oh-my-knowledge 0.32.0 → 0.33.0
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 +11 -9
- package/README.zh.md +11 -9
- package/dist/analysis/saturation.d.ts +8 -0
- package/dist/analysis/saturation.d.ts.map +1 -1
- package/dist/analysis/saturation.js +12 -4
- package/dist/analysis/saturation.js.map +1 -1
- package/dist/authoring/evolver.d.ts +43 -2
- package/dist/authoring/evolver.d.ts.map +1 -1
- package/dist/authoring/evolver.js +91 -8
- package/dist/authoring/evolver.js.map +1 -1
- package/dist/authoring/generator.d.ts +29 -1
- package/dist/authoring/generator.d.ts.map +1 -1
- package/dist/authoring/generator.js +113 -0
- package/dist/authoring/generator.js.map +1 -1
- package/dist/cli/commands/doctor.d.ts +2 -0
- package/dist/cli/commands/doctor.d.ts.map +1 -1
- package/dist/cli/commands/doctor.js +22 -3
- package/dist/cli/commands/doctor.js.map +1 -1
- package/dist/cli/commands/eval/index.d.ts.map +1 -1
- package/dist/cli/commands/eval/index.js +3 -2
- package/dist/cli/commands/eval/index.js.map +1 -1
- package/dist/cli/commands/evolve.d.ts +1 -0
- package/dist/cli/commands/evolve.d.ts.map +1 -1
- package/dist/cli/commands/evolve.js +16 -0
- package/dist/cli/commands/evolve.js.map +1 -1
- package/dist/cli/commands/observe/index.d.ts.map +1 -1
- package/dist/cli/commands/observe/index.js +6 -2
- package/dist/cli/commands/observe/index.js.map +1 -1
- package/dist/cli/commands/sample.d.ts +2 -0
- package/dist/cli/commands/sample.d.ts.map +1 -1
- package/dist/cli/commands/sample.js +98 -6
- package/dist/cli/commands/sample.js.map +1 -1
- package/dist/cli/lib/cmd-flags.d.ts +5 -0
- package/dist/cli/lib/cmd-flags.d.ts.map +1 -1
- package/dist/cli/lib/i18n-dict/common.d.ts +1 -1
- package/dist/cli/lib/i18n-dict/common.d.ts.map +1 -1
- package/dist/cli/lib/i18n-dict/common.js +18 -2
- package/dist/cli/lib/i18n-dict/common.js.map +1 -1
- package/dist/cli/lib/i18n-dict/evolve.d.ts +1 -1
- package/dist/cli/lib/i18n-dict/evolve.d.ts.map +1 -1
- package/dist/cli/lib/i18n-dict/evolve.js +8 -0
- package/dist/cli/lib/i18n-dict/evolve.js.map +1 -1
- package/dist/cli/lib/parse-run-config/variant-resolution.d.ts +8 -3
- package/dist/cli/lib/parse-run-config/variant-resolution.d.ts.map +1 -1
- package/dist/cli/lib/parse-run-config/variant-resolution.js +16 -10
- package/dist/cli/lib/parse-run-config/variant-resolution.js.map +1 -1
- package/dist/cli/lib/resolve-skill-input.d.ts.map +1 -1
- package/dist/cli/lib/resolve-skill-input.js +7 -5
- package/dist/cli/lib/resolve-skill-input.js.map +1 -1
- package/dist/cli/lib/update-check.d.ts +65 -0
- package/dist/cli/lib/update-check.d.ts.map +1 -1
- package/dist/cli/lib/update-check.js +220 -31
- package/dist/cli/lib/update-check.js.map +1 -1
- package/dist/cli/lib/update-fetch-worker.d.ts +2 -0
- package/dist/cli/lib/update-fetch-worker.d.ts.map +1 -0
- package/dist/cli/lib/update-fetch-worker.js +35 -0
- package/dist/cli/lib/update-fetch-worker.js.map +1 -0
- package/dist/doctor/health/load-custom-dimensions.d.ts +2 -0
- package/dist/doctor/health/load-custom-dimensions.d.ts.map +1 -0
- package/dist/doctor/health/load-custom-dimensions.js +31 -0
- package/dist/doctor/health/load-custom-dimensions.js.map +1 -0
- package/dist/eval-core/bootstrap.d.ts +8 -0
- package/dist/eval-core/bootstrap.d.ts.map +1 -1
- package/dist/eval-core/bootstrap.js +11 -3
- package/dist/eval-core/bootstrap.js.map +1 -1
- package/dist/eval-core/evaluation-reporting.d.ts.map +1 -1
- package/dist/eval-core/evaluation-reporting.js +4 -4
- package/dist/eval-core/evaluation-reporting.js.map +1 -1
- package/dist/eval-core/verdict.d.ts +21 -3
- package/dist/eval-core/verdict.d.ts.map +1 -1
- package/dist/eval-core/verdict.js +54 -4
- package/dist/eval-core/verdict.js.map +1 -1
- package/dist/eval-workflows/evaluation-preparation.d.ts.map +1 -1
- package/dist/eval-workflows/evaluation-preparation.js +39 -11
- package/dist/eval-workflows/evaluation-preparation.js.map +1 -1
- package/dist/eval-workflows/run-evaluation.d.ts +1 -1
- package/dist/eval-workflows/run-evaluation.d.ts.map +1 -1
- package/dist/eval-workflows/run-evaluation.js +7 -7
- package/dist/eval-workflows/run-evaluation.js.map +1 -1
- package/dist/inputs/eval-config.d.ts.map +1 -1
- package/dist/inputs/eval-config.js +1 -0
- package/dist/inputs/eval-config.js.map +1 -1
- package/dist/inputs/skill-loader.d.ts +22 -0
- package/dist/inputs/skill-loader.d.ts.map +1 -1
- package/dist/inputs/skill-loader.js +171 -12
- package/dist/inputs/skill-loader.js.map +1 -1
- package/dist/observability/skill-health-analyzer.d.ts +13 -0
- package/dist/observability/skill-health-analyzer.d.ts.map +1 -1
- package/dist/observability/skill-health-analyzer.js +17 -2
- package/dist/observability/skill-health-analyzer.js.map +1 -1
- package/dist/renderer/layout.d.ts.map +1 -1
- package/dist/renderer/layout.js +2 -0
- package/dist/renderer/layout.js.map +1 -1
- package/dist/renderer/skill-detail-renderer.d.ts.map +1 -1
- package/dist/renderer/skill-detail-renderer.js +98 -19
- package/dist/renderer/skill-detail-renderer.js.map +1 -1
- package/dist/renderer/skill-health-renderer.d.ts.map +1 -1
- package/dist/renderer/skill-health-renderer.js +33 -4
- package/dist/renderer/skill-health-renderer.js.map +1 -1
- package/dist/renderer/skill-list-renderer.d.ts.map +1 -1
- package/dist/renderer/skill-list-renderer.js +18 -6
- package/dist/renderer/skill-list-renderer.js.map +1 -1
- package/dist/renderer/summary.d.ts.map +1 -1
- package/dist/renderer/summary.js +20 -8
- package/dist/renderer/summary.js.map +1 -1
- package/dist/server/report-server.d.ts.map +1 -1
- package/dist/server/report-server.js +9 -2
- package/dist/server/report-server.js.map +1 -1
- package/dist/server/skill-index.d.ts.map +1 -1
- package/dist/server/skill-index.js +3 -0
- package/dist/server/skill-index.js.map +1 -1
- package/dist/server/skill-insights.d.ts.map +1 -1
- package/dist/server/skill-insights.js +33 -5
- package/dist/server/skill-insights.js.map +1 -1
- package/dist/types/eval.d.ts +1 -0
- package/dist/types/eval.d.ts.map +1 -1
- package/dist/types/skill-index.d.ts +3 -0
- package/dist/types/skill-index.d.ts.map +1 -1
- package/package.json +4 -4
|
@@ -0,0 +1,35 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* 升级检查的后台刷新 worker。由 update-check.ts 的 fetchAndStore 以 detached + unref
|
|
3
|
+
* spawn,独立于父 CLI 进程运行:抓 registry latest 写回 ~/.oh-my-knowledge/update-check.json,
