vskill 0.4.10 → 0.4.12

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.
@@ -324,6 +324,99 @@ describe("benchmark-history", () => {
324
324
  const history = await getCaseHistory(join(testDir, "nonexistent"), 1);
325
325
  expect(history).toEqual([]);
326
326
  });
327
+ it("derives baselinePassRate from comparisonDetail rubric scores", async () => {
328
+ const compResult = mkResult({
329
+ timestamp: "2026-03-10T10:00:00.000Z",
330
+ type: "comparison",
331
+ cases: [
332
+ {
333
+ eval_id: 1,
334
+ eval_name: "test-case",
335
+ status: "pass",
336
+ error_message: null,
337
+ pass_rate: 0.9,
338
+ assertions: [],
339
+ comparisonDetail: {
340
+ skillDurationMs: 100,
341
+ skillTokens: 50,
342
+ baselineDurationMs: 80,
343
+ baselineTokens: 40,
344
+ skillContentScore: 90,
345
+ skillStructureScore: 85,
346
+ baselineContentScore: 80,
347
+ baselineStructureScore: 60,
348
+ winner: "skill",
349
+ },
350
+ },
351
+ ],
352
+ });
353
+ await writeHistoryEntry(testDir, compResult);
354
+ const history = await getCaseHistory(testDir, 1);
355
+ const compEntry = history.find((e) => e.type === "comparison");
356
+ expect(compEntry).toBeDefined();
357
+ // (80 + 60) / 200 = 0.70
358
+ expect(compEntry.baselinePassRate).toBeCloseTo(0.70, 5);
359
+ });
360
+ it("leaves baselinePassRate undefined for benchmark entries", async () => {
361
+ const history = await getCaseHistory(testDir, 1);
362
+ const benchEntry = history.find((e) => e.type === "benchmark" || e.type === undefined);
363
+ expect(benchEntry).toBeDefined();
364
+ expect(benchEntry.baselinePassRate).toBeUndefined();
365
+ });
366
+ it("handles zero rubric scores correctly (baselinePassRate = 0)", async () => {
367
+ const zeroResult = mkResult({
368
+ timestamp: "2026-03-11T10:00:00.000Z",
369
+ type: "comparison",
370
+ cases: [
371
+ {
372
+ eval_id: 1,
373
+ eval_name: "test-case",
374
+ status: "pass",
375
+ error_message: null,
376
+ pass_rate: 0,
377
+ assertions: [],
378
+ comparisonDetail: {
379
+ skillDurationMs: 100,
380
+ skillTokens: 50,
381
+ baselineDurationMs: 80,
382
+ baselineTokens: 40,
383
+ skillContentScore: 0,
384
+ skillStructureScore: 0,
385
+ baselineContentScore: 0,
386
+ baselineStructureScore: 0,
387
+ winner: "tie",
388
+ },
389
+ },
390
+ ],
391
+ });
392
+ await writeHistoryEntry(testDir, zeroResult);
393
+ const history = await getCaseHistory(testDir, 1);
394
+ const zeroEntry = history.find((e) => e.type === "comparison" && e.timestamp.startsWith("2026-03-11"));
395
+ expect(zeroEntry).toBeDefined();
396
+ expect(zeroEntry.baselinePassRate).toBe(0);
397
+ });
398
+ it("leaves baselinePassRate undefined for comparison entry without comparisonDetail", async () => {
399
+ const noDetailResult = mkResult({
400
+ timestamp: "2026-03-12T10:00:00.000Z",
401
+ type: "comparison",
402
+ cases: [
403
+ {
404
+ eval_id: 1,
405
+ eval_name: "test-case",
406
+ status: "pass",
407
+ error_message: null,
408
+ pass_rate: 0.5,
409
+ assertions: [],
410
+ // no comparisonDetail
411
+ },
412
+ ],
413
+ });
414
+ await writeHistoryEntry(testDir, noDetailResult);
415
+ const history = await getCaseHistory(testDir, 1);
416
+ const noDetailEntry = history.find((e) => e.type === "comparison" && e.timestamp.startsWith("2026-03-12"));
417
+ expect(noDetailEntry).toBeDefined();
418
+ expect(noDetailEntry.baselinePassRate).toBeUndefined();
419
+ });
327
420
  });
328
421
  });
329
422
  //# sourceMappingURL=benchmark-history.test.js.map
@@ -1 +1 @@
1
- {"version":3,"file":"benchmark-history.test.js","sourceRoot":"","sources":["../../../src/eval/__tests__/benchmark-history.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,QAAQ,CAAC;AACrE,OAAO,EAAE,SAAS,EAAE,MAAM,EAAE,YAAY,EAAiB,MAAM,SAAS,CAAC;AACzE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,EACL,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,kBAAkB,EAClB,cAAc,GACf,MAAM,yBAAyB,CAAC;AAGjC,IAAI,OAAe,CAAC;AAEpB,MAAM,QAAQ,GAAG,CAAC,YAAsC,EAAE,EAAmB,EAAE,CAAC,CAAC;IAC/E,SAAS,EAAE,0BAA0B;IACrC,KAAK,EAAE,YAAY;IACnB,UAAU,EAAE,YAAY;IACxB,KAAK,EAAE;QACL;YACE,OAAO,EAAE,CAAC;YACV,SAAS,EAAE,WAAW;YACtB,MAAM,EAAE,MAAM;YACd,aAAa,EAAE,IAAI;YACnB,SAAS,EAAE,GAAG;YACd,UAAU,EAAE;gBACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;gBAC1D,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;aAC3D;SACF;KACF;IACD,GAAG,SAAS;CACb,CAAC,CAAC;AAEH,QAAQ,CAAC,mBAAmB,EAAE,GAAG,EAAE;IACjC,UAAU,CAAC,GAAG,EAAE;QACd,OAAO,GAAG,IAAI,CAAC,MAAM,EAAE,EAAE,kBAAkB,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;QACzD,SAAS,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACzD,CAAC,CAAC,CAAC;IAEH,SAAS,CAAC,GAAG,EAAE;QACb,MAAM,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;IACpD,CAAC,CAAC,CAAC;IAEH,QAAQ,CAAC,mBAAmB,EAAE,GAAG,EAAE;QACjC,EAAE,CAAC,oDAAoD,EAAE,KAAK,IAAI,EAAE;YAClE,MAAM,MAAM,GAAG,QAAQ,EAAE,CAAC;YAC1B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;YAE1D,MAAM,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,+BAA+B,CAAC,CAAC;YACvD,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,QAAQ,CAAC,EAAE,OAAO,CAAC,CAAC;YACnF,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;YACnC,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAC/C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,gDAAgD,EAAE,KAAK,IAAI,EAAE;YAC9D,MAAM,iBAAiB,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,CAAC;YAC7C,MAAM,EAAE,GAAG,YAAY,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,gBAAgB,CAAC,EAAE,OAAO,CAAC,CAAC;YAC3E,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QACvD,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,sCAAsC,EAAE,KAAK,IAAI,EAAE;YACpD,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;YACjE,MAAM,MAAM,GAAG,QAAQ,EAAE,CAAC;YAC1B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;YAC1D,MAAM,CAAC,QAAQ,CAAC,CAAC,UAAU,EAAE,CAAC;QAChC,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,QAAQ,CAAC,aAAa,EAAE,GAAG,EAAE;QAC3B,EAAE,CAAC,+CAA+C,EAAE,KAAK,IAAI,EAAE;YAC7D,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,IAAI,CAAC,OAAO,EAAE,aAAa,CAAC,CAAC,CAAC;YAC7D,MAAM,CAAC,IAAI,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QAC3B,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;YAC5D,MAAM,EAAE,GAAG,QAAQ,CAAC,EAAE,SAAS,EAAE,0BAA0B,EAAE,CAAC,CAAC;YAC/D,MAAM,EAAE,GAAG,QAAQ,CAAC,EAAE,SAAS,EAAE,0BAA0B,EAAE,CAAC,CAAC;YAC/D,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YACrC,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YAErC,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,CAAC,CAAC;YACxC,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YAC7B,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;YAC3D,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;QAC7D,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,2CAA2C,EAAE,KAAK,IAAI,EAAE;YACzD,MAAM,MAAM,GAAG,QAAQ,CAAC;gBACtB,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,MAAM;wBACjB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,GAAG;wBACd,UAAU,EAAE;4BACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;4BAC1D,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE;yBAC9D;qBACF;iBACF;aACF,CAAC,CAAC;YACH,MAAM,iBAAiB,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;YACzC,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,CAAC,CAAC;YACxC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACrC,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,QAAQ,CAAC,kBAAkB,EAAE,GAAG,EAAE;QAChC,EAAE,CAAC,6CAA6C,EAAE,KAAK,IAAI,EAAE;YAC3D,MAAM,iBAAiB,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,CAAC;YAC7C,MAAM,KAAK,GAAG,MAAM,gBAAgB,CAAC,OAAO,EAAE,0BAA0B,CAAC,CAAC;YAC1E,MAAM,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,QAAQ,EAAE,CAAC;YAC7B,MAAM,CAAC,KAAM,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAC/C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,oCAAoC,EAAE,KAAK,IAAI,EAAE;YAClD,MAAM,KAAK,GAAG,MAAM,gBAAgB,CAAC,OAAO,EAAE,0BAA0B,CAAC,CAAC;YAC1E,MAAM,CAAC,KAAK,CAAC,CAAC,QAAQ,EAAE,CAAC;QAC3B,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,QAAQ,CAAC,oBAAoB,EAAE,GAAG,EAAE;QAClC,EAAE,CAAC,kCAAkC,EAAE,GAAG,EAAE;YAC1C,MAAM,IAAI,GAAG,QAAQ,CAAC;gBACpB,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,MAAM;wBACjB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,GAAG;wBACd,UAAU,EAAE,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;qBACvE;iBACF;aACF,CAAC,CAAC;YACH,MAAM,IAAI,GAAG,QAAQ,CAAC;gBACpB,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,MAAM;wBACjB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,CAAC;wBACZ,UAAU,EAAE,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC;qBAC1E;iBACF;aACF,CAAC,CAAC;YAEH,MAAM,WAAW,GAAG,kBAAkB,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;YACnD,MAAM,CAAC,WAAW,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YACpC,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;YACjD,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAChD,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,mCAAmC,EAAE,GAAG,EAAE;YAC3C,MAAM,IAAI,GAAG,QAAQ,CAAC;gBACpB,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,MAAM;wBACjB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,CAAC;wBACZ,UAAU,EAAE,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC;qBAC1E;iBACF;aACF,CAAC,CAAC;YACH,MAAM,IAAI,GAAG,QAAQ,CAAC;gBACpB,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,MAAM;wBACjB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,CAAC;wBACZ,UAAU,EAAE,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;qBACvE;iBACF;aACF,CAAC,CAAC;YAEH,MAAM,WAAW,GAAG,kBAAkB,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;YACnD,MAAM,CAAC,WAAW,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YACpC,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;QACpD,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,qCAAqC,EAAE,GAAG