oh-my-knowledge 0.20.1 → 0.22.0
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.
- package/README.md +10 -4
- package/README.zh.md +9 -4
- package/dist/src/analysis/coverage-analyzer.d.ts +1 -1
- package/dist/src/analysis/coverage-analyzer.d.ts.map +1 -1
- package/dist/src/analysis/failure-clusterer.d.ts +1 -1
- package/dist/src/analysis/failure-clusterer.d.ts.map +1 -1
- package/dist/src/analysis/gap-analyzer.d.ts +2 -2
- package/dist/src/analysis/gap-analyzer.d.ts.map +1 -1
- package/dist/src/analysis/gap-analyzer.js +4 -4
- package/dist/src/analysis/hedging-classifier.d.ts +1 -1
- package/dist/src/analysis/hedging-classifier.d.ts.map +1 -1
- package/dist/src/analysis/report-diagnostics.d.ts +1 -1
- package/dist/src/analysis/report-diagnostics.d.ts.map +1 -1
- package/dist/src/analysis/report-diagnostics.js +7 -7
- package/dist/src/analysis/sample-diagnostics.d.ts +1 -1
- package/dist/src/analysis/sample-diagnostics.d.ts.map +1 -1
- package/dist/src/analysis/sample-diagnostics.js +15 -15
- package/dist/src/authoring/evolver.d.ts +1 -1
- package/dist/src/authoring/evolver.d.ts.map +1 -1
- package/dist/src/authoring/evolver.js +5 -5
- package/dist/src/authoring/evolver.js.map +1 -1
- package/dist/src/authoring/generator.d.ts +1 -1
- package/dist/src/authoring/generator.d.ts.map +1 -1
- package/dist/src/authoring/generator.js +8 -8
- package/dist/src/authoring/generator.js.map +1 -1
- package/dist/src/cli/i18n-dict.d.ts +56 -0
- package/dist/src/cli/i18n-dict.d.ts.map +1 -0
- package/dist/src/cli/i18n-dict.js +934 -0
- package/dist/src/cli/i18n-dict.js.map +1 -0
- package/dist/src/cli/i18n.d.ts +24 -0
- package/dist/src/cli/i18n.d.ts.map +1 -0
- package/dist/src/cli/i18n.js +53 -0
- package/dist/src/cli/i18n.js.map +1 -0
- package/dist/src/cli.js +320 -413
- package/dist/src/cli.js.map +1 -1
- package/dist/src/eval-core/cache.d.ts +7 -3
- package/dist/src/eval-core/cache.d.ts.map +1 -1
- package/dist/src/eval-core/cache.js +14 -4
- package/dist/src/eval-core/cache.js.map +1 -1
- package/dist/src/eval-core/dependency-checker.d.ts +16 -1
- package/dist/src/eval-core/dependency-checker.d.ts.map +1 -1
- package/dist/src/eval-core/dependency-checker.js +79 -4
- package/dist/src/eval-core/dependency-checker.js.map +1 -1
- package/dist/src/eval-core/evaluation-execution.d.ts +3 -3
- package/dist/src/eval-core/evaluation-execution.d.ts.map +1 -1
- package/dist/src/eval-core/evaluation-execution.js +4 -2
- package/dist/src/eval-core/evaluation-execution.js.map +1 -1
- package/dist/src/eval-core/evaluation-job.d.ts +2 -2
- package/dist/src/eval-core/evaluation-job.d.ts.map +1 -1
- package/dist/src/eval-core/evaluation-reporting.d.ts +1 -1
- package/dist/src/eval-core/evaluation-reporting.d.ts.map +1 -1
- package/dist/src/eval-core/evaluation-reporting.js +4 -0
- package/dist/src/eval-core/evaluation-reporting.js.map +1 -1
- package/dist/src/eval-core/execution-strategy.d.ts +1 -1
- package/dist/src/eval-core/execution-strategy.d.ts.map +1 -1
- package/dist/src/eval-core/execution-strategy.js +35 -2
- package/dist/src/eval-core/execution-strategy.js.map +1 -1
- package/dist/src/eval-core/layer-gates.d.ts +17 -0
- package/dist/src/eval-core/layer-gates.d.ts.map +1 -0
- package/dist/src/eval-core/{ci-gates.js → layer-gates.js} +5 -5
- package/dist/src/eval-core/layer-gates.js.map +1 -0
- package/dist/src/eval-core/schema.d.ts +1 -1
- package/dist/src/eval-core/schema.d.ts.map +1 -1
- package/dist/src/eval-core/schema.js +17 -3
- package/dist/src/eval-core/schema.js.map +1 -1
- package/dist/src/eval-core/task-planner.d.ts +1 -1
- package/dist/src/eval-core/task-planner.d.ts.map +1 -1
- package/dist/src/eval-core/task-planner.js +1 -1
- package/dist/src/eval-core/task-planner.js.map +1 -1
- package/dist/src/eval-core/verdict.d.ts +5 -2
- package/dist/src/eval-core/verdict.d.ts.map +1 -1
- package/dist/src/eval-core/verdict.js +47 -8
- package/dist/src/eval-core/verdict.js.map +1 -1
- package/dist/src/eval-workflows/each-evaluation-workflow.d.ts +15 -8
- package/dist/src/eval-workflows/each-evaluation-workflow.d.ts.map +1 -1
- package/dist/src/eval-workflows/each-evaluation-workflow.js +3 -2
- package/dist/src/eval-workflows/each-evaluation-workflow.js.map +1 -1
- package/dist/src/eval-workflows/evaluation-pipeline.d.ts +33 -4
- package/dist/src/eval-workflows/evaluation-pipeline.d.ts.map +1 -1
- package/dist/src/eval-workflows/evaluation-pipeline.js +81 -2
- package/dist/src/eval-workflows/evaluation-pipeline.js.map +1 -1
- package/dist/src/eval-workflows/evaluation-preparation.d.ts +10 -6
- package/dist/src/eval-workflows/evaluation-preparation.d.ts.map +1 -1
- package/dist/src/eval-workflows/evaluation-preparation.js +2 -2
- package/dist/src/eval-workflows/evaluation-preparation.js.map +1 -1
- package/dist/src/eval-workflows/run-evaluation.d.ts +16 -5
- package/dist/src/eval-workflows/run-evaluation.d.ts.map +1 -1
- package/dist/src/eval-workflows/run-evaluation.js +34 -5
- package/dist/src/eval-workflows/run-evaluation.js.map +1 -1
- package/dist/src/executors/anthropic-api.d.ts +1 -1
- package/dist/src/executors/anthropic-api.d.ts.map +1 -1
- package/dist/src/executors/anthropic-api.js +1 -1
- package/dist/src/executors/anthropic-api.js.map +1 -1
- package/dist/src/executors/claude-cli.d.ts +2 -2
- package/dist/src/executors/claude-cli.d.ts.map +1 -1
- package/dist/src/executors/claude-cli.js +21 -1
- package/dist/src/executors/claude-cli.js.map +1 -1
- package/dist/src/executors/claude-sdk-trace.d.ts +1 -1
- package/dist/src/executors/claude-sdk-trace.d.ts.map +1 -1
- package/dist/src/executors/claude-sdk.d.ts +15 -2
- package/dist/src/executors/claude-sdk.d.ts.map +1 -1
- package/dist/src/executors/claude-sdk.js +19 -1
- package/dist/src/executors/claude-sdk.js.map +1 -1
- package/dist/src/executors/gemini.d.ts +1 -1
- package/dist/src/executors/gemini.d.ts.map +1 -1
- package/dist/src/executors/index.d.ts +1 -1
- package/dist/src/executors/index.d.ts.map +1 -1
- package/dist/src/executors/openai-api.d.ts +1 -1
- package/dist/src/executors/openai-api.d.ts.map +1 -1
- package/dist/src/executors/openai-api.js +1 -1
- package/dist/src/executors/openai-api.js.map +1 -1
- package/dist/src/executors/openai-cli.d.ts +1 -1
- package/dist/src/executors/openai-cli.d.ts.map +1 -1
- package/dist/src/executors/script.d.ts +1 -1
- package/dist/src/executors/script.d.ts.map +1 -1
- package/dist/src/executors/script.js +10 -1
- package/dist/src/executors/script.js.map +1 -1
- package/dist/src/executors/shared.d.ts +1 -1
- package/dist/src/executors/shared.d.ts.map +1 -1
- package/dist/src/grading/assertions.d.ts +1 -1
- package/dist/src/grading/assertions.d.ts.map +1 -1
- package/dist/src/grading/debias-validate.d.ts +1 -1
- package/dist/src/grading/debias-validate.d.ts.map +1 -1
- package/dist/src/grading/debias-validate.js +3 -3
- package/dist/src/grading/gold-cli.d.ts +1 -1
- package/dist/src/grading/gold-cli.d.ts.map +1 -1
- package/dist/src/grading/gold-cli.js +3 -3
- package/dist/src/grading/gold-cli.js.map +1 -1
- package/dist/src/grading/index.d.ts +1 -1
- package/dist/src/grading/index.d.ts.map +1 -1
- package/dist/src/grading/judge.d.ts +1 -1
- package/dist/src/grading/judge.d.ts.map +1 -1
- package/dist/src/grading/layered-scores.d.ts +1 -1
- package/dist/src/grading/layered-scores.d.ts.map +1 -1
- package/dist/src/inputs/eval-config.d.ts +1 -1
- package/dist/src/inputs/eval-config.d.ts.map +1 -1
- package/dist/src/inputs/eval-config.js +37 -18
- package/dist/src/inputs/eval-config.js.map +1 -1
- package/dist/src/inputs/load-samples.d.ts +1 -1
- package/dist/src/inputs/load-samples.d.ts.map +1 -1
- package/dist/src/inputs/load-samples.js +4 -4
- package/dist/src/inputs/load-samples.js.map +1 -1
- package/dist/src/inputs/mcp-resolver.d.ts +1 -1
- package/dist/src/inputs/mcp-resolver.d.ts.map +1 -1
- package/dist/src/inputs/mcp-resolver.js +4 -4
- package/dist/src/inputs/mcp-resolver.js.map +1 -1
- package/dist/src/inputs/skill-loader.d.ts +11 -2
- package/dist/src/inputs/skill-loader.d.ts.map +1 -1
- package/dist/src/inputs/skill-loader.js +30 -7
- package/dist/src/inputs/skill-loader.js.map +1 -1
- package/dist/src/inputs/url-fetcher.d.ts +1 -1
- package/dist/src/inputs/url-fetcher.d.ts.map +1 -1
- package/dist/src/inputs/url-fetcher.js +2 -2
- package/dist/src/inputs/url-fetcher.js.map +1 -1
- package/dist/src/observability/skill-health-analyzer.d.ts +1 -1
- package/dist/src/observability/skill-health-analyzer.d.ts.map +1 -1
- package/dist/src/observability/trace-adapter.d.ts +1 -1
- package/dist/src/observability/trace-adapter.d.ts.map +1 -1
- package/dist/src/renderer/html-renderer.d.ts +1 -1
- package/dist/src/renderer/html-renderer.d.ts.map +1 -1
- package/dist/src/renderer/html-renderer.js +3 -3
- package/dist/src/renderer/html-renderer.js.map +1 -1
- package/dist/src/renderer/layout.d.ts +1 -1
- package/dist/src/renderer/layout.d.ts.map +1 -1
- package/dist/src/renderer/layout.js +1 -1
- package/dist/src/renderer/layout.js.map +1 -1
- package/dist/src/renderer/skill-health-renderer.d.ts +1 -1
- package/dist/src/renderer/skill-health-renderer.d.ts.map +1 -1
- package/dist/src/renderer/summary.d.ts +1 -1
- package/dist/src/renderer/summary.d.ts.map +1 -1
- package/dist/src/renderer/summary.js +22 -20
- package/dist/src/renderer/summary.js.map +1 -1
- package/dist/src/renderer/table.d.ts +1 -1
- package/dist/src/renderer/table.d.ts.map +1 -1
- package/dist/src/renderer/trends.d.ts +1 -1
- package/dist/src/renderer/trends.d.ts.map +1 -1
- package/dist/src/renderer/trends.js +1 -1
- package/dist/src/renderer/trends.js.map +1 -1
- package/dist/src/server/job-store.d.ts +1 -1
- package/dist/src/server/job-store.d.ts.map +1 -1
- package/dist/src/server/report-server.d.ts +1 -1
- package/dist/src/server/report-server.d.ts.map +1 -1
- package/dist/src/server/report-server.js +20 -20
- package/dist/src/server/report-server.js.map +1 -1
- package/dist/src/server/report-store.d.ts +1 -1
- package/dist/src/server/report-store.d.ts.map +1 -1
- package/dist/src/types/eval.d.ts +9 -0
- package/dist/src/types/eval.d.ts.map +1 -1
- package/dist/src/types/executor.d.ts +1 -0
- package/dist/src/types/executor.d.ts.map +1 -1
- package/dist/src/types/report.d.ts +10 -0
- package/dist/src/types/report.d.ts.map +1 -1
- package/package.json +2 -2
- package/dist/src/eval-core/ci-gates.d.ts +0 -17
- package/dist/src/eval-core/ci-gates.d.ts.map +0 -1
- package/dist/src/eval-core/ci-gates.js.map +0 -1
- package/dist/src/types.d.ts +0 -2
- package/dist/src/types.d.ts.map +0 -1
- package/dist/src/types.js +0 -6
- package/dist/src/types.js.map +0 -1
|
@@ -26,27 +26,28 @@
|
|
|
26
26
|
* proven optimal. Each rule's source (NIST AI 800-3 / Krippendorff thresholds /
|
|
27
27
|
* empirical) is documented inline so users can audit and override.
|
|
28
28
|
*/
|
|
29
|
-
import {
|
|
29
|
+
import { evaluateLayerGates } from './layer-gates.js';
|
|
30
30
|
/**
|
|
31
31
|
* Compute a verdict for a finished report. Pure function — no I/O.
|
|
32
32
|
*/
|
|
33
33
|
export function computeVerdict(report, options = {}) {
|
|
34
|
-
const {
|
|
34
|
+
const { gateThreshold = 3.5, triviallySmallDiff = 0.1 } = options;
|
|
35
35
|
const variants = report.meta?.variants ?? [];
|
|
36
36
|
const summary = report.summary ?? {};
|
|
37
37
|
const sampleCount = report.meta?.sampleCount ?? 0;
|
|
38
38
|
if (variants.length < 2) {
|
|
39
39
|
// Single-variant — no comparison possible. Just report whether the variant
|
|
40
40
|
// passes its own three-layer gate.
|
|
41
|
-
const gate =
|
|
41
|
+
const gate = evaluateLayerGates(summary, gateThreshold);
|
|
42
42
|
return {
|
|
43
43
|
level: 'SOLO',
|
|
44
44
|
headline: gate.allPass
|
|
45
|
-
? `SOLO · single variant, three-layer gate PASS @ threshold ${
|
|
45
|
+
? `SOLO · single variant, three-layer gate PASS @ threshold ${gateThreshold}`
|
|
46
46
|
: `SOLO · single variant, three-layer gate FAIL — see ci output`,
|
|
47
47
|
rationale: {
|
|
48
48
|
layerWinners: gate.lines.join('; '),
|
|
49
49
|
sampleSize: `N=${sampleCount}`,
|
|
50
|
+
stability: formatStability(report),
|
|
50
51
|
},
|
|
51
52
|
variants,
|
|
52
53
|
};
|
|
@@ -58,7 +59,7 @@ export function computeVerdict(report, options = {}) {
|
|
|
58
59
|
const pairs = explicitPairs.length > 0
|
|
59
60
|
? explicitPairs
|
|
60
61
|
: variants.slice(1).map((treatment) => ({ control: variants[0], treatment }));
|
|
61
|
-
const perPair = pairs.map((p) => verdictForPair(p, summary, sampleCount, report,
|
|
62
|
+
const perPair = pairs.map((p) => verdictForPair(p, summary, sampleCount, report, gateThreshold, triviallySmallDiff));
|
|
62
63
|
// Worst-case roll-up: REGRESS dominates, then CAUTIOUS, then NOISE/UNDERPOWERED, then PROGRESS.