|
|
4
|
+
* 供下次运行展示用。父进程不等它,所以这里阻塞式 await fetch 没关系。全程 fail-silent。
|
|
5
|
+
*
|
|
6
|
+
* argv: [node, update-fetch-worker.js, cachePath, registry, pkgName]
|
|
7
|
+
*/
|
|
8
|
+
import { readCache, writeCache } from './update-check.js';
|
|
9
|
+
async function main() {
|
|
10
|
+
const [cachePath, registry, name] = process.argv.slice(2);
|
|
11
|
+
if (!cachePath || !registry || !name)
|
|
12
|
+
return;
|
|
13
|
+
try {
|
|
14
|
+
const res = await fetch(`${registry.replace(/\/$/, '')}/${name}/latest`, {
|
|
15
|
+
signal: AbortSignal.timeout(3000),
|
|
16
|
+
});
|
|
17
|
+
if (!res.ok)
|
|
18
|
+
return;
|
|
19
|
+
const data = (await res.json());
|
|
20
|
+
if (!data.version)
|
|
21
|
+
return;
|
|
22
|
+
const prev = readCache(cachePath); // 重读以保留展示侧写的 notify 节流字段,别覆盖
|
|
23
|
+
writeCache(cachePath, {
|
|
24
|
+
latestVersion: data.version,
|
|
25
|
+
lastCheckedAt: new Date().toISOString(),
|
|
26
|
+
lastNotifiedVersion: prev?.lastNotifiedVersion,
|
|
27
|
+
lastNotifiedAt: prev?.lastNotifiedAt,
|
|
28
|
+
});
|
|
29
|
+
}
|
|
30
|
+
catch {
|
|
31
|
+
/* 静默:网络 / 超时 / 解析 / 磁盘任一失败都不影响 */
|
|
32
|
+
}
|
|
33
|
+
}
|
|
34
|
+
void main();
|
|
35
|
+
//# sourceMappingURL=update-fetch-worker.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"update-fetch-worker.js","sourceRoot":"","sources":["../../../src/cli/lib/update-fetch-worker.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,OAAO,EAAE,SAAS,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE1D,KAAK,UAAU,IAAI;IACjB,MAAM,CAAC,SAAS,EAAE,QAAQ,EAAE,IAAI,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IAC1D,IAAI,CAAC,SAAS,IAAI,CAAC,QAAQ,IAAI,CAAC,IAAI;QAAE,OAAO;IAC7C,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,QAAQ,CAAC,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,IAAI,IAAI,SAAS,EAAE;YACvE,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,IAAI,CAAC;SAClC,CAAC,CAAC;QACH,IAAI,CAAC,GAAG,CAAC,EAAE;YAAE,OAAO;QACpB,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAyB,CAAC;QACxD,IAAI,CAAC,IAAI,CAAC,OAAO;YAAE,OAAO;QAC1B,MAAM,IAAI,GAAG,SAAS,CAAC,SAAS,CAAC,CAAC,CAAC,6BAA6B;QAChE,UAAU,CAAC,SAAS,EAAE;YACpB,aAAa,EAAE,IAAI,CAAC,OAAO;YAC3B,aAAa,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACvC,mBAAmB,EAAE,IAAI,EAAE,mBAAmB;YAC9C,cAAc,EAAE,IAAI,EAAE,cAAc;SACrC,CAAC,CAAC;IACL,CAAC;IAAC,MAAM,CAAC;QACP,kCAAkC;IACpC,CAAC;AACH,CAAC;AAED,KAAK,IAAI,EAAE,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"load-custom-dimensions.d.ts","sourceRoot":"","sources":["../../../src/doctor/health/load-custom-dimensions.ts"],"names":[],"mappings":"AAeA,wBAAgB,+BAA+B,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAyBxE"}
|
|
@@ -0,0 +1,31 @@
|
|
|
1
|
+
import { readFileSync, existsSync } from 'node:fs';
|
|
2
|
+
import yaml from 'js-yaml';
|
|
3
|
+
import { registerHealthDimension } from './dimension-registry.js';
|
|
4
|
+
const VALID_SEVERITIES = new Set(['fatal', 'warn', 'info']);
|
|
5
|
+
export function loadAndRegisterCustomDimensions(filePath) {
|
|
6
|
+
if (!existsSync(filePath)) {
|
|
7
|
+
throw new Error(`自定义维度配置文件不存在: ${filePath}`);
|
|
8
|
+
}
|
|
9
|
+
const raw = readFileSync(filePath, 'utf-8');
|
|
10
|
+
const doc = yaml.load(raw);
|
|
11
|
+
if (!doc?.dimensions || !Array.isArray(doc.dimensions))
|
|
12
|
+
return 0;
|
|
13
|
+
let count = 0;
|
|
14
|
+
for (const dim of doc.dimensions) {
|
|
15
|
+
if (!dim.id || !dim.displayName || !dim.promptSection) {
|
|
16
|
+
throw new Error(`自定义维度缺少必填字段(id/displayName/promptSection): ${JSON.stringify(dim).slice(0, 100)}`);
|
|
17
|
+
}
|
|
18
|
+
const severity = VALID_SEVERITIES.has(dim.severity ?? '') ? dim.severity : 'warn';
|
|
19
|
+
const spec = {
|
|
20
|
+
id: dim.id,
|
|
21
|
+
displayName: dim.displayName,
|
|
22
|
+
labelKey: `cli.doctor.health.dim.${dim.id}`,
|
|
23
|
+
severity,
|
|
24
|
+
promptSection: dim.promptSection,
|
|
25
|
+
};
|
|
26
|
+
registerHealthDimension(spec);
|
|
27
|
+
count++;
|
|
28
|
+
}
|
|
29
|
+
return count;
|
|
30
|
+
}
|
|
31
|
+
//# sourceMappingURL=load-custom-dimensions.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"load-custom-dimensions.js","sourceRoot":"","sources":["../../../src/doctor/health/load-custom-dimensions.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACnD,OAAO,IAAI,MAAM,SAAS,CAAC;AAC3B,OAAO,EAAE,uBAAuB,EAAE,MAAM,yBAAyB,CAAC;AAWlE,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC;AAE5D,MAAM,UAAU,+BAA+B,CAAC,QAAgB;IAC9D,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC,EAAE,CAAC;QAC1B,MAAM,IAAI,KAAK,CAAC,iBAAiB,QAAQ,EAAE,CAAC,CAAC;IAC/C,CAAC;IACD,MAAM,GAAG,GAAG,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IAC5C,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,CAA2C,CAAC;IACrE,IAAI,CAAC,GAAG,EAAE,UAAU,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC;QAAE,OAAO,CAAC,CAAC;IAEjE,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,MAAM,GAAG,IAAI,GAAG,CAAC,UAAU,EAAE,CAAC;QACjC,IAAI,CAAC,GAAG,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,WAAW,IAAI,CAAC,GAAG,CAAC,aAAa,EAAE,CAAC;YACtD,MAAM,IAAI,KAAK,CAAC,8CAA8C,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;QACrG,CAAC;QACD,MAAM,QAAQ,GAAmB,gBAAgB,CAAC,GAAG,CAAC,GAAG,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,QAA0B,CAAC,CAAC,CAAC,MAAM,CAAC;QACpH,MAAM,IAAI,GAAwB;YAChC,EAAE,EAAE,GAAG,CAAC,EAAE;YACV,WAAW,EAAE,GAAG,CAAC,WAAW;YAC5B,QAAQ,EAAE,yBAAyB,GAAG,CAAC,EAAE,EAAE;YAC3C,QAAQ;YACR,aAAa,EAAE,GAAG,CAAC,aAAa;SACjC,CAAC;QACF,uBAAuB,CAAC,IAAI,CAAC,CAAC;QAC9B,KAAK,EAAE,CAAC;IACV,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC"}
|
|
@@ -34,6 +34,14 @@ export interface BootstrapDiffCI extends BootstrapCI {
|
|
|
34
34
|
/** Whether 0 is inside the CI — when false, the difference is statistically significant. */
|
|
35
35
|
significant: boolean;
|
|
36
36
|
}
|
|
37
|
+
/**
|
|
38
|
+
* Default number of bootstrap resamples. Every eval path uses this unless
|
|
39
|
+
* `--bootstrap-samples` overrides it. Single source of truth: the docs cite
|
|
40
|
+
* it and `test/scripts/doc-constants-drift.test.ts` guards doc ↔ code parity.
|
|
41
|
+
*/
|
|
42
|
+
export declare const DEFAULT_BOOTSTRAP_SAMPLES = 1000;
|
|
43
|
+
/** Default significance level; 0.05 → 95% CI. */
|
|
44
|
+
export declare const DEFAULT_BOOTSTRAP_ALPHA = 0.05;
|
|
37
45
|
/**
|
|
38
46
|
* Bootstrap confidence interval for the mean of a single sample.
