureClass: z.ZodNullable<z.ZodEnum<{
125
+ policy_failure: "policy_failure";
126
+ grader_failure: "grader_failure";
127
+ environment_failure: "environment_failure";
128
+ infrastructure_failure: "infrastructure_failure";
129
+ timeout: "timeout";
130
+ cancelled: "cancelled";
131
+ }>>;
132
+ outputHash: z.ZodNullable<z.ZodString>;
133
+ traceHash: z.ZodString;
134
+ artifactRefs: z.ZodArray<z.ZodObject<{
135
+ id: z.ZodString;
136
+ contentHash: z.ZodString;
137
+ mediaType: z.ZodDefault<z.ZodNullable<z.ZodString>>;
138
+ sizeBytes: z.ZodDefault<z.ZodNullable<z.ZodNumber>>;
139
+ }, z.core.$strict>>;
140
+ graderEvidenceRefs: z.ZodArray<z.ZodObject<{
141
+ id: z.ZodString;
142
+ contentHash: z.ZodString;
143
+ mediaType: z.ZodDefault<z.ZodNullable<z.ZodString>>;
144
+ sizeBytes: z.ZodDefault<z.ZodNullable<z.ZodNumber>>;
145
+ }, z.core.$strict>>;
146
+ startedAt: z.ZodString;
147
+ completedAt: z.ZodString;
148
+ latencyMs: z.ZodNumber;
149
+ costUsd: z.ZodNullable<z.ZodNumber>;
150
+ legacyAttemptRef: z.ZodDefault<z.ZodNullable<z.ZodString>>;
151
+ metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
152
+ }, z.core.$strict>;
153
+ export declare const AttemptReceiptSchema: z.ZodObject<{
154
+ schemaVersion: z.ZodLiteral<"openpond.attemptReceipt.v1">;
155
+ id: z.ZodString;
156
+ runManifest: z.ZodObject<{
157
+ id: z.ZodString;
158
+ contentHash: z.ZodString;
159
+ }, z.core.$strict>;
160
+ taskId: z.ZodString;
161
+ seed: z.ZodString;
162
+ terminal: z.ZodBoolean;
163
+ failureClass: z.ZodNullable<z.ZodEnum<{
164
+ policy_failure: "policy_failure";
165
+ grader_failure: "grader_failure";
166
+ environment_failure: "environment_failure";
167
+ infrastructure_failure: "infrastructure_failure";
168
+ timeout: "timeout";
169
+ cancelled: "cancelled";
170
+ }>>;
171
+ outputHash: z.ZodNullable<z.ZodString>;
172
+ traceHash: z.ZodString;
173
+ artifactRefs: z.ZodArray<z.ZodObject<{
174
+ id: z.ZodString;
175
+ contentHash: z.ZodString;
176
+ mediaType: z.ZodDefault<z.ZodNullable<z.ZodString>>;
177
+ sizeBytes: z.ZodDefault<z.ZodNullable<z.ZodNumber>>;
178
+ }, z.core.$strict>>;
179
+ graderEvidenceRefs: z.ZodArray<z.ZodObject<{
180
+ id: z.ZodString;
181
+ contentHash: z.ZodString;
182
+ mediaType: z.ZodDefault<z.ZodNullable<z.ZodString>>;
183
+ sizeBytes: z.ZodDefault<z.ZodNullable<z.ZodNumber>>;
184
+ }, z.core.$strict>>;
185
+ startedAt: z.ZodString;
186
+ completedAt: z.ZodString;
187
+ latencyMs: z.ZodNumber;
188
+ costUsd: z.ZodNullable<z.ZodNumber>;
189
+ legacyAttemptRef: z.ZodDefault<z.ZodNullable<z.ZodString>>;
190
+ metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
191
+ contentHash: z.ZodString;
192
+ }, z.core.$strict>;
193
+ export declare const EvaluationResultContentSchema: z.ZodObject<{
194
+ schemaVersion: z.ZodLiteral<"openpond.evaluationResult.v1">;
195
+ id: z.ZodString;
196
+ runManifest: z.ZodObject<{
197
+ id: z.ZodString;
198
+ contentHash: z.ZodString;
199
+ }, z.core.$strict>;
200
+ harnessRelease: z.ZodObject<{
201
+ id: z.ZodString;
202
+ contentHash: z.ZodString;
203
+ }, z.core.$strict>;
204
+ tasksetRelease: z.ZodObject<{