|
|
63
64
|
const order = ['REGRESS', 'CAUTIOUS', 'UNDERPOWERED', 'NOISE', 'PROGRESS'];
|
|
64
65
|
let topLevel = 'PROGRESS';
|
|
@@ -75,6 +76,7 @@ export function computeVerdict(report, options = {}) {
|
|
|
75
76
|
: 'no pairwise comparison available — was --bootstrap used?';
|
|
76
77
|
const layerWinners = formatLayerWinners(summary, variants);
|
|
77
78
|
const sampleSize = formatSampleSize(report);
|
|
79
|
+
const stability = formatStability(report);
|
|
78
80
|
const judgeAgreement = formatJudgeAgreement(report);
|
|
79
81
|
const shipRecommendation = recommendation(topLevel, perPair);
|
|
80
82
|
return {
|
|
@@ -87,6 +89,7 @@ export function computeVerdict(report, options = {}) {
|
|
|
87
89
|
significance,
|
|
88
90
|
layerWinners,
|
|
89
91
|
sampleSize,
|
|
92
|
+
stability,
|
|
90
93
|
judgeAgreement,
|
|
91
94
|
shipRecommendation,
|
|
92
95
|
},
|
|
@@ -104,11 +107,11 @@ export function computeVerdict(report, options = {}) {
|
|
|
104
107
|
* 5. If diff CI is positive AND treatment passes all gates → PROGRESS.
|
|
105
108
|
* 6. If diff CI is positive but treatment broke a gate → CAUTIOUS.
|
|
106
109
|
*/
|
|
107
|
-
function verdictForPair(pair, summary, sampleCount, report,
|
|
110
|
+
function verdictForPair(pair, summary, sampleCount, report, gateThreshold, triviallySmallDiff) {
|
|
108
111
|
const { control, treatment, diffBootstrapCI: diff } = pair;
|
|
109
112
|
// Layer-gate check: did any layer fall below threshold for either variant?
|
|
110
|
-
const cGate =
|
|
111
|
-
const tGate =
|
|
113
|
+
const cGate = evaluateLayerGates({ [control]: summary[control] }, gateThreshold);
|
|
114
|
+
const tGate = evaluateLayerGates({ [treatment]: summary[treatment] }, gateThreshold);
|
|
112
115
|
// No bootstrap CI available → fall back to point-estimate diff comparison.
|
|
113
116
|
if (!diff) {
|
|
114
117
|
const cMean = avgComposite(summary[control]);
|
|
@@ -227,6 +230,40 @@ function formatSampleSize(report) {
|
|
|
227
230
|
return `N=${n}, saturated @ N=${v.atN ?? '?'} (${v.confidence} confidence)`;
|
|
228
231
|
return `N=${n}, not yet saturated (${v.confidence} confidence)`;
|
|
229
232
|
}
|
|
233
|
+
/**
|
|
234
|
+
* Stability rationale. 三种状态:
|
|
235
|
+
* - --repeat ≥ 2 + 有 variance 数据: 报告 CV (variation coefficient) 主指标
|
|
236
|
+
* - --repeat ≥ 2 但 variance 缺失: 异常,标 "—" 提示数据丢失
|
|
237
|
+
* - --repeat < 2: 显式说"未测量,需 --repeat ≥ 2",而不是默默不提
|
|
238
|
+
*
|
|
239
|
+
* 单轮场景关键:不是"稳定 = 100%"(常见误读),而是"测不到稳定性"。
|
|
240
|
+
* Verdict 必须诚实交代这个盲区,不能让用户以为没说就是 OK。
|
|
241
|
+
*/
|
|
242
|
+
function formatStability(report) {
|
|
243
|
+
const runs = report.variance?.runs ?? report.meta?.request?.repeat ?? 1;
|
|
244
|
+
if (runs < 2) {
|
|
245
|
+
return '稳定性未测量(单轮评测,需 --repeat ≥ 2 才能测 CV)';
|
|
246
|
+
}
|
|
247
|
+
const variance = report.variance?.perVariant;
|
|
248
|
+
if (!variance || Object.keys(variance).length === 0) {
|
|
249
|
+
return `runs=${runs} 但 variance 数据缺失`;
|
|
250
|
+
}
|
|
251
|
+
// CV = stddev / mean,取所有 variant 的中位数(单 variant 易有 NaN/0,聚合更稳)
|
|
252
|
+
const cvs = [];
|
|
253
|
+
for (const v of Object.values(variance)) {
|
|
254
|
+
if (typeof v.stddev === 'number' && typeof v.mean === 'number' && v.mean > 0) {
|
|
255
|
+
cvs.push(v.stddev / v.mean);
|
|
256
|
+
}
|
|
257
|
+
}
|
|
258
|
+
if (cvs.length === 0)
|
|
259
|
+
return `runs=${runs}, CV 计算失败(stddev/mean 数据缺失)`;
|
|
260
|
+
cvs.sort((a, b) => a - b);
|
|
261
|
+
const median = cvs[Math.floor(cvs.length / 2)];
|
|
262
|
+
const cvPct = (median * 100).toFixed(1);
|
|
263
|
+
// 阈值参考 terminology-spec §5:<5% 稳 / 5-15% 中 / >15% 不稳
|
|
264
|
+
const verdict = median < 0.05 ? '稳定' : median < 0.15 ? '中等' : '不稳';
|
|
265
|
+
return `CV=${cvPct}% (${verdict}, runs=${runs}; 阈值 <5%=稳/5-15%=中/>15%=不稳)`;
|
|
266
|
+
}
|
|
230
267
|
function formatJudgeAgreement(report) {
|
|
231
268
|
const a = report.meta?.humanAgreement;
|
|
232
269
|
if (!a)
|
|
@@ -267,6 +304,8 @@ export function formatVerdictText(result, options = {}) {
|
|
|
267
304
|
lines.push(` Layer winners: ${result.rationale.layerWinners}`);
|
|
268
305
|
if (result.rationale.sampleSize)
|
|
269
306
|
lines.push(` Sample size: ${result.rationale.sampleSize}`);
|
|
307
|
+
if (result.rationale.stability)
|
|
308
|
+
lines.push(` Stability: ${result.rationale.stability}`);
|
|
270
309
|
if (result.rationale.judgeAgreement)
|
|
271
310
|
lines.push(` Judge α: ${result.rationale.judgeAgreement}`);
|
|
272
311
|
if (result.rationale.shipRecommendation)
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"verdict.js","sourceRoot":"","sources":["../../../src/eval-core/verdict.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAGH,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAuChD;;GAEG;AACH,MAAM,UAAU,cAAc,CAAC,MAAc,EAAE,UAA0B,EAAE;IACzE,MAAM,EAAE,WAAW,GAAG,GAAG,EAAE,kBAAkB,GAAG,GAAG,EAAE,GAAG,OAAO,CAAC;IAChE,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,EAAE,QAAQ,IAAI,EAAE,CAAC;IAC7C,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC;IACrC,MAAM,WAAW,GAAG,MAAM,CAAC,IAAI,EAAE,WAAW,IAAI,CAAC,CAAC;IAElD,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,2EAA2E;QAC3E,mCAAmC;QACnC,MAAM,IAAI,GAAG,eAAe,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;QACnD,OAAO;YACL,KAAK,EAAE,MAAM;YACb,QAAQ,EAAE,IAAI,CAAC,OAAO;gBACpB,CAAC,CAAC,4DAA4D,WAAW,EAAE;gBAC3E,CAAC,CAAC,8DAA8D;YAClE,SAAS,EAAE;gBACT,YAAY,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC;gBACnC,UAAU,EAAE,KAAK,WAAW,EAAE;aAC/B;YACD,QAAQ;SACT,CAAC;IACJ,CAAC;IAED,kFAAkF;IAClF,iFAAiF;IACjF,+EAA+E;IAC/E,MAAM,aAAa,GAAG,MAAM,CAAC,IAAI,EAAE,eAAe,IAAI,EAAE,CAAC;IACzD,MAAM,KAAK,GAA4B,aAAa,CAAC,MAAM,GAAG,CAAC;QAC7D,CAAC,CAAC,aAAa;QACf,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC;IAChF,MAAM,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,cAAc,CAAC,CAAC,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,WAAW,EAAE,kBAAkB,CAAC,CAAC,CAAC;IAEnH,gGAAgG;IAChG,MAAM,KAAK,GAAmB,CAAC,SAAS,EAAE,UAAU,EAAE,cAAc,EAAE,OAAO,EAAE,UAAU,CAAC,CAAC;IAC3F,IAAI,QAAQ,GAAiB,UAAU,CAAC;IACxC,KAAK,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;QAC1B,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,KAAK,CAAC,EAAE,CAAC;YAC3C,QAAQ,GAAG,KAAK,CAAC;YACjB,MAAM;QACR,CAAC;IACH,CAAC;IAED,0EAA0E;IAC1E,MAAM,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,QAAQ,CAAC,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IAE/E,MAAM,YAAY,GAAG,cAAc;QACjC,CAAC,CAAC,kBAAkB,CAAC,cAAc,CAAC;QACpC,CAAC,CAAC,0DAA0D,CAAC;IAE/D,MAAM,YAAY,GAAG,kBAAkB,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;IAC3D,MAAM,UAAU,GAAG,gBAAgB,CAAC,MAAM,CAAC,CAAC;IAC5C,MAAM,cAAc,GAAG,oBAAoB,CAAC,MAAM,CAAC,CAAC;IACpD,MAAM,kBAAkB,GAAG,cAAc,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IAE7D,OAAO;QACL,KAAK,EAAE,QAAQ;QACf,QAAQ,EAAE,cAAc;YACtB,CAAC,CAAC,GAAG,QAAQ,MAAM,cAAc,CAAC,SAAS,OAAO,cAAc,CAAC,OAAO,KAAK,cAAc,CAAC,QAAQ,EAAE;YACtG,CAAC,CAAC,GAAG,QAAQ,MAAM,QAAQ,CAAC,MAAM,WAAW;QAC/C,OAAO;QACP,SAAS,EAAE;YACT,YAAY;YACZ,YAAY;YACZ,UAAU;YACV,cAAc;YACd,kBAAkB;SACnB;QACD,QAAQ;KACT,CAAC;AACJ,CAAC;AAED;;;;;;;;;;GAUG;AACH,SAAS,cAAc,CACrB,IAA2B,EAC3B,OAAuC,EACvC,WAAmB,EACnB,MAAc,EACd,WAAmB,EACnB,kBAA0B;IAE1B,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,eAAe,EAAE,IAAI,EAAE,GAAG,IAAI,CAAC;IAE3D,2EAA2E;IAC3E,MAAM,KAAK,GAAG,eAAe,CAAC,EAAE,CAAC,OAAO,CAAC,EAAE,OAAO,CAAC,OAAO,CAAC,EAAE,EAAE,WAAW,CAAC,CAAC;IAC5E,MAAM,KAAK,GAAG,eAAe,CAAC,EAAE,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC,SAAS,CAAC,EAAE,EAAE,WAAW,CAAC,CAAC;IAEhF,2EAA2E;IAC3E,IAAI,CAAC,IAAI,EAAE,CAAC;QACV,MAAM,KAAK,GAAG,YAAY,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;QAC7C,MAAM,KAAK,GAAG,YAAY,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC;QAC/C,MAAM,KAAK,GAAG,KAAK,GAAG,KAAK,CAAC;QAC5B,MAAM,QAAQ,GAAG,KAAK,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,kCAAkC,CAAC;QACjG,IAAI,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,kBAAkB,EAAE,CAAC;YACzC,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC;QAC1D,CAAC;QACD,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS;YACzC,QAAQ;SACT,CAAC;IACJ,CAAC;IAED,MAAM,YAAY,GAAG,KAAK,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,QAAQ,QAAQ,IAAI,CAAC,GAAG,KAAK,IAAI,CAAC,IAAI,GAAG,CAAC;IAEzG,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC;QACtB,wFAAwF;QACxF,gFAAgF;QAChF,MAAM,UAAU,GAAG,MAAM,CAAC,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC,SAAS,CAAC,CAAC;QACtE,MAAM,YAAY,GAChB,WAAW,GAAG,EAAE;YAChB,CAAC,UAAU,IAAI,CAAC,UAAU,CAAC,SAAS,IAAI,UAAU,CAAC,UAAU,KAAK,MAAM,CAAC,CAAC;QAC5E,IAAI,YAAY,EAAE,CAAC;YACjB,OAAO;gBACL,OAAO;gBACP,SAAS;gBACT,KAAK,EAAE,cAAc;gBACrB,QAAQ,EAAE,GAAG,YAAY,QAAQ,WAAW,mBAAmB;aAChE,CAAC;QACJ,CAAC;QACD,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,YAAY,eAAe,EAAE,CAAC;IAC1F,CAAC;IAED,uBAAuB;IACvB,IAAI,IAAI,CAAC,QAAQ,GAAG,CAAC,EAAE,CAAC;QACtB,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,SAAS,EAAE,QAAQ,EAAE,GAAG,YAAY,oBAAoB,EAAE,CAAC;IACjG,CAAC;IAED,wDAAwD;IACxD,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;QACnB,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,UAAU;YACjB,QAAQ,EAAE,GAAG,YAAY,8CAA8C;SACxE,CAAC;IACJ,CAAC;IAED,wBAAwB;IACxB,IAAI,IAAI,CAAC,QAAQ,GAAG,kBAAkB,EAAE,CAAC;QACvC,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,UAAU;YACjB,QAAQ,EAAE,GAAG,YAAY,qCAAqC;SAC/D,CAAC;IACJ,CAAC;IAED,4EAA4E;IAC5E,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;QACnB,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,UAAU;YACjB,QAAQ,EAAE,GAAG,YAAY,2CAA2C;SACrE,CAAC;IACJ,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,UAAU,EAAE,QAAQ,EAAE,GAAG,YAAY,cAAc,EAAE,CAAC;AAC5F,CAAC;AAED,SAAS,YAAY,CAAC,CAA6B;IACjD,IAAI,CAAC,CAAC;QAAE,OAAO,CAAC,CAAC;IACjB,IAAI,OAAO,CAAC,CAAC,iBAAiB,KAAK,QAAQ;QAAE,OAAO,CAAC,CAAC,iBAAiB,CAAC;IACxE,6EAA6E;IAC7E,MAAM,MAAM,GAAG,CAAC,CAAC,CAAC,YAAY,EAAE,CAAC,CAAC,gBAAgB,EAAE,CAAC,CAAC,aAAa,CAAC,CAAC,MAAM,CACzE,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAC1C,CAAC;IACF,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAClC,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;AAC3D,CAAC;AAED,SAAS,kBAAkB,CAAC,CAA4C;IACtE,OAAO,GAAG,CAAC,CAAC,KAAK,MAAM,CAAC,CAAC,QAAQ,EAAE,CAAC;AACtC,CAAC;AAED,SAAS,kBAAkB,CAAC,OAAuC,EAAE,QAAkB;IACrF,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,GAAG,CAAC;IACpC,MAAM,OAAO,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;IAC5B,MAAM,SAAS,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;IAC9B,MAAM,CAAC,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;IAC3B,MAAM,CAAC,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IAC7B,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC;QAAE,OAAO,GAAG,CAAC;IACzB,MAAM,MAAM,GAA4D;QACtE,CAAC,MAAM,EAAE,CAAC,CAAC,YAAY,EAAE,CAAC,CAAC,YAAY,CAAC;QACxC,CAAC,UAAU,EAAE,CAAC,CAAC,gBAAgB,EAAE,CAAC,CAAC,gBAAgB,CAAC;QACpD,CAAC,OAAO,EAAE,CAAC,CAAC,aAAa,EAAE,CAAC,CAAC,aAAa,CAAC;KAC5C,CAAC;IACF,OAAO,MAAM;SACV,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,EAAE,EAAE,EAAE,CAAC,EAAE,EAAE;QACtB,IAAI,OAAO,EAAE,KAAK,QAAQ,IAAI,OAAO,EAAE,KAAK,QAAQ;YAAE,OAAO,GAAG,IAAI,IAAI,CAAC;QACzE,MAAM,KAAK,GAAG,EAAE,GAAG,EAAE,CAAC;QACtB,MAAM,IAAI,GAAG,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;QACnC,OAAO,GAAG,IAAI,KAAK,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;IACnF,CAAC,CAAC;SACD,IAAI,CAAC,IAAI,CAAC,CAAC;AAChB,CAAC;AAED,SAAS,gBAAgB,CAAC,MAAc;IACtC,MAAM,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,WAAW,IAAI,CAAC,CAAC;IACxC,MAAM,GAAG,GAAG,MAAM,CAAC,QAAQ,EAAE,UAAU,EAAE,QAAQ,CAAC;IAClD,IAAI,CAAC,GAAG;QAAE,OAAO,KAAK,CAAC,EAAE,CAAC;IAC1B,