|
|
39
47
|
*
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"bootstrap.d.ts","sourceRoot":"","sources":["../../src/eval-core/bootstrap.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG;AAEH,MAAM,WAAW,WAAW;IAC1B,6BAA6B;IAC7B,GAAG,EAAE,MAAM,CAAC;IACZ,6BAA6B;IAC7B,IAAI,EAAE,MAAM,CAAC;IACb,mFAAmF;IACnF,QAAQ,EAAE,MAAM,CAAC;IACjB,+CAA+C;IAC/C,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,eAAgB,SAAQ,WAAW;IAClD,4FAA4F;IAC5F,WAAW,EAAE,OAAO,CAAC;CACtB;AA+
|
|
1
|
+
{"version":3,"file":"bootstrap.d.ts","sourceRoot":"","sources":["../../src/eval-core/bootstrap.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG;AAEH,MAAM,WAAW,WAAW;IAC1B,6BAA6B;IAC7B,GAAG,EAAE,MAAM,CAAC;IACZ,6BAA6B;IAC7B,IAAI,EAAE,MAAM,CAAC;IACb,mFAAmF;IACnF,QAAQ,EAAE,MAAM,CAAC;IACjB,+CAA+C;IAC/C,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,eAAgB,SAAQ,WAAW;IAClD,4FAA4F;IAC5F,WAAW,EAAE,OAAO,CAAC;CACtB;AAED;;;;GAIG;AACH,eAAO,MAAM,yBAAyB,OAAO,CAAC;AAE9C,iDAAiD;AACjD,eAAO,MAAM,uBAAuB,OAAO,CAAC;AA+C5C;;;;;;;;GAQG;AACH,wBAAgB,eAAe,CAC7B,MAAM,EAAE,MAAM,EAAE,EAChB,KAAK,SAA0B,EAC/B,OAAO,SAA4B,EACnC,IAAI,CAAC,EAAE,MAAM,GACZ,WAAW,CAsBb;AAED;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,eAAe,CAC7B,OAAO,EAAE,MAAM,EAAE,EACjB,OAAO,EAAE,MAAM,EAAE,EACjB,KAAK,SAA0B,EAC/B,OAAO,SAA4B,EACnC,IAAI,CAAC,EAAE,MAAM,GACZ,eAAe,CAyBjB;AAED;;;;;;;GAOG;AACH,wBAAgB,mBAAmB,CACjC,MAAM,EAAE,MAAM,EAAE,EAChB,QAAQ,EAAE,CAAC,SAAS,EAAE,MAAM,EAAE,KAAK,MAAM,EACzC,KAAK,SAA0B,EAC/B,OAAO,SAA4B,EACnC,IAAI,CAAC,EAAE,MAAM,GACZ,WAAW,CAgBb"}
|
|
@@ -20,6 +20,14 @@
|
|
|
20
20
|
* Reproducibility: pass a fixed `seed` to get deterministic CIs across runs.
|
|
21
21
|
* Without a seed we use Math.random() — fine for production but not for tests.
|
|
22
22
|
*/
|
|
23
|
+
/**
|
|
24
|
+
* Default number of bootstrap resamples. Every eval path uses this unless
|
|
25
|
+
* `--bootstrap-samples` overrides it. Single source of truth: the docs cite
|
|
26
|
+
* it and `test/scripts/doc-constants-drift.test.ts` guards doc ↔ code parity.
|
|
27
|
+
*/
|
|
28
|
+
export const DEFAULT_BOOTSTRAP_SAMPLES = 1000;
|
|
29
|
+
/** Default significance level; 0.05 → 95% CI. */
|
|
30
|
+
export const DEFAULT_BOOTSTRAP_ALPHA = 0.05;
|
|
23
31
|
/** Mulberry32 PRNG — seedable, deterministic for tests. */
|
|
24
32
|
function mulberry32(seed) {
|
|
25
33
|
let s = seed >>> 0;
|
|
@@ -75,7 +83,7 @@ function sortedQuantile(sorted, q) {
|
|
|
75
83
|
* @param seed Optional seed for deterministic CIs (tests).
|
|
76
84
|
* @returns { low, high, estimate, samples }; all numbers rounded to 4 decimals.
|
|
77
85
|
*/
|
|
78
|
-
export function bootstrapMeanCI(scores, alpha =
|
|
86
|
+
export function bootstrapMeanCI(scores, alpha = DEFAULT_BOOTSTRAP_ALPHA, samples = DEFAULT_BOOTSTRAP_SAMPLES, seed) {
|
|
79
87
|
if (scores.length === 0) {
|
|
80
88
|
return { low: 0, high: 0, estimate: 0, samples: 0 };
|
|
81
89
|
}
|
|
@@ -114,7 +122,7 @@ export function bootstrapMeanCI(scores, alpha = 0.05, samples = 1000, seed) {
|
|
|
114
122
|
* @param seed Optional seed.
|
|
115
123
|
* @returns BootstrapDiffCI with low/high of (B - A) and significant flag.
|
|
116
124
|
*/
|
|
117
|
-
export function bootstrapDiffCI(scoresA, scoresB, alpha =
|
|
125
|
+
export function bootstrapDiffCI(scoresA, scoresB, alpha = DEFAULT_BOOTSTRAP_ALPHA, samples = DEFAULT_BOOTSTRAP_SAMPLES, seed) {
|
|
118
126
|
if (scoresA.length === 0 || scoresB.length === 0) {
|
|
119
127
|
return { low: 0, high: 0, estimate: 0, samples: 0, significant: false };
|
|
120
128
|
}
|
|
@@ -150,7 +158,7 @@ export function bootstrapDiffCI(scoresA, scoresB, alpha = 0.05, samples = 1000,
|
|
|
150
158
|
* @param scores Original sample.
|
|
151
159
|
* @param metricFn Function reducing a resampled array to a scalar.
|
|
152
160
|
*/
|
|
153
|
-
export function bootstrapWithMetric(scores, metricFn, alpha =
|
|
161
|
+
export function bootstrapWithMetric(scores, metricFn, alpha = DEFAULT_BOOTSTRAP_ALPHA, samples = DEFAULT_BOOTSTRAP_SAMPLES, seed) {
|
|
154
162
|
if (scores.length === 0)
|
|
155
163
|
return { low: 0, high: 0, estimate: 0, samples: 0 };
|
|
156
164
|
const rng = makeRng(seed);
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"bootstrap.js","sourceRoot":"","sources":["../../src/eval-core/bootstrap.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG;AAkBH,2DAA2D;AAC3D,SAAS,UAAU,CAAC,IAAY;IAC9B,IAAI,CAAC,GAAG,IAAI,KAAK,CAAC,CAAC;IACnB,OAAO;QACL,CAAC,GAAG,CAAC,CAAC,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC;QAC3B,IAAI,CAAC,GAAG,CAAC,CAAC;QACV,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACrC,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC;QAC1C,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,GAAG,UAAU,CAAC;IAC/C,CAAC,CAAC;AACJ,CAAC;AAED,SAAS,OAAO,CAAC,IAAa;IAC5B,IAAI,IAAI,IAAI,IAAI;QAAE,OAAO,IAAI,CAAC,MAAM,CAAC;IACrC,OAAO,UAAU,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED,+EAA+E;AAC/E,SAAS,eAAe,CAAC,MAAc,EAAE,CAAS,EAAE,GAAiB;IACnE,MAAM,OAAO,GAAa,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC;IACvC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3B,OAAO,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,GAAG,MAAM,CAAC,CAAC;IAC1C,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,IAAI,CAAC,GAAa;IACzB,IAAI,GAAG,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAC/B,IAAI,GAAG,GAAG,CAAC,CAAC;IACZ,KAAK,MAAM,CAAC,IAAI,GAAG;QAAE,GAAG,IAAI,CAAC,CAAC;IAC9B,OAAO,GAAG,GAAG,GAAG,CAAC,MAAM,CAAC;AAC1B,CAAC;AAED,6DAA6D;AAC7D,SAAS,cAAc,CAAC,MAAgB,EAAE,CAAS;IACjD,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAClC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,MAAM,CAAC,CAAC,CAAC,CAAC;IAC1C,MAAM,GAAG,GAAG,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACpC,MAAM,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAC3B,MAAM,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC1B,IAAI,EAAE,KAAK,EAAE;QAAE,OAAO,MAAM,CAAC,EAAE,CAAC,CAAC;IACjC,MAAM,IAAI,GAAG,GAAG,GAAG,EAAE,CAAC;IACtB,OAAO,MAAM,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,MAAM,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC;AACrD,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,eAAe,CAC7B,MAAgB,EAChB,KAAK,GAAG,
|
|
1
|
+
{"version":3,"file":"bootstrap.js","sourceRoot":"","sources":["../../src/eval-core/bootstrap.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG;AAkBH;;;;GAIG;AACH,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC;AAE9C,iDAAiD;AACjD,MAAM,CAAC,MAAM,uBAAuB,GAAG,IAAI,CAAC;AAE5C,2DAA2D;AAC3D,SAAS,UAAU,CAAC,IAAY;IAC9B,IAAI,CAAC,GAAG,IAAI,KAAK,CAAC,CAAC;IACnB,OAAO;QACL,CAAC,GAAG,CAAC,CAAC,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC;QAC3B,IAAI,CAAC,GAAG,CAAC,CAAC;QACV,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACrC,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC;QAC1C,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,GAAG,UAAU,CAAC;IAC/C,CAAC,CAAC;AACJ,CAAC;AAED,SAAS,OAAO,CAAC,IAAa;IAC5B,IAAI,IAAI,IAAI,IAAI;QAAE,OAAO,IAAI,CAAC,MAAM,CAAC;IACrC,OAAO,UAAU,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED,+EAA+E;AAC/E,SAAS,eAAe,CAAC,MAAc,EAAE,CAAS,EAAE,GAAiB;IACnE,MAAM,OAAO,GAAa,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC;IACvC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3B,OAAO,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,GAAG,MAAM,CAAC,CAAC;IAC1C,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,IAAI,CAAC,GAAa;IACzB,IAAI,GAAG,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAC/B,IAAI,GAAG,GAAG,CAAC,CAAC;IACZ,KAAK,MAAM,CAAC,IAAI,GAAG;QAAE,GAAG,IAAI,CAAC,CAAC;IAC9B,OAAO,GAAG,GAAG,GAAG,CAAC,MAAM,CAAC;AAC1B,CAAC;AAED,6DAA6D;AAC7D,SAAS,cAAc,CAAC,MAAgB,EAAE,CAAS;IACjD,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAClC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,MAAM,CAAC,CAAC,CAAC,CAAC;IAC1C,MAAM,GAAG,GAAG,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACpC,MAAM,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAC3B,MAAM,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC1B,IAAI,EAAE,KAAK,EAAE;QAAE,OAAO,MAAM,CAAC,EAAE,CAAC,CAAC;IACjC,MAAM,IAAI,GAAG,GAAG,GAAG,EAAE,CAAC;IACtB,OAAO,MAAM,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,MAAM,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC;AACrD,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,eAAe,CAC7B,MAAgB,EAChB,KAAK,GAAG,uBAAuB,EAC/B,OAAO,GAAG,yBAAyB,EACnC,IAAa;IAEb,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,EAAE,GAAG,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;IACtD,CAAC;IACD,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;IAC9E,CAAC;IACD,MAAM,GAAG,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC1B,MAAM,aAAa,GAAa,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC;IACnD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,EAAE,CAAC,EAAE,EAAE,CAAC;QACjC,MAAM,GAAG,GAAG,eAAe,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;QAC/D,IAAI,GAAG,GAAG,CAAC,CAAC;QACZ,KAAK,MAAM,CAAC,IAAI,GAAG;YAAE,GAAG,IAAI,MAAM,CAAC,CAAC,CAAC,CAAC;QACtC,aAAa,CAAC,CAAC,CAAC,GAAG,GAAG,GAAG,MAAM,CAAC,MAAM,CAAC;IACzC,CAAC;IACD,aAAa,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACpC,OAAO;QACL,GAAG,EAAE,MAAM,CAAC,cAAc,CAAC,aAAa,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC;QACrD,IAAI,EAAE,MAAM,CAAC,cAAc,CAAC,aAAa,EAAE,CAAC,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC;QAC1D,QAAQ,EAAE,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAC9B,OAAO;KACR,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;GAcG;AACH,MAAM,UAAU,eAAe,CAC7B,OAAiB,EACjB,OAAiB,EACjB,KAAK,GAAG,uBAAuB,EAC/B,OAAO,GAAG,yBAAyB,EACnC,IAAa;IAEb,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACjD,OAAO,EAAE,GAAG,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,WAAW,EAAE,KAAK,EAAE,CAAC;IAC1E,CAAC;IACD,MAAM,GAAG,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC1B,MAAM,SAAS,GAAa,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC;IAC/C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,EAAE,CAAC,EAAE,EAAE,CAAC;QACjC,MAAM,IAAI,GAAG,eAAe,CAAC,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;QAClE,MAAM,IAAI,GAAG,eAAe,CAAC,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;QAClE,IAAI,IAAI,GAAG,CAAC,CAAC;QACb,IAAI,IAAI,GAAG,CAAC,CAAC;QACb,KAAK,MAAM,CAAC,IAAI,IAAI;YAAE,IAAI,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;QACzC,KAAK,MAAM,CAAC,IAAI,IAAI;YAAE,IAAI,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;QACzC,SAAS,CAAC,CAAC,CAAC,GAAG,IAAI,GAAG,OAAO,CAAC,MAAM,GAAG,IAAI,GAAG,OAAO,CAAC,MAAM,CAAC;IAC/D,CAAC;IACD,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAChC,MAAM,GAAG,GAAG,MAAM,CAAC,cAAc,CAAC,SAAS,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC;IACzD,MAAM,IAAI,GAAG,MAAM,CAAC,cAAc,CAAC,SAAS,EAAE,CAAC,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC;IAC9D,OAAO;QACL,GAAG;QACH,IAAI;QACJ,QAAQ,EAAE,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,IAAI,CAAC,OAAO,CAAC,CAAC;QAC/C,OAAO;QACP,WAAW,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC;KACtC,CAAC;AACJ,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,mBAAmB,CACjC,MAAgB,EAChB,QAAyC,EACzC,KAAK,GAAG,uBAAuB,EAC/B,OAAO,GAAG,yBAAyB,EACnC,IAAa;IAEb,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,GAAG,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;IAC7E,MAAM,GAAG,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC1B,MAAM,YAAY,GAAa,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC;IAClD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,EAAE,CAAC,EAAE,EAAE,CAAC;QACjC,MAAM,GAAG,GAAG,eAAe,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;QAC/D,MAAM,SAAS,GAAG,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;QAC5C,YAAY,CAAC,CAAC,CAAC,GAAG,QAAQ,CAAC,SAAS,CAAC,CAAC;IACxC,CAAC;IACD,YAAY,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACnC,OAAO;QACL,GAAG,EAAE,MAAM,CAAC,cAAc,CAAC,YAAY,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC;QACpD,IAAI,EAAE,MAAM,CAAC,cAAc,CAAC,YAAY,EAAE,CAAC,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC;QACzD,QAAQ,EAAE,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;QAClC,OAAO;KACR,CAAC;AACJ,CAAC;AAED,SAAS,MAAM,CAAC,CAAS;IACvB,OAAO,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,KAAK,CAAC,GAAG,KAAK,CAAC;AACvC,CAAC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"evaluation-reporting.d.ts","sourceRoot":"","sources":["../../src/eval-core/evaluation-reporting.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"evaluation-reporting.d.ts","sourceRoot":"","sources":["../../src/eval-core/evaluation-reporting.ts"],"names":[],"mappings":"AAgBA,OAAO,KAAK,EACV,QAAQ,EACR,MAAM,EACN,MAAM,EACN,IAAI,EACJ,aAAa,EAGb,OAAO,EACP,aAAa,EACb,iBAAiB,EACjB,aAAa,EACd,MAAM,mBAAmB,CAAC;AAgB3B,eAAO,MAAM,kBAAkB,EAAE,MAAuD,CAAC;AAEzF,wBAAgB,UAAU,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAE9C;AAeD;;;;GAIG;AACH,wBAAgB,UAAU,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,CASjD;AAED,wBAAgB,aAAa,IAAI,MAAM,CAEtC;AAED,wBAAgB,UAAU,IAAI,OAAO,GAAG,IAAI,CAS3C;AAqDD,UAAU,sBAAsB;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC,CAAC;IACvD,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,OAAO,CAAC,EAAE,iBAAiB,CAAC;IAC5B,GAAG,CAAC,EAAE,aAAa,CAAC;IACpB,GAAG,CAAC,EAAE,aAAa,CAAC;IACpB,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAED,wBAAgB,eAAe,CAAC,EAC9B,KAAK,EACL,QAAQ,EACR,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,OAAO,EACP,KAAK,EACL,OAAO,EACP,YAAY,EACZ,SAAS,EACT,OAAO,EACP,GAAG,EACH,GAAG,EACH,YAAY,GACb,EAAE,sBAAsB,GAAG,MAAM,CA2IjC;AAED,wBAAgB,cAAc,CAAC,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,EAAE,SAAS,EAAE,MAAM,GAAG,IAAI,CAwC1F;AAED,MAAM,WAAW,iBAAiB;IAChC,EAAE,EAAE,MAAM,CAAC;CACZ;AAED,wBAAgB,aAAa,CAAC,MAAM,EAAE,iBAAiB,EAAE,SAAS,EAAE,MAAM,GAAG,IAAI,GAAG,MAAM,GAAG,IAAI,CAMhG;AAED,wBAAgB,aAAa,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,MAAM,CASxD"}
|
|
@@ -7,7 +7,7 @@ import { fileURLToPath } from 'node:url';
|
|
|
7
7
|
import { buildVariantSummary } from './schema.js';
|
|
8
8
|
import { buildVariantConfig, resolveExecutionStrategy } from './execution-strategy.js';
|
|
9
9
|
import { getJudgePromptHash } from '../grading/judge.js';
|
|
10
|
-
import { bootstrapMeanCI, bootstrapDiffCI } from './bootstrap.js';
|
|
10
|
+
import { bootstrapMeanCI, bootstrapDiffCI, DEFAULT_BOOTSTRAP_ALPHA, DEFAULT_BOOTSTRAP_SAMPLES, } from './bootstrap.js';
|
|
11
11
|
import { getExecutorRuntimeFingerprint } from '../executors/runtime-fingerprint.js';
|
|
12
12
|
const __dirname = dirname(fileURLToPath(import.meta.url));
|
|
13
13
|
function findPackageJson(startDir) {
|
|
@@ -112,7 +112,7 @@ export function aggregateReport({ runId, variants, model, judgeModel, noJudge, e
|
|
|
112
112
|
// Bootstrap CI (per-variant mean) when --bootstrap requested. Adds bootstrapCI to
|
|
113
113
|
// each VariantSummary; legacy t-interval (in summary's other fields) is preserved.