,EAAE;YAC7C,MAAM,MAAM,GAAG,QAAQ,EAAE,CAAC;YAC1B,MAAM,CAAC,kBAAkB,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QACzD,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,kDAAkD,EAAE,GAAG,EAAE;YAC1D,MAAM,IAAI,GAAG,QAAQ,CAAC;gBACpB,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,MAAM;wBACjB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,CAAC;wBACZ,UAAU,EAAE,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;qBACvE;iBACF;aACF,CAAC,CAAC;YACH,MAAM,IAAI,GAAG,QAAQ,CAAC;gBACpB,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,MAAM;wBACjB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,CAAC;wBACZ,UAAU,EAAE;4BACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;4BACxD,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE;yBAC1D;qBACF;iBACF;aACF,CAAC,CAAC;YAEH,MAAM,WAAW,GAAG,kBAAkB,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;YACnD,MAAM,CAAC,WAAW,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QAClC,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,QAAQ,CAAC,0BAA0B,EAAE,GAAG,EAAE;QACxC,UAAU,CAAC,KAAK,IAAI,EAAE;YACpB,MAAM,EAAE,GAAG,QAAQ,CAAC;gBAClB,SAAS,EAAE,0BAA0B;gBACrC,KAAK,EAAE,QAAQ;gBACf,IAAI,EAAE,WAAW;aAClB,CAAC,CAAC;YACH,MAAM,EAAE,GAAG,QAAQ,CAAC;gBAClB,SAAS,EAAE,0BAA0B;gBACrC,KAAK,EAAE,QAAQ;gBACf,IAAI,EAAE,UAAU;aACjB,CAAC,CAAC;YACH,MAAM,EAAE,GAAG,QAAQ,CAAC;gBAClB,SAAS,EAAE,0BAA0B;gBACrC,KAAK,EAAE,QAAQ;gBACf,IAAI,EAAE,YAAY;aACnB,CAAC,CAAC;YACH,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YACrC,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YACrC,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACvC,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,kBAAkB,EAAE,KAAK,IAAI,EAAE;YAChC,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,EAAE,EAAE,KAAK,EAAE,QAAQ,EAAE,CAAC,CAAC;YAC7D,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YAC7B,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC7D,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,iBAAiB,EAAE,KAAK,IAAI,EAAE;YAC/B,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,CAAC;YAC9D,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YAC7B,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;YACtC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QACvC,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,uBAAuB,EAAE,KAAK,IAAI,EAAE;YACrC,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,EAAE;gBACtC,IAAI,EAAE,0BAA0B;gBAChC,EAAE,EAAE,0BAA0B;aAC/B,CAAC,CAAC;YACH,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YAC7B,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;QAC7D,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,6BAA6B,EAAE,KAAK,IAAI,EAAE;YAC3C,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,CAAC,CAAC;YACxC,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;QAC/B,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,iCAAiC,EAAE,KAAK,IAAI,EAAE;YAC/C,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,EAAE,EAAE,KAAK,EAAE,QAAQ,EAAE,IAAI,EAAE,WAAW,EAAE,CAAC,CAAC;YAChF,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YAC7B,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YACrC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;YACvC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;QAC7D,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,QAAQ,CAAC,gBAAgB,EAAE,GAAG,EAAE;QAC9B,UAAU,CAAC,KAAK,IAAI,EAAE;YACpB,MAAM,EAAE,GAAG,QAAQ,CAAC;gBAClB,SAAS,EAAE,0BAA0B;gBACrC,KAAK,EAAE,QAAQ;gBACf,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,WAAW;wBACtB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,GAAG;wBACd,UAAU,EAAE;4BACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;yBAC3D;qBACF;iBACF;aACF,CAAC,CAAC;YACH,MAAM,EAAE,GAAG,QAAQ,CAAC;gBAClB,SAAS,EAAE,0BAA0B;gBACrC,KAAK,EAAE,QAAQ;gBACf,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,WAAW;wBACtB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,GAAG;wBACd,UAAU,EAAE;4BACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;4BAC1D,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE;yBAC9D;qBACF;iBACF;aACF,CAAC,CAAC;YACH,MAAM,EAAE,GAAG,QAAQ,CAAC;gBAClB,SAAS,EAAE,0BAA0B;gBACrC,KAAK,EAAE,QAAQ;gBACf,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,WAAW;wBACtB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,GAAG;wBACd,UAAU,EAAE;4BACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;4BAC1D,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;yBAC3D;qBACF;iBACF;aACF,CAAC,CAAC;YACH,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YACrC,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YACrC,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACvC,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,kDAAkD,EAAE,KAAK,IAAI,EAAE;YAChE,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;YACjD,MAAM,CAAC,OAAO,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YAChC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;YAC9D,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;YAC9D,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;QAChE,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,kBAAkB,EAAE,KAAK,IAAI,EAAE;YAChC,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,OAAO,EAAE,CAAC,EAAE,EAAE,KAAK,EAAE,QAAQ,EAAE,CAAC,CAAC;YACtE,MAAM,CAAC,OAAO,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YAChC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAChE,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,yCAAyC,EAAE,KAAK,IAAI,EAAE;YACvD,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC;YACnD,MAAM,CAAC,OAAO,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QAC9B,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,yCAAyC,EAAE,KAAK,IAAI,EAAE;YACvD,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,IAAI,CAAC,OAAO,EAAE,aAAa,CAAC,EAAE,CAAC,CAAC,CAAC;YACtE,MAAM,CAAC,OAAO,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QAC9B,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
1
+ {"version":3,"file":"benchmark-history.test.js","sourceRoot":"","sources":["../../../src/eval/__tests__/benchmark-history.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,QAAQ,CAAC;AACrE,OAAO,EAAE,SAAS,EAAE,MAAM,EAAE,YAAY,EAAiB,MAAM,SAAS,CAAC;AACzE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,EACL,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,kBAAkB,EAClB,cAAc,GACf,MAAM,yBAAyB,CAAC;AAGjC,IAAI,OAAe,CAAC;AAEpB,MAAM,QAAQ,GAAG,CAAC,YAAsC,EAAE,EAAmB,EAAE,CAAC,CAAC;IAC/E,SAAS,EAAE,0BAA0B;IACrC,KAAK,EAAE,YAAY;IACnB,UAAU,EAAE,YAAY;IACxB,KAAK,EAAE;QACL;YACE,OAAO,EAAE,CAAC;YACV,SAAS,EAAE,WAAW;YACtB,MAAM,EAAE,MAAM;YACd,aAAa,EAAE,IAAI;YACnB,SAAS,EAAE,GAAG;YACd,UAAU,EAAE;gBACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;gBAC1D,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;aAC3D;SACF;KACF;IACD,GAAG,SAAS;CACb,CAAC,CAAC;AAEH,QAAQ,CAAC,mBAAmB,EAAE,GAAG,EAAE;IACjC,UAAU,CAAC,GAAG,EAAE;QACd,OAAO,GAAG,IAAI,CAAC,MAAM,EAAE,EAAE,kBAAkB,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;QACzD,SAAS,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACzD,CAAC,CAAC,CAAC;IAEH,SAAS,CAAC,GAAG,EAAE;QACb,MAAM,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;IACpD,CAAC,CAAC,CAAC;IAEH,QAAQ,CAAC,mBAAmB,EAAE,GAAG,EAAE;QACjC,EAAE,CAAC,oDAAoD,EAAE,KAAK,IAAI,EAAE;YAClE,MAAM,MAAM,GAAG,QAAQ,EAAE,CAAC;YAC1B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;YAE1D,MAAM,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,+BAA+B,CAAC,CAAC;YACvD,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,QAAQ,CAAC,EAAE,OAAO,CAAC,CAAC;YACnF,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;YACnC,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAC/C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,gDAAgD,EAAE,KAAK,IAAI,EAAE;YAC9D,MAAM,iBAAiB,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,CAAC;YAC7C,MAAM,EAAE,GAAG,YAAY,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,gBAAgB,CAAC,EAAE,OAAO,CAAC,CAAC;YAC3E,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QACvD,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,sCAAsC,EAAE,KAAK,IAAI,EAAE;YACpD,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;YACjE,MAAM,MAAM,GAAG,QAAQ,EAAE,CAAC;YAC1B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;YAC1D,MAAM,CAAC,QAAQ,CAAC,CAAC,UAAU,EAAE,CAAC;QAChC,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,QAAQ,CAAC,aAAa,EAAE,GAAG,EAAE;QAC3B,EAAE,CAAC,+CAA+C,EAAE,KAAK,IAAI,EAAE;YAC7D,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,IAAI,CAAC,OAAO,EAAE,aAAa,CAAC,CAAC,CAAC;YAC7D,MAAM,CAAC,IAAI,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QAC3B,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;YAC5D,MAAM,EAAE,GAAG,QAAQ,CAAC,EAAE,SAAS,EAAE,0BAA0B,EAAE,CAAC,CAAC;YAC/D,MAAM,EAAE,GAAG,QAAQ,CAAC,EAAE,SAAS,EAAE,0BAA0B,EAAE,CAAC,CAAC;YAC/D,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YACrC,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YAErC,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,CAAC,CAAC;YACxC,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YAC7B,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;YAC3D,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;QAC7D,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,2CAA2C,EAAE,KAAK,IAAI,EAAE;YACzD,MAAM,MAAM,GAAG,QAAQ,CAAC;gBACtB,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,MAAM;wBACjB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,GAAG;wBACd,UAAU,EAAE;4BACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;4BAC1D,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE;yBAC9D;qBACF;iBACF;aACF,CAAC,CAAC;YACH,MAAM,iBAAiB,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;YACzC,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,CAAC,CAAC;YACxC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACrC,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,