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,EAAE,QAAQ,IAAI,EAAE,CAAC;IAC7C,MAAM,SAAS,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;IAC9B,MAAM,CAAC,GAAG,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACjD,IAAI,CAAC,CAAC;QAAE,OAAO,KAAK,CAAC,EAAE,CAAC;IACxB,IAAI,CAAC,CAAC,SAAS;QAAE,OAAO,KAAK,CAAC,mBAAmB,CAAC,CAAC,GAAG,IAAI,GAAG,KAAK,CAAC,CAAC,UAAU,cAAc,CAAC;IAC7F,OAAO,KAAK,CAAC,wBAAwB,CAAC,CAAC,UAAU,cAAc,CAAC;AAClE,CAAC;AAED,SAAS,oBAAoB,CAAC,MAAc;IAC1C,MAAM,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,cAAc,CAAC;IACtC,IAAI,CAAC,CAAC;QAAE,OAAO,SAAS,CAAC;IACzB,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC;QACnC,CAAC,CAAC,WAAW;QACb,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,GAAG,CAAC,CAAC,CAAC,QAAQ;YAC3B,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,KAAK,CAAC,CAAC,CAAC,YAAY;gBACjC,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,GAAG,CAAC,CAAC,CAAC,MAAM;oBACzB,CAAC,CAAC,MAAM,CAAC;IACX,OAAO,KAAK,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,OAAO,aAAa,CAAC,CAAC,aAAa,EAAE,CAAC;AAC3G,CAAC;AAED,SAAS,cAAc,CAAC,KAAmB,EAAE,QAAwC;IACnF,QAAQ,KAAK,EAAE,CAAC;QACd,KAAK,UAAU;YACb,OAAO,sEAAsE,CAAC;QAChF,KAAK,UAAU;YACb,OAAO,uIAAuI,CAAC;QACjJ,KAAK,SAAS;YACZ,OAAO,mFAAmF,CAAC;QAC7F,KAAK,OAAO;YACV,OAAO,0EAA0E,CAAC;QACpF,KAAK,cAAc;YACjB,OAAO,6DAA6D,CAAC;QACvE,KAAK,MAAM;YACT,OAAO,2FAA2F,CAAC;IACvG,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,iBAAiB,CAAC,MAAqB,EAAE,UAAiC,EAAE;IAC1F,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,CAAC,IAAI,CAAC,YAAY,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;IACvC,KAAK,CAAC,IAAI,CAAC,KAAK,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC;IACnC,IAAI,MAAM,CAAC,SAAS,CAAC,YAAY;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,SAAS,CAAC,YAAY,EAAE,CAAC,CAAC;IACnG,IAAI,MAAM,CAAC,SAAS,CAAC,UAAU;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,SAAS,CAAC,UAAU,EAAE,CAAC,CAAC;IAC/F,IAAI,MAAM,CAAC,SAAS,CAAC,cAAc;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,SAAS,CAAC,cAAc,EAAE,CAAC,CAAC;IACvG,IAAI,MAAM,CAAC,SAAS,CAAC,kBAAkB;QAAE,KAAK,CAAC,IAAI,CAAC,KAAK,MAAM,CAAC,SAAS,CAAC,kBAAkB,EAAE,CAAC,CAAC;IAChG,IAAI,OAAO,CAAC,OAAO,IAAI,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACnE,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;QACjC,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;YAC/B,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,KAAK,KAAK,CAAC,CAAC,SAAS,OAAO,CAAC,CAAC,OAAO,MAAM,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC;QAC/E,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC"}
|
|
1
|
+
{"version":3,"file":"verdict.js","sourceRoot":"","sources":["../../../src/eval-core/verdict.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAGH,OAAO,EAAE,kBAAkB,EAAE,MAAM,kBAAkB,CAAC;AA0CtD;;GAEG;AACH,MAAM,UAAU,cAAc,CAAC,MAAc,EAAE,UAA0B,EAAE;IACzE,MAAM,EAAE,aAAa,GAAG,GAAG,EAAE,kBAAkB,GAAG,GAAG,EAAE,GAAG,OAAO,CAAC;IAClE,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,EAAE,QAAQ,IAAI,EAAE,CAAC;IAC7C,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC;IACrC,MAAM,WAAW,GAAG,MAAM,CAAC,IAAI,EAAE,WAAW,IAAI,CAAC,CAAC;IAElD,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,2EAA2E;QAC3E,mCAAmC;QACnC,MAAM,IAAI,GAAG,kBAAkB,CAAC,OAAO,EAAE,aAAa,CAAC,CAAC;QACxD,OAAO;YACL,KAAK,EAAE,MAAM;YACb,QAAQ,EAAE,IAAI,CAAC,OAAO;gBACpB,CAAC,CAAC,4DAA4D,aAAa,EAAE;gBAC7E,CAAC,CAAC,8DAA8D;YAClE,SAAS,EAAE;gBACT,YAAY,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC;gBACnC,UAAU,EAAE,KAAK,WAAW,EAAE;gBAC9B,SAAS,EAAE,eAAe,CAAC,MAAM,CAAC;aACnC;YACD,QAAQ;SACT,CAAC;IACJ,CAAC;IAED,kFAAkF;IAClF,iFAAiF;IACjF,+EAA+E;IAC/E,MAAM,aAAa,GAAG,MAAM,CAAC,IAAI,EAAE,eAAe,IAAI,EAAE,CAAC;IACzD,MAAM,KAAK,GAA4B,aAAa,CAAC,MAAM,GAAG,CAAC;QAC7D,CAAC,CAAC,aAAa;QACf,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC;IAChF,MAAM,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,cAAc,CAAC,CAAC,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,aAAa,EAAE,kBAAkB,CAAC,CAAC,CAAC;IAErH,gGAAgG;IAChG,MAAM,KAAK,GAAmB,CAAC,SAAS,EAAE,UAAU,EAAE,cAAc,EAAE,OAAO,EAAE,UAAU,CAAC,CAAC;IAC3F,IAAI,QAAQ,GAAiB,UAAU,CAAC;IACxC,KAAK,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;QAC1B,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,KAAK,CAAC,EAAE,CAAC;YAC3C,QAAQ,GAAG,KAAK,CAAC;YACjB,MAAM;QACR,CAAC;IACH,CAAC;IAED,0EAA0E;IAC1E,MAAM,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,QAAQ,CAAC,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IAE/E,MAAM,YAAY,GAAG,cAAc;QACjC,CAAC,CAAC,kBAAkB,CAAC,cAAc,CAAC;QACpC,CAAC,CAAC,0DAA0D,CAAC;IAE/D,MAAM,YAAY,GAAG,kBAAkB,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;IAC3D,MAAM,UAAU,GAAG,gBAAgB,CAAC,MAAM,CAAC,CAAC;IAC5C,MAAM,SAAS,GAAG,eAAe,CAAC,MAAM,CAAC,CAAC;IAC1C,MAAM,cAAc,GAAG,oBAAoB,CAAC,MAAM,CAAC,CAAC;IACpD,MAAM,kBAAkB,GAAG,cAAc,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IAE7D,OAAO;QACL,KAAK,EAAE,QAAQ;QACf,QAAQ,EAAE,cAAc;YACtB,CAAC,CAAC,GAAG,QAAQ,MAAM,cAAc,CAAC,SAAS,OAAO,cAAc,CAAC,OAAO,KAAK,cAAc,CAAC,QAAQ,EAAE;YACtG,CAAC,CAAC,GAAG,QAAQ,MAAM,QAAQ,CAAC,MAAM,WAAW;QAC/C,OAAO;QACP,SAAS,EAAE;YACT,YAAY;YACZ,YAAY;YACZ,UAAU;YACV,SAAS;YACT,cAAc;YACd,kBAAkB;SACnB;QACD,QAAQ;KACT,CAAC;AACJ,CAAC;AAED;;;;;;;;;;GAUG;AACH,SAAS,cAAc,CACrB,IAA2B,EAC3B,OAAuC,EACvC,WAAmB,EACnB,MAAc,EACd,aAAqB,EACrB,kBAA0B;IAE1B,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,eAAe,EAAE,IAAI,EAAE,GAAG,IAAI,CAAC;IAE3D,2EAA2E;IAC3E,MAAM,KAAK,GAAG,kBAAkB,CAAC,EAAE,CAAC,OAAO,CAAC,EAAE,OAAO,CAAC,OAAO,CAAC,EAAE,EAAE,aAAa,CAAC,CAAC;IACjF,MAAM,KAAK,GAAG,kBAAkB,CAAC,EAAE,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC,SAAS,CAAC,EAAE,EAAE,aAAa,CAAC,CAAC;IAErF,2EAA2E;IAC3E,IAAI,CAAC,IAAI,EAAE,CAAC;QACV,MAAM,KAAK,GAAG,YAAY,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;QAC7C,MAAM,KAAK,GAAG,YAAY,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC;QAC/C,MAAM,KAAK,GAAG,KAAK,GAAG,KAAK,CAAC;QAC5B,MAAM,QAAQ,GAAG,KAAK,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,kCAAkC,CAAC;QACjG,IAAI,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,kBAAkB,EAAE,CAAC;YACzC,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC;QAC1D,CAAC;QACD,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS;YACzC,QAAQ;SACT,CAAC;IACJ,CAAC;IAED,MAAM,YAAY,GAAG,KAAK,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,QAAQ,QAAQ,IAAI,CAAC,GAAG,KAAK,IAAI,CAAC,IAAI,GAAG,CAAC;IAEzG,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC;QACtB,wFAAwF;QACxF,gFAAgF;QAChF,MAAM,UAAU,GAAG,MAAM,CAAC,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC,SAAS,CAAC,CAAC;QACtE,MAAM,YAAY,GAChB,WAAW,GAAG,EAAE;YAChB,CAAC,UAAU,IAAI,CAAC,UAAU,CAAC,SAAS,IAAI,UAAU,CAAC,UAAU,KAAK,MAAM,CAAC,CAAC;QAC5E,IAAI,YAAY,EAAE,CAAC;YACjB,OAAO;gBACL,OAAO;gBACP,SAAS;gBACT,KAAK,EAAE,cAAc;gBACrB,QAAQ,EAAE,GAAG,YAAY,QAAQ,WAAW,mBAAmB;aAChE,CAAC;QACJ,CAAC;QACD,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,YAAY,eAAe,EAAE,CAAC;IAC1F,CAAC;IAED,uBAAuB;IACvB,IAAI,IAAI,CAAC,QAAQ,GAAG,CAAC,EAAE,CAAC;QACtB,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,SAAS,EAAE,QAAQ,EAAE,GAAG,YAAY,oBAAoB,EAAE,CAAC;IACjG,CAAC;IAED,wDAAwD;IACxD,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;QACnB,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,UAAU;YACjB,QAAQ,EAAE,GAAG,YAAY,8CAA8C;SACxE,CAAC;IACJ,CAAC;IAED,wBAAwB;IACxB,IAAI,IAAI,CAAC,QAAQ,GAAG,kBAAkB,EAAE,CAAC;QACvC,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,UAAU;YACjB,QAAQ,EAAE,GAAG,YAAY,qCAAqC;SAC/D,CAAC;IACJ,CAAC;IAED,4EAA4E;IAC5E,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;QACnB,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,UAAU;YACjB,QAAQ,EAAE,GAAG,YAAY,2CAA2C;SACrE,CAAC;IACJ,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,UAAU,EAAE,QAAQ,EAAE,GAAG,YAAY,cAAc,EAAE,CAAC;AAC5F,CAAC;AAED,SAAS,YAAY,CAAC,CAA6B;IACjD,IAAI,CAAC,CAAC;QAAE,OAAO,CAAC,CAAC;IACjB,IAAI,OAAO,CAAC,CAAC,iBAAiB,KAAK,QAAQ;QAAE,OAAO,CAAC,CAAC,iBAAiB,CAAC;IACxE,6EAA6E;IAC7E,MAAM,MAAM,GAAG,CAAC,CAAC,CAAC,YAAY,EAAE,CAAC,CAAC,gBAAgB,EAAE,CAAC,CAAC,aAAa,CAAC,CAAC,MAAM,CACzE,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAC1C,CAAC;IACF,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAClC,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;AAC3D,CAAC;AAED,SAAS,kBAAkB,CAAC,CAA4C;IACtE,OAAO,GAAG,CAAC,CAAC,KAAK,MAAM,CAAC,CAAC,QAAQ,EAAE,CAAC;AACtC,CAAC;AAED,SAAS,kBAAkB,CAAC,OAAuC,EAAE,QAAkB;IACrF,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,GAAG,CAAC;IACpC,MAAM,OAAO,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;IAC5B,MAAM,SAAS,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;IAC9B,MAAM,CAAC,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;IAC3B,MAAM,CAAC,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IAC7B,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC;QAAE,OAAO,GAAG,CAAC;IACzB,MAAM,MAAM,GAA4D;QACtE,CAAC,MAAM,EAAE,CAAC,CAAC,YAAY,EAAE,CAAC,CAAC,YAAY,CAAC;QACxC,CAAC,UAAU,EAAE,CAAC,CAAC,gBAAgB,EAAE,CAAC,CAAC,gBAAgB,CAAC;QACpD,CAAC,OAAO,EAAE,CAAC,CAAC,aAAa,EAAE,CAAC,CAAC,aAAa,CAAC;KAC5C,CAAC;IACF,OAAO,MAAM;SACV,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,EAAE,EAAE,EAAE,CAAC,EAAE,EAAE;QACtB,IAAI,OAAO,EAAE,KAAK,QAAQ,IAAI,OAAO,EAAE,KAAK,QAAQ;YAAE,OAAO,GAAG,IAAI,IAAI,CAAC;QACzE,MAAM,KAAK,GAAG,EAAE,GAAG,EAAE,CAAC;QACtB,MAAM,IAAI,GAAG,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;QACnC,OAAO,GAAG,IAAI,KAAK,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;IACnF,CAAC,CAAC;SACD,IAAI,CAAC,IAAI,CAAC,CAAC;AAChB,CAAC;AAED,SAAS,gBAAgB,CAAC,MAAc;IACtC,MAAM,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,WAAW,IAAI,CAAC,CAAC;IACxC,MAAM,GAAG,GAAG,MAAM,CAAC,QAAQ,EAAE,UAAU,EAAE,QAAQ,CAAC;IAClD,IAAI,CAAC,GAAG;QAAE,OAAO,KAAK,CAAC,EAAE,CAAC;IAC1B,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,EAAE,QAAQ,IAAI,EAAE,CAAC;IAC7C,MAAM,SAAS,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;IAC9B,MAAM,CAAC,GAAG,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACjD,IAAI,CAAC,CAAC;QAAE,OAAO,KAAK,CAAC,EAAE,CAAC;IACxB,IAAI,CAAC,CAAC,SAAS;QAAE,OAAO,KAAK,CAAC,mBAAmB,CAAC,CAAC,GAAG,IAAI,GAAG,KAAK,CAAC,CAAC,UAAU,cAAc,CAAC;IAC7F,OAAO,KAAK,CAAC,wBAAwB,CAAC,CAAC,UAAU,cAAc,CAAC;AAClE,CAAC;AAED;;;;;;;;GAQG;AACH,SAAS,eAAe,CAAC,MAAc;IACrC,MAAM,IAAI,GAAG,MAAM,CAAC,QAAQ,EAAE,IAAI,IAAI,MAAM,CAAC,IAAI,EAAE,OAAO,EAAE,MAAM,IAAI,CAAC,CAAC;IACxE,IAAI,IAAI,GAAG,CAAC,EAAE,CAAC;QACb,OAAO,oCAAoC,CAAC;IAC9C,CAAC;IACD,MAAM,QAAQ,GAAG,MAAM,CAAC,QAAQ,EAAE,UAAU,CAAC;IAC7C,IAAI,CAAC,QAAQ,IAAI,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpD,OAAO,QAAQ,IAAI,kBAAkB,CAAC;IACxC,CAAC;IACD,+DAA+D;IAC/D,MAAM,GAAG,GAAa,EAAE,CAAC;IACzB,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;QACxC,IAAI,OAAO,CAAC,CAAC,MAAM,KAAK,QAAQ,IAAI,OAAO,CAAC,CAAC,IAAI,KAAK,QAAQ,IAAI,CAAC,CAAC,IAAI,GAAG,CAAC,EAAE,CAAC;YAC7E,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC;QAC9B,CAAC;IACH,CAAC;IACD,IAAI,GAAG,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,QAAQ,IAAI,6BAA6B,CAAC;IACvE,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC1B,MAAM,MAAM,GAAG,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC;IAC/C,MAAM,KAAK,GAAG,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;IACxC,qDAAqD;IACrD,MAAM,OAAO,GAAG,MAAM,GAAG,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC;IACnE,OAAO,MAAM,KAAK,MAAM,OAAO,UAAU,IAAI,6BAA6B,CAAC;AAC7E,CAAC;AAED,SAAS,oBAAoB,CAAC,MAAc;IAC1C,MAAM,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,cAAc,CAAC;IACtC,IAAI,CAAC,CAAC;QAAE,OAAO,SAAS,CAAC;IACzB,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC;QACnC,CAAC,CAAC,WAAW;QACb,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,GAAG,CAAC,CAAC,CAAC,QAAQ;YAC3B,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,KAAK,CAAC,CAAC,CAAC,YAAY;gBACjC,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,GAAG,CAAC,CAAC,CAAC,MAAM;oBACzB,CAAC,CAAC,MAAM,CAAC;IACX,OAAO,KAAK,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,OAAO,aAAa,CAAC,CAAC,aAAa,EAAE,CAAC;AAC3G,CAAC;AAED,SAAS,cAAc,CAAC,KAAmB,EAAE,QAAwC;IACnF,QAAQ,KAAK,EAAE,CAAC;QACd,KAAK,UAAU;YACb,OAAO,sEAAsE,CAAC;QAChF,KAAK,UAAU;YACb,OAAO,uIAAuI,CAAC;QACjJ,KAAK,SAAS;YACZ,OAAO,mFAAmF,CAAC;QAC7F,KAAK,OAAO;YACV,OAAO,0EAA0E,CAAC;QACpF,KAAK,cAAc;YACjB,OAAO,6DAA6D,CAAC;QACvE,KAAK,MAAM;YACT,OAAO,2FAA2F,CAAC;IACvG,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,iBAAiB,CAAC,MAAqB,EAAE,UAAiC,EAAE;IAC1F,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,CAAC,IAAI,CAAC,YAAY,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;IACvC,KAAK,CAAC,IAAI,CAAC,KAAK,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC;IACnC,IAAI,MAAM,CAAC,SAAS,CAAC,YAAY;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,SAAS,CAAC,YAAY,EAAE,CAAC,CAAC;IACnG,IAAI,MAAM,CAAC,SAAS,CAAC,UAAU;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,SAAS,CAAC,UAAU,EAAE,CAAC,CAAC;IAC/F,IAAI,MAAM,CAAC,SAAS,CAAC,SAAS;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,SAAS,CAAC,SAAS,EAAE,CAAC,CAAC;IAC7F,IAAI,MAAM,CAAC,SAAS,CAAC,cAAc;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,SAAS,CAAC,cAAc,EAAE,CAAC,CAAC;IACvG,IAAI,MAAM,CAAC,SAAS,CAAC,kBAAkB;QAAE,KAAK,CAAC,IAAI,CAAC,KAAK,MAAM,CAAC,SAAS,CAAC,kBAAkB,EAAE,CAAC,CAAC;IAChG,IAAI,OAAO,CAAC,OAAO,IAAI,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACnE,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;QACjC,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;YAC/B,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,KAAK,KAAK,CAAC,CAAC,SAAS,OAAO,CAAC,CAAC,OAAO,MAAM,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC;QAC/E,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import type { Artifact, JobStore, ProgressCallback, Report, VarianceData, VariantResult, VariantSummary } from '../types.js';
|
|
1
|
+
import type { Artifact, JobStore, ProgressCallback, Report, VarianceData, VariantResult, VariantSummary } from '../types/index.js';
|
|
2
2
|
interface RunSingleEvaluationOptions {
|
|
3
3
|
samplesPath: string;
|
|
4
4
|
skillDir: string;
|
|
@@ -20,7 +20,7 @@ interface RunSingleEvaluationOptions {
|
|
|
20
20
|
/** Forwarded to grade(); each sample × dimension is judged N times. Default 1. */
|
|
21
21
|
judgeRepeat?: number;
|
|
22
22
|
/** Forwarded to pipeline; ≥ 2 entries triggers multi-judge ensemble mode. */
|
|
23
|
-
judgeModels?: import('../types.js').JudgeConfig[];
|
|
23
|
+
judgeModels?: import('../types/index.js').JudgeConfig[];
|
|
24
24
|
/** v0.21 Phase 3a length-debias toggle. Default true. */
|
|
25
25
|
lengthDebias?: boolean;
|
|
26
26
|
}
|
|
@@ -79,9 +79,9 @@ export declare function buildEachReport({ skillDir, skillEntries, skillResults,
|
|
|
79
79
|
artifactHashes: {
|
|
80
80
|
[k: string]: string;
|
|
81
81
|
};
|
|
82
|
-
request: import("../types.js").EvaluationRequest;
|
|
83
|
-
run: import("../types.js").EvaluationRun;
|
|
84
|
-
job: import("../types.js").EvaluationJob;
|
|
82
|
+
request: import("../types/eval.js").EvaluationRequest;
|
|
83
|
+
run: import("../types/eval.js").EvaluationRun;
|
|
84
|
+
job: import("../types/eval.js").EvaluationJob;
|
|
85
85
|
};
|
|
86
86
|
summary: {};
|
|
87
87
|
results: never[];
|
|
@@ -98,9 +98,9 @@ export declare function buildEachReport({ skillDir, skillEntries, skillResults,
|
|
|
98
98
|
};
|
|
99
99
|
artifacts: EachSkillResult[];
|
|
100
100
|
};
|
|
101
|
-
job: import("../types.js").EvaluationJob;
|
|
101
|
+
job: import("../types/eval.js").EvaluationJob;
|
|
102
102
|
};
|
|
103
|
-
export declare function executeEachEvaluationRuns({ skillDir, skillEntries, model, judgeModel, outputDir, project, owner, tags, noJudge, concurrency, timeoutMs, executorName, judgeExecutorName, jobStore, persistJob, onProgress, onSkillProgress, skipPreflight, mcpConfig, verbose, repeat, judgeRepeat, judgeModels, lengthDebias, runSingleEvaluation, }: {
|
|
103
|
+
export declare function executeEachEvaluationRuns({ skillDir, skillEntries, model, judgeModel, outputDir, project, owner, tags, noJudge, concurrency, timeoutMs, executorName, judgeExecutorName, jobStore, persistJob, onProgress, onSkillProgress, skipPreflight, mcpConfig, verbose, repeat, judgeRepeat, judgeModels, lengthDebias, strictBaseline, variantAllowedSkills, runSingleEvaluation, }: {
|
|
104
104
|
skillDir: string;
|
|
105
105
|
skillEntries: Array<{
|
|
106
106
|
name: string;
|
|
@@ -132,8 +132,15 @@ export declare function executeEachEvaluationRuns({ skillDir, skillEntries, mode
|
|
|
132
132
|
verbose?: boolean;
|
|
133
133
|
repeat?: number;
|
|
134
134
|
judgeRepeat?: number;
|
|
135
|
-
judgeModels?: import('../types.js').JudgeConfig[];
|
|
135
|
+
judgeModels?: import('../types/index.js').JudgeConfig[];
|
|
136
136
|
lengthDebias?: boolean;
|
|
137
|
+
/** v0.22 — strict-baseline default. Forwarded to per-skill resolveArtifacts so each
|
|
138
|
+
* skill's baseline gets allowedSkills=[] when default true. */
|
|
139
|
+
strictBaseline?: boolean;
|
|
140
|
+
/** v0.22 — explicit per-variant allowedSkills override. each mode rarely uses this
|
|
141
|
+
* (since it auto-pairs baseline + skill), but if user has yaml override on either
|
|
142
|
+
* "baseline" or "skill" name, we honor it. */
|
|
143
|
+
variantAllowedSkills?: Record<string, string[]>;
|
|
137
144
|
runSingleEvaluation: (options: RunSingleEvaluationOptions) => Promise<{
|
|
138
145
|
report: Report;
|
|
139
146
|
filePath: string | null;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"each-evaluation-workflow.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/each-evaluation-workflow.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,QAAQ,EAAE,QAAQ,EAAE,gBAAgB,EAAE,MAAM,EAAE,YAAY,EAAE,aAAa,EAAE,cAAc,EAAE,MAAM,
|
|
1
|
+
{"version":3,"file":"each-evaluation-workflow.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/each-evaluation-workflow.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,QAAQ,EAAE,QAAQ,EAAE,gBAAgB,EAAE,MAAM,EAAE,YAAY,EAAE,aAAa,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AAEnI,UAAU,0BAA0B;IAClC,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,IAAI,CAAC;IAChB,OAAO,EAAE,OAAO,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,EAAE,IAAI,CAAC;IACf,UAAU,EAAE,KAAK,CAAC;IAClB,UAAU,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACpC,aAAa,EAAE,OAAO,CAAC;IACvB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,kFAAkF;IAClF,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,6EAA6E;IAC7E,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,yDAAyD;IACzD,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IACxC,0DAA0D;IAC1D,QAAQ,CAAC,EAAE,YAAY,CAAC;IACxB,OAAO,EAAE,KAAK,CAAC;QACb,SAAS,EAAE,MAAM,CAAC;QAClB,QAAQ,EAAE;YACR,QAAQ,EAAE,aAAa,CAAC;YACxB,KAAK,EAAE,aAAa,CAAC;SACtB,CAAC;KACH,CAAC,CAAC;CACJ;AAsBD,wBAAgB,eAAe,CAAC,EAC9B,QAAQ,EACR,YAAY,EACZ,YAAY,EACZ,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,WAAW,EACX,SAAS,EACT,YAAY,EACZ,MAAM,GACP,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC9E,YAAY,EAAE,eAAe,EAAE,CAAC;IAChC,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,MAAM,EAAE,OAAO,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAwEA;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,QAAQ,EACR,YAAY,EACZ,KAAK,EACL,UAAU,EACV,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,WAAe,EACf,SAAS,EACT,YAAY,EACZ,iBAAiB,EACjB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,eAAsB,EACtB,aAAqB,EACrB,SAAS,EACT,OAAe,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,YAAY,EACZ,cAAc,EACd,oBAAoB,EACpB,mBAAmB,GACpB,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC9E,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACrC,eAAe,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;IAC5G,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB;oEACgE;IAChE,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB;;mDAE+C;IAC/C,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;IAChD,mBAAmB,EAAE,CAAC,OAAO,EAAE,0BAA0B,KAAK,OAAO,CAAC;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;KAAE,CAAC,CAAC;CACpH,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CA0FvD"}
|
|
@@ -92,7 +92,7 @@ export function buildEachReport({ skillDir, skillEntries, skillResults, model, j
|
|
|
92
92
|
job,
|
|
93
93
|
};
|
|
94
94
|
}
|
|
95
|
-
export async function executeEachEvaluationRuns({ skillDir, skillEntries, model, judgeModel, outputDir = DEFAULT_OUTPUT_DIR, project, owner, tags, noJudge = false, concurrency = 1, timeoutMs, executorName, judgeExecutorName, jobStore = null, persistJob = true, onProgress = null, onSkillProgress = null, skipPreflight = false, mcpConfig, verbose = false, repeat, judgeRepeat, judgeModels, lengthDebias, runSingleEvaluation, }) {
|
|
95
|
+
export async function executeEachEvaluationRuns({ skillDir, skillEntries, model, judgeModel, outputDir = DEFAULT_OUTPUT_DIR, project, owner, tags, noJudge = false, concurrency = 1, timeoutMs, executorName, judgeExecutorName, jobStore = null, persistJob = true, onProgress = null, onSkillProgress = null, skipPreflight = false, mcpConfig, verbose = false, repeat, judgeRepeat, judgeModels, lengthDebias, strictBaseline, variantAllowedSkills, runSingleEvaluation, }) {
|
|
96
96
|
const skillResults = [];
|
|
97
97
|
let totalCostUSD = 0;
|
|
98
98
|
for (let i = 0; i < skillEntries.length; i++) {
|
|
@@ -100,7 +100,8 @@ export async function executeEachEvaluationRuns({ skillDir, skillEntries, model,
|
|
|
100
100
|
onSkillProgress?.({ phase: 'start', skill: entry.name, current: i + 1, total: skillEntries.length });
|
|
101
101
|
// each mode 的实验结构固定为"baseline (control) vs skill (treatment)"。
|
|
102
102
|
// 显式在 artifact 上填 experimentRole,下游 buildVariantConfig 直接读取。
|
|
103
|
-
|
|
103
|
+
// v0.22 — pass strictBaseline + variantAllowedSkills opts so isolation 默认生效。
|
|
104
|
+
const skillArtifacts = resolveArtifacts(resolve(skillDir), ['baseline', entry.skillPath], { strictBaseline, variantAllowedSkills }).map((artifact) => {
|
|
104
105
|
if (artifact.name === entry.skillPath) {
|
|
105
106
|
return { ...artifact, name: 'skill', experimentRole: 'treatment' };
|
|
106
107
|