|
|
114
114
|
const bootstrapEnabled = request?.bootstrap === true;
|
|
115
|
-
const bootstrapSamples = request?.bootstrapSamples ??
|
|
115
|
+
const bootstrapSamples = request?.bootstrapSamples ?? DEFAULT_BOOTSTRAP_SAMPLES;
|
|
116
116
|
let pairComparisons;
|
|
117
117
|
if (bootstrapEnabled) {
|
|
118
118
|
for (const variant of variants) {
|
|
@@ -121,7 +121,7 @@ export function aggregateReport({ runId, variants, model, judgeModel, noJudge, e
|
|
|
121
121
|
.filter((e) => typeof e.compositeScore === 'number' && e.compositeScore > 0)
|
|
122
122
|
.map((e) => e.compositeScore);
|
|
123
123
|
if (compositeScores.length >= 2) {
|
|
124
|
-
summary[variant].bootstrapCI = bootstrapMeanCI(compositeScores,
|
|
124
|
+
summary[variant].bootstrapCI = bootstrapMeanCI(compositeScores, DEFAULT_BOOTSTRAP_ALPHA, bootstrapSamples);
|
|
125
125
|
}
|
|
126
126
|
}
|
|
127
127
|
// Pairwise treatment-vs-control comparisons. Convention: variants[0] is control;
|
|
@@ -143,7 +143,7 @@ export function aggregateReport({ runId, variants, model, judgeModel, noJudge, e
|
|
|
143
143
|
pairComparisons.push({
|
|
144
144
|
control: controlName,
|
|
145
145
|
treatment: treatmentName,
|
|
146
|
-
diffBootstrapCI: bootstrapDiffCI(controlScores, treatmentScores,
|
|
146
|
+
diffBootstrapCI: bootstrapDiffCI(controlScores, treatmentScores, DEFAULT_BOOTSTRAP_ALPHA, bootstrapSamples),
|
|
147
147
|
});
|
|
148
148
|
}
|
|
149
149
|
}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"evaluation-reporting.js","sourceRoot":"","sources":["../../src/eval-core/evaluation-reporting.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,SAAS,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAC1C,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AACzC,OAAO,EAAE,mBAAmB,EAAE,MAAM,aAAa,CAAC;AAClD,OAAO,EAAE,kBAAkB,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AACvF,OAAO,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AACzD,OAAO,EAAE,eAAe,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AAClE,OAAO,EAAE,6BAA6B,EAAE,MAAM,qCAAqC,CAAC;AAepF,MAAM,SAAS,GAAG,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAE1D,SAAS,eAAe,CAAC,QAAgB;IACvC,IAAI,GAAG,GAAG,QAAQ,CAAC;IACnB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,cAAc,CAAC,CAAC;QAC5C,IAAI,UAAU,CAAC,SAAS,CAAC;YAAE,OAAO,SAAS,CAAC;QAC5C,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC;IACrB,CAAC;IACD,OAAO,IAAI,CAAC,QAAQ,EAAE,IAAI,EAAE,cAAc,CAAC,CAAC;AAC9C,CAAC;AAED,MAAM,GAAG,GAAwB,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,eAAe,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC,CAAwB,CAAC;AAEtH,MAAM,CAAC,MAAM,kBAAkB,GAAW,IAAI,CAAC,OAAO,EAAE,EAAE,kBAAkB,EAAE,SAAS,CAAC,CAAC;AAEzF,MAAM,UAAU,UAAU,CAAC,GAAW;IACpC,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;AACrE,CAAC;AAED;;;;;GAKG;AACH,SAAS,kBAAkB,CAAC,KAAc;IACxC,IAAI,KAAK,KAAK,IAAI,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IAC9E,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,OAAO,GAAG,GAAG,KAAK,CAAC,GAAG,CAAC,kBAAkB,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC;IACrF,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,CAAC,KAAgC,CAAC,CAAC,IAAI,EAAE,CAAC;IACrE,OAAO,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,GAAG,GAAG,GAAG,kBAAkB,CAAE,KAAiC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC;AACvI,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,UAAU,CAAC,MAAc;IACvC,MAAM,UAAU,GAAG,kBAAkB,CAAC;QACpC,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,IAAI;QAC7B,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;QACrC,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;QACrC,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,IAAI;KAC9B,CAAC,CAAC;IACH,OAAO,UAAU,CAAC,UAAU,CAAC,CAAC;AAChC,CAAC;AAED,MAAM,UAAU,aAAa;IAC3B,OAAO,GAAG,CAAC,OAAO,CAAC;AACrB,CAAC;AAED,MAAM,UAAU,UAAU;IACxB,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,WAAW,EAAE,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxF,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,WAAW,EAAE,cAAc,EAAE,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxG,MAAM,KAAK,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,QAAQ,EAAE,aAAa,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC;QACtG,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC;IACpE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,aAAa,CAAC,QAA0E;IAC/F,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IACvC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAC1C,MAAM,KAAK,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;IACxB,OAAO,MAAM,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,WAAW,KAAK,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;AAClG,CAAC;AAED,SAAS,qBAAqB,CAAC,QAA0E;IACvG,OAAO,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;AACpC,CAAC;AAED,SAAS,8BAA8B,CAAC,EACtC,QAAQ,EACR,KAAK,EACL,YAAY,EACZ,KAAK,EACL,SAAS,EACT,OAAO,GAQR;IACC,MAAM,QAAQ,GAAqE,EAAE,CAAC;IACtF,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,IAAI,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC;YAAE,SAAS;QACrC,MAAM,aAAa,GAAG,wBAAwB,CAAC,IAAI,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,CAAC,CAAC;QACvF,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE;YAC1E,QAAQ,EAAE,aAAa,CAAC,KAAK,CAAC,QAAQ;SACvC,CAAC,CAAC;IACL,CAAC;IAED,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,IAAI,QAAQ,CAAC,OAAO,CAAC;YAAE,SAAS;QAChC,MAAM,QAAQ,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,OAAO,CAAC,CAAC;QAC3D,MAAM,gBAAgB,GAAG,QAAQ,EAAE,IAAI,KAAK,UAAU;YACpD,CAAC,CAAC,IAAI;YACN,CAAC,CAAC,QAAQ,EAAE,OAAO,IAAI,QAAQ,CAAC,MAAM,KAAK,KAAK;gBAC9C,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;gBAC3B,CAAC,CAAC,OAAO,EAAE,QAAQ,CAAC;QACxB,QAAQ,CAAC,OAAO,CAAC,GAAG,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE;YACrE,QAAQ,EAAE,gBAAgB;SAC3B,CAAC,CAAC;IACL,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC;AAoBD,MAAM,UAAU,eAAe,CAAC,EAC9B,KAAK,EACL,QAAQ,EACR,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,OAAO,EACP,KAAK,EACL,OAAO,EACP,YAAY,EACZ,SAAS,EACT,OAAO,EACP,GAAG,EACH,GAAG,EACH,YAAY,GACW;IACvB,MAAM,OAAO,GAAmC,EAAE,CAAC;IACnD,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACxF,OAAO,CAAC,OAAO,CAAC,GAAG,mBAAmB,CAAC,OAAO,CAAC,CAAC;IAClD,CAAC;IAED,kFAAkF;IAClF,mFAAmF;IACnF,MAAM,gBAAgB,GAAG,OAAO,EAAE,SAAS,KAAK,IAAI,CAAC;IACrD,MAAM,gBAAgB,GAAG,OAAO,EAAE,gBAAgB,IAAI,IAAI,CAAC;IAC3D,IAAI,eAAoD,CAAC;IACzD,IAAI,gBAAgB,EAAE,CAAC;QACrB,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;YAC/B,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YAC9E,MAAM,eAAe,GAAG,OAAO;iBAC5B,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,CAAC,cAAe,GAAG,CAAC,CAAC;iBAC5E,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAe,CAAC,CAAC;YACjC,IAAI,eAAe,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;gBAChC,OAAO,CAAC,OAAO,CAAC,CAAC,WAAW,GAAG,eAAe,CAAC,eAAe,EAAE,IAAI,EAAE,gBAAgB,CAAC,CAAC;YAC1F,CAAC;QACH,CAAC;QAED,iFAAiF;QACjF,8DAA8D;QAC9D,IAAI,QAAQ,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;YACzB,eAAe,GAAG,EAAE,CAAC;YACrB,MAAM,WAAW,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;YAChC,MAAM,cAAc,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YACzF,MAAM,aAAa,GAAG,cAAc;iBACjC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,CAAC,cAAe,GAAG,CAAC,CAAC;iBAC5E,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAe,CAAC,CAAC;YACjC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBACzC,MAAM,aAAa,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;gBAClC,MAAM,gBAAgB,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;gBAC7F,MAAM,eAAe,GAAG,gBAAgB;qBACrC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,CAAC,cAAe,GAAG,CAAC,CAAC;qBAC5E,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAe,CAAC,CAAC;gBACjC,IAAI,aAAa,CAAC,MAAM,IAAI,CAAC,IAAI,eAAe,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;oBAC7D,eAAe,CAAC,IAAI,CAAC;wBACnB,OAAO,EAAE,WAAW;wBACpB,SAAS,EAAE,aAAa;wBACxB,eAAe,EAAE,eAAe,CAAC,aAAa,EAAE,eAAe,EAAE,IAAI,EAAE,gBAAgB,CAAC;qBACzF,CAAC,CAAC;gBACL,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IAED,MAAM,cAAc,GAAG,MAAM,CAAC,WAAW,CACvC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAC3G,CAAC;IAEF,MAAM,YAAY,GAAG,MAAM,CAAC,WAAW,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1F,MAAM,WAAW,GAAG,OAAO,EAAE,WAAW,IAAI,OAAO,CAAC,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;IACtG,MAAM,cAAc,GAAG,EAAE,QAAQ,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC;IACvD,MAAM,gBAAgB,GAAG,8BAA8B,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,YAAY,EAAE,KAAK,EAAE,SAAS,EAAE,OAAO,EAAE,CAAC,CAAC;IACtH,MAAM,eAAe,GAAG,aAAa,CAAC,gBAAgB,CAAC;WAClD,qBAAqB,CAAC,gBAAgB,CAAC;WACvC,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE,cAAc,CAAC,CAAC;IACxE,oFAAoF;IACpF,iFAAiF;IACjF,sFAAsF;IACtF,MAAM,aAAa,GAAG,OAAO,EAAE,WAAW,IAAI,CAAC,EAAE,QAAQ,EAAE,YAAY,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;IAC9F,MAAM,eAAe,GAAoD,aAAa,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC;QAClG,QAAQ,EAAE,EAAE,CAAC,QAAQ;QACrB,KAAK,EAAE,EAAE,CAAC,KAAK;QACf,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,6BAA6B,CAAC,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC,KAAK,EAAE,cAAc,CAAC,EAAE,CAAC;KACtG,CAAC,CAAC,CAAC;IACJ,2DAA2D;IAC3D,0EAA0E;IAC1E,4EAA4E;IAC5E,MAAM,cAAc,GAAG,OAAO,EAAE,YAAY,KAAK,KAAK,CAAC;IACvD,MAAM,cAAc,GAAiC,EAAE,CAAC;IACxD,IAAI,cAAc;QAAE,cAAc,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAClD,MAAM,iBAAiB,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CACjE,OAAO,CAAC,gBAAgB,KAAK,KAAK,IAAI,OAAO,CAAC,iBAAiB,KAAK,KAAK,CAAC,CAAC;IAE7E,OAAO;QACL,IAAI,EAAE,YAAY;QAClB,EAAE,EAAE,KAAK;QACT,IAAI,EAAE;YACJ,QAAQ;YACR,KAAK;YACL,QAAQ,EAAE,YAAY;YACtB,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACtD,WAAW,EAAE,OAAO,CAAC,MAAM;YAC3B,SAAS,EAAE,KAAK,CAAC,MAAM;YACvB,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YAC7C,GAAG,CAAC,iBAAiB,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,KAAK,EAAE,CAAC;YAC1D,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACnC,UAAU,EAAE,aAAa,EAAE;YAC3B,WAAW,EAAE,OAAO,CAAC,OAAO;YAC5B,cAAc;YACd,YAAY;YACZ,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,kBAAkB,CAAC,cAAc,CAAC,EAAE,CAAC;YAC3E,eAAe;YACf,gBAAgB;YAChB,WAAW,EAAE,eAAe;YAC5B,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACrC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACvC,GAAG,CAAC,cAAc,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACpE,GAAG,CAAC,gBAAgB,CAAC,CAAC,CAAC,EAAE,mBAAmB,EAAE,MAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACrE,GAAG,CAAC,eAAe,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAC/C,cAAc,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,kBAAkB,CAAC,QAAQ,CAAC,CAAC;YACzE,6EAA6E;YAC7E,6DAA6D;YAC7D,4CAA4C;YAC5C,cAAc,EAAE,MAAM,CAAC,WAAW,CAChC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,aAAa,IAAI,IAAI,CAAC,CAAC,CACxD;YACD,OAAO;YACP,GAAG;YACH,GAAG;YACH,OAAO,EAAE,UAAU,EAAE;YACrB,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAChD;QACD,OAAO;QACP,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,EAAE,WAAW,CAAC,EAAE,EAAE,CAAC,CAAC;YAClE,SAAS;YACT,QAAQ,EAAE,WAAW;SACtB,CAAC,CAAC;QACH,mDAAmD;QACnD,iEAAiE;QACjE,0CAA0C;QAC1C,eAAe,EAAE,MAAM,CAAC,WAAW,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE;gBACnE,SAAS,EAAE,CAAC,CAAC,SAAS;gBACtB,MAAM,EAAE,CAAC,CAAC,MAAM;gBAChB,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACzC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC5C,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAChF,GAAG,CAAC,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC5D,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAChF,GAAG,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACrD,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAClD,GAAG,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACrD,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC1C,CAAC,CAAC,CAAC;KACL,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,MAAc,EAAE,QAAkB,EAAE,SAAiB;IAClF,MAAM,MAAM,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,YAAY,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC;IACnE,IAAI,IAAI,GAAG,QAAQ,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC;IAC/D,MAAM,YAAY,GAAG,GAAW,EAAE;QAChC,IAAI,IAAI,CAAC,CAAC;QACV,IAAI,GAAG,IAAI,GAAG,UAAU,GAAG,CAAC,CAAC;QAC7B,IAAI,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,GAAG,IAAI,KAAK,EAAE,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;QACpD,KAAK,IAAI,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,KAAK,CAAC,EAAE,EAAE,GAAG,KAAK,CAAC,CAAC;QAC5D,OAAO,CAAC,CAAC,KAAK,GAAG,KAAK,KAAK,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,UAAU,CAAC;IACrD,CAAC,CAAC;IACF,MAAM,QAAQ,GAAG,CAAC,GAAG,QAAQ,CAAC,CAAC;IAC/B,KAAK,IAAI,CAAC,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC7C,MAAM,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,EAAE,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QAC/C,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1D,CAAC;IACD,MAAM,QAAQ,GAA2B,MAAM,CAAC,WAAW,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC;IAChH,MAAM,UAAU,GAA2B,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,EAAE,OAAO,CAAC,EAAE,EAAE,CAAC,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC;IAEpI,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC;IACzB,MAAM,CAAC,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;IAChC,MAAM,CAAC,IAAI,CAAC,QAAQ,GAAG,MAAM,CAAC;IAC9B,IAAI,MAAM,CAAC,IAAI,CAAC,gBAAgB,EAAE,CAAC;QACjC,MAAM,CAAC,IAAI,CAAC,gBAAgB,GAAG,MAAM,CAAC,WAAW,CAC/C,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,OAAO,CAAC,EAAE,EAAE,CAAC,CAAC,UAAU,CAAC,OAAO,CAAC,IAAI,OAAO,EAAE,OAAO,CAAC,CAAC,CACpH,CAAC;IACJ,CAAC;IAED,MAAM,UAAU,GAAmC,EAAE,CAAC;IACtD,KAAK,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC;QAC9D,UAAU,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,GAAG,KAAK,CAAC;IAC1C,CAAC;IACD,MAAM,CAAC,OAAO,GAAG,UAAU,CAAC;IAE5B,KAAK,MAAM,MAAM,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;QACpC,MAAM,WAAW,GAAkC,EAAE,CAAC;QACtD,KAAK,MAAM,CAAC,OAAO,EAAE,IAAI,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC9D,WAAW,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,GAAG,IAAI,CAAC;QAC1C,CAAC;QACD,MAAM,CAAC,QAAQ,GAAG,WAAW,CAAC;IAChC,CAAC;AACH,CAAC;AAMD,MAAM,UAAU,aAAa,CAAC,MAAyB,EAAE,SAAwB;IAC/E,IAAI,CAAC,SAAS;QAAE,OAAO,IAAI,CAAC;IAC5B,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC;QAAE,SAAS,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACtE,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,EAAE,GAAG,MAAM,CAAC,EAAE,OAAO,CAAC,CAAC;IACtD,aAAa,CAAC,QAAQ,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IACzD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,QAAkB;IAC9C,MAAM,CAAC,GAAG,IAAI,IAAI,EAAE,CAAC;IACrB,MAAM,GAAG,GAAG,CAAC,CAAS,EAAU,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;IAC9D,MAAM,IAAI,GAAG,GAAG,CAAC,CAAC,WAAW,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,EAAE,CAAC;IAC7E,MAAM,IAAI,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,UAAU,EAAE,CAAC,EAAE,CAAC;IAC1D,MAAM,WAAW,GAAG,QAAQ;SACzB,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC,UAAU,CAAC,mBAAmB,EAAE,GAAG,CAAC,CAAC;SACzF,IAAI,CAAC,MAAM,CAAC,CAAC;IAChB,OAAO,GAAG,WAAW,IAAI,IAAI,IAAI,IAAI,EAAE,CAAC;AAC1C,CAAC"}