QAAQ,CAAC,kBAAkB,EAAE,GAAG,EAAE;QAChC,EAAE,CAAC,6CAA6C,EAAE,KAAK,IAAI,EAAE;YAC3D,MAAM,iBAAiB,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,CAAC;YAC7C,MAAM,KAAK,GAAG,MAAM,gBAAgB,CAAC,OAAO,EAAE,0BAA0B,CAAC,CAAC;YAC1E,MAAM,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,QAAQ,EAAE,CAAC;YAC7B,MAAM,CAAC,KAAM,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAC/C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,oCAAoC,EAAE,KAAK,IAAI,EAAE;YAClD,MAAM,KAAK,GAAG,MAAM,gBAAgB,CAAC,OAAO,EAAE,0BAA0B,CAAC,CAAC;YAC1E,MAAM,CAAC,KAAK,CAAC,CAAC,QAAQ,EAAE,CAAC;QAC3B,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,QAAQ,CAAC,oBAAoB,EAAE,GAAG,EAAE;QAClC,EAAE,CAAC,kCAAkC,EAAE,GAAG,EAAE;YAC1C,MAAM,IAAI,GAAG,QAAQ,CAAC;gBACpB,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,MAAM;wBACjB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,GAAG;wBACd,UAAU,EAAE,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;qBACvE;iBACF;aACF,CAAC,CAAC;YACH,MAAM,IAAI,GAAG,QAAQ,CAAC;gBACpB,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,MAAM;wBACjB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,CAAC;wBACZ,UAAU,EAAE,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC;qBAC1E;iBACF;aACF,CAAC,CAAC;YAEH,MAAM,WAAW,GAAG,kBAAkB,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;YACnD,MAAM,CAAC,WAAW,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YACpC,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;YACjD,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAChD,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,mCAAmC,EAAE,GAAG,EAAE;YAC3C,MAAM,IAAI,GAAG,QAAQ,CAAC;gBACpB,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,MAAM;wBACjB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,CAAC;wBACZ,UAAU,EAAE,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC;qBAC1E;iBACF;aACF,CAAC,CAAC;YACH,MAAM,IAAI,GAAG,QAAQ,CAAC;gBACpB,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,MAAM;wBACjB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,CAAC;wBACZ,UAAU,EAAE,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;qBACvE;iBACF;aACF,CAAC,CAAC;YAEH,MAAM,WAAW,GAAG,kBAAkB,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;YACnD,MAAM,CAAC,WAAW,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YACpC,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;QACpD,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,qCAAqC,EAAE,GAAG,EAAE;YAC7C,MAAM,MAAM,GAAG,QAAQ,EAAE,CAAC;YAC1B,MAAM,CAAC,kBAAkB,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QACzD,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,kDAAkD,EAAE,GAAG,EAAE;YAC1D,MAAM,IAAI,GAAG,QAAQ,CAAC;gBACpB,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,MAAM;wBACjB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,CAAC;wBACZ,UAAU,EAAE,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;qBACvE;iBACF;aACF,CAAC,CAAC;YACH,MAAM,IAAI,GAAG,QAAQ,CAAC;gBACpB,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,MAAM;wBACjB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,CAAC;wBACZ,UAAU,EAAE;4BACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;4BACxD,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE;yBAC1D;qBACF;iBACF;aACF,CAAC,CAAC;YAEH,MAAM,WAAW,GAAG,kBAAkB,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;YACnD,MAAM,CAAC,WAAW,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QAClC,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,QAAQ,CAAC,0BAA0B,EAAE,GAAG,EAAE;QACxC,UAAU,CAAC,KAAK,IAAI,EAAE;YACpB,MAAM,EAAE,GAAG,QAAQ,CAAC;gBAClB,SAAS,EAAE,0BAA0B;gBACrC,KAAK,EAAE,QAAQ;gBACf,IAAI,EAAE,WAAW;aAClB,CAAC,CAAC;YACH,MAAM,EAAE,GAAG,QAAQ,CAAC;gBAClB,SAAS,EAAE,0BAA0B;gBACrC,KAAK,EAAE,QAAQ;gBACf,IAAI,EAAE,UAAU;aACjB,CAAC,CAAC;YACH,MAAM,EAAE,GAAG,QAAQ,CAAC;gBAClB,SAAS,EAAE,0BAA0B;gBACrC,KAAK,EAAE,QAAQ;gBACf,IAAI,EAAE,YAAY;aACnB,CAAC,CAAC;YACH,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YACrC,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YACrC,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACvC,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,kBAAkB,EAAE,KAAK,IAAI,EAAE;YAChC,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,EAAE,EAAE,KAAK,EAAE,QAAQ,EAAE,CAAC,CAAC;YAC7D,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YAC7B,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC7D,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,iBAAiB,EAAE,KAAK,IAAI,EAAE;YAC/B,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,CAAC;YAC9D,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YAC7B,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;YACtC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QACvC,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,uBAAuB,EAAE,KAAK,IAAI,EAAE;YACrC,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,EAAE;gBACtC,IAAI,EAAE,0BAA0B;gBAChC,EAAE,EAAE,0BAA0B;aAC/B,CAAC,CAAC;YACH,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YAC7B,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;QAC7D,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,6BAA6B,EAAE,KAAK,IAAI,EAAE;YAC3C,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,CAAC,CAAC;YACxC,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;QAC/B,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,iCAAiC,EAAE,KAAK,IAAI,EAAE;YAC/C,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,EAAE,EAAE,KAAK,EAAE,QAAQ,EAAE,IAAI,EAAE,WAAW,EAAE,CAAC,CAAC;YAChF,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YAC7B,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YACrC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;YACvC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;QAC7D,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,QAAQ,CAAC,gBAAgB,EAAE,GAAG,EAAE;QAC9B,UAAU,CAAC,KAAK,IAAI,EAAE;YACpB,MAAM,EAAE,GAAG,QAAQ,CAAC;gBAClB,SAAS,EAAE,0BAA0B;gBACrC,KAAK,EAAE,QAAQ;gBACf,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,WAAW;wBACtB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,GAAG;wBACd,UAAU,EAAE;4BACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;yBAC3D;qBACF;iBACF;aACF,CAAC,CAAC;YACH,MAAM,EAAE,GAAG,QAAQ,CAAC;gBAClB,SAAS,EAAE,0BAA0B;gBACrC,KAAK,EAAE,QAAQ;gBACf,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,WAAW;wBACtB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,GAAG;wBACd,UAAU,EAAE;4BACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;4BAC1D,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE;yBAC9D;qBACF;iBACF;aACF,CAAC,CAAC;YACH,MAAM,EAAE,GAAG,QAAQ,CAAC;gBAClB,SAAS,EAAE,0BAA0B;gBACrC,KAAK,EAAE,QAAQ;gBACf,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,WAAW;wBACtB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,GAAG;wBACd,UAAU,EAAE;4BACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;4BAC1D,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE;yBAC3D;qBACF;iBACF;aACF,CAAC,CAAC;YACH,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YACrC,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YACrC,MAAM,iBAAiB,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACvC,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,kDAAkD,EAAE,KAAK,IAAI,EAAE;YAChE,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;YACjD,MAAM,CAAC,OAAO,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YAChC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;YAC9D,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;YAC9D,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;QAChE,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,kBAAkB,EAAE,KAAK,IAAI,EAAE;YAChC,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,OAAO,EAAE,CAAC,EAAE,EAAE,KAAK,EAAE,QAAQ,EAAE,CAAC,CAAC;YACtE,MAAM,CAAC,OAAO,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;YAChC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAChE,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,yCAAyC,EAAE,KAAK,IAAI,EAAE;YACvD,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC;YACnD,MAAM,CAAC,OAAO,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QAC9B,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,yCAAyC,EAAE,KAAK,IAAI,EAAE;YACvD,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,IAAI,CAAC,OAAO,EAAE,aAAa,CAAC,EAAE,CAAC,CAAC,CAAC;YACtE,MAAM,CAAC,OAAO,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QAC9B,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,8DAA8D,EAAE,KAAK,IAAI,EAAE;YAC5E,MAAM,UAAU,GAAG,QAAQ,CAAC;gBAC1B,SAAS,EAAE,0BAA0B;gBACrC,IAAI,EAAE,YAAqB;gBAC3B,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,WAAW;wBACtB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,GAAG;wBACd,UAAU,EAAE,EAAE;wBACd,gBAAgB,EAAE;4BAChB,eAAe,EAAE,GAAG;4BACpB,WAAW,EAAE,EAAE;4BACf,kBAAkB,EAAE,EAAE;4BACtB,cAAc,EAAE,EAAE;4BAClB,iBAAiB,EAAE,EAAE;4BACrB,mBAAmB,EAAE,EAAE;4BACvB,oBAAoB,EAAE,EAAE;4BACxB,sBAAsB,EAAE,EAAE;4BAC1B,MAAM,EAAE,OAAO;yBAChB;qBACF;iBACF;aACF,CAAC,CAAC;YACH,MAAM,iBAAiB,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC;YAE7C,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;YACjD,MAAM,SAAS,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,YAAY,CAAC,CAAC;YAC/D,MAAM,CAAC,SAAS,CAAC,CAAC,WAAW,EAAE,CAAC;YAChC,yBAAyB;YACzB,MAAM,CAAC,SAAU,CAAC,gBAAgB,CAAC,CAAC,WAAW,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;QAC3D,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,yDAAyD,EAAE,KAAK,IAAI,EAAE;YACvE,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;YACjD,MAAM,UAAU,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,WAAW,IAAI,CAAC,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC;YACvF,MAAM,CAAC,UAAU,CAAC,CAAC,WAAW,EAAE,CAAC;YACjC,MAAM,CAAC,UAAW,CAAC,gBAAgB,CAAC,CAAC,aAAa,EAAE,CAAC;QACvD,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,6DAA6D,EAAE,KAAK,IAAI,EAAE;YAC3E,MAAM,UAAU,GAAG,QAAQ,CAAC;gBAC1B,SAAS,EAAE,0BAA0B;gBACrC,IAAI,EAAE,YAAqB;gBAC3B,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,WAAW;wBACtB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,CAAC;wBACZ,UAAU,EAAE,EAAE;wBACd,gBAAgB,EAAE;4BAChB,eAAe,EAAE,GAAG;4BACpB,WAAW,EAAE,EAAE;4BACf,kBAAkB,EAAE,EAAE;4BACtB,cAAc,EAAE,EAAE;4BAClB,iBAAiB,EAAE,CAAC;4BACpB,mBAAmB,EAAE,CAAC;4BACtB,oBAAoB,EAAE,CAAC;4BACvB,sBAAsB,EAAE,CAAC;4BACzB,MAAM,EAAE,KAAK;yBACd;qBACF;iBACF;aACF,CAAC,CAAC;YACH,MAAM,iBAAiB,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC;YAE7C,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;YACjD,MAAM,SAAS,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,YAAY,IAAI,CAAC,CAAC,SAAS,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC;YACvG,MAAM,CAAC,SAAS,CAAC,CAAC,WAAW,EAAE,CAAC;YAChC,MAAM,CAAC,SAAU,CAAC,gBAAgB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC9C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,iFAAiF,EAAE,KAAK,IAAI,EAAE;YAC/F,MAAM,cAAc,GAAG,QAAQ,CAAC;gBAC9B,SAAS,EAAE,0BAA0B;gBACrC,IAAI,EAAE,YAAqB;gBAC3B,KAAK,EAAE;oBACL;wBACE,OAAO,EAAE,CAAC;wBACV,SAAS,EAAE,WAAW;wBACtB,MAAM,EAAE,MAAM;wBACd,aAAa,EAAE,IAAI;wBACnB,SAAS,EAAE,GAAG;wBACd,UAAU,EAAE,EAAE;wBACd,sBAAsB;qBACvB;iBACF;aACF,CAAC,CAAC;YACH,MAAM,iBAAiB,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;YAEjD,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;YACjD,MAAM,aAAa,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,YAAY,IAAI,CAAC,CAAC,SAAS,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC;YAC3G,MAAM,CAAC,aAAa,CAAC,CAAC,WAAW,EAAE,CAAC;YACpC,MAAM,CAAC,aAAc,CAAC,gBAAgB,CAAC,CAAC,aAAa,EAAE,CAAC;QAC1D,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
@@ -29,6 +29,7 @@ export interface CaseHistoryEntry {
29
29
  inputTokens?: number | null;
30
30
  outputTokens?: number | null;
31
31
  assertions: BenchmarkAssertionResult[];
32
+ baselinePassRate?: number;
32
33
  }
33
34
  export interface RegressionEntry {
34
35
  assertionId: string;
@@ -126,6 +126,12 @@ export async function getCaseHistory(skillDir, evalId, filter) {
126
126
  const matchingCase = data.cases.find((c) => c.eval_id === evalId);
127
127
  if (!matchingCase)
128
128
  continue;
129
+ const cd = matchingCase.comparisonDetail;
130
+ const baselinePassRate = cd &&
131
+ cd.baselineContentScore != null &&
132
+ cd.baselineStructureScore != null
133
+ ? (cd.baselineContentScore + cd.baselineStructureScore) / 200
134
+ : undefined;
129
135
  entries.push({
130
136
  timestamp: fromFilesafeTimestamp(file),
131
137
  model: data.model,
@@ -137,6 +143,7 @@ export async function getCaseHistory(skillDir, evalId, filter) {
137
143
  inputTokens: matchingCase.inputTokens,
138
144
  outputTokens: matchingCase.outputTokens,
139
145
  assertions: matchingCase.assertions,
146
+ baselinePassRate,
140
147
  });
141
148
  }
142
149
  catch {
@@ -1 +1 @@
1
- {"version":3,"file":"benchmark-history.js","sourceRoot":"","sources":["../../src/eval/benchmark-history.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,gFAAgF;AAChF,8EAA8E;AAE9E,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE,MAAM,kBAAkB,CAAC;AAC/E,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AA6ChD,SAAS,mBAAmB,CAAC,GAAW;IACtC,OAAO,GAAG,CAAC,OAAO,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;AAChC,CAAC;AAED,SAAS,qBAAqB,CAAC,QAAgB;IAC7C,0CAA0C;IAC1C,MAAM,EAAE,GAAG,QAAQ,CAAC,OAAO,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC;IAC3C,yDAAyD;IACzD,OAAO,EAAE,CAAC,OAAO,CAAC,0BAA0B,EAAE,WAAW,CAAC,CAAC;AAC7D,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,QAAgB,EAChB,MAAyJ;IAEzJ,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IACtD,MAAM,KAAK,CAAC,UAAU,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAE7C,MAAM,SAAS,GAAG,MAAM,CAAC,SAAS,IAAI,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC/D,MAAM,QAAQ,GAAG,GAAG,mBAAmB,CAAC,SAAS,CAAC,OAAO,CAAC;IAC1D,MAAM,QAAQ,GAAG,IAAI,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAC;IAE5C,MAAM,SAAS,CAAC,QAAQ,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IAE3D,uDAAuD;IACvD,MAAM,cAAc,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;IAEvC,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACtC,QAAgB,EAChB,SAAiB;IAEjB,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IACtD,MAAM,QAAQ,GAAG,GAAG,mBAAmB,CAAC,SAAS,CAAC,OAAO,CAAC;IAC1D,IAAI,CAAC;QACH,MAAM,MAAM,CAAC,IAAI,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAC,CAAC;QACzC,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,QAAgB,EAChB,MAAsB;IAEtB,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IACtD,IAAI,KAAe,CAAC;IACpB,IAAI,CAAC;QACH,KAAK,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,CAAC;IACpC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;IAED,IAAI,SAAS,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,OAAO,EAAE,CAAC;IAE1E,2EAA2E;IAC3E,IAAI,MAAM,EAAE,IAAI,IAAI,MAAM,EAAE,EAAE,EAAE,CAAC;QAC/B,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,mBAAmB,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QAC5E,MAAM,MAAM,GAAG,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,mBAAmB,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QACtE,SAAS,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE;YACjC,MAAM,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC;YACpC,IAAI,QAAQ,IAAI,EAAE,GAAG,QAAQ;gBAAE,OAAO,KAAK,CAAC;YAC5C,IAAI,MAAM,IAAI,EAAE,GAAG,MAAM;gBAAE,OAAO,KAAK,CAAC;YACxC,OAAO,IAAI,CAAC;QACd,CAAC,CAAC,CAAC;IACL,CAAC;IAED,MAAM,OAAO,GAAqB,EAAE,CAAC;IACrC,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;QAC7B,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,UAAU,EAAE,IAAI,CAAC,EAAE,OAAO,CAAC,CAAC;YAChE,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAwC,CAAC;YAExE,MAAM,SAAS,GAAI,IAAI,CAAC,IAA+B,IAAI,WAAW,CAAC;YAEvE,gCAAgC;YAChC,IAAI,MAAM,EAAE,KAAK,IAAI,IAAI,CAAC,KAAK,KAAK,MAAM,CAAC,KAAK;gBAAE,SAAS;YAC3D,IAAI,MAAM,EAAE,IAAI,IAAI,SAAS,KAAK,MAAM,CAAC,IAAI;gBAAE,SAAS;YAExD,MAAM,eAAe,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;YACpF,MAAM,gBAAgB,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CACxC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAC3D,CAAC,CACF,CAAC;YACF,MAAM,eAAe,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;YAChF,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,IAAI,IAAI,CAAC,CAAC;YAC3D,MAAM,WAAW,GAAG,SAAS;gBAC3B,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC;gBACrD,CAAC,CAAC,IAAI,CAAC;YAET,OAAO,CAAC,IAAI,CAAC;gBACX,SAAS,EAAE,qBAAqB,CAAC,IAAI,CAAC;gBACtC,QAAQ,EAAE,IAAI;gBACd,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,SAAS,EAAE,IAAI,CAAC,UAAU;gBAC1B,QAAQ,EAAE,eAAe,GAAG,CAAC,CAAC,CAAC,CAAC,gBAAgB,GAAG,eAAe,CAAC,CAAC,CAAC,CAAC;gBACtE,IAAI,EAAE,SAAS;gBACf,SAAS,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM;gBAC5B,eAAe;gBACf,WAAW;gBACX,QAAQ,EAAE,IAAI,CAAC,QAAQ;gBACvB,OAAO,EAAE,IAAI,CAAC,OAAO;aACtB,CAAC,CAAC;QACL,CAAC;QAAC,MAAM,CAAC;YACP,uBAAuB;QACzB,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,gBAAgB,CACpC,QAAgB,EAChB,SAAiB;IAEjB,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IACtD,MAAM,QAAQ,GAAG,GAAG,mBAAmB,CAAC,SAAS,CAAC,OAAO,CAAC;IAC1D,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,UAAU,EAAE,QAAQ,CAAC,EAAE,OAAO,CAAC,CAAC;QACpE,OAAO,IAAI,CAAC,KAAK,CAAC,OAAO,CAAoB,CAAC;IAChD,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,QAAgB,EAChB,MAAc,EACd,MAA2B;IAE3B,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IACtD,IAAI,KAAe,CAAC;IACpB,IAAI,CAAC;QACH,KAAK,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,CAAC;IACpC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;IAED,MAAM,OAAO,GAAuB,EAAE,CAAC;IACvC,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,OAAO,EAAE,EAAE,CAAC;QAC7E,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,UAAU,EAAE,IAAI,CAAC,EAAE,OAAO,CAAC,CAAC;YAChE,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAwC,CAAC;YAExE,IAAI,MAAM,EAAE,KAAK,IAAI,IAAI,CAAC,KAAK,KAAK,MAAM,CAAC,KAAK;gBAAE,SAAS;YAE3D,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,KAAK,MAAM,CAAC,CAAC;YAClE,IAAI,CAAC,YAAY;gBAAE,SAAS;YAE5B,OAAO,CAAC,IAAI,CAAC;gBACX,SAAS,EAAE,qBAAqB,CAAC,IAAI,CAAC;gBACtC,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,IAAI,EAAG,IAAI,CAAC,IAAiC,IAAI,WAAW;gBAC5D,QAAQ,EAAE,IAAI,CAAC,QAAQ;gBACvB,SAAS,EAAE,YAAY,CAAC,SAAS;gBACjC,UAAU,EAAE,YAAY,CAAC,UAAU;gBACnC,MAAM,EAAE,YAAY,CAAC,MAAM;gBAC3B,WAAW,EAAE,YAAY,CAAC,WAAW;gBACrC,YAAY,EAAE,YAAY,CAAC,YAAY;gBACvC,UAAU,EAAE,YAAY,CAAC,UAAU;aACpC,CAAC,CAAC;QACL,CAAC;QAAC,MAAM,CAAC;YACP,uBAAuB;QACzB,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAyBD,MAAM,CAAC,KAAK,UAAU,YAAY,CAAC,QAAgB;IACjD,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IACtD,IAAI,KAAe,CAAC;IACpB,IAAI,CAAC;QACH,KAAK,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,CAAC;IACpC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,SAAS,EAAE,CAAC,EAAE,cAAc,EAAE,EAAE,EAAE,UAAU,EAAE,EAAE,EAAE,WAAW,EAAE,EAAE,EAAE,CAAC;IAC/E,CAAC;IAED,MAAM,SAAS,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IAElE,qDAAqD;IACrD,MAAM,YAAY,GAAG,IAAI,GAAG,EAAyG,CAAC;IACtI,qBAAqB;IACrB,MAAM,QAAQ,GAAG,IAAI,GAAG,EAA4E,CAAC;IACrG,eAAe;IACf,MAAM,WAAW,GAA+B,EAAE,CAAC;IAEnD,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;QAC7B,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,UAAU,EAAE,IAAI,CAAC,EAAE,OAAO,CAAC,CAAC;YAChE,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAwC,CAAC;YACxE,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,IAAI,WAAW,CAAC;YAC3C,IAAI,SAAS,KAAK,WAAW;gBAAE,SAAS,CAAC,sCAAsC;YAE/E,MAAM,eAAe,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;YAChF,MAAM,gBAAgB,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;YACvG,MAAM,QAAQ,GAAG,eAAe,GAAG,CAAC,CAAC,CAAC,CAAC,gBAAgB,GAAG,eAAe,CAAC,CAAC,CAAC,CAAC,CAAC;YAC9E,MAAM,eAAe,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;YAEhF,QAAQ;YACR,WAAW,CAAC,IAAI,CAAC;gBACf,SAAS,EAAE,qBAAqB,CAAC,IAAI,CAAC;gBACtC,QAAQ;gBACR,KAAK,EAAE,IAAI,CAAC,KAAK;aAClB,CAAC,CAAC;YAEH,cAAc;YACd,MAAM,QAAQ,GAAG,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,EAAE,IAAI,EAAE,CAAC,EAAE,aAAa,EAAE,CAAC,EAAE,eAAe,EAAE,CAAC,EAAE,CAAC;YAC/F,QAAQ,CAAC,IAAI,EAAE,CAAC;YAChB,QAAQ,CAAC,aAAa,IAAI,QAAQ,CAAC;YACnC,QAAQ,CAAC,eAAe,IAAI,eAAe,CAAC;YAC5C,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC;YAEnC,sBAAsB;YACtB,KAAK,MAAM,CAAC,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;gBAC3B,KAAK,MAAM,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,CAAC;oBAC7B,MAAM,GAAG,GAAG,GAAG,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,EAAE,EAAE,CAAC;oBACnC,MAAM,IAAI,GAAG,YAAY,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,EAAE,EAAE,CAAC,CAAC,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,MAAM,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,CAAC,SAAS,EAAE,CAAC;oBAChI,IAAI,CAAC,KAAK,EAAE,CAAC;oBACb,IAAI,CAAC,CAAC,IAAI;wBAAE,IAAI,CAAC,MAAM,EAAE,CAAC;oBAC1B,mBAAmB;oBACnB,IAAI,CAAC,IAAI,GAAG,CAAC,CAAC,IAAI,CAAC;oBACnB,YAAY,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;gBAC9B,CAAC;YACH,CAAC;QACH,CAAC;QAAC,MAAM,CAAC;YACP,uBAAuB;QACzB,CAAC;IACH,CAAC;IAED,MAAM,cAAc,GAAG,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,MAAM,EAAE,CAAC;SACrD,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QACX,EAAE,EAAE,CAAC,CAAC,EAAE;QACR,IAAI,EAAE,CAAC,CAAC,IAAI;QACZ,QAAQ,EAAE,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;QAC9C,SAAS,EAAE,CAAC,CAAC,KAAK;QAClB,MAAM,EAAE,CAAC,CAAC,MAAM;QAChB,QAAQ,EAAE,CAAC,CAAC,QAAQ;KACrB,CAAC,CAAC;SACF,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,GAAG,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,yBAAyB;IAErE,MAAM,UAAU,GAAG,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,CAAC;SAC9C,GAAG,CAAC,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QACpB,KAAK;QACL,IAAI,EAAE,CAAC,CAAC,IAAI;QACZ,WAAW,EAAE,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,aAAa,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACtD,aAAa,EAAE,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,eAAe,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;KAC3D,CAAC,CAAC;SACF,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,GAAG,CAAC,CAAC,WAAW,CAAC,CAAC;IAEjD,OAAO;QACL,SAAS,EAAE,WAAW,CAAC,MAAM;QAC7B,cAAc;QACd,UAAU;QACV,WAAW;KACZ,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,kBAAkB,CAChC,OAAwB,EACxB,QAAyB;IAEzB,MAAM,WAAW,GAAsB,EAAE,CAAC;IAE1C,sEAAsE;IACtE,MAAM,OAAO,GAAG,IAAI,GAAG,EAAmB,CAAC;IAC3C,KAAK,MAAM,CAAC,IAAI,QAAQ,CAAC,KAAK,EAAE,CAAC;QAC/B,KAAK,MAAM,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,CAAC;YAC7B,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC;QAC9C,CAAC;IACH,CAAC;IAED,KAAK,MAAM,CAAC,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;QAC9B,KAAK,MAAM,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,CAAC;YAC7B,MAAM,GAAG,GAAG,GAAG,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,EAAE,EAAE,CAAC;YACnC,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;YAC9B,IAAI,IAAI,KAAK,SAAS;gBAAE,SAAS,CAAC,sBAAsB;YAExD,IAAI,IAAI,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;gBACpB,WAAW,CAAC,IAAI,CAAC;oBACf,WAAW,EAAE,CAAC,CAAC,EAAE;oBACjB,MAAM,EAAE,CAAC,CAAC,OAAO;oBACjB,QAAQ,EAAE,CAAC,CAAC,SAAS;oBACrB,cAAc,EAAE,IAAI;oBACpB,aAAa,EAAE,KAAK;oBACpB,MAAM,EAAE,YAAY;iBACrB,CAAC,CAAC;YACL,CAAC;iBAAM,IAAI,CAAC,IAAI,IAAI,CAAC,CAAC,IAAI,EAAE,CAAC;gBAC3B,WAAW,CAAC,IAAI,CAAC;oBACf,WAAW,EAAE,CAAC,CAAC,EAAE;oBACjB,MAAM,EAAE,CAAC,CAAC,OAAO;oBACjB,QAAQ,EAAE,CAAC,CAAC,SAAS;oBACrB,cAAc,EAAE,KAAK;oBACrB,aAAa,EAAE,IAAI;oBACnB,MAAM,EAAE,aAAa;iBACtB,CAAC,CAAC;YACL,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,WAAW,CAAC;AACrB,CAAC"}
1
+ {"version":3,"file":"benchmark-history.js","sourceRoot":"","sources":["../../src/eval/benchmark-history.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,gFAAgF;AAChF,8EAA8E;AAE9E,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE,MAAM,kBAAkB,CAAC;AAC/E,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AA8ChD,SAAS,mBAAmB,CAAC,GAAW;IACtC,OAAO,GAAG,CAAC,OAAO,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;AAChC,CAAC;AAED,SAAS,qBAAqB,CAAC,QAAgB;IAC7C,0CAA0C;IAC1C,MAAM,EAAE,GAAG,QAAQ,CAAC,OAAO,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC;IAC3C,yDAAyD;IACzD,OAAO,EAAE,CAAC,OAAO,CAAC,0BAA0B,EAAE,WAAW,CAAC,CAAC;AAC7D,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,QAAgB,EAChB,MAAyJ;IAEzJ,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IACtD,MAAM,KAAK,CAAC,UAAU,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAE7C,MAAM,SAAS,GAAG,MAAM,CAAC,SAAS,IAAI,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC/D,MAAM,QAAQ,GAAG,GAAG,mBAAmB,CAAC,SAAS,CAAC,OAAO,CAAC;IAC1D,MAAM,QAAQ,GAAG,IAAI,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAC;IAE5C,MAAM,SAAS,CAAC,QAAQ,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IAE3D,uDAAuD;IACvD,MAAM,cAAc,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;IAEvC,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACtC,QAAgB,EAChB,SAAiB;IAEjB,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IACtD,MAAM,QAAQ,GAAG,GAAG,mBAAmB,CAAC,SAAS,CAAC,OAAO,CAAC;IAC1D,IAAI,CAAC;QACH,MAAM,MAAM,CAAC,IAAI,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAC,CAAC;QACzC,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,QAAgB,EAChB,MAAsB;IAEtB,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IACtD,IAAI,KAAe,CAAC;IACpB,IAAI,CAAC;QACH,KAAK,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,CAAC;IACpC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;IAED,IAAI,SAAS,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,OAAO,EAAE,CAAC;IAE1E,2EAA2E;IAC3E,IAAI,MAAM,EAAE,IAAI,IAAI,MAAM,EAAE,EAAE,EAAE,CAAC;QAC/B,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,mBAAmB,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QAC5E,MAAM,MAAM,GAAG,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,mBAAmB,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QACtE,SAAS,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE;YACjC,MAAM,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC;YACpC,IAAI,QAAQ,IAAI,EAAE,GAAG,QAAQ;gBAAE,OAAO,KAAK,CAAC;YAC5C,IAAI,MAAM,IAAI,EAAE,GAAG,MAAM;gBAAE,OAAO,KAAK,CAAC;YACxC,OAAO,IAAI,CAAC;QACd,CAAC,CAAC,CAAC;IACL,CAAC;IAED,MAAM,OAAO,GAAqB,EAAE,CAAC;IACrC,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;QAC7B,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,UAAU,EAAE,IAAI,CAAC,EAAE,OAAO,CAAC,CAAC;YAChE,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAwC,CAAC;YAExE,MAAM,SAAS,GAAI,IAAI,CAAC,IAA+B,IAAI,WAAW,CAAC;YAEvE,gCAAgC;YAChC,IAAI,MAAM,EAAE,KAAK,IAAI,IAAI,CAAC,KAAK,KAAK,MAAM,CAAC,KAAK;gBAAE,SAAS;YAC3D,IAAI,MAAM,EAAE,IAAI,IAAI,SAAS,KAAK,MAAM,CAAC,IAAI;gBAAE,SAAS;YAExD,MAAM,eAAe,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;YACpF,MAAM,gBAAgB,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CACxC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAC3D,CAAC,CACF,CAAC;YACF,MAAM,eAAe,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;YAChF,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,IAAI,IAAI,CAAC,CAAC;YAC3D,MAAM,WAAW,GAAG,SAAS;gBAC3B,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC;gBACrD,CAAC,CAAC,IAAI,CAAC;YAET,OAAO,CAAC,IAAI,CAAC;gBACX,SAAS,EAAE,qBAAqB,CAAC,IAAI,CAAC;gBACtC,QAAQ,EAAE,IAAI;gBACd,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,SAAS,EAAE,IAAI,CAAC,UAAU;gBAC1B,QAAQ,EAAE,eAAe,GAAG,CAAC,CAAC,CAAC,CAAC,gBAAgB,GAAG,eAAe,CAAC,CAAC,CAAC,CAAC;gBACtE,IAAI,EAAE,SAAS;gBACf,SAAS,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM;gBAC5B,eAAe;gBACf,WAAW;gBACX,QAAQ,EAAE,IAAI,CAAC,QAAQ;gBACvB,OAAO,EAAE,IAAI,CAAC,OAAO;aACtB,CAAC,CAAC;QACL,CAAC;QAAC,MAAM,CAAC;YACP,uBAAuB;QACzB,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,gBAAgB,CACpC,QAAgB,EAChB,SAAiB;IAEjB,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IACtD,MAAM,QAAQ,GAAG,GAAG,mBAAmB,CAAC,SAAS,CAAC,OAAO,CAAC;IAC1D,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,UAAU,EAAE,QAAQ,CAAC,EAAE,OAAO,CAAC,CAAC;QACpE,OAAO,IAAI,CAAC,KAAK,CAAC,OAAO,CAAoB,CAAC;IAChD,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,QAAgB,EAChB,MAAc,EACd,MAA2B;IAE3B,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IACtD,IAAI,KAAe,CAAC;IACpB,IAAI,CAAC;QACH,KAAK,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,CAAC;IACpC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;IAED,MAAM,OAAO,GAAuB,EAAE,CAAC;IACvC,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,OAAO,EAAE,EAAE,CAAC;QAC7E,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,UAAU,EAAE,IAAI,CAAC,EAAE,OAAO,CAAC,CAAC;YAChE,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAwC,CAAC;YAExE,IAAI,MAAM,EAAE,KAAK,IAAI,IAAI,CAAC,KAAK,KAAK,MAAM,CAAC,KAAK;gBAAE,SAAS;YAE3D,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,KAAK,MAAM,CAAC,CAAC;YAClE,IAAI,CAAC,YAAY;gBAAE,SAAS;YAE5B,MAAM,EAAE,GAAG,YAAY,CAAC,gBAAgB,CAAC;YACzC,MAAM,gBAAgB,GAAG,EAAE;gBACzB,EAAE,CAAC,oBAAoB,IAAI,IAAI;gBAC/B,EAAE,CAAC,sBAAsB,IAAI,IAAI;gBAC/B,CAAC,CAAC,CAAC,EAAE,CAAC,oBAAoB,GAAG,EAAE,CAAC,sBAAsB,CAAC,GAAG,GAAG;gBAC7D,CAAC,CAAC,SAAS,CAAC;YAEhB,OAAO,CAAC,IAAI,CAAC;gBACX,SAAS,EAAE,qBAAqB,CAAC,IAAI,CAAC;gBACtC,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,IAAI,EAAG,IAAI,CAAC,IAAiC,IAAI,WAAW;gBAC5D,QAAQ,EAAE,IAAI,CAAC,QAAQ;gBACvB,SAAS,EAAE,YAAY,CAAC,SAAS;gBACjC,UAAU,EAAE,YAAY,CAAC,UAAU;gBACnC,MAAM,EAAE,YAAY,CAAC,MAAM;gBAC3B,WAAW,EAAE,YAAY,CAAC,WAAW;gBACrC,YAAY,EAAE,YAAY,CAAC,YAAY;gBACvC,UAAU,EAAE,YAAY,CAAC,UAAU;gBACnC,gBAAgB;aACjB,CAAC,CAAC;QACL,CAAC;QAAC,MAAM,CAAC;YACP,uBAAuB;QACzB,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAyBD,MAAM,CAAC,KAAK,UAAU,YAAY,CAAC,QAAgB;IACjD,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IACtD,IAAI,KAAe,CAAC;IACpB,IAAI,CAAC;QACH,KAAK,