}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"each-evaluation-workflow.js","sourceRoot":"","sources":["../../../src/eval-workflows/each-evaluation-workflow.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,kBAAkB,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACxG,OAAO,EAAE,sBAAsB,EAAE,mBAAmB,EAAE,kBAAkB,EAAE,qBAAqB,EAAE,MAAM,gCAAgC,CAAC;AACxI,OAAO,EAAE,kBAAkB,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC9E,OAAO,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AA8C7D,SAAS,iBAAiB,CAAC,YAA+B,EAAE,YAAoB;IAC9E,OAAO;QACL,cAAc,EAAE,YAAY,CAAC,MAAM;QACnC,YAAY,EAAE,YAAY,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC;QAC7E,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC7C,SAAS,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;YACpC,MAAM,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC;YAC/C,MAAM,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;YAC5C,MAAM,aAAa,GAAG,eAAe,EAAE,iBAAiB,IAAI,eAAe,EAAE,WAAW,IAAI,IAAI,CAAC;YACjG,MAAM,aAAa,GAAG,eAAe,EAAE,iBAAiB,IAAI,eAAe,EAAE,WAAW,IAAI,IAAI,CAAC;YACjG,IAAI,OAAO,aAAa,KAAK,QAAQ,IAAI,OAAO,aAAa,KAAK,QAAQ,IAAI,aAAa,IAAI,CAAC,EAAE,CAAC;gBACjG,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,aAAa,EAAE,aAAa,EAAE,WAAW,EAAE,GAAG,EAAE,CAAC;YAC9E,CAAC;YACD,MAAM,KAAK,GAAG,CAAC,CAAC,aAAa,GAAG,aAAa,CAAC,GAAG,aAAa,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YACjF,MAAM,WAAW,GAAG,aAAa,IAAI,aAAa,CAAC,CAAC,CAAC,IAAI,KAAK,GAAG,CAAC,CAAC,CAAC,GAAG,KAAK,GAAG,CAAC;YAChF,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,aAAa,EAAE,aAAa,EAAE,WAAW,EAAE,CAAC;QACzE,CAAC,CAAC;KACH,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,EAC9B,QAAQ,EACR,YAAY,EACZ,YAAY,EACZ,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,WAAW,EACX,SAAS,EACT,YAAY,EACZ,MAAM,GAkBP;IACC,MAAM,KAAK,GAAG,aAAa,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC;IACtC,MAAM,OAAO,GAAG,sBAAsB,CAAC;QACrC,WAAW,EAAE,EAAE;QACf,QAAQ;QACR,SAAS,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;YACtC,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,IAAI,EAAE,OAAO;YACb,MAAM,EAAE,WAAW;YACnB,OAAO,EAAE,IAAI;YACb,OAAO,EAAE,KAAK,CAAC,SAAS;SACzB,CAAC,CAAC;QACH,KAAK;QACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;QACvC,QAAQ,EAAE,YAAY;QACtB,aAAa,EAAE,iBAAiB,IAAI,YAAY;QAChD,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO,EAAE,KAAK;QACd,MAAM;QACN,KAAK,EAAE,KAAK;QACZ,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM;QACN,IAAI,EAAE,IAAI;KACX,CAAC,CAAC;IACH,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,MAAM,EAAE,GAAG,EAAE,UAAU,EAAE,SAAS,EAAE,GAAG,mBAAmB,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC7E,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,GAAG,GAAG,qBAAqB,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IAC1D,MAAM,GAAG,GAAG,kBAAkB,CAAC;QAC7B,KAAK,EAAE,OAAO,KAAK,EAAE;QACrB,KAAK;QACL,QAAQ,EAAE,KAAK;QACf,OAAO;QACP,SAAS;QACT,SAAS;QACT,UAAU;KACX,CAAC,CAAC;IACH,MAAM,gBAAgB,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;IAEzF,OAAO;QACL,MAAM,EAAE;YACN,EAAE,EAAE,KAAK;YACT,IAAI,EAAE,IAAI;YACV,IAAI,EAAE;gBACJ,QAAQ,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC;gBACjD,KAAK;gBACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;gBACvC,QAAQ,EAAE,YAAY;gBACtB,WAAW,EAAE,gBAAgB;gBAC7B,SAAS,EAAE,gBAAgB,GAAG,CAAC;gBAC/B,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;gBAC7C,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;gBACnC,UAAU,EAAE,EAAE;gBACd,WAAW,EAAE,OAAO,CAAC,OAAO;gBAC5B,cAAc,EAAE,MAAM,CAAC,WAAW,CAChC,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,YAAY,IAAI,UAAU,CAAC,CAAC,CAC5E;gBACD,OAAO;gBACP,GAAG;gBACH,GAAG;aACJ;YACD,OAAO,EAAE,EAAE;YACX,OAAO,EAAE,EAAE;YACX,QAAQ,EAAE,iBAAiB,CAAC,YAAY,EAAE,YAAY,CAAC;YACvD,SAAS,EAAE,YAAY;SACP;QAClB,GAAG;KACJ,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,EAC9C,QAAQ,EACR,YAAY,EACZ,KAAK,EACL,UAAU,EACV,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAO,GAAG,KAAK,EACf,WAAW,GAAG,CAAC,EACf,SAAS,EACT,YAAY,EACZ,iBAAiB,EACjB,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,eAAe,GAAG,IAAI,EACtB,aAAa,GAAG,KAAK,EACrB,SAAS,EACT,OAAO,GAAG,KAAK,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,YAAY,EACZ,mBAAmB,
|
|
1
|
+
{"version":3,"file":"each-evaluation-workflow.js","sourceRoot":"","sources":["../../../src/eval-workflows/each-evaluation-workflow.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,kBAAkB,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACxG,OAAO,EAAE,sBAAsB,EAAE,mBAAmB,EAAE,kBAAkB,EAAE,qBAAqB,EAAE,MAAM,gCAAgC,CAAC;AACxI,OAAO,EAAE,kBAAkB,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC9E,OAAO,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AA8C7D,SAAS,iBAAiB,CAAC,YAA+B,EAAE,YAAoB;IAC9E,OAAO;QACL,cAAc,EAAE,YAAY,CAAC,MAAM;QACnC,YAAY,EAAE,YAAY,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC;QAC7E,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC7C,SAAS,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;YACpC,MAAM,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC;YAC/C,MAAM,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;YAC5C,MAAM,aAAa,GAAG,eAAe,EAAE,iBAAiB,IAAI,eAAe,EAAE,WAAW,IAAI,IAAI,CAAC;YACjG,MAAM,aAAa,GAAG,eAAe,EAAE,iBAAiB,IAAI,eAAe,EAAE,WAAW,IAAI,IAAI,CAAC;YACjG,IAAI,OAAO,aAAa,KAAK,QAAQ,IAAI,OAAO,aAAa,KAAK,QAAQ,IAAI,aAAa,IAAI,CAAC,EAAE,CAAC;gBACjG,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,aAAa,EAAE,aAAa,EAAE,WAAW,EAAE,GAAG,EAAE,CAAC;YAC9E,CAAC;YACD,MAAM,KAAK,GAAG,CAAC,CAAC,aAAa,GAAG,aAAa,CAAC,GAAG,aAAa,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YACjF,MAAM,WAAW,GAAG,aAAa,IAAI,aAAa,CAAC,CAAC,CAAC,IAAI,KAAK,GAAG,CAAC,CAAC,CAAC,GAAG,KAAK,GAAG,CAAC;YAChF,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,aAAa,EAAE,aAAa,EAAE,WAAW,EAAE,CAAC;QACzE,CAAC,CAAC;KACH,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,EAC9B,QAAQ,EACR,YAAY,EACZ,YAAY,EACZ,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,WAAW,EACX,SAAS,EACT,YAAY,EACZ,MAAM,GAkBP;IACC,MAAM,KAAK,GAAG,aAAa,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC;IACtC,MAAM,OAAO,GAAG,sBAAsB,CAAC;QACrC,WAAW,EAAE,EAAE;QACf,QAAQ;QACR,SAAS,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;YACtC,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,IAAI,EAAE,OAAO;YACb,MAAM,EAAE,WAAW;YACnB,OAAO,EAAE,IAAI;YACb,OAAO,EAAE,KAAK,CAAC,SAAS;SACzB,CAAC,CAAC;QACH,KAAK;QACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;QACvC,QAAQ,EAAE,YAAY;QACtB,aAAa,EAAE,iBAAiB,IAAI,YAAY;QAChD,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO,EAAE,KAAK;QACd,MAAM;QACN,KAAK,EAAE,KAAK;QACZ,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM;QACN,IAAI,EAAE,IAAI;KACX,CAAC,CAAC;IACH,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,MAAM,EAAE,GAAG,EAAE,UAAU,EAAE,SAAS,EAAE,GAAG,mBAAmB,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC7E,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,GAAG,GAAG,qBAAqB,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IAC1D,MAAM,GAAG,GAAG,kBAAkB,CAAC;QAC7B,KAAK,EAAE,OAAO,KAAK,EAAE;QACrB,KAAK;QACL,QAAQ,EAAE,KAAK;QACf,OAAO;QACP,SAAS;QACT,SAAS;QACT,UAAU;KACX,CAAC,CAAC;IACH,MAAM,gBAAgB,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;IAEzF,OAAO;QACL,MAAM,EAAE;YACN,EAAE,EAAE,KAAK;YACT,IAAI,EAAE,IAAI;YACV,IAAI,EAAE;gBACJ,QAAQ,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC;gBACjD,KAAK;gBACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;gBACvC,QAAQ,EAAE,YAAY;gBACtB,WAAW,EAAE,gBAAgB;gBAC7B,SAAS,EAAE,gBAAgB,GAAG,CAAC;gBAC/B,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;gBAC7C,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;gBACnC,UAAU,EAAE,EAAE;gBACd,WAAW,EAAE,OAAO,CAAC,OAAO;gBAC5B,cAAc,EAAE,MAAM,CAAC,WAAW,CAChC,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,YAAY,IAAI,UAAU,CAAC,CAAC,CAC5E;gBACD,OAAO;gBACP,GAAG;gBACH,GAAG;aACJ;YACD,OAAO,EAAE,EAAE;YACX,OAAO,EAAE,EAAE;YACX,QAAQ,EAAE,iBAAiB,CAAC,YAAY,EAAE,YAAY,CAAC;YACvD,SAAS,EAAE,YAAY;SACP;QAClB,GAAG;KACJ,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,EAC9C,QAAQ,EACR,YAAY,EACZ,KAAK,EACL,UAAU,EACV,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAO,GAAG,KAAK,EACf,WAAW,GAAG,CAAC,EACf,SAAS,EACT,YAAY,EACZ,iBAAiB,EACjB,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,eAAe,GAAG,IAAI,EACtB,aAAa,GAAG,KAAK,EACrB,SAAS,EACT,OAAO,GAAG,KAAK,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,YAAY,EACZ,cAAc,EACd,oBAAoB,EACpB,mBAAmB,GAkCpB;IACC,MAAM,YAAY,GAAsB,EAAE,CAAC;IAC3C,IAAI,YAAY,GAAG,CAAC,CAAC;IAErB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,YAAY,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC7C,MAAM,KAAK,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC;QAC9B,eAAe,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,YAAY,CAAC,MAAM,EAAE,CAAC,CAAC;QAErG,+DAA+D;QAC/D,6DAA6D;QAC7D,6EAA6E;QAC7E,MAAM,cAAc,GAAG,gBAAgB,CACrC,OAAO,CAAC,QAAQ,CAAC,EACjB,CAAC,UAAU,EAAE,KAAK,CAAC,SAAS,CAAC,EAC7B,EAAE,cAAc,EAAE,oBAAoB,EAAE,CACzC,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE;YACjB,IAAI,QAAQ,CAAC,IAAI,KAAK,KAAK,CAAC,SAAS,EAAE,CAAC;gBACtC,OAAO,EAAE,GAAG,QAAQ,EAAE,IAAI,EAAE,OAAO,EAAE,cAAc,EAAE,WAAoB,EAAE,CAAC;YAC9E,CAAC;YACD,OAAO,EAAE,GAAG,QAAQ,EAAE,cAAc,EAAE,SAAkB,EAAE,CAAC;QAC7D,CAAC,CAAC,CAAC;QACH,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM,mBAAmB,CAAC;YAC3C,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,QAAQ;YACR,SAAS,EAAE,cAAc;YACzB,KAAK;YACL,UAAU;YACV,SAAS,EAAE,IAAI;YACf,OAAO;YACP,WAAW;YACX,SAAS;YACT,YAAY;YACZ,iBAAiB;YACjB,QAAQ,EAAE,IAAI;YACd,UAAU,EAAE,KAAK;YACjB,UAAU;YACV,aAAa,EAAE,aAAa,IAAI,CAAC,GAAG,CAAC;YACrC,SAAS;YACT,OAAO;YACP,WAAW;YACX,WAAW;YACX,YAAY;SACb,CAAC,CAAC;QAEH,YAAY,CAAC,IAAI,CAAC;YAChB,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,YAAY,EAAE,MAAM,CAAC,IAAI,CAAC,cAAc,EAAE,KAAK,IAAI,EAAE;YACrD,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,WAAW;YACpC,OAAO,EAAE;gBACP,QAAQ,EAAE,MAAM,CAAC,OAAO,CAAC,QAAQ,IAAK,EAAqB;gBAC3D,KAAK,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,IAAI,EAAE,CAAmB;aACtD;YACD,iEAAiE;YACjE,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACzD,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;gBACvC,SAAS,EAAE,MAAM,CAAC,SAAS;gBAC3B,QAAQ,EAAE;oBACR,QAAQ,EAAE,MAAM,CAAC,QAAQ,CAAC,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC;oBACjE,KAAK,EAAE,MAAM,CAAC,QAAQ,CAAC,KAAK;iBAC7B;aACF,CAAC,CAAC;SACJ,CAAC,CAAC;QAEH,YAAY,IAAI,MAAM,CAAC,IAAI,CAAC,YAAY,CAAC;QACzC,eAAe,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,YAAY,CAAC,MAAM,EAAE,CAAC,CAAC;IACtG,CAAC;IAED,MAAM,EAAE,MAAM,EAAE,cAAc,EAAE,GAAG,EAAE,GAAG,eAAe,CAAC;QACtD,QAAQ;QACR,YAAY;QACZ,YAAY;QACZ,KAAK;QACL,UAAU;QACV,OAAO;QACP,YAAY;QACZ,iBAAiB;QACjB,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM,EAAE,KAAK;QACb,WAAW;QACX,SAAS;QACT,YAAY;QACZ,MAAM;KACP,CAAC,CAAC;IACH,MAAM,QAAQ,GAAG,aAAa,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC;IAC1D,MAAM,gBAAgB,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,QAAQ,IAAI,kBAAkB,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAChG,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IAClE,OAAO,EAAE,MAAM,EAAE,cAAc,EAAE,QAAQ,EAAE,CAAC;AAC9C,CAAC"}
|
|
@@ -1,5 +1,30 @@
|
|
|
1
1
|
import type { DependencyRequirements } from '../eval-core/dependency-checker.js';
|
|
2
|
-
import type { Artifact, ExecutorFn, JobStore, ProgressCallback, Report, Sample, Task, VariantResult } from '../types.js';
|
|
2
|
+
import type { Artifact, ExecutorFn, JobStore, ProgressCallback, Report, Sample, Task, VariantResult } from '../types/index.js';
|
|
3
|
+
/**
|
|
4
|
+
* Compute structural power warnings (pure function, no I/O — for testing).
|
|
5
|
+
*
|
|
6
|
+
* Not MDE / power-analysis predictions (we don't have σ pre-run; predicting
|
|
7
|
+
* "CI half-width ~ ±0.4" before any data exists is hand-wave). These are
|
|
8
|
+
* **hard-floor + experience-based** thresholds:
|
|
9
|
+
* - n < 5: any conclusion unreliable, CI uselessly wide
|
|
10
|
+
* - 5 ≤ n < 20: only large effects (Cohen's d > 0.8) detectable
|
|
11
|
+
* - repeat=1: stability cannot be measured at all
|
|
12
|
+
*
|
|
13
|
+
* Real power claims happen post-hoc via `bench verdict` UNDERPOWERED state +
|
|
14
|
+
* saturation curves. This is the upfront "you might be wasting the run"
|
|
15
|
+
* heads-up, not a gate.
|
|
16
|
+
*/
|
|
17
|
+
export declare function buildPowerWarnings(sampleCount: number, repeat: number): string[];
|
|
18
|
+
/**
|
|
19
|
+
* Pre-flight warning emitted when user explicitly opts out of strict-baseline
|
|
20
|
+
* (--no-strict-baseline) AND there are baseline-kind variants AND ~/.claude/skills/
|
|
21
|
+
* has content. baseline 会被 SDK 全发现污染 → verdict / Δ 不可信。
|
|
22
|
+
*
|
|
23
|
+
* 默认 strict 时(strictBaseline === true / undefined)不出 warn。
|
|
24
|
+
*
|
|
25
|
+
* Exported for tests.