205
+ id: z.ZodString;
206
+ contentHash: z.ZodString;
207
+ }, z.core.$strict>;
208
+ model: z.ZodObject<{
209
+ provider: z.ZodString;
210
+ model: z.ZodString;
211
+ revision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
212
+ artifactHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
213
+ tokenizerRevision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
214
+ chatTemplateHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
215
+ }, z.core.$strict>;
216
+ receiptRefs: z.ZodArray<z.ZodObject<{
217
+ id: z.ZodString;
218
+ contentHash: z.ZodString;
219
+ }, z.core.$strict>>;
220
+ attemptCount: z.ZodNumber;
221
+ rewardEligibleCount: z.ZodNumber;
222
+ terminalCount: z.ZodNumber;
223
+ meanScore: z.ZodNullable<z.ZodNumber>;
224
+ failureCounts: z.ZodRecord<z.ZodEnum<{
225
+ policy_failure: "policy_failure";
226
+ grader_failure: "grader_failure";
227
+ environment_failure: "environment_failure";
228
+ infrastructure_failure: "infrastructure_failure";
229
+ timeout: "timeout";
230
+ cancelled: "cancelled";
231
+ }>, z.ZodNumber>;
232
+ metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
233
+ }, z.core.$strict>;
234
+ export declare const EvaluationResultSchema: z.ZodObject<{
235
+ schemaVersion: z.ZodLiteral<"openpond.evaluationResult.v1">;
236
+ id: z.ZodString;
237
+ runManifest: z.ZodObject<{
238
+ id: z.ZodString;
239
+ contentHash: z.ZodString;
240
+ }, z.core.$strict>;
241
+ harnessRelease: z.ZodObject<{
242
+ id: z.ZodString;
243
+ contentHash: z.ZodString;
244
+ }, z.core.$strict>;
245
+ tasksetRelease: z.ZodObject<{
246
+ id: z.ZodString;
247
+ contentHash: z.ZodString;
248
+ }, z.core.$strict>;
249
+ model: z.ZodObject<{
250
+ provider: z.ZodString;
251
+ model: z.ZodString;
252
+ revision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
253
+ artifactHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
254
+ tokenizerRevision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
255
+ chatTemplateHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
256
+ }, z.core.$strict>;
257
+ receiptRefs: z.ZodArray<z.ZodObject<{
258
+ id: z.ZodString;
259
+ contentHash: z.ZodString;
260
+ }, z.core.$strict>>;
261
+ attemptCount: z.ZodNumber;
262
+ rewardEligibleCount: z.ZodNumber;
263
+ terminalCount: z.ZodNumber;
264
+ meanScore: z.ZodNullable<z.ZodNumber>;
265
+ failureCounts: z.ZodRecord<z.ZodEnum<{
266
+ policy_failure: "policy_failure";
267
+ grader_failure: "grader_failure";
268
+ environment_failure: "environment_failure";
269
+ infrastructure_failure: "infrastructure_failure";
270
+ timeout: "timeout";
271
+ cancelled: "cancelled";
272
+ }>, z.ZodNumber>;
273
+ metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
274
+ contentHash: z.ZodString;
275
+ }, z.core.$strict>;
276
+ export declare function createRunManifest(input: z.input<typeof RunManifestContentSchema>): RunManifest;
277
+ export declare function createAttemptReceipt(input: z.input<typeof AttemptReceiptContentSchema>): AttemptReceipt;
278
+ export declare function verifyAttemptReceipt(receipt: AttemptReceipt): boolean;
279
