@vitest-evals/core 0.16.1 → 0.17.0

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.
@@ -136,7 +136,7 @@ type VitestJsonReport = z.infer<typeof VitestJsonReportSchema>;
136
136
  declare function parseVitestJsonReport(input: unknown): VitestJsonReport;
137
137
 
138
138
  /** Current schema version for collected report workspaces. */
139
- declare const REPORT_WORKSPACE_SCHEMA_VERSION = 1;
139
+ declare const REPORT_WORKSPACE_SCHEMA_VERSION = 2;
140
140
  /** One collected Vitest JSON report source in a multi-run workspace. */
141
141
  declare const ReportRunSchema: z.ZodObject<{
142
142
  id: z.ZodString;
@@ -189,6 +189,104 @@ declare const ReportCaseSchema: z.ZodObject<{
189
189
  scores: z.ZodOptional<z.ZodArray<z.ZodObject<{
190
190
  name: z.ZodOptional<z.ZodString>;
191
191
  score: z.ZodPipe<z.ZodTransform<unknown, unknown>, z.ZodOptional<z.ZodNullable<z.ZodNumber>>>;
192
+ judgeRuns: z.ZodOptional<z.ZodArray<z.ZodObject<{
193
+ output: z.ZodOptional<z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>;
194
+ session: z.ZodObject<{
195
+ events: z.ZodArray<z.ZodDiscriminatedUnion<[z.ZodObject<{
196
+ type: z.ZodLiteral<"message">;
197
+ role: z.ZodEnum<{
198
+ system: "system";
199
+ user: "user";
200
+ assistant: "assistant";
201
+ }>;
202
+ content: z.ZodOptional<z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>;
203
+ metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
204
+ }, z.core.$strict>, z.ZodObject<{
205
+ type: z.ZodLiteral<"tool_call">;
206
+ id: z.ZodString;
207
+ name: z.ZodString;
208
+ arguments: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
209
+ startedAt: z.ZodOptional<z.ZodString>;
210
+ finishedAt: z.ZodOptional<z.ZodString>;
211
+ durationMs: z.ZodOptional<z.ZodNumber>;
212
+ metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
213
+ }, z.core.$strict>, z.ZodObject<{
214
+ type: z.ZodLiteral<"tool_result">;
215
+ toolCallId: z.ZodString;
216
+ name: z.ZodOptional<z.ZodString>;
217
+ content: z.ZodOptional<z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>;
218
+ error: z.ZodOptional<z.ZodObject<{
219
+ message: z.ZodString;
220
+ type: z.ZodOptional<z.ZodString>;
221
+ }, z.core.$catchall<z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>>;
222
+ startedAt: z.ZodOptional<z.ZodString>;
223
+ finishedAt: z.ZodOptional<z.ZodString>;
224
+ durationMs: z.ZodOptional<z.ZodNumber>;
225
+ metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
226
+ }, z.core.$strict>], "type">>;
227
+ provider: z.ZodOptional<z.ZodString>;
228
+ model: z.ZodOptional<z.ZodString>;
229
+ metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
230
+ }, z.core.$strict>;
231
+ usage: z.ZodObject<{
232
+ provider: z.ZodOptional<z.ZodString>;
233
+ model: z.ZodOptional<z.ZodString>;
234
+ inputTokens: z.ZodOptional<z.ZodNumber>;
235
+ outputTokens: z.ZodOptional<z.ZodNumber>;
236
+ reasoningTokens: z.ZodOptional<z.ZodNumber>;
237
+ totalTokens: z.ZodOptional<z.ZodNumber>;
238
+ costUsd: z.ZodOptional<z.ZodNumber>;
239
+ toolCalls: z.ZodOptional<z.ZodNumber>;
240
+ retries: z.ZodOptional<z.ZodNumber>;
241
+ metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
242
+ }, z.core.$strict>;
243
+ timings: z.ZodOptional<z.ZodObject<{
244
+ totalMs: z.ZodOptional<z.ZodNumber>;
245
+ metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
246
+ }, z.core.$strict>>;
247
+ artifacts: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
248
+ traces: z.ZodOptional<z.ZodArray<z.ZodObject<{
249
+ id: z.ZodOptional<z.ZodString>;
250
+ name: z.ZodOptional<z.ZodString>;
251
+ startedAt: z.ZodOptional<z.ZodString>;
252
+ finishedAt: z.ZodOptional<z.ZodString>;
253
+ durationMs: z.ZodOptional<z.ZodNumber>;
254
+ metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
255
+ spans: z.ZodArray<z.ZodObject<{
256
+ id: z.ZodOptional<z.ZodString>;
257
+ traceId: z.ZodOptional<z.ZodString>;
258
+ parentId: z.ZodOptional<z.ZodString>;
259
+ name: z.ZodString;
260