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,CAAC;IACpC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,SAAS,EAAE,CAAC,EAAE,cAAc,EAAE,EAAE,EAAE,UAAU,EAAE,EAAE,EAAE,WAAW,EAAE,EAAE,EAAE,CAAC;IAC/E,CAAC;IAED,MAAM,SAAS,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IAElE,qDAAqD;IACrD,MAAM,YAAY,GAAG,IAAI,GAAG,EAAyG,CAAC;IACtI,qBAAqB;IACrB,MAAM,QAAQ,GAAG,IAAI,GAAG,EAA4E,CAAC;IACrG,eAAe;IACf,MAAM,WAAW,GAA+B,EAAE,CAAC;IAEnD,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;QAC7B,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,UAAU,EAAE,IAAI,CAAC,EAAE,OAAO,CAAC,CAAC;YAChE,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAwC,CAAC;YACxE,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,IAAI,WAAW,CAAC;YAC3C,IAAI,SAAS,KAAK,WAAW;gBAAE,SAAS,CAAC,sCAAsC;YAE/E,MAAM,eAAe,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;YAChF,MAAM,gBAAgB,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;YACvG,MAAM,QAAQ,GAAG,eAAe,GAAG,CAAC,CAAC,CAAC,CAAC,gBAAgB,GAAG,eAAe,CAAC,CAAC,CAAC,CAAC,CAAC;YAC9E,MAAM,eAAe,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;YAEhF,QAAQ;YACR,WAAW,CAAC,IAAI,CAAC;gBACf,SAAS,EAAE,qBAAqB,CAAC,IAAI,CAAC;gBACtC,QAAQ;gBACR,KAAK,EAAE,IAAI,CAAC,KAAK;aAClB,CAAC,CAAC;YAEH,cAAc;YACd,MAAM,QAAQ,GAAG,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,EAAE,IAAI,EAAE,CAAC,EAAE,aAAa,EAAE,CAAC,EAAE,eAAe,EAAE,CAAC,EAAE,CAAC;YAC/F,QAAQ,CAAC,IAAI,EAAE,CAAC;YAChB,QAAQ,CAAC,aAAa,IAAI,QAAQ,CAAC;YACnC,QAAQ,CAAC,eAAe,IAAI,eAAe,CAAC;YAC5C,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC;YAEnC,sBAAsB;YACtB,KAAK,MAAM,CAAC,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;gBAC3B,KAAK,MAAM,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,CAAC;oBAC7B,MAAM,GAAG,GAAG,GAAG,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,EAAE,EAAE,CAAC;oBACnC,MAAM,IAAI,GAAG,YAAY,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,EAAE,EAAE,CAAC,CAAC,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,MAAM,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,CAAC,SAAS,EAAE,CAAC;oBAChI,IAAI,CAAC,KAAK,EAAE,CAAC;oBACb,IAAI,CAAC,CAAC,IAAI;wBAAE,IAAI,CAAC,MAAM,EAAE,CAAC;oBAC1B,mBAAmB;oBACnB,IAAI,CAAC,IAAI,GAAG,CAAC,CAAC,IAAI,CAAC;oBACnB,YAAY,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;gBAC9B,CAAC;YACH,CAAC;QACH,CAAC;QAAC,MAAM,CAAC;YACP,uBAAuB;QACzB,CAAC;IACH,CAAC;IAED,MAAM,cAAc,GAAG,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,MAAM,EAAE,CAAC;SACrD,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QACX,EAAE,EAAE,CAAC,CAAC,EAAE;QACR,IAAI,EAAE,CAAC,CAAC,IAAI;QACZ,QAAQ,EAAE,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;QAC9C,SAAS,EAAE,CAAC,CAAC,KAAK;QAClB,MAAM,EAAE,CAAC,CAAC,MAAM;QAChB,QAAQ,EAAE,CAAC,CAAC,QAAQ;KACrB,CAAC,CAAC;SACF,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,GAAG,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,yBAAyB;IAErE,MAAM,UAAU,GAAG,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,CAAC;SAC9C,GAAG,CAAC,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QACpB,KAAK;QACL,IAAI,EAAE,CAAC,CAAC,IAAI;QACZ,WAAW,EAAE,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,aAAa,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACtD,aAAa,EAAE,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,eAAe,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;KAC3D,CAAC,CAAC;SACF,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,GAAG,CAAC,CAAC,WAAW,CAAC,CAAC;IAEjD,OAAO;QACL,SAAS,EAAE,WAAW,CAAC,MAAM;QAC7B,cAAc;QACd,UAAU;QACV,WAAW;KACZ,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,kBAAkB,CAChC,OAAwB,EACxB,QAAyB;IAEzB,MAAM,WAAW,GAAsB,EAAE,CAAC;IAE1C,sEAAsE;IACtE,MAAM,OAAO,GAAG,IAAI,GAAG,EAAmB,CAAC;IAC3C,KAAK,MAAM,CAAC,IAAI,QAAQ,CAAC,KAAK,EAAE,CAAC;QAC/B,KAAK,MAAM,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,CAAC;YAC7B,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC;QAC9C,CAAC;IACH,CAAC;IAED,KAAK,MAAM,CAAC,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;QAC9B,KAAK,MAAM,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,CAAC;YAC7B,MAAM,GAAG,GAAG,GAAG,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,EAAE,EAAE,CAAC;YACnC,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;YAC9B,IAAI,IAAI,KAAK,SAAS;gBAAE,SAAS,CAAC,sBAAsB;YAExD,IAAI,IAAI,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;gBACpB,WAAW,CAAC,IAAI,CAAC;oBACf,WAAW,EAAE,CAAC,CAAC,EAAE;oBACjB,MAAM,EAAE,CAAC,CAAC,OAAO;oBACjB,QAAQ,EAAE,CAAC,CAAC,SAAS;oBACrB,cAAc,EAAE,IAAI;oBACpB,aAAa,EAAE,KAAK;oBACpB,MAAM,EAAE,YAAY;iBACrB,CAAC,CAAC;YACL,CAAC;iBAAM,IAAI,CAAC,IAAI,IAAI,CAAC,CAAC,IAAI,EAAE,CAAC;gBAC3B,WAAW,CAAC,IAAI,CAAC;oBACf,WAAW,EAAE,CAAC,CAAC,EAAE;oBACjB,MAAM,EAAE,CAAC,CAAC,OAAO;oBACjB,QAAQ,EAAE,CAAC,CAAC,SAAS;oBACrB,cAAc,EAAE,KAAK;oBACrB,aAAa,EAAE,IAAI;oBACnB,MAAM,EAAE,aAAa;iBACtB,CAAC,CAAC;YACL,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,WAAW,CAAC;AACrB,CAAC"}
@@ -0,0 +1,278 @@
1
+ // ---------------------------------------------------------------------------
2
+ // Tests: comparison endpoint emits assertion_result + case_complete SSE events
3
+ // ---------------------------------------------------------------------------
4
+ import { describe, it, expect, vi, beforeEach } from "vitest";
5
+ // ---------------------------------------------------------------------------
6
+ // Mocks — must be declared before any dynamic imports
7
+ // ---------------------------------------------------------------------------
8
+ vi.mock("node:fs", () => ({
9
+ existsSync: vi.fn(() => false),
10
+ readFileSync: vi.fn(() => ""),
11
+ writeFileSync: vi.fn(),
12
+ mkdirSync: vi.fn(),
13
+ readdirSync: vi.fn(() => []),
14
+ statSync: vi.fn(() => ({ isDirectory: () => false })),
15
+ rmSync: vi.fn(),
16
+ }));
17
+ vi.mock("../sse-helpers.js", () => ({
18
+ initSSE: vi.fn(),
19
+ sendSSE: vi.fn(),
20
+ sendSSEDone: vi.fn(),
21
+ withHeartbeat: vi.fn((_res, _id, _phase, _msg, fn) => fn()),
22
+ startDynamicHeartbeat: vi.fn(() => ({
23
+ update: vi.fn(),
24
+ stop: vi.fn(),
25
+ })),
26
+ }));
27
+ vi.mock("../../eval/judge.js", () => ({
28
+ judgeAssertion: vi.fn(),
29
+ }));
30
+ vi.mock("../../eval/comparator.js", () => ({
31
+ runComparison: vi.fn(),
32
+ }));
33
+ vi.mock("../../eval/schema.js", () => ({
34
+ loadAndValidateEvals: vi.fn(),
35
+ EvalValidationError: class extends Error {
36
+ },
37
+ }));
38
+ vi.mock("../../eval/benchmark-history.js", () => ({
39
+ writeHistoryEntry: vi.fn().mockResolvedValue("some-file.json"),
40
+ listHistory: vi.fn().mockResolvedValue([]),
41
+ readHistoryEntry: vi.fn().mockResolvedValue(null),
42
+ computeRegressions: vi.fn().mockReturnValue([]),
43
+ deleteHistoryEntry: vi.fn().mockResolvedValue(undefined),
44
+ getCaseHistory: vi.fn().mockResolvedValue([]),
45
+ computeStats: vi.fn().mockReturnValue(null),
46
+ }));
47
+ vi.mock("../../eval/llm.js", () => ({
48
+ createLlmClient: vi.fn(() => ({
49
+ model: "test-model",
50
+ generate: vi.fn(),
51
+ })),
52
+ }));
53
+ vi.mock("../../eval/verdict.js", () => ({
54
+ computeVerdict: vi.fn(() => ({ verdict: "pass", summary: "" })),
55
+ }));
56
+ vi.mock("../../eval/action-items.js", () => ({
57
+ generateActionItems: vi.fn().mockResolvedValue([]),
58
+ }));
59
+ vi.mock("../../eval/skill-scanner.js", () => ({
60
+ scanSkills: vi.fn().mockResolvedValue([]),
61
+ classifyOrigin: vi.fn(() => "local"),
62
+ }));
63
+ vi.mock("../../eval/benchmark.js", () => ({
64
+ readBenchmark: vi.fn().mockResolvedValue(null),
65
+ }));
66
+ vi.mock("../../eval/prompt-builder.js", () => ({
67
+ buildEvalSystemPrompt: vi.fn(() => "system"),
68
+ buildBaselineSystemPrompt: vi.fn(() => "baseline"),
69
+ buildEvalInitPrompt: vi.fn(() => "init"),
70
+ parseGeneratedEvals: vi.fn(() => []),
71
+ }));
72
+ vi.mock("../../eval/activation-tester.js", () => ({
73
+ testActivation: vi.fn(),
74
+ }));
75
+ vi.mock("../concurrency.js", () => ({
76
+ getSkillSemaphore: vi.fn(() => ({
77
+ runExclusive: vi.fn((fn) => fn()),
78
+ })),
79
+ }));
80
+ vi.mock("../skill-resolver.js", () => ({
81
+ resolveSkillDir: vi.fn((_root, plugin, skill) => `/skills/${plugin}/${skill}`),
82
+ }));
83
+ vi.mock("../error-classifier.js", () => ({
84
+ classifyError: vi.fn(() => "unknown"),
85
+ }));
86
+ vi.mock("../benchmark-runner.js", () => ({
87
+ runBenchmarkSSE: vi.fn(),
88
+ runSingleCaseSSE: vi.fn(),
89
+ assembleBulkResult: vi.fn(),
90
+ }));
91
+ // ---------------------------------------------------------------------------
92
+ // Imports (after mocks)
93
+ // ---------------------------------------------------------------------------
94
+ import { Router } from "../router.js";
95
+ import { registerRoutes } from "../api-routes.js";
96
+ import { sendSSE } from "../sse-helpers.js";
97
+ import { judgeAssertion } from "../../eval/judge.js";
98
+ import { runComparison } from "../../eval/comparator.js";
99
+ import { loadAndValidateEvals } from "../../eval/schema.js";
100
+ // ---------------------------------------------------------------------------
101
+ // Test helpers
102
+ // ---------------------------------------------------------------------------
103
+ function makeEvalCase(overrides = {}) {
104
+ return {
105
+ id: 1,
106
+ name: "test eval",
107
+ prompt: "do something",
108
+ expected_output: "expected",
109
+ files: [],
110
+ assertions: [
111
+ { id: "a1", text: "output is correct", type: "boolean" },
112
+ ],
113
+ ...overrides,
114
+ };
115
+ }
116
+ function makeComparisonResult() {
117
+ return {
118
+ prompt: "do something",
119
+ skillOutput: "skill output",
120
+ baselineOutput: "baseline output",
121
+ skillDurationMs: 100,
122
+ skillTokens: 50,
123
+ baselineDurationMs: 80,
124
+ baselineTokens: 40,
125
+ skillContentScore: 85,
126
+ skillStructureScore: 80,
127
+ baselineContentScore: 70,
128