|
|
26
|
+
*/
|
|
27
|
+
export declare function buildIsolationWarnings(artifacts: Artifact[], strictBaseline: boolean | undefined): string[];
|
|
3
28
|
export interface EvaluationPipelineOptions {
|
|
4
29
|
samplesPath: string;
|
|
5
30
|
skillDir: string;
|
|
@@ -37,7 +62,7 @@ export interface EvaluationPipelineOptions {
|
|
|
37
62
|
/** 透传到 meta.request.judgeRepeat 与 grade(),每条 sample × dimension judge N 次 */
|
|
38
63
|
judgeRepeat?: number;
|
|
39
64
|
/** Multi-judge ensemble configs (≥ 2 entries triggers ensemble mode). */
|
|
40
|
-
judgeModels?: import('../types.js').JudgeConfig[];
|
|
65
|
+
judgeModels?: import('../types/index.js').JudgeConfig[];
|
|
41
66
|
/** --bootstrap. */
|
|
42
67
|
bootstrap?: boolean;
|
|
43
68
|
/** --bootstrap-samples N. Default 1000. */
|
|
@@ -45,9 +70,13 @@ export interface EvaluationPipelineOptions {
|
|
|
45
70
|
/** v0.21 length-debias toggle. Default true; --no-debias-length flips to false. */
|
|
46
71
|
lengthDebias?: boolean;
|
|
47
72
|
/** v0.22 — hard budget caps. */
|
|
48
|
-
budget?: import('../types.js').EvalBudget;
|
|
73
|
+
budget?: import('../types/index.js').EvalBudget;
|
|
74
|
+
/** v0.22 — strict-baseline default state (only used to decide whether to emit
|
|
75
|
+
* isolation-disabled pre-flight warnings). True/undefined = default behavior
|
|
76
|
+
* (no warning); false = user explicitly disabled, warn if ~/.claude/skills/ has content. */
|
|
77
|
+
strictBaseline?: boolean;
|
|
49
78
|
}
|
|
50
|
-
export declare function executeEvaluationPipeline({ samplesPath, skillDir, samples, tasks, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, executor, judgeExecutor, outputDir, project, owner, tags, blind, concurrency, timeoutMs, noCache, jobStore, persistJob, onProgress, skipPreflight, verbose, retry, existingResults, requires, layeredStats, repeat, each, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, }: EvaluationPipelineOptions): Promise<{
|
|
79
|
+
export declare function executeEvaluationPipeline({ samplesPath, skillDir, samples, tasks, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, executor, judgeExecutor, outputDir, project, owner, tags, blind, concurrency, timeoutMs, noCache, jobStore, persistJob, onProgress, skipPreflight, verbose, retry, existingResults, requires, layeredStats, repeat, each, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, strictBaseline, }: EvaluationPipelineOptions): Promise<{
|
|
51
80
|
report: Report;
|
|
52
81
|
filePath: string | null;
|
|
53
82
|
}>;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"evaluation-pipeline.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"evaluation-pipeline.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AAgBjF,OAAO,KAAK,EACV,QAAQ,EAIR,UAAU,EACV,QAAQ,EACR,gBAAgB,EAChB,MAAM,EACN,MAAM,EACN,IAAI,EACJ,aAAa,EACd,MAAM,mBAAmB,CAAC;AA4J3B;;;;;;;;;;;;;GAaG;AACH,wBAAgB,kBAAkB,CAAC,WAAW,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,MAAM,EAAE,CAiBhF;AAQD;;;;;;;;GAQG;AACH,wBAAgB,sBAAsB,CACpC,SAAS,EAAE,QAAQ,EAAE,EACrB,cAAc,EAAE,OAAO,GAAG,SAAS,GAClC,MAAM,EAAE,CAsBV;AA2DD,MAAM,WAAW,yBAAyB;IACxC,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,QAAQ,EAAE,UAAU,CAAC;IACrB,aAAa,EAAE,UAAU,CAAC;IAC1B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACrC,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,eAAe,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC,CAAC;IAChE,QAAQ,CAAC,EAAE,sBAAsB,CAAC;IAClC,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,8BAA8B;IAC9B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,4BAA4B;IAC5B,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,6EAA6E;IAC7E,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,yEAAyE;IACzE,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,mBAAmB;IACnB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,2CAA2C;IAC3C,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,mFAAmF;IACnF,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,gCAAgC;IAChC,MAAM,CAAC,EAAE,OAAO,mBAAmB,EAAE,UAAU,CAAC;IAChD;;iGAE6F;IAC7F,cAAc,CAAC,EAAE,OAAO,CAAC;CAC1B;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,WAAW,EACX,QAAQ,EACR,OAAO,EACP,KAAK,EACL,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,QAAQ,EACR,aAAa,EACb,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAa,EACb,WAAe,EACf,SAAS,EACT,OAAe,EACf,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,aAAqB,EACrB,OAAe,EACf,KAAS,EACT,eAAe,EACf,QAAQ,EACR,YAAoB,EACpB,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAmB,EACnB,MAAM,EACN,cAAc,GACf,EAAE,yBAAyB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAuIlF"}
|
|
@@ -1,5 +1,7 @@
|
|
|
1
1
|
import { createHash } from 'node:crypto';
|
|
2
|
-
import { existsSync, readFileSync } from 'node:fs';
|
|
2
|
+
import { existsSync, readFileSync, readdirSync } from 'node:fs';
|
|
3
|
+
import { homedir } from 'node:os';
|
|
4
|
+
import { join } from 'node:path';
|
|
3
5
|
import { analyzeResults } from '../analysis/report-diagnostics.js';
|
|
4
6
|
import { computeReportCoverage } from '../analysis/coverage-analyzer.js';
|
|
5
7
|
import { computeReportGapRates } from '../analysis/gap-analyzer.js';
|
|
@@ -95,6 +97,76 @@ function computeTestSetHash(samplesPath) {
|
|
|
95
97
|
return null;
|
|
96
98
|
}
|
|
97
99
|
}
|
|
100
|
+
/**
|
|
101
|
+
* Compute structural power warnings (pure function, no I/O — for testing).
|
|
102
|
+
*
|
|
103
|
+
* Not MDE / power-analysis predictions (we don't have σ pre-run; predicting
|
|
104
|
+
* "CI half-width ~ ±0.4" before any data exists is hand-wave). These are
|
|
105
|
+
* **hard-floor + experience-based** thresholds:
|
|
106
|
+
* - n < 5: any conclusion unreliable, CI uselessly wide
|
|
107
|
+
* - 5 ≤ n < 20: only large effects (Cohen's d > 0.8) detectable
|
|
108
|
+
* - repeat=1: stability cannot be measured at all
|
|
109
|
+
*
|
|
110
|
+
* Real power claims happen post-hoc via `bench verdict` UNDERPOWERED state +
|
|
111
|
+
* saturation curves. This is the upfront "you might be wasting the run"
|
|
112
|
+
* heads-up, not a gate.
|
|
113
|
+
*/
|
|
114
|
+
export function buildPowerWarnings(sampleCount, repeat) {
|
|
115
|
+
const warnings = [];
|
|
116
|
+
if (sampleCount < 5) {
|
|
117
|
+
warnings.push(`⚠ N=${sampleCount} < 5 (exploration-only): any conclusion is unreliable, CI will be uselessly wide. Decisions need ≥20 cases.`);
|
|
118
|
+
}
|
|
119
|
+
else if (sampleCount < 20) {
|
|
120
|
+
warnings.push(`⚠ N=${sampleCount} < 20 (large-effect-only, Cohen's d > 0.8): medium effects (d ≈ 0.5) hard to detect. For confident decisions consider ≥20 cases.`);
|
|
121
|
+
}
|
|
122
|
+
if (repeat < 2) {
|
|
123
|
+
warnings.push(`⚠ --repeat=1: single-run cannot measure stability (CV will be marked "not measured"). Use --repeat 3+ to detect within-variant variance.`);
|
|
124
|
+
}
|
|
125
|
+
return warnings;
|
|
126
|
+
}
|
|
127
|
+
function emitPowerWarnings(sampleCount, repeat) {
|
|
128
|
+
for (const w of buildPowerWarnings(sampleCount, repeat)) {
|
|
129
|
+
process.stderr.write(`${w}\n`);
|
|
130
|
+
}
|
|
131
|
+
}
|
|
132
|
+
/**
|
|
133
|
+
* Pre-flight warning emitted when user explicitly opts out of strict-baseline
|
|
134
|
+
* (--no-strict-baseline) AND there are baseline-kind variants AND ~/.claude/skills/
|
|
135
|
+
* has content. baseline 会被 SDK 全发现污染 → verdict / Δ 不可信。
|
|
136
|
+
*
|
|
137
|
+
* 默认 strict 时(strictBaseline === true / undefined)不出 warn。
|
|
138
|
+
*
|
|
139
|
+
* Exported for tests.
|
|
140
|
+
*/
|
|
141
|
+
export function buildIsolationWarnings(artifacts, strictBaseline) {
|
|
142
|
+
// Only warn when user explicitly disabled isolation.
|
|
143
|
+
if (strictBaseline !== false)
|
|
144
|
+
return [];
|
|
145
|
+
const hasBaselineKind = artifacts.some((a) => a.kind === 'baseline');
|
|
146
|
+
if (!hasBaselineKind)
|
|
147
|
+
return [];
|
|
148
|
+
// Check ~/.claude/skills/ for content (avoid hard-coding home — read at runtime).
|
|
149
|
+
const skillsDir = join(homedir(), '.claude', 'skills');
|
|
150
|
+
if (!existsSync(skillsDir))
|
|
151
|
+
return [];
|
|
152
|
+
let skillCount = 0;
|
|
153
|
+
try {
|
|
154
|
+
skillCount = readdirSync(skillsDir).filter((entry) => !entry.startsWith('.')).length;
|
|
155
|
+
}
|
|
156
|
+
catch {
|
|
157
|
+
return [];
|
|
158
|
+
}
|
|
159
|
+
if (skillCount === 0)
|
|
160
|
+
return [];
|
|
161
|
+
return [
|
|
162
|
+
`⚠ baseline 隔离已关闭(--no-strict-baseline)。检测到 ~/.claude/skills/ 内有 ${skillCount} 个 skill, baseline variant 可能被 auto-discovery 污染。除非你确认要这种比较,建议恢复默认 strict 模式。`,
|
|
163
|
+
];
|
|
164
|
+
}
|
|
165
|
+
function emitIsolationWarnings(artifacts, strictBaseline) {
|
|
166
|
+
for (const w of buildIsolationWarnings(artifacts, strictBaseline)) {
|
|
167
|
+
process.stderr.write(`${w}\n`);
|
|
168
|
+
}
|
|
169
|
+
}
|
|
98
170
|
function finalizeEvaluationReport({ report, results, artifacts, variantNames, blind, samplesPath, }) {
|
|
99
171
|
report.analysis = analyzeResults(report);
|
|
100
172
|
const hasToolData = Object.values(results).some((sampleResults) => (Object.values(sampleResults).some((variantResult) => variantResult.toolCalls && variantResult.toolCalls.length > 0)));
|
|
@@ -126,7 +198,7 @@ function finalizeEvaluationReport({ report, results, artifacts, variantNames, bl
|
|
|
126
198
|
}
|
|
127
199
|
return report;
|
|
128
200
|
}
|
|
129
|
-
export async function executeEvaluationPipeline({ samplesPath, skillDir, samples, tasks, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, executor, judgeExecutor, outputDir = DEFAULT_OUTPUT_DIR, project, owner, tags, blind = false, concurrency = 1, timeoutMs, noCache = false, jobStore = null, persistJob = true, onProgress = null, skipPreflight = false, verbose = false, retry = 0, existingResults, requires, layeredStats = false, repeat, each, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias = true, budget, }) {
|
|
201
|
+
export async function executeEvaluationPipeline({ samplesPath, skillDir, samples, tasks, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, executor, judgeExecutor, outputDir = DEFAULT_OUTPUT_DIR, project, owner, tags, blind = false, concurrency = 1, timeoutMs, noCache = false, jobStore = null, persistJob = true, onProgress = null, skipPreflight = false, verbose = false, retry = 0, existingResults, requires, layeredStats = false, repeat, each, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias = true, budget, strictBaseline, }) {
|
|
130
202
|
const variantNames = artifacts.map((artifact) => artifact.name);
|
|
131
203
|
const runState = await initializeEvaluationRunState({
|
|
132
204
|
samplesPath,
|
|
@@ -171,6 +243,13 @@ export async function executeEvaluationPipeline({ samplesPath, skillDir, samples
|
|
|
171
243
|
throw new Error(formatDependencyErrors(depResult.missing));
|
|
172
244
|
}
|
|
173
245
|
}
|
|
246
|
+
// Structural power warnings — print to stderr after preflight passes, before
|
|
247
|
+
// tasks start. These are *not* MDE / power-analysis predictions (we don't have
|
|
248
|
+
// σ before the run); they're hard-floor + experience-based thresholds. Verdict
|
|
249
|
+
// gate (computeVerdict) handles real power claims post-hoc.
|
|
250
|
+
emitPowerWarnings(samples.length, repeat ?? 1);
|
|
251
|
+
// v0.22 — Isolation pre-flight warning (--no-strict-baseline + ~/.claude/skills/ non-empty)
|
|
252
|
+
emitIsolationWarnings(artifacts, strictBaseline);
|
|
174
253
|
// Pre-build a per-executor map for ensemble judges. Each unique executor name
|
|
175
254
|
// gets one ExecutorFn, shared across all judges using that executor. The default
|
|
176
255
|
// judge's executor is also included so single-judge fallbacks have it available.
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"evaluation-pipeline.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACnD,OAAO,EAAE,cAAc,EAAE,MAAM,mCAAmC,CAAC;AACnE,OAAO,EAAE,qBAAqB,EAAE,MAAM,kCAAkC,CAAC;AACzE,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,eAAe,EAAE,cAAc,EAAE,kBAAkB,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACzI,OAAO,EAAE,YAAY,EAAE,SAAS,EAAE,MAAM,sCAAsC,CAAC;AAC/E,OAAO,EAAE,qBAAqB,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AAEnG,OAAO,EACL,kBAAkB,EAClB,gBAAgB,GACjB,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC3D,OAAO,EACL,sBAAsB,EACtB,eAAe,EACf,mBAAmB,EACnB,eAAe,EACf,kBAAkB,EAClB,qBAAqB,EACrB,cAAc,EACd,iBAAiB,GAClB,MAAM,gCAAgC,CAAC;AA4BxC,KAAK,UAAU,4BAA4B,CAAC,EAC1C,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,WAAW,EACX,SAAS,EACT,OAAO,EACP,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAK,EACL,QAAQ,EACR,UAAU,EACV,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,GA4BP;IACC,MAAM,OAAO,GAAG,sBAAsB,CAAC;QACrC,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;QACvC,QAAQ,EAAE,YAAY;QACtB,aAAa,EAAE,iBAAiB;QAChC,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO;QACP,MAAM,EAAE,KAAK;QACb,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM;QACN,IAAI;QACJ,WAAW;QACX,WAAW;QACX,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;KACP,CAAC,CAAC;IACH,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,MAAM,EAAE,GAAG,EAAE,UAAU,EAAE,SAAS,EAAE,GAAG,mBAAmB,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC7E,MAAM,KAAK,GAAG,OAAO,KAAK,EAAE,CAAC;IAC7B,MAAM,gBAAgB,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,QAAQ,IAAI,kBAAkB,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAChG,MAAM,SAAS,GAAG,eAAe,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC,CAAC;IACjE,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IACpE,MAAM,UAAU,GAAG,cAAc,CAAC,SAAS,EAAE,KAAK,EAAE,SAAS,CAAC,CAAC;IAC/D,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,KAAK,EAAE,UAAU,CAAC,CAAC;IACrE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,UAAU,EAAE,UAAU,EAAE,gBAAgB,EAAE,CAAC;AACnG,CAAC;AAED,SAAS,qBAAqB,CAAC,KAAyB;IACtD,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,GAAG,GAAG,qBAAqB,CAAC,KAAK,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IAChE,MAAM,GAAG,GAAG,kBAAkB,CAAC;QAC7B,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,QAAQ,EAAE,KAAK,CAAC,KAAK;QACrB,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,UAAU;KACX,CAAC,CAAC;IACH,OAAO,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC;AACtB,CAAC;AAED,KAAK,UAAU,oBAAoB,CAAC,KAAyB,EAAE,GAAkB;IAC/E,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC3B,MAAM,KAAK,CAAC,gBAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IACtD,CAAC;AACH,CAAC;AAED,KAAK,UAAU,gBAAgB,CAAC,KAAyB,EAAE,GAAY;IACrE,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,SAAS,GAAG,eAAe,CAAC;QAChC,GAAG,EAAE,EAAE,GAAG,KAAK,CAAC,UAAU,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,UAAU,EAAE,SAAS,EAAE;QACnG,KAAK,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;QACvD,UAAU;KACX,CAAC,CAAC;IACH,KAAK,iBAAiB,CAAC,KAAK,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IACrD,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC3B,MAAM,KAAK,CAAC,gBAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC5D,CAAC;AACH,CAAC;AAED;;;;GAIG;AACH,SAAS,kBAAkB,CAAC,WAAmB;IAC7C,IAAI,CAAC,WAAW,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC;QAAE,OAAO,IAAI,CAAC;IAC1D,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;QACnD,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IACzE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,wBAAwB,CAAC,EAChC,MAAM,EACN,OAAO,EACP,SAAS,EACT,YAAY,EACZ,KAAK,EACL,WAAW,GAQZ;IACC,MAAM,CAAC,QAAQ,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC;IAEzC,MAAM,WAAW,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,aAAa,EAAE,EAAE,CAAC,CACjE,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,aAAa,EAAE,EAAE,CAAC,aAAa,CAAC,SAAS,IAAI,aAAa,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CACpH,CAAC,CAAC;IACH,IAAI,WAAW,EAAE,CAAC;QAChB,MAAM,gBAAgB,GAAG,MAAM,CAAC,WAAW,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC5G,MAAM,YAAY,GAAG,MAAM,CAAC,WAAW,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,GAAG,IAAI,IAAI,CAAC,CAAC,CAAC,CAAC;QAC5G,MAAM,QAAQ,GAAG,qBAAqB,CAAC,MAAM,EAAE,gBAAgB,EAAE,YAAY,CAAC,CAAC;QAC/E,IAAI,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACrC,MAAM,CAAC,QAAS,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACvC,CAAC;IACH,CAAC;IAED,6EAA6E;IAC7E,2EAA2E;IAC3E,gFAAgF;IAChF,MAAM,UAAU,GAAG,qBAAqB,CAAC,MAAM,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;IACvE,IAAI,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACvC,MAAM,WAAW,GAAG,kBAAkB,CAAC,WAAW,CAAC,CAAC;QACpD,KAAK,MAAM,OAAO,IAAI,YAAY,EAAE,CAAC;YACnC,MAAM,EAAE,GAAG,UAAU,CAAC,OAAO,CAAC,CAAC;YAC/B,IAAI,CAAC,EAAE;gBAAE,SAAS;YAClB,EAAE,CAAC,WAAW,GAAG,WAAW,CAAC;YAC7B,EAAE,CAAC,WAAW,GAAG,WAAW,CAAC;QAC/B,CAAC;QACD,MAAM,CAAC,QAAS,CAAC,UAAU,GAAG,UAAU,CAAC;IAC3C,CAAC;IAED,IAAI,KAAK,EAAE,CAAC;QACV,cAAc,CAAC,MAAM,EAAE,YAAY,EAAE,GAAG,YAAY,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,WAAW,EAAE,CAAC,CAAC;IACnF,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAkDD,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,EAC9C,WAAW,EACX,QAAQ,EACR,OAAO,EACP,KAAK,EACL,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,QAAQ,EACR,aAAa,EACb,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAK,GAAG,KAAK,EACb,WAAW,GAAG,CAAC,EACf,SAAS,EACT,OAAO,GAAG,KAAK,EACf,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,aAAa,GAAG,KAAK,EACrB,OAAO,GAAG,KAAK,EACf,KAAK,GAAG,CAAC,EACT,eAAe,EACf,QAAQ,EACR,YAAY,GAAG,KAAK,EACpB,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,GAAG,IAAI,EACnB,MAAM,GACoB;IAC1B,MAAM,YAAY,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IAChE,MAAM,QAAQ,GAAG,MAAM,4BAA4B,CAAC;QAClD,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,UAAU;QACV,OAAO;QACP,YAAY;QACZ,iBAAiB;QACjB,WAAW;QACX,SAAS;QACT,OAAO;QACP,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,KAAK,EAAE,aAAa,CAAC,YAAY,CAAC;QAClC,QAAQ;QACR,UAAU;QACV,MAAM;QACN,IAAI;QACJ,WAAW;QACX,WAAW;QACX,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;KACP,CAAC,CAAC;IAEH,IAAI,CAAC;QACH,IAAI,CAAC,aAAa,EAAE,CAAC;YACnB,IAAI,UAAU;gBAAE,UAAU,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC;YAC1E,MAAM,SAAS,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC;YACjC,IAAI,CAAC,OAAO;gBAAE,MAAM,SAAS,CAAC,aAAa,EAAE,UAAU,CAAC,CAAC;YAEzD,+EAA+E;YAC/E,MAAM,aAAa,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAC;YACxG,MAAM,GAAG,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,GAAG,IAAI,QAAQ,IAAI,OAAO,CAAC,GAAG,EAAE,CAAC;YAC3E,MAAM,SAAS,GAAG,MAAM,qBAAqB,CAAC,aAAa,EAAE,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,SAAS,CAAC,CAAC;YAChG,IAAI,CAAC,SAAS,CAAC,EAAE,EAAE,CAAC;gBAClB,MAAM,IAAI,KAAK,CAAC,sBAAsB,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC;YAC7D,CAAC;QACH,CAAC;QAED,8EAA8E;QAC9E,iFAAiF;QACjF,iFAAiF;QACjF,IAAI,cAAsD,CAAC;QAC3D,IAAI,WAAW,IAAI,WAAW,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;YAC3C,MAAM,EAAE,cAAc,EAAE,GAAG,MAAM,MAAM,CAAC,uBAAuB,CAAC,CAAC;YACjE,cAAc,GAAG,EAAE,CAAC;YACpB,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;YAC/B,KAAK,MAAM,EAAE,IAAI,WAAW,EAAE,CAAC;gBAC7B,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC;oBAC3B,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;oBACtB,cAAc,CAAC,EAAE,CAAC,QAAQ,CAAC,GAAG,cAAc,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;gBAC5D,CAAC;YACH,CAAC;QACH,CAAC;QAED,MAAM,EAAE,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,eAAe,EAAE,GAAG,MAAM,YAAY,CAAC;YAC7E,KAAK;YACL,QAAQ;YACR,aAAa;YACb,KAAK;YACL,UAAU;YACV,OAAO;YACP,WAAW;YACX,WAAW;YACX,SAAS;YACT,OAAO;YACP,OAAO;YACP,UAAU;YACV,KAAK;YACL,eAAe;YACf,WAAW;YACX,WAAW;YACX,cAAc;YACd,YAAY;YACZ,MAAM;SACP,CAAC,CAAC;QACH,IAAI,OAAO,GAAG,CAAC,IAAI,UAAU,EAAE,CAAC;YAC9B,UAAU,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,CAAC,MAAM,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,SAAS,EAAE,EAAE,EAAE,OAAO,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;QACzH,CAAC;QAED,MAAM,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,qBAAqB,CAAC,QAAQ,CAAC,CAAC;QACrD,MAAM,MAAM,GAAG,wBAAwB,CAAC;YACtC,MAAM,EAAE,eAAe,CAAC;gBACtB,KAAK,EAAE,QAAQ,CAAC,KAAK;gBACrB,QAAQ,EAAE,YAAY;gBACtB,KAAK;gBACL,UAAU;gBACV,OAAO;gBACP,YAAY;gBACZ,OAAO;gBACP,KAAK;gBACL,OAAO;gBACP,YAAY;gBACZ,SAAS;gBACT,OAAO,EAAE,QAAQ,CAAC,OAAO;gBACzB,GAAG;gBACH,GAAG;gBACH,YAAY;aACb,CAAC;YACF,OAAO;YACP,SAAS;YACT,YAAY;YACZ,KAAK;YACL,WAAW;SACZ,CAAC,CAAC;QACH,IAAI,eAAe,EAAE,CAAC;YACpB,MAAM,CAAC,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC;QACrC,CAAC;QACD,IAAI,MAAM,EAAE,CAAC;YACX,MAAM,CAAC,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QAC9B,CAAC;QACD,MAAM,QAAQ,GAAG,aAAa,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;QAClD,MAAM,oBAAoB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QAC1C,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC;IAC9B,CAAC;IAAC,OAAO,GAAY,EAAE,CAAC;QACtB,MAAM,gBAAgB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QACtC,MAAM,GAAG,CAAC;IACZ,CAAC;YAAS,CAAC;QACT,MAAM,cAAc,EAAE,CAAC;IACzB,CAAC;AACH,CAAC"}
|
|
1
|
+
{"version":3,"file":"evaluation-pipeline.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAChE,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,cAAc,EAAE,MAAM,mCAAmC,CAAC;AACnE,OAAO,EAAE,qBAAqB,EAAE,MAAM,kCAAkC,CAAC;AACzE,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,eAAe,EAAE,cAAc,EAAE,kBAAkB,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACzI,OAAO,EAAE,YAAY,EAAE,SAAS,EAAE,MAAM,sCAAsC,CAAC;AAC/E,OAAO,EAAE,qBAAqB,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AAEnG,OAAO,EACL,kBAAkB,EAClB,gBAAgB,GACjB,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC3D,OAAO,EACL,sBAAsB,EACtB,eAAe,EACf,mBAAmB,EACnB,eAAe,EACf,kBAAkB,EAClB,qBAAqB,EACrB,cAAc,EACd,iBAAiB,GAClB,MAAM,gCAAgC,CAAC;AA4BxC,KAAK,UAAU,4BAA4B,CAAC,EAC1C,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,WAAW,EACX,SAAS,EACT,OAAO,EACP,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAK,EACL,QAAQ,EACR,UAAU,EACV,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,GA4BP;IACC,MAAM,OAAO,GAAG,sBAAsB,CAAC;QACrC,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;QACvC,QAAQ,EAAE,YAAY;QACtB,aAAa,EAAE,iBAAiB;QAChC,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO;QACP,MAAM,EAAE,KAAK;QACb,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM;QACN,IAAI;QACJ,WAAW;QACX,WAAW;QACX,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;KACP,CAAC,CAAC;IACH,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,MAAM,EAAE,GAAG,EAAE,UAAU,EAAE,SAAS,EAAE,GAAG,mBAAmB,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC7E,MAAM,KAAK,GAAG,OAAO,KAAK,EAAE,CAAC;IAC7B,MAAM,gBAAgB,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,QAAQ,IAAI,kBAAkB,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAChG,MAAM,SAAS,GAAG,eAAe,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC,CAAC;IACjE,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IACpE,MAAM,UAAU,GAAG,cAAc,CAAC,SAAS,EAAE,KAAK,EAAE,SAAS,CAAC,CAAC;IAC/D,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,KAAK,EAAE,UAAU,CAAC,CAAC;IACrE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,UAAU,EAAE,UAAU,EAAE,gBAAgB,EAAE,CAAC;AACnG,CAAC;AAED,SAAS,qBAAqB,CAAC,KAAyB;IACtD,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,GAAG,GAAG,qBAAqB,CAAC,KAAK,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IAChE,MAAM,GAAG,GAAG,kBAAkB,CAAC;QAC7B,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,QAAQ,EAAE,KAAK,CAAC,KAAK;QACrB,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,UAAU;KACX,CAAC,CAAC;IACH,OAAO,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC;AACtB,CAAC;AAED,KAAK,UAAU,oBAAoB,CAAC,KAAyB,EAAE,GAAkB;IAC/E,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC3B,MAAM,KAAK,CAAC,gBAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IACtD,CAAC;AACH,CAAC;AAED,KAAK,UAAU,gBAAgB,CAAC,KAAyB,EAAE,GAAY;IACrE,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,SAAS,GAAG,eAAe,CAAC;QAChC,GAAG,EAAE,EAAE,GAAG,KAAK,CAAC,UAAU,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,UAAU,EAAE,SAAS,EAAE;QACnG,KAAK,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;QACvD,UAAU;KACX,CAAC,CAAC;IACH,KAAK,iBAAiB,CAAC,KAAK,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IACrD,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC3B,MAAM,KAAK,CAAC,gBAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC5D,CAAC;AACH,CAAC;AAED;;;;GAIG;AACH,SAAS,kBAAkB,CAAC,WAAmB;IAC7C,IAAI,CAAC,WAAW,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC;QAAE,OAAO,IAAI,CAAC;IAC1D,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;QACnD,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IACzE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;;GAaG;AACH,MAAM,UAAU,kBAAkB,CAAC,WAAmB,EAAE,MAAc;IACpE,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;QACpB,QAAQ,CAAC,IAAI,CACX,OAAO,WAAW,6GAA6G,CAChI,CAAC;IACJ,CAAC;SAAM,IAAI,WAAW,GAAG,EAAE,EAAE,CAAC;QAC5B,QAAQ,CAAC,IAAI,CACX,OAAO,WAAW,kIAAkI,CACrJ,CAAC;IACJ,CAAC;IACD,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;QACf,QAAQ,CAAC,IAAI,CACX,0IAA0I,CAC3I,CAAC;IACJ,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,SAAS,iBAAiB,CAAC,WAAmB,EAAE,MAAc;IAC5D,KAAK,MAAM,CAAC,IAAI,kBAAkB,CAAC,WAAW,EAAE,MAAM,CAAC,EAAE,CAAC;QACxD,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACjC,CAAC;AACH,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,sBAAsB,CACpC,SAAqB,EACrB,cAAmC;IAEnC,qDAAqD;IACrD,IAAI,cAAc,KAAK,KAAK;QAAE,OAAO,EAAE,CAAC;IAExC,MAAM,eAAe,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,UAAU,CAAC,CAAC;IACrE,IAAI,CAAC,eAAe;QAAE,OAAO,EAAE,CAAC;IAEhC,kFAAkF;IAClF,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC;IACvD,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC;QAAE,OAAO,EAAE,CAAC;IAEtC,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,CAAC;QACH,UAAU,GAAG,WAAW,CAAC,SAAS,CAAC,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC;IACvF,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;IACD,IAAI,UAAU,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAEhC,OAAO;QACL,mEAAmE,UAAU,+EAA+E;KAC7J,CAAC;AACJ,CAAC;AAED,SAAS,qBAAqB,CAAC,SAAqB,EAAE,cAAmC;IACvF,KAAK,MAAM,CAAC,IAAI,sBAAsB,CAAC,SAAS,EAAE,cAAc,CAAC,EAAE,CAAC;QAClE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACjC,CAAC;AACH,CAAC;AAED,SAAS,wBAAwB,CAAC,EAChC,MAAM,EACN,OAAO,EACP,SAAS,EACT,YAAY,EACZ,KAAK,EACL,WAAW,GAQZ;IACC,MAAM,CAAC,QAAQ,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC;IAEzC,MAAM,WAAW,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,aAAa,EAAE,EAAE,CAAC,CACjE,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,aAAa,EAAE,EAAE,CAAC,aAAa,CAAC,SAAS,IAAI,aAAa,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CACpH,CAAC,CAAC;IACH,IAAI,WAAW,EAAE,CAAC;QAChB,MAAM,gBAAgB,GAAG,MAAM,CAAC,WAAW,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC5G,MAAM,YAAY,GAAG,MAAM,CAAC,WAAW,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,GAAG,IAAI,IAAI,CAAC,CAAC,CAAC,CAAC;QAC5G,MAAM,QAAQ,GAAG,qBAAqB,CAAC,MAAM,EAAE,gBAAgB,EAAE,YAAY,CAAC,CAAC;QAC/E,IAAI,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACrC,MAAM,CAAC,QAAS,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACvC,CAAC;IACH,CAAC;IAED,6EAA6E;IAC7E,2EAA2E;IAC3E,gFAAgF;IAChF,MAAM,UAAU,GAAG,qBAAqB,CAAC,MAAM,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;IACvE,IAAI,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACvC,MAAM,WAAW,GAAG,kBAAkB,CAAC,WAAW,CAAC,CAAC;QACpD,KAAK,MAAM,OAAO,IAAI,YAAY,EAAE,CAAC;YACnC,MAAM,EAAE,GAAG,UAAU,CAAC,OAAO,CAAC,CAAC;YAC/B,IAAI,CAAC,EAAE;gBAAE,SAAS;YAClB,EAAE,CAAC,WAAW,GAAG,WAAW,CAAC;YAC7B,EAAE,CAAC,WAAW,GAAG,WAAW,CAAC;QAC/B,CAAC;QACD,MAAM,CAAC,QAAS,CAAC,UAAU,GAAG,UAAU,CAAC;IAC3C,CAAC;IAED,IAAI,KAAK,EAAE,CAAC;QACV,cAAc,CAAC,MAAM,EAAE,YAAY,EAAE,GAAG,YAAY,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,WAAW,EAAE,CAAC,CAAC;IACnF,