+ kind: z.ZodOptional<z.ZodEnum<{
261
+ custom: "custom";
262
+ tool: "tool";
263
+ model: "model";
264
+ run: "run";
265
+ agent: "agent";
266
+ guardrail: "guardrail";
267
+ handoff: "handoff";
268
+ }>>;
269
+ startedAt: z.ZodOptional<z.ZodString>;
270
+ finishedAt: z.ZodOptional<z.ZodString>;
271
+ durationMs: z.ZodOptional<z.ZodNumber>;
272
+ status: z.ZodOptional<z.ZodEnum<{
273
+ error: "error";
274
+ ok: "ok";
275
+ }>>;
276
+ error: z.ZodOptional<z.ZodObject<{
277
+ message: z.ZodString;
278
+ type: z.ZodOptional<z.ZodString>;
279
+ }, z.core.$catchall<z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>>;
280
+ attributes: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
281
+ events: z.ZodOptional<z.ZodArray<z.ZodObject<{
282
+ name: z.ZodString;
283
+ timestamp: z.ZodOptional<z.ZodString>;
284
+ attributes: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
285
+ }, z.core.$strict>>>;
286
+ }, z.core.$strict>>;
287
+ }, z.core.$strict>>>;
288
+ errors: z.ZodArray<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
289
+ }, z.core.$strict>>>;
192
290
  metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
193
291
  }, z.core.$strict>>>;
194
292
  avgScore: z.ZodPipe<z.ZodTransform<unknown, unknown>, z.ZodOptional<z.ZodNullable<z.ZodNumber>>>;
@@ -261,6 +359,7 @@ declare const ReportCaseSchema: z.ZodObject<{
261
359
  outputTokens: z.ZodOptional<z.ZodNumber>;
262
360
  reasoningTokens: z.ZodOptional<z.ZodNumber>;
263
361
  totalTokens: z.ZodOptional<z.ZodNumber>;
362
+ costUsd: z.ZodOptional<z.ZodNumber>;
264
363
  toolCalls: z.ZodOptional<z.ZodNumber>;
265
364
  retries: z.ZodOptional<z.ZodNumber>;
266
365
  metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
@@ -318,7 +417,6 @@ declare const ReportCaseSchema: z.ZodObject<{
318
417
  type ReportCase = z.infer<typeof ReportCaseSchema>;
319
418
  /** Full multi-run report workspace consumed by rich report UIs. */
320
419
  declare const ReportWorkspaceSchema: z.ZodObject<{
321
- schemaVersion: z.ZodLiteral<1>;
322
420
  runs: z.ZodArray<z.ZodObject<{
323
421
  id: z.ZodString;
324
422
  source: z.ZodOptional<z.ZodString>;
@@ -367,6 +465,104 @@ declare const ReportWorkspaceSchema: z.ZodObject<{
367
465
  scores: z.ZodOptional<z.ZodArray<z.ZodObject<{
368
466
  name: z.ZodOptional<z.ZodString>;
369
467
  score: z.ZodPipe<z.ZodTransform<unknown, unknown>, z.ZodOptional<z.ZodNullable<z.ZodNumber>>>;
468
+ judgeRuns: z.ZodOptional<z.ZodArray<z.ZodObject<{
469
+ output: z.ZodOptional<z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>;
470
+ session: z.ZodObject<{
471
+ events: z.ZodArray<z.ZodDiscriminatedUnion<[z.ZodObject<{
472
+ type: z.ZodLiteral<"message">;
473
+ role: z.ZodEnum<{
474
+ system: "system";
475
+ user: "user";
476
+ assistant: "assistant";
477
+ }>;
478
+ content: z.ZodOptional<z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>;
479
+ metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
480
+ }, z.core.$strict>, z.ZodObject<{
481
+ type: z.ZodLiteral<"tool_call">;
482
+ id: z.ZodString;
483
+ name: z.ZodString;
484
+ arguments: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
485
+ startedAt: z.ZodOptional<z.ZodString>;
486
+ finishedAt: z.ZodOptional<z.ZodString>;
487
+ durationMs: z.ZodOptional<z.ZodNumber>;
488
+ metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
489
+ }, z.core.$strict>, z.ZodObject<{
490
+ type: z.ZodLiteral<"tool_result">;
491
+ toolCallId: z.ZodString;
492
+ name: z.ZodOptional<z.ZodString>;
493
+ content: z.ZodOptional<z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>;
494
+ error: z.ZodOptional<z.ZodObject<{
495
+ message: z.ZodString;
496
+ type: z.ZodOptional<z.ZodString>;
497
+ }, z.core.$catchall<z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>>;
498
+ startedAt: z.ZodOptional<z.ZodString>;
499
+ finishedAt: z.ZodOptional<z.ZodString>;
500
+ durationMs: z.ZodOptional<z.ZodNumber>;
501
+ metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