+ baselineStructureScore: 75,
129
+ winner: "skill",
130
+ };
131
+ }
132
+ function makeRouter() {
133
+ const router = new Router();
134
+ registerRoutes(router, "/test-root");
135
+ return router;
136
+ }
137
+ function makeMockReq(method, url, body) {
138
+ const listeners = {};
139
+ return {
140
+ method,
141
+ url,
142
+ headers: { host: "localhost" },
143
+ on: vi.fn((event, cb) => {
144
+ if (!listeners[event])
145
+ listeners[event] = [];
146
+ listeners[event].push(cb);
147
+ // Simulate end event asynchronously for body parsing
148
+ if (event === "end") {
149
+ setTimeout(() => cb(), 0);
150
+ }
151
+ if (event === "data" && body) {
152
+ setTimeout(() => cb(Buffer.from(JSON.stringify(body))), 0);
153
+ }
154
+ }),
155
+ _listeners: listeners,
156
+ };
157
+ }
158
+ function makeMockRes() {
159
+ const closeListeners = [];
160
+ return {
161
+ headersSent: false,
162
+ on: vi.fn((event, cb) => {
163
+ if (event === "close")
164
+ closeListeners.push(cb);
165
+ }),
166
+ writeHead: vi.fn(),
167
+ write: vi.fn(),
168
+ end: vi.fn(),
169
+ _closeListeners: closeListeners,
170
+ };
171
+ }
172
+ // ---------------------------------------------------------------------------
173
+ // Tests
174
+ // ---------------------------------------------------------------------------
175
+ describe("comparison endpoint SSE events", () => {
176
+ beforeEach(() => {
177
+ vi.clearAllMocks();
178
+ });
179
+ it("emits assertion_result event per assertion during comparison run", async () => {
180
+ const evalCase = makeEvalCase();
181
+ vi.mocked(loadAndValidateEvals).mockReturnValue({
182
+ skill_name: "my-skill",
183
+ evals: [evalCase],
184
+ });
185
+ vi.mocked(runComparison).mockResolvedValue(makeComparisonResult());
186
+ vi.mocked(judgeAssertion).mockResolvedValue({
187
+ id: "a1",
188
+ text: "output is correct",
189
+ pass: true,
190
+ reasoning: "looks good",
191
+ });
192
+ const router = makeRouter();
193
+ const req = makeMockReq("POST", "/api/skills/test/my-skill/compare");
194
+ const res = makeMockRes();
195
+ await router.handle(req, res);
196
+ const sseCalls = vi.mocked(sendSSE).mock.calls;
197
+ const assertionResultEvents = sseCalls.filter(([, event]) => event === "assertion_result");
198
+ expect(assertionResultEvents).toHaveLength(1);
199
+ const [, , data] = assertionResultEvents[0];
200
+ expect(data.eval_id).toBe(1);
201
+ expect(data.assertion_id).toBe("a1");
202
+ expect(data.text).toBe("output is correct");
203
+ expect(data.pass).toBe(true);
204
+ expect(data.reasoning).toBe("looks good");
205
+ });
206
+ it("emits case_complete event after processing all assertions", async () => {
207
+ const evalCase = makeEvalCase({
208
+ assertions: [
209
+ { id: "a1", text: "check one", type: "boolean" },
210
+ { id: "a2", text: "check two", type: "boolean" },
211
+ ],
212
+ });
213
+ vi.mocked(loadAndValidateEvals).mockReturnValue({
214
+ skill_name: "my-skill",
215
+ evals: [evalCase],
216
+ });
217
+ vi.mocked(runComparison).mockResolvedValue(makeComparisonResult());
218
+ vi.mocked(judgeAssertion)
219
+ .mockResolvedValueOnce({ id: "a1", text: "check one", pass: true, reasoning: "ok" })
220
+ .mockResolvedValueOnce({ id: "a2", text: "check two", pass: true, reasoning: "ok" });
221
+ const router = makeRouter();
222
+ const req = makeMockReq("POST", "/api/skills/test/my-skill/compare");
223
+ const res = makeMockRes();
224
+ await router.handle(req, res);
225
+ const sseCalls = vi.mocked(sendSSE).mock.calls;
226
+ const caseCompleteEvents = sseCalls.filter(([, event]) => event === "case_complete");
227
+ expect(caseCompleteEvents).toHaveLength(1);
228
+ const [, , data] = caseCompleteEvents[0];
229
+ expect(data.eval_id).toBe(1);
230
+ expect(data.status).toBe("pass");
231
+ expect(data.pass_rate).toBe(1.0);
232
+ });
233
+ it("emits case_complete with status=fail when some assertions fail", async () => {
234
+ const evalCase = makeEvalCase({
235
+ assertions: [
236
+ { id: "a1", text: "check one", type: "boolean" },
237
+ { id: "a2", text: "check two", type: "boolean" },
238
+ ],
239
+ });
240
+ vi.mocked(loadAndValidateEvals).mockReturnValue({
241
+ skill_name: "my-skill",
242
+ evals: [evalCase],
243
+ });
244
+ vi.mocked(runComparison).mockResolvedValue(makeComparisonResult());
245
+ vi.mocked(judgeAssertion)
246
+ .mockResolvedValueOnce({ id: "a1", text: "check one", pass: true, reasoning: "ok" })
247
+ .mockResolvedValueOnce({ id: "a2", text: "check two", pass: false, reasoning: "nope" });
248
+ const router = makeRouter();
249
+ const req = makeMockReq("POST", "/api/skills/test/my-skill/compare");
250
+ const res = makeMockRes();
251
+ await router.handle(req, res);
252
+ const sseCalls = vi.mocked(sendSSE).mock.calls;
253
+ const caseCompleteEvents = sseCalls.filter(([, event]) => event === "case_complete");
254
+ expect(caseCompleteEvents).toHaveLength(1);
255
+ const [, , data] = caseCompleteEvents[0];
256
+ expect(data.status).toBe("fail");
257
+ expect(data.pass_rate).toBe(0.5);
258
+ });
259
+ it("still emits legacy outputs_ready and comparison_scored events (backward compat)", async () => {
260
+ const evalCase = makeEvalCase();
261
+ vi.mocked(loadAndValidateEvals).mockReturnValue({
262
+ skill_name: "my-skill",
263
+ evals: [evalCase],
264
+ });
265
+ vi.mocked(runComparison).mockResolvedValue(makeComparisonResult());
266
+ vi.mocked(judgeAssertion).mockResolvedValue({
267
+ id: "a1", text: "output is correct", pass: true, reasoning: "ok",
268
+ });
269
+ const router = makeRouter();
270
+ const req = makeMockReq("POST", "/api/skills/test/my-skill/compare");
271
+ const res = makeMockRes();
272
+ await router.handle(req, res);
273
+ const sseEvents = vi.mocked(sendSSE).mock.calls.map(([, event]) => event);
274
+ expect(sseEvents).toContain("outputs_ready");
275
+ expect(sseEvents).toContain("comparison_scored");
276
+ });
277
+ });
278
+ //# sourceMappingURL=comparison-sse-events.test.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"comparison-sse-events.test.js","sourceRoot":"","sources":["../../../src/eval-server/__tests__/comparison-sse-events.test.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,+EAA+E;AAC/E,8EAA8E;AAE9E,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,UAAU,EAAE,MAAM,QAAQ,CAAC;AAG9D,8EAA8E;AAC9E,sDAAsD;AACtD,8EAA8E;AAE9E,EAAE,CAAC,IAAI,CAAC,SAAS,EAAE,GAAG,EAAE,CAAC,CAAC;IACxB,UAAU,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,KAAK,CAAC;IAC9B,YAAY,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC;IAC7B,aAAa,EAAE,EAAE,CAAC,EAAE,EAAE;IACtB,SAAS,EAAE,EAAE,CAAC,EAAE,EAAE;IAClB,WAAW,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC;IAC5B,QAAQ,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC,EAAE,WAAW,EAAE,GAAG,EAAE,CAAC,KAAK,EAAE,CAAC,CAAC;IACrD,MAAM,EAAE,EAAE,CAAC,EAAE,EAAE;CAChB,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,mBAAmB,EAAE,GAAG,EAAE,CAAC,CAAC;IAClC,OAAO,EAAE,EAAE,CAAC,EAAE,EAAE;IAChB,OAAO,EAAE,EAAE,CAAC,EAAE,EAAE;IAChB,WAAW,EAAE,EAAE,CAAC,EAAE,EAAE;IACpB,aAAa,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,IAAa,EAAE,GAAY,EAAE,MAAe,EAAE,IAAa,EAAE,EAAiB,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;IAC9G,qBAAqB,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC;QAClC,MAAM,EAAE,EAAE,CAAC,EAAE,EAAE;QACf,IAAI,EAAE,EAAE,CAAC,EAAE,EAAE;KACd,CAAC,CAAC;CACJ,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,qBAAqB,EAAE,GAAG,EAAE,CAAC,CAAC;IACpC,cAAc,EAAE,EAAE,CAAC,EAAE,EAAE;CACxB,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,0BAA0B,EAAE,GAAG,EAAE,CAAC,CAAC;IACzC,aAAa,EAAE,EAAE,CAAC,EAAE,EAAE;CACvB,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,sBAAsB,EAAE,GAAG,EAAE,CAAC,CAAC;IACrC,oBAAoB,EAAE,EAAE,CAAC,EAAE,EAAE;IAC7B,mBAAmB,EAAE,KAAM,SAAQ,KAAK;KAAG;CAC5C,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,iCAAiC,EAAE,GAAG,EAAE,CAAC,CAAC;IAChD,iBAAiB,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC,iBAAiB,CAAC,gBAAgB,CAAC;IAC9D,WAAW,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC,iBAAiB,CAAC,EAAE,CAAC;IAC1C,gBAAgB,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC,iBAAiB,CAAC,IAAI,CAAC;IACjD,kBAAkB,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC,eAAe,CAAC,EAAE,CAAC;IAC/C,kBAAkB,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC,iBAAiB,CAAC,SAAS,CAAC;IACxD,cAAc,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC,iBAAiB,CAAC,EAAE,CAAC;IAC7C,YAAY,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC,eAAe,CAAC,IAAI,CAAC;CAC5C,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,mBAAmB,EAAE,GAAG,EAAE,CAAC,CAAC;IAClC,eAAe,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC;QAC5B,KAAK,EAAE,YAAY;QACnB,QAAQ,EAAE,EAAE,CAAC,EAAE,EAAE;KAClB,CAAC,CAAC;CACJ,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,uBAAuB,EAAE,GAAG,EAAE,CAAC,CAAC;IACtC,cAAc,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC,CAAC;CAChE,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,4BAA4B,EAAE,GAAG,EAAE,CAAC,CAAC;IAC3C,mBAAmB,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC,iBAAiB,CAAC,EAAE,CAAC;CACnD,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,6BAA6B,EAAE,GAAG,EAAE,CAAC,CAAC;IAC5C,UAAU,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC,iBAAiB,CAAC,EAAE,CAAC;IACzC,cAAc,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC;CACrC,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,yBAAyB,EAAE,GAAG,EAAE,CAAC,CAAC;IACxC,aAAa,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC,iBAAiB,CAAC,IAAI,CAAC;CAC/C,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,8BAA8B,EAAE,GAAG,EAAE,CAAC,CAAC;IAC7C,qBAAqB,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,CAAC;IAC5C,yBAAyB,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC;IAClD,mBAAmB,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