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAsDD,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,EAC9C,WAAW,EACX,QAAQ,EACR,OAAO,EACP,KAAK,EACL,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,QAAQ,EACR,aAAa,EACb,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAK,GAAG,KAAK,EACb,WAAW,GAAG,CAAC,EACf,SAAS,EACT,OAAO,GAAG,KAAK,EACf,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,aAAa,GAAG,KAAK,EACrB,OAAO,GAAG,KAAK,EACf,KAAK,GAAG,CAAC,EACT,eAAe,EACf,QAAQ,EACR,YAAY,GAAG,KAAK,EACpB,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,GAAG,IAAI,EACnB,MAAM,EACN,cAAc,GACY;IAC1B,MAAM,YAAY,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IAChE,MAAM,QAAQ,GAAG,MAAM,4BAA4B,CAAC;QAClD,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,UAAU;QACV,OAAO;QACP,YAAY;QACZ,iBAAiB;QACjB,WAAW;QACX,SAAS;QACT,OAAO;QACP,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,KAAK,EAAE,aAAa,CAAC,YAAY,CAAC;QAClC,QAAQ;QACR,UAAU;QACV,MAAM;QACN,IAAI;QACJ,WAAW;QACX,WAAW;QACX,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;KACP,CAAC,CAAC;IAEH,IAAI,CAAC;QACH,IAAI,CAAC,aAAa,EAAE,CAAC;YACnB,IAAI,UAAU;gBAAE,UAAU,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC;YAC1E,MAAM,SAAS,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC;YACjC,IAAI,CAAC,OAAO;gBAAE,MAAM,SAAS,CAAC,aAAa,EAAE,UAAU,CAAC,CAAC;YAEzD,+EAA+E;YAC/E,MAAM,aAAa,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAC;YACxG,MAAM,GAAG,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,GAAG,IAAI,QAAQ,IAAI,OAAO,CAAC,GAAG,EAAE,CAAC;YAC3E,MAAM,SAAS,GAAG,MAAM,qBAAqB,CAAC,aAAa,EAAE,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,SAAS,CAAC,CAAC;YAChG,IAAI,CAAC,SAAS,CAAC,EAAE,EAAE,CAAC;gBAClB,MAAM,IAAI,KAAK,CAAC,sBAAsB,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC;YAC7D,CAAC;QACH,CAAC;QAED,6EAA6E;QAC7E,+EAA+E;QAC/E,+EAA+E;QAC/E,4DAA4D;QAC5D,iBAAiB,CAAC,OAAO,CAAC,MAAM,EAAE,MAAM,IAAI,CAAC,CAAC,CAAC;QAC/C,4FAA4F;QAC5F,qBAAqB,CAAC,SAAS,EAAE,cAAc,CAAC,CAAC;QAEjD,8EAA8E;QAC9E,iFAAiF;QACjF,iFAAiF;QACjF,IAAI,cAAsD,CAAC;QAC3D,IAAI,WAAW,IAAI,WAAW,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;YAC3C,MAAM,EAAE,cAAc,EAAE,GAAG,MAAM,MAAM,CAAC,uBAAuB,CAAC,CAAC;YACjE,cAAc,GAAG,EAAE,CAAC;YACpB,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;YAC/B,KAAK,MAAM,EAAE,IAAI,WAAW,EAAE,CAAC;gBAC7B,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC;oBAC3B,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;oBACtB,cAAc,CAAC,EAAE,CAAC,QAAQ,CAAC,GAAG,cAAc,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;gBAC5D,CAAC;YACH,CAAC;QACH,CAAC;QAED,MAAM,EAAE,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,eAAe,EAAE,GAAG,MAAM,YAAY,CAAC;YAC7E,KAAK;YACL,QAAQ;YACR,aAAa;YACb,KAAK;YACL,UAAU;YACV,OAAO;YACP,WAAW;YACX,WAAW;YACX,SAAS;YACT,OAAO;YACP,OAAO;YACP,UAAU;YACV,KAAK;YACL,eAAe;YACf,WAAW;YACX,WAAW;YACX,cAAc;YACd,YAAY;YACZ,MAAM;SACP,CAAC,CAAC;QACH,IAAI,OAAO,GAAG,CAAC,IAAI,UAAU,EAAE,CAAC;YAC9B,UAAU,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,CAAC,MAAM,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,SAAS,EAAE,EAAE,EAAE,OAAO,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;QACzH,CAAC;QAED,MAAM,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,qBAAqB,CAAC,QAAQ,CAAC,CAAC;QACrD,MAAM,MAAM,GAAG,wBAAwB,CAAC;YACtC,MAAM,EAAE,eAAe,CAAC;gBACtB,KAAK,EAAE,QAAQ,CAAC,KAAK;gBACrB,QAAQ,EAAE,YAAY;gBACtB,KAAK;gBACL,UAAU;gBACV,OAAO;gBACP,YAAY;gBACZ,OAAO;gBACP,KAAK;gBACL,OAAO;gBACP,YAAY;gBACZ,SAAS;gBACT,OAAO,EAAE,QAAQ,CAAC,OAAO;gBACzB,GAAG;gBACH,GAAG;gBACH,YAAY;aACb,CAAC;YACF,OAAO;YACP,SAAS;YACT,YAAY;YACZ,KAAK;YACL,WAAW;SACZ,CAAC,CAAC;QACH,IAAI,eAAe,EAAE,CAAC;YACpB,MAAM,CAAC,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC;QACrC,CAAC;QACD,IAAI,MAAM,EAAE,CAAC;YACX,MAAM,CAAC,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QAC9B,CAAC;QACD,MAAM,QAAQ,GAAG,aAAa,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;QAClD,MAAM,oBAAoB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QAC1C,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC;IAC9B,CAAC;IAAC,OAAO,GAAY,EAAE,CAAC;QACtB,MAAM,gBAAgB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QACtC,MAAM,GAAG,CAAC;IACZ,CAAC;YAAS,CAAC;QACT,MAAM,cAAc,EAAE,CAAC;IACzB,CAAC;AACH,CAAC"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import type { DependencyRequirements } from '../eval-core/dependency-checker.js';
|
|
2
|
-
import type { Artifact, Sample, Task, VariantSpec } from '../types.js';
|
|
2
|
+
import type { Artifact, Sample, Task, VariantSpec } from '../types/index.js';
|
|
3
3
|
export interface PreparedEvaluationRun {
|
|
4
4
|
samples: Sample[];
|
|
5
5
|
artifacts: Artifact[];
|
|
@@ -7,13 +7,17 @@ export interface PreparedEvaluationRun {
|
|
|
7
7
|
variantNames: string[];
|
|
8
8
|
requires?: DependencyRequirements;
|
|
9
9
|
}
|
|
10
|
-
export declare function prepareEvaluationRun({ samplesPath, skillDir, variantSpecs, artifacts, dryRun, mcpConfig, }: {
|
|
10
|
+
export declare function prepareEvaluationRun({ samplesPath, skillDir, variantSpecs, artifacts, dryRun, mcpConfig, strictBaseline, variantAllowedSkills, }: {
|
|
11
11
|
samplesPath: string;
|
|
12
12
|
skillDir: string;
|
|
13
13
|
variantSpecs: VariantSpec[];
|
|
14
14
|
artifacts?: Artifact[];
|
|
15
15
|
dryRun: boolean;
|
|
16
16
|
mcpConfig?: string;
|
|
17
|
+
/** v0.22 — default true, baseline-kind 自动 allowedSkills=[]。 */
|
|
18
|
+
strictBaseline?: boolean;
|
|
19
|
+
/** v0.22 — eval.yaml 显式 allowedSkills per variant (overrides strictBaseline default). */
|
|
20
|
+
variantAllowedSkills?: Record<string, string[]>;
|
|
17
21
|
}): Promise<PreparedEvaluationRun>;
|
|
18
22
|
export declare function buildDryRunTaskReport({ model, judgeModel, executorName, samplesPath, skillDir, tasks, variantNames, }: {
|
|
19
23
|
model: string;
|
|
@@ -35,11 +39,11 @@ export declare function buildDryRunTaskReport({ model, judgeModel, executorName,
|
|
|
35
39
|
tasks: {
|
|
36
40
|
sample_id: string;
|
|
37
41
|
variant: string;
|
|
38
|
-
artifactKind: import("../types.js").ArtifactKind;
|
|
42
|
+
artifactKind: import("../types/eval.js").ArtifactKind;
|
|
39
43
|
artifactSource: "baseline" | "variant-name" | "file-path" | "git" | "inline" | "custom";
|
|
40
|
-
executionStrategy: import("../types.js").ExecutionStrategyKind;
|
|
41
|
-
experimentType: import("../types.js").ExperimentType;
|
|
42
|
-
experimentRole: import("../types.js").ExperimentRole;
|
|
44
|
+
executionStrategy: import("../types/eval.js").ExecutionStrategyKind;
|
|
45
|
+
experimentType: import("../types/eval.js").ExperimentType;
|
|
46
|
+
experimentRole: import("../types/eval.js").ExperimentRole;
|
|
43
47
|
cwd: string | null;
|
|
44
48
|
promptPreview: string;
|
|
45
49
|
hasRubric: boolean;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"evaluation-preparation.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-preparation.ts"],"names":[],"mappings":"AAOA,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AACjF,OAAO,KAAK,EAAE,QAAQ,EAAc,MAAM,EAAE,IAAI,EAAE,WAAW,EAAE,MAAM,
|
|
1
|
+
{"version":3,"file":"evaluation-preparation.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-preparation.ts"],"names":[],"mappings":"AAOA,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AACjF,OAAO,KAAK,EAAE,QAAQ,EAAc,MAAM,EAAE,IAAI,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAEzF,MAAM,WAAW,qBAAqB;IACpC,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,QAAQ,CAAC,EAAE,sBAAsB,CAAC;CACnC;AAED,wBAAsB,oBAAoB,CAAC,EACzC,WAAW,EACX,QAAQ,EACR,YAAY,EACZ,SAAS,EACT,MAAM,EACN,SAAS,EACT,cAAc,EACd,oBAAoB,GACrB,EAAE;IACD,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,WAAW,EAAE,CAAC;IAC5B,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,MAAM,EAAE,OAAO,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,+DAA+D;IAC/D,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,yFAAyF;IACzF,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CACjD,GAAG,OAAO,CAAC,qBAAqB,CAAC,CA6CjC;AAED,wBAAgB,qBAAqB,CAAC,EACpC,KAAK,EACL,UAAU,EACV,YAAY,EACZ,WAAW,EACX,QAAQ,EACR,KAAK,EACL,YAAY,GACb,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB;;;;;;;;;;;;;;;;;;;;;;;;EA6BA;AAED,wBAAgB,wBAAwB,CAAC,YAAY,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,CAAC;;;;;;;;EAerH"}
|
|
@@ -5,11 +5,11 @@ import { resolveArtifacts } from '../inputs/skill-loader.js';
|
|
|
5
5
|
import { buildVariantConfig } from '../eval-core/execution-strategy.js';
|
|
6
6
|
import { loadMcpConfig, resolveMcpUrls } from '../inputs/mcp-resolver.js';
|
|
7
7
|
import { resolveUrls } from '../inputs/url-fetcher.js';
|
|
8
|
-
export async function prepareEvaluationRun({ samplesPath, skillDir, variantSpecs, artifacts, dryRun, mcpConfig, }) {
|
|
8
|
+
export async function prepareEvaluationRun({ samplesPath, skillDir, variantSpecs, artifacts, dryRun, mcpConfig, strictBaseline, variantAllowedSkills, }) {
|
|
9
9
|
const { samples, requires } = loadSamples(samplesPath);
|
|
10
10
|
// Build expressions from specs (preserving order) and resolve to artifacts.
|
|
11
11
|
const variantExpressions = variantSpecs.map((spec) => spec.expr);
|
|
12
|
-
const resolvedArtifacts = artifacts || resolveArtifacts(resolve(skillDir), variantExpressions);
|
|
12
|
+
const resolvedArtifacts = artifacts || resolveArtifacts(resolve(skillDir), variantExpressions, { strictBaseline, variantAllowedSkills });
|
|
13
13
|
// Attach experimentRole to each artifact by matching spec.name to artifact.name.
|
|
14
14
|
const roleByName = {};
|
|
15
15
|
for (const spec of variantSpecs)
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"evaluation-preparation.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-preparation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,uBAAuB,EAAE,MAAM,8BAA8B,CAAC;AACvE,OAAO,EAAE,WAAW,EAAE,MAAM,2BAA2B,CAAC;AACxD,OAAO,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AAC7D,OAAO,EAAE,kBAAkB,EAAE,MAAM,oCAAoC,CAAC;AACxE,OAAO,EAAE,aAAa,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC1E,OAAO,EAAE,WAAW,EAAE,MAAM,0BAA0B,CAAC;AAYvD,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,EACzC,WAAW,EACX,QAAQ,EACR,YAAY,EACZ,SAAS,EACT,MAAM,EACN,SAAS,
|
|
1
|
+
{"version":3,"file":"evaluation-preparation.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-preparation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,uBAAuB,EAAE,MAAM,8BAA8B,CAAC;AACvE,OAAO,EAAE,WAAW,EAAE,MAAM,2BAA2B,CAAC;AACxD,OAAO,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AAC7D,OAAO,EAAE,kBAAkB,EAAE,MAAM,oCAAoC,CAAC;AACxE,OAAO,EAAE,aAAa,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC1E,OAAO,EAAE,WAAW,EAAE,MAAM,0BAA0B,CAAC;AAYvD,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,EACzC,WAAW,EACX,QAAQ,EACR,YAAY,EACZ,SAAS,EACT,MAAM,EACN,SAAS,EACT,cAAc,EACd,oBAAoB,GAYrB;IACC,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,WAAW,CAAC,WAAW,CAAC,CAAC;IAEvD,4EAA4E;IAC5E,MAAM,kBAAkB,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACjE,MAAM,iBAAiB,GAAG,SAAS,IAAI,gBAAgB,CACrD,OAAO,CAAC,QAAQ,CAAC,EACjB,kBAAkB,EAClB,EAAE,cAAc,EAAE,oBAAoB,EAAE,CACzC,CAAC;IAEF,iFAAiF;IACjF,MAAM,UAAU,GAAwC,EAAE,CAAC;IAC3D,KAAK,MAAM,IAAI,IAAI,YAAY;QAAE,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC;IACnE,KAAK,MAAM,QAAQ,IAAI,iBAAiB,EAAE,CAAC;QACzC,IAAI,QAAQ,CAAC,cAAc;YAAE,SAAS,CAAE,+CAA+C;QACvF,MAAM,IAAI,GAAG,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;QACvC,IAAI,IAAI;YAAE,QAAQ,CAAC,cAAc,GAAG,IAAI,CAAC;IAC3C,CAAC;IAED,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,MAAM,UAAU,GAAsB,aAAa,CAAC,SAAS,CAAC,CAAC;QAC/D,MAAM,WAAW,GAAG,UAAU,CAAC,CAAC,CAAC,MAAM,cAAc,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC,IAAI,GAAG,EAAU,CAAC;QAC/F,MAAM,WAAW,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;IAC1C,CAAC;IAED,IAAI,iBAAiB,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACnC,MAAM,IAAI,KAAK,CACb,sBAAsB;cACpB,qBAAqB,OAAO,CAAC,QAAQ,CAAC,IAAI;cAC1C,sCAAsC;cACtC,wEAAwE,CAC3E,CAAC;IACJ,CAAC;IAED,MAAM,KAAK,GAAG,uBAAuB,CAAC,OAAO,EAAE,iBAAiB,CAAC,CAAC;IAClE,MAAM,YAAY,GAAG,iBAAiB,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IAExE,OAAO;QACL,OAAO;QACP,SAAS,EAAE,iBAAiB;QAC5B,KAAK;QACL,YAAY;QACZ,QAAQ;KACT,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,EACpC,KAAK,EACL,UAAU,EACV,YAAY,EACZ,WAAW,EACX,QAAQ,EACR,KAAK,EACL,YAAY,GASb;IACC,OAAO;QACL,MAAM,EAAE,IAAa;QACrB,KAAK;QACL,UAAU;QACV,QAAQ,EAAE,YAAY;QACtB,QAAQ,EAAE,YAAY;QACtB,WAAW;QACX,QAAQ;QACR,UAAU,EAAE,KAAK,CAAC,MAAM;QACxB,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;YACxB,MAAM,MAAM,GAAG,kBAAkB,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YACjD,OAAO;gBACL,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,OAAO,EAAE,IAAI,CAAC,OAAO;gBACrB,YAAY,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI;gBAChC,cAAc,EAAE,IAAI,CAAC,QAAQ,CAAC,MAAM;gBACpC,iBAAiB,EAAE,MAAM,CAAC,iBAAiB;gBAC3C,cAAc,EAAE,MAAM,CAAC,cAAc;gBACrC,cAAc,EAAE,MAAM,CAAC,cAAc;gBACrC,GAAG,EAAE,IAAI,CAAC,GAAG;gBACb,aAAa,EAAE,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC;gBACxC,SAAS,EAAE,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC;gBAC/B,aAAa,EAAE,OAAO,CAAC,IAAI,CAAC,UAAU,EAAE,MAAM,CAAC;gBAC/C,aAAa,EAAE,OAAO,CAAC,IAAI,CAAC,UAAU,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC;gBAC9E,SAAS,EAAE,OAAO,CAAC,IAAI,CAAC,eAAe,CAAC;aACzC,CAAC;QACJ,CAAC,CAAC;KACH,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,wBAAwB,CAAC,YAA6E;IACpH,MAAM,SAAS,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;QAC3C,MAAM,EAAE,OAAO,EAAE,GAAG,WAAW,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;QACnD,OAAO;YACL,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,WAAW,EAAE,OAAO,CAAC,MAAM;YAC3B,SAAS,EAAE,OAAO,CAAC,MAAM,GAAG,CAAC;SAC9B,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,OAAO;QACL,SAAS;QACT,UAAU,EAAE,SAAS,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,QAAQ,EAAE,EAAE,CAAC,GAAG,GAAG,QAAQ,CAAC,SAAS,EAAE,CAAC,CAAC;KAC7E,CAAC;AACJ,CAAC"}
|