502
+ }, z.core.$strict>], "type">>;
503
+ provider: z.ZodOptional<z.ZodString>;
504
+ model: z.ZodOptional<z.ZodString>;
505
+ metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
506
+ }, z.core.$strict>;
507
+ usage: z.ZodObject<{
508
+ provider: z.ZodOptional<z.ZodString>;
509
+ model: z.ZodOptional<z.ZodString>;
510
+ inputTokens: z.ZodOptional<z.ZodNumber>;
511
+ outputTokens: z.ZodOptional<z.ZodNumber>;
512
+ reasoningTokens: z.ZodOptional<z.ZodNumber>;
513
+ totalTokens: z.ZodOptional<z.ZodNumber>;
514
+ costUsd: z.ZodOptional<z.ZodNumber>;
515
+ toolCalls: z.ZodOptional<z.ZodNumber>;
516
+ retries: z.ZodOptional<z.ZodNumber>;
517
+ metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
518
+ }, z.core.$strict>;
519
+ timings: z.ZodOptional<z.ZodObject<{
520
+ totalMs: z.ZodOptional<z.ZodNumber>;
521
+ metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
522
+ }, z.core.$strict>>;
523
+ artifacts: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
524
+ traces: z.ZodOptional<z.ZodArray<z.ZodObject<{
525
+ id: z.ZodOptional<z.ZodString>;
526
+ name: z.ZodOptional<z.ZodString>;
527
+ startedAt: z.ZodOptional<z.ZodString>;
528
+ finishedAt: z.ZodOptional<z.ZodString>;
529
+ durationMs: z.ZodOptional<z.ZodNumber>;
530
+ metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
531
+ spans: z.ZodArray<z.ZodObject<{
532
+ id: z.ZodOptional<z.ZodString>;
533
+ traceId: z.ZodOptional<z.ZodString>;
534
+ parentId: z.ZodOptional<z.ZodString>;
535
+ name: z.ZodString;
536
+ kind: z.ZodOptional<z.ZodEnum<{
537
+ custom: "custom";
538
+ tool: "tool";
539
+ model: "model";
540
+ run: "run";
541
+ agent: "agent";
542
+ guardrail: "guardrail";
543
+ handoff: "handoff";
544
+ }>>;
545
+ startedAt: z.ZodOptional<z.ZodString>;
546
+ finishedAt: z.ZodOptional<z.ZodString>;
547
+ durationMs: z.ZodOptional<z.ZodNumber>;
548
+ status: z.ZodOptional<z.ZodEnum<{
549
+ error: "error";
550
+ ok: "ok";
551
+ }>>;
552
+ error: z.ZodOptional<z.ZodObject<{
553
+ message: z.ZodString;
554
+ type: z.ZodOptional<z.ZodString>;
555
+ }, z.core.$catchall<z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>>;
556
+ attributes: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
557
+ events: z.ZodOptional<z.ZodArray<z.ZodObject<{
558
+ name: z.ZodString;
559
+ timestamp: z.ZodOptional<z.ZodString>;
560
+ attributes: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
561
+ }, z.core.$strict>>>;
562
+ }, z.core.$strict>>;
563
+ }, z.core.$strict>>>;
564
+ errors: z.ZodArray<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
565
+ }, z.core.$strict>>>;
370
566
  metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
371
567
  }, z.core.$strict>>>;
372
568
  avgScore: z.ZodPipe<z.ZodTransform<unknown, unknown>, z.ZodOptional<z.ZodNullable<z.ZodNumber>>>;
@@ -439,6 +635,7 @@ declare const ReportWorkspaceSchema: z.ZodObject<{
439
635
  outputTokens: z.ZodOptional<z.ZodNumber>;
440
636
  reasoningTokens: z.ZodOptional<z.ZodNumber>;
441
637
  totalTokens: z.ZodOptional<z.ZodNumber>;
638
+ costUsd: z.ZodOptional<z.ZodNumber>;
442
639
  toolCalls: z.ZodOptional<z.ZodNumber>;
443
640
  retries: z.ZodOptional<z.ZodNumber>;
444
641
  metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodType<JsonValue, unknown, z.core.$ZodTypeInternals<JsonValue, unknown>>>>;
@@ -492,6 +689,7 @@ declare const ReportWorkspaceSchema: z.ZodObject<{
492
689
  }, z.core.$strict>>;
493
690
  }, z.core.$strict>>;
494
691
  }, z.core.$strict>>;
692
+ schemaVersion: z.ZodLiteral<2>;
495
693
  }, z.core.$strict>;
496
694
  /** Full multi-run report workspace consumed by rich report UIs. */
497
695
  type ReportWorkspace = z.infer<typeof ReportWorkspaceSchema>;
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@vitest-evals/core",
3
- "version": "0.16.1",
3
+ "version": "0.17.0",
4
4
  "description": "Shared primitives, schemas, and report collection helpers for vitest-evals.",
5
5
  "repository": {
6
6
  "type": "git",