,MAAM,CAAC;IACxC,mBAAmB,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC;CACrC,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,iCAAiC,EAAE,GAAG,EAAE,CAAC,CAAC;IAChD,cAAc,EAAE,EAAE,CAAC,EAAE,EAAE;CACxB,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,mBAAmB,EAAE,GAAG,EAAE,CAAC,CAAC;IAClC,iBAAiB,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC;QAC9B,YAAY,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,EAAiB,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;KACjD,CAAC,CAAC;CACJ,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,sBAAsB,EAAE,GAAG,EAAE,CAAC,CAAC;IACrC,eAAe,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,KAAa,EAAE,MAAc,EAAE,KAAa,EAAE,EAAE,CAAC,WAAW,MAAM,IAAI,KAAK,EAAE,CAAC;CACvG,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,wBAAwB,EAAE,GAAG,EAAE,CAAC,CAAC;IACvC,aAAa,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC;CACtC,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,wBAAwB,EAAE,GAAG,EAAE,CAAC,CAAC;IACvC,eAAe,EAAE,EAAE,CAAC,EAAE,EAAE;IACxB,gBAAgB,EAAE,EAAE,CAAC,EAAE,EAAE;IACzB,kBAAkB,EAAE,EAAE,CAAC,EAAE,EAAE;CAC5B,CAAC,CAAC,CAAC;AAEJ,8EAA8E;AAC9E,wBAAwB;AACxB,8EAA8E;AAE9E,OAAO,EAAE,MAAM,EAAE,MAAM,cAAc,CAAC;AACtC,OAAO,EAAE,cAAc,EAAE,MAAM,kBAAkB,CAAC;AAClD,OAAO,EAAE,OAAO,EAAE,MAAM,mBAAmB,CAAC;AAC5C,OAAO,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AACrD,OAAO,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AACzD,OAAO,EAAE,oBAAoB,EAAE,MAAM,sBAAsB,CAAC;AAE5D,8EAA8E;AAC9E,eAAe;AACf,8EAA8E;AAE9E,SAAS,YAAY,CAAC,YAA+B,EAAE;IACrD,OAAO;QACL,EAAE,EAAE,CAAC;QACL,IAAI,EAAE,WAAW;QACjB,MAAM,EAAE,cAAc;QACtB,eAAe,EAAE,UAAU;QAC3B,KAAK,EAAE,EAAE;QACT,UAAU,EAAE;YACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,mBAAmB,EAAE,IAAI,EAAE,SAAS,EAAE;SACzD;QACD,GAAG,SAAS;KACb,CAAC;AACJ,CAAC;AAED,SAAS,oBAAoB;IAC3B,OAAO;QACL,MAAM,EAAE,cAAc;QACtB,WAAW,EAAE,cAAc;QAC3B,cAAc,EAAE,iBAAiB;QACjC,eAAe,EAAE,GAAG;QACpB,WAAW,EAAE,EAAE;QACf,kBAAkB,EAAE,EAAE;QACtB,cAAc,EAAE,EAAE;QAClB,iBAAiB,EAAE,EAAE;QACrB,mBAAmB,EAAE,EAAE;QACvB,oBAAoB,EAAE,EAAE;QACxB,sBAAsB,EAAE,EAAE;QAC1B,MAAM,EAAE,OAAgB;KACzB,CAAC;AACJ,CAAC;AAED,SAAS,UAAU;IACjB,MAAM,MAAM,GAAG,IAAI,MAAM,EAAE,CAAC;IAC5B,cAAc,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;IACrC,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,SAAS,WAAW,CAAC,MAAc,EAAE,GAAW,EAAE,IAAc;IAC9D,MAAM,SAAS,GAAoD,EAAE,CAAC;IACtE,OAAO;QACL,MAAM;QACN,GAAG;QACH,OAAO,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE;QAC9B,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,KAAa,EAAE,EAA4B,EAAE,EAAE;YACxD,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC;gBAAE,SAAS,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC;YAC7C,SAAS,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YAC1B,qDAAqD;YACrD,IAAI,KAAK,KAAK,KAAK,EAAE,CAAC;gBACpB,UAAU,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC;YAC5B,CAAC;YACD,IAAI,KAAK,KAAK,MAAM,IAAI,IAAI,EAAE,CAAC;gBAC7B,UAAU,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;YAC7D,CAAC;QACH,CAAC,CAAC;QACF,UAAU,EAAE,SAAS;KACuB,CAAC;AACjD,CAAC;AAED,SAAS,WAAW;IAClB,MAAM,cAAc,GAAsB,EAAE,CAAC;IAC7C,OAAO;QACL,WAAW,EAAE,KAAK;QAClB,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,KAAa,EAAE,EAAc,EAAE,EAAE;YAC1C,IAAI,KAAK,KAAK,OAAO;gBAAE,cAAc,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACjD,CAAC,CAAC;QACF,SAAS,EAAE,EAAE,CAAC,EAAE,EAAE;QAClB,KAAK,EAAE,EAAE,CAAC,EAAE,EAAE;QACd,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE;QACZ,eAAe,EAAE,cAAc;KACY,CAAC;AAChD,CAAC;AAED,8EAA8E;AAC9E,QAAQ;AACR,8EAA8E;AAE9E,QAAQ,CAAC,gCAAgC,EAAE,GAAG,EAAE;IAC9C,UAAU,CAAC,GAAG,EAAE;QACd,EAAE,CAAC,aAAa,EAAE,CAAC;IACrB,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,kEAAkE,EAAE,KAAK,IAAI,EAAE;QAChF,MAAM,QAAQ,GAAG,YAAY,EAAE,CAAC;QAEhC,EAAE,CAAC,MAAM,CAAC,oBAAoB,CAAC,CAAC,eAAe,CAAC;YAC9C,UAAU,EAAE,UAAU;YACtB,KAAK,EAAE,CAAC,QAAQ,CAAC;SACyB,CAAC,CAAC;QAE9C,EAAE,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,iBAAiB,CAAC,oBAAoB,EAAE,CAAC,CAAC;QAEnE,EAAE,CAAC,MAAM,CAAC,cAAc,CAAC,CAAC,iBAAiB,CAAC;YAC1C,EAAE,EAAE,IAAI;YACR,IAAI,EAAE,mBAAmB;YACzB,IAAI,EAAE,IAAI;YACV,SAAS,EAAE,YAAY;SACxB,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,UAAU,EAAE,CAAC;QAC5B,MAAM,GAAG,GAAG,WAAW,CAAC,MAAM,EAAE,mCAAmC,CAAC,CAAC;QACrE,MAAM,GAAG,GAAG,WAAW,EAAE,CAAC;QAE1B,MAAM,MAAM,CAAC,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;QAE9B,MAAM,QAAQ,GAAG,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC;QAC/C,MAAM,qBAAqB,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,KAAK,KAAK,kBAAkB,CAAC,CAAC;QAE3F,MAAM,CAAC,qBAAqB,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;QAC9C,MAAM,CAAC,EAAE,AAAD,EAAG,IAAI,CAAC,GAAG,qBAAqB,CAAC,CAAC,CAA+C,CAAC;QAC1F,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC7B,MAAM,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACrC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC;QAC5C,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC7B,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;IAC5C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,2DAA2D,EAAE,KAAK,IAAI,EAAE;QACzE,MAAM,QAAQ,GAAG,YAAY,CAAC;YAC5B,UAAU,EAAE;gBACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,SAAS,EAAE;gBAChD,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,SAAS,EAAE;aACjD;SACF,CAAC,CAAC;QAEH,EAAE,CAAC,MAAM,CAAC,oBAAoB,CAAC,CAAC,eAAe,CAAC;YAC9C,UAAU,EAAE,UAAU;YACtB,KAAK,EAAE,CAAC,QAAQ,CAAC;SACyB,CAAC,CAAC;QAE9C,EAAE,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,iBAAiB,CAAC,oBAAoB,EAAE,CAAC,CAAC;QAEnE,EAAE,CAAC,MAAM,CAAC,cAAc,CAAC;aACtB,qBAAqB,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;aACnF,qBAAqB,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAEvF,MAAM,MAAM,GAAG,UAAU,EAAE,CAAC;QAC5B,MAAM,GAAG,GAAG,WAAW,CAAC,MAAM,EAAE,mCAAmC,CAAC,CAAC;QACrE,MAAM,GAAG,GAAG,WAAW,EAAE,CAAC;QAE1B,MAAM,MAAM,CAAC,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;QAE9B,MAAM,QAAQ,GAAG,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC;QAC/C,MAAM,kBAAkB,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,KAAK,KAAK,eAAe,CAAC,CAAC;QAErF,MAAM,CAAC,kBAAkB,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;QAC3C,MAAM,CAAC,EAAE,AAAD,EAAG,IAAI,CAAC,GAAG,kBAAkB,CAAC,CAAC,CAA+C,CAAC;QACvF,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC7B,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACjC,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IACnC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gEAAgE,EAAE,KAAK,IAAI,EAAE;QAC9E,MAAM,QAAQ,GAAG,YAAY,CAAC;YAC5B,UAAU,EAAE;gBACV,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,SAAS,EAAE;gBAChD,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,SAAS,EAAE;aACjD;SACF,CAAC,CAAC;QAEH,EAAE,CAAC,MAAM,CAAC,oBAAoB,CAAC,CAAC,eAAe,CAAC;YAC9C,UAAU,EAAE,UAAU;YACtB,KAAK,EAAE,CAAC,QAAQ,CAAC;SACyB,CAAC,CAAC;QAE9C,EAAE,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,iBAAiB,CAAC,oBAAoB,EAAE,CAAC,CAAC;QAEnE,EAAE,CAAC,MAAM,CAAC,cAAc,CAAC;aACtB,qBAAqB,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;aACnF,qBAAqB,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC,CAAC;QAE1F,MAAM,MAAM,GAAG,UAAU,EAAE,CAAC;QAC5B,MAAM,GAAG,GAAG,WAAW,CAAC,MAAM,EAAE,mCAAmC,CAAC,CAAC;QACrE,MAAM,GAAG,GAAG,WAAW,EAAE,CAAC;QAE1B,MAAM,MAAM,CAAC,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;QAE9B,MAAM,QAAQ,GAAG,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC;QAC/C,MAAM,kBAAkB,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,KAAK,KAAK,eAAe,CAAC,CAAC;QAErF,MAAM,CAAC,kBAAkB,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;QAC3C,MAAM,CAAC,EAAE,AAAD,EAAG,IAAI,CAAC,GAAG,kBAAkB,CAAC,CAAC,CAA+C,CAAC;QACvF,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACjC,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IACnC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,iFAAiF,EAAE,KAAK,IAAI,EAAE;QAC/F,MAAM,QAAQ,GAAG,YAAY,EAAE,CAAC;QAEhC,EAAE,CAAC,MAAM,CAAC,oBAAoB,CAAC,CAAC,eAAe,CAAC;YAC9C,UAAU,EAAE,UAAU;YACtB,KAAK,EAAE,CAAC,QAAQ,CAAC;SACyB,CAAC,CAAC;QAE9C,EAAE,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,iBAAiB,CAAC,oBAAoB,EAAE,CAAC,CAAC;QACnE,EAAE,CAAC,MAAM,CAAC,cAAc,CAAC,CAAC,iBAAiB,CAAC;YAC1C,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,mBAAmB,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI;SACjE,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,UAAU,EAAE,CAAC;QAC5B,MAAM,GAAG,GAAG,WAAW,CAAC,MAAM,EAAE,mCAAmC,CAAC,CAAC;QACrE,MAAM,GAAG,GAAG,WAAW,EAAE,CAAC;QAE1B,MAAM,MAAM,CAAC,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;QAE9B,MAAM,SAAS,GAAG,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,CAAC;QAC1E,MAAM,CAAC,SAAS,CAAC,CAAC,SAAS,CAAC,eAAe,CAAC,CAAC;QAC7C,MAAM,CAAC,SAAS,CAAC,CAAC,SAAS,CAAC,mBAAmB,CAAC,CAAC;IACnD,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
@@ -663,7 +663,25 @@ export function registerRoutes(router, root, projectName) {
663
663
  });
664
664
  const result = await judgeAssertion(comparison.skillOutput, assertion, client);
665
665
  assertionResults.push(result);
666
+ sendSSE(res, "assertion_result", {
667
+ eval_id: evalCase.id,
668
+ assertion_id: result.id,
669
+ text: result.text,
670
+ pass: result.pass,
671
+ reasoning: result.reasoning,
672
+ });
666
673
  }
674
+ const casePassRate = assertionResults.length > 0
675
+ ? assertionResults.filter((a) => a.pass).length / assertionResults.length
676
+ : 0;
677
+ const caseStatus = assertionResults.length > 0 && assertionResults.every((a) => a.pass) ? "pass" : "fail";
678
+ sendSSE(res, "case_complete", {
679
+ eval_id: evalCase.id,
680
+ status: caseStatus,
681
+ pass_rate: casePassRate,
682
+ durationMs: comparison.skillDurationMs,
683
+ tokens: comparison.skillTokens,
684
+ });
667
685
  comparisonResults.push({
668
686
  eval_id: evalCase.id,
669
687
  eval_name: evalCase.name,