+ export declare function aggregateEvaluationReceipts(input: {
280
+ id: string;
281
+ manifest: RunManifest;
282
+ receipts: AttemptReceipt[];
283
+ metadata?: Record<string, unknown>;
284
+ }): EvaluationResult;
285
+ export declare function assertComparableRunManifests(base: RunManifest, candidate: RunManifest): void;
286
+ export declare function rewardEligibleReceipts(receipts: AttemptReceipt[]): AttemptReceipt[];
287
+ export type ModelRef = z.infer<typeof ModelRefSchema>;
288
+ export type RuntimeTargetBinding = z.infer<typeof RuntimeTargetBindingSchema>;
289
+ export type RunLimits = z.infer<typeof RunLimitsSchema>;
290
+ export type RunManifest = z.infer<typeof RunManifestSchema>;
291
+ export type AttemptReceipt = z.infer<typeof AttemptReceiptSchema>;
292
+ export type EvaluationResult = z.infer<typeof EvaluationResultSchema>;
293
+ //# sourceMappingURL=runs.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"runs.d.ts","sourceRoot":"","sources":["../../src/runs.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAaxB,eAAO,MAAM,cAAc;;;;;;;kBAOhB,CAAC;AAEZ,eAAO,MAAM,0BAA0B;;;;;;;;;;kBAK5B,CAAC;AAEZ,eAAO,MAAM,eAAe;;;;;kBAKjB,CAAC;AAEZ,eAAO,MAAM,wBAAwB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAW1B,CAAC;AACZ,eAAO,MAAM,iBAAiB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAA+E,CAAC;AAE9G,eAAO,MAAM,2BAA2B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAkB7B,CAAC;AACZ,eAAO,MAAM,oBAAoB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAAkF,CAAC;AAEpH,eAAO,MAAM,6BAA6B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAc/B,CAAC;AACZ,eAAO,MAAM,sBAAsB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAAoF,CAAC;AAExH,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,wBAAwB,CAAC,GAAG,WAAW,CAG9F;AAED,wBAAgB,oBAAoB,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,2BAA2B,CAAC,GAAG,cAAc,CAGvG;AAED,wBAAgB,oBAAoB,CAAC,OAAO,EAAE,cAAc,GAAG,OAAO,CAKrE;AAED,wBAAgB,2BAA2B,CAAC,KAAK,EAAE;IACjD,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,WAAW,CAAC;IACtB,QAAQ,EAAE,cAAc,EAAE,CAAC;IAC3B,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC,GAAG,gBAAgB,CAgCnB;AAED,wBAAgB,4BAA4B,CAAC,IAAI,EAAE,WAAW,EAAE,SAAS,EAAE,WAAW,GAAG,IAAI,CAO5F;AAED,wBAAgB,sBAAsB,CAAC,QAAQ,EAAE,cAAc,EAAE,GAAG,cAAc,EAAE,CASnF;AAED,MAAM,MAAM,QAAQ,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,cAAc,CAAC,CAAC;AACtD,MAAM,MAAM,oBAAoB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,0BAA0B,CAAC,CAAC;AAC9E,MAAM,MAAM,SAAS,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,eAAe,CAAC,CAAC;AACxD,MAAM,MAAM,WAAW,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,iBAAiB,CAAC,CAAC;AAC5D,MAAM,MAAM,cAAc,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,oBAAoB,CAAC,CAAC;AAClE,MAAM,MAAM,gBAAgB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,sBAAsB,CAAC,CAAC"}
@@ -0,0 +1,2 @@
1
+ export declare function sha256Hex(value: string | Uint8Array): string;
2
+ //# sourceMappingURL=sha256.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"sha256.d.ts","sourceRoot":"","sources":["../../src/sha256.ts"],"names":[],"mappings":"AAgCA,wBAAgB,SAAS,CAAC,KAAK,EAAE,MAAM,GAAG,UAAU,GAAG,MAAM,CAgE5D"}