|
|
1
|
+
{"version":3,"file":"evaluation-reporting.js","sourceRoot":"","sources":["../../src/eval-core/evaluation-reporting.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,SAAS,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAC1C,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AACzC,OAAO,EAAE,mBAAmB,EAAE,MAAM,aAAa,CAAC;AAClD,OAAO,EAAE,kBAAkB,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AACvF,OAAO,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AACzD,OAAO,EACL,eAAe,EACf,eAAe,EACf,uBAAuB,EACvB,yBAAyB,GAC1B,MAAM,gBAAgB,CAAC;AACxB,OAAO,EAAE,6BAA6B,EAAE,MAAM,qCAAqC,CAAC;AAepF,MAAM,SAAS,GAAG,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAE1D,SAAS,eAAe,CAAC,QAAgB;IACvC,IAAI,GAAG,GAAG,QAAQ,CAAC;IACnB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,cAAc,CAAC,CAAC;QAC5C,IAAI,UAAU,CAAC,SAAS,CAAC;YAAE,OAAO,SAAS,CAAC;QAC5C,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC;IACrB,CAAC;IACD,OAAO,IAAI,CAAC,QAAQ,EAAE,IAAI,EAAE,cAAc,CAAC,CAAC;AAC9C,CAAC;AAED,MAAM,GAAG,GAAwB,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,eAAe,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC,CAAwB,CAAC;AAEtH,MAAM,CAAC,MAAM,kBAAkB,GAAW,IAAI,CAAC,OAAO,EAAE,EAAE,kBAAkB,EAAE,SAAS,CAAC,CAAC;AAEzF,MAAM,UAAU,UAAU,CAAC,GAAW;IACpC,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;AACrE,CAAC;AAED;;;;;GAKG;AACH,SAAS,kBAAkB,CAAC,KAAc;IACxC,IAAI,KAAK,KAAK,IAAI,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IAC9E,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,OAAO,GAAG,GAAG,KAAK,CAAC,GAAG,CAAC,kBAAkB,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC;IACrF,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,CAAC,KAAgC,CAAC,CAAC,IAAI,EAAE,CAAC;IACrE,OAAO,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,GAAG,GAAG,GAAG,kBAAkB,CAAE,KAAiC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC;AACvI,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,UAAU,CAAC,MAAc;IACvC,MAAM,UAAU,GAAG,kBAAkB,CAAC;QACpC,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,IAAI;QAC7B,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;QACrC,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;QACrC,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,IAAI;KAC9B,CAAC,CAAC;IACH,OAAO,UAAU,CAAC,UAAU,CAAC,CAAC;AAChC,CAAC;AAED,MAAM,UAAU,aAAa;IAC3B,OAAO,GAAG,CAAC,OAAO,CAAC;AACrB,CAAC;AAED,MAAM,UAAU,UAAU;IACxB,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,WAAW,EAAE,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxF,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,WAAW,EAAE,cAAc,EAAE,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxG,MAAM,KAAK,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,QAAQ,EAAE,aAAa,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC;QACtG,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC;IACpE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,aAAa,CAAC,QAA0E;IAC/F,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IACvC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAC1C,MAAM,KAAK,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;IACxB,OAAO,MAAM,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,WAAW,KAAK,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;AAClG,CAAC;AAED,SAAS,qBAAqB,CAAC,QAA0E;IACvG,OAAO,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;AACpC,CAAC;AAED,SAAS,8BAA8B,CAAC,EACtC,QAAQ,EACR,KAAK,EACL,YAAY,EACZ,KAAK,EACL,SAAS,EACT,OAAO,GAQR;IACC,MAAM,QAAQ,GAAqE,EAAE,CAAC;IACtF,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,IAAI,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC;YAAE,SAAS;QACrC,MAAM,aAAa,GAAG,wBAAwB,CAAC,IAAI,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,CAAC,CAAC;QACvF,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE;YAC1E,QAAQ,EAAE,aAAa,CAAC,KAAK,CAAC,QAAQ;SACvC,CAAC,CAAC;IACL,CAAC;IAED,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,IAAI,QAAQ,CAAC,OAAO,CAAC;YAAE,SAAS;QAChC,MAAM,QAAQ,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,OAAO,CAAC,CAAC;QAC3D,MAAM,gBAAgB,GAAG,QAAQ,EAAE,IAAI,KAAK,UAAU;YACpD,CAAC,CAAC,IAAI;YACN,CAAC,CAAC,QAAQ,EAAE,OAAO,IAAI,QAAQ,CAAC,MAAM,KAAK,KAAK;gBAC9C,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;gBAC3B,CAAC,CAAC,OAAO,EAAE,QAAQ,CAAC;QACxB,QAAQ,CAAC,OAAO,CAAC,GAAG,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE;YACrE,QAAQ,EAAE,gBAAgB;SAC3B,CAAC,CAAC;IACL,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC;AAoBD,MAAM,UAAU,eAAe,CAAC,EAC9B,KAAK,EACL,QAAQ,EACR,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,OAAO,EACP,KAAK,EACL,OAAO,EACP,YAAY,EACZ,SAAS,EACT,OAAO,EACP,GAAG,EACH,GAAG,EACH,YAAY,GACW;IACvB,MAAM,OAAO,GAAmC,EAAE,CAAC;IACnD,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACxF,OAAO,CAAC,OAAO,CAAC,GAAG,mBAAmB,CAAC,OAAO,CAAC,CAAC;IAClD,CAAC;IAED,kFAAkF;IAClF,mFAAmF;IACnF,MAAM,gBAAgB,GAAG,OAAO,EAAE,SAAS,KAAK,IAAI,CAAC;IACrD,MAAM,gBAAgB,GAAG,OAAO,EAAE,gBAAgB,IAAI,yBAAyB,CAAC;IAChF,IAAI,eAAoD,CAAC;IACzD,IAAI,gBAAgB,EAAE,CAAC;QACrB,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;YAC/B,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YAC9E,MAAM,eAAe,GAAG,OAAO;iBAC5B,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,CAAC,cAAe,GAAG,CAAC,CAAC;iBAC5E,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAe,CAAC,CAAC;YACjC,IAAI,eAAe,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;gBAChC,OAAO,CAAC,OAAO,CAAC,CAAC,WAAW,GAAG,eAAe,CAAC,eAAe,EAAE,uBAAuB,EAAE,gBAAgB,CAAC,CAAC;YAC7G,CAAC;QACH,CAAC;QAED,iFAAiF;QACjF,8DAA8D;QAC9D,IAAI,QAAQ,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;YACzB,eAAe,GAAG,EAAE,CAAC;YACrB,MAAM,WAAW,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;YAChC,MAAM,cAAc,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YACzF,MAAM,aAAa,GAAG,cAAc;iBACjC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,CAAC,cAAe,GAAG,CAAC,CAAC;iBAC5E,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAe,CAAC,CAAC;YACjC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBACzC,MAAM,aAAa,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;gBAClC,MAAM,gBAAgB,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;gBAC7F,MAAM,eAAe,GAAG,gBAAgB;qBACrC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,CAAC,cAAe,GAAG,CAAC,CAAC;qBAC5E,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAe,CAAC,CAAC;gBACjC,IAAI,aAAa,CAAC,MAAM,IAAI,CAAC,IAAI,eAAe,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;oBAC7D,eAAe,CAAC,IAAI,CAAC;wBACnB,OAAO,EAAE,WAAW;wBACpB,SAAS,EAAE,aAAa;wBACxB,eAAe,EAAE,eAAe,CAAC,aAAa,EAAE,eAAe,EAAE,uBAAuB,EAAE,gBAAgB,CAAC;qBAC5G,CAAC,CAAC;gBACL,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IAED,MAAM,cAAc,GAAG,MAAM,CAAC,WAAW,CACvC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAC3G,CAAC;IAEF,MAAM,YAAY,GAAG,MAAM,CAAC,WAAW,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1F,MAAM,WAAW,GAAG,OAAO,EAAE,WAAW,IAAI,OAAO,CAAC,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;IACtG,MAAM,cAAc,GAAG,EAAE,QAAQ,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC;IACvD,MAAM,gBAAgB,GAAG,8BAA8B,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,YAAY,EAAE,KAAK,EAAE,SAAS,EAAE,OAAO,EAAE,CAAC,CAAC;IACtH,MAAM,eAAe,GAAG,aAAa,CAAC,gBAAgB,CAAC;WAClD,qBAAqB,CAAC,gBAAgB,CAAC;WACvC,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE,cAAc,CAAC,CAAC;IACxE,oFAAoF;IACpF,iFAAiF;IACjF,sFAAsF;IACtF,MAAM,aAAa,GAAG,OAAO,EAAE,WAAW,IAAI,CAAC,EAAE,QAAQ,EAAE,YAAY,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;IAC9F,MAAM,eAAe,GAAoD,aAAa,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC;QAClG,QAAQ,EAAE,EAAE,CAAC,QAAQ;QACrB,KAAK,EAAE,EAAE,CAAC,KAAK;QACf,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,6BAA6B,CAAC,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC,KAAK,EAAE,cAAc,CAAC,EAAE,CAAC;KACtG,CAAC,CAAC,CAAC;IACJ,2DAA2D;IAC3D,0EAA0E;IAC1E,4EAA4E;IAC5E,MAAM,cAAc,GAAG,OAAO,EAAE,YAAY,KAAK,KAAK,CAAC;IACvD,MAAM,cAAc,GAAiC,EAAE,CAAC;IACxD,IAAI,cAAc;QAAE,cAAc,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAClD,MAAM,iBAAiB,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CACjE,OAAO,CAAC,gBAAgB,KAAK,KAAK,IAAI,OAAO,CAAC,iBAAiB,KAAK,KAAK,CAAC,CAAC;IAE7E,OAAO;QACL,IAAI,EAAE,YAAY;QAClB,EAAE,EAAE,KAAK;QACT,IAAI,EAAE;YACJ,QAAQ;YACR,KAAK;YACL,QAAQ,EAAE,YAAY;YACtB,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACtD,WAAW,EAAE,OAAO,CAAC,MAAM;YAC3B,SAAS,EAAE,KAAK,CAAC,MAAM;YACvB,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YAC7C,GAAG,CAAC,iBAAiB,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,KAAK,EAAE,CAAC;YAC1D,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACnC,UAAU,EAAE,aAAa,EAAE;YAC3B,WAAW,EAAE,OAAO,CAAC,OAAO;YAC5B,cAAc;YACd,YAAY;YACZ,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,kBAAkB,CAAC,cAAc,CAAC,EAAE,CAAC;YAC3E,eAAe;YACf,gBAAgB;YAChB,WAAW,EAAE,eAAe;YAC5B,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACrC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACvC,GAAG,CAAC,cAAc,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACpE,GAAG,CAAC,gBAAgB,CAAC,CAAC,CAAC,EAAE,mBAAmB,EAAE,MAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACrE,GAAG,CAAC,eAAe,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAC/C,cAAc,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,kBAAkB,CAAC,QAAQ,CAAC,CAAC;YACzE,6EAA6E;YAC7E,6DAA6D;YAC7D,4CAA4C;YAC5C,cAAc,EAAE,MAAM,CAAC,WAAW,CAChC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,aAAa,IAAI,IAAI,CAAC,CAAC,CACxD;YACD,OAAO;YACP,GAAG;YACH,GAAG;YACH,OAAO,EAAE,UAAU,EAAE;YACrB,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAChD;QACD,OAAO;QACP,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,EAAE,WAAW,CAAC,EAAE,EAAE,CAAC,CAAC;YAClE,SAAS;YACT,QAAQ,EAAE,WAAW;SACtB,CAAC,CAAC;QACH,mDAAmD;QACnD,iEAAiE;QACjE,0CAA0C;QAC1C,eAAe,EAAE,MAAM,CAAC,WAAW,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE;gBACnE,SAAS,EAAE,CAAC,CAAC,SAAS;gBACtB,MAAM,EAAE,CAAC,CAAC,MAAM;gBAChB,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACzC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC5C,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAChF,GAAG,CAAC,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC5D,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAChF,GAAG,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACrD,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAClD,GAAG,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACrD,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC1C,CAAC,CAAC,CAAC;KACL,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,MAAc,EAAE,QAAkB,EAAE,SAAiB;IAClF,MAAM,MAAM,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,YAAY,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC;IACnE,IAAI,IAAI,GAAG,QAAQ,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC;IAC/D,MAAM,YAAY,GAAG,GAAW,EAAE;QAChC,IAAI,IAAI,CAAC,CAAC;QACV,IAAI,GAAG,IAAI,GAAG,UAAU,GAAG,CAAC,CAAC;QAC7B,IAAI,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,GAAG,IAAI,KAAK,EAAE,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;QACpD,KAAK,IAAI,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,KAAK,CAAC,EAAE,EAAE,GAAG,KAAK,CAAC,CAAC;QAC5D,OAAO,CAAC,CAAC,KAAK,GAAG,KAAK,KAAK,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,UAAU,CAAC;IACrD,CAAC,CAAC;IACF,MAAM,QAAQ,GAAG,CAAC,GAAG,QAAQ,CAAC,CAAC;IAC/B,KAAK,IAAI,CAAC,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC7C,MAAM,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,EAAE,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QAC/C,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1D,CAAC;IACD,MAAM,QAAQ,GAA2B,MAAM,CAAC,WAAW,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC;IAChH,MAAM,UAAU,GAA2B,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,EAAE,OAAO,CAAC,EAAE,EAAE,CAAC,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC;IAEpI,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC;IACzB,MAAM,CAAC,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;IAChC,MAAM,CAAC,IAAI,CAAC,QAAQ,GAAG,MAAM,CAAC;IAC9B,IAAI,MAAM,CAAC,IAAI,CAAC,gBAAgB,EAAE,CAAC;QACjC,MAAM,CAAC,IAAI,CAAC,gBAAgB,GAAG,MAAM,CAAC,WAAW,CAC/C,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,OAAO,CAAC,EAAE,EAAE,CAAC,CAAC,UAAU,CAAC,OAAO,CAAC,IAAI,OAAO,EAAE,OAAO,CAAC,CAAC,CACpH,CAAC;IACJ,CAAC;IAED,MAAM,UAAU,GAAmC,EAAE,CAAC;IACtD,KAAK,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC;QAC9D,UAAU,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,GAAG,KAAK,CAAC;IAC1C,CAAC;IACD,MAAM,CAAC,OAAO,GAAG,UAAU,CAAC;IAE5B,KAAK,MAAM,MAAM,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;QACpC,MAAM,WAAW,GAAkC,EAAE,CAAC;QACtD,KAAK,MAAM,CAAC,OAAO,EAAE,IAAI,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC9D,WAAW,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,GAAG,IAAI,CAAC;QAC1C,CAAC;QACD,MAAM,CAAC,QAAQ,GAAG,WAAW,CAAC;IAChC,CAAC;AACH,CAAC;AAMD,MAAM,UAAU,aAAa,CAAC,MAAyB,EAAE,SAAwB;IAC/E,IAAI,CAAC,SAAS;QAAE,OAAO,IAAI,CAAC;IAC5B,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC;QAAE,SAAS,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACtE,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,EAAE,GAAG,MAAM,CAAC,EAAE,OAAO,CAAC,CAAC;IACtD,aAAa,CAAC,QAAQ,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IACzD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,QAAkB;IAC9C,MAAM,CAAC,GAAG,IAAI,IAAI,EAAE,CAAC;IACrB,MAAM,GAAG,GAAG,CAAC,CAAS,EAAU,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;IAC9D,MAAM,IAAI,GAAG,GAAG,CAAC,CAAC,WAAW,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,EAAE,CAAC;IAC7E,MAAM,IAAI,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,UAAU,EAAE,CAAC,EAAE,CAAC;IAC1D,MAAM,WAAW,GAAG,QAAQ;SACzB,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC,UAAU,CAAC,mBAAmB,EAAE,GAAG,CAAC,CAAC;SACzF,IAAI,CAAC,MAAM,CAAC,CAAC;IAChB,OAAO,GAAG,WAAW,IAAI,IAAI,IAAI,IAAI,EAAE,CAAC;AAC1C,CAAC"}
|
|
@@ -6,9 +6,10 @@
|
|
|
6
6
|
* v0.21 makes the data trustworthy. closes the last mile: senior
|
|
7
7
|
* engineers' #1 complaint is "the report has all the right numbers but I
|
|
8
8
|
* still need 30 minutes to read it before I can decide". This module
|
|
9
|
-
* aggregates the
|
|
10
|
-
*
|
|
11
|
-
* available) Krippendorff α against gold — and emits one
|
|
9
|
+
* aggregates the data sources omk already produces — Bootstrap CI on the
|
|
10
|
+
* pairwise diff, three-layer CI gate, saturation curve, inter-judge ensemble
|
|
11
|
+
* agreement, and (when available) Krippendorff α against gold — and emits one
|
|
12
|
+
* of six verdicts:
|
|
12
13
|
*
|
|
13
14
|
* - **PROGRESS** diff CI shows real positive shift, no layer regressed
|
|
14
15
|
* - **CAUTIOUS** positive shift but at least one layer broke its gate,
|
|
@@ -27,6 +28,23 @@
|
|
|
27
28
|
* empirical) is documented inline so users can audit and override.
|
|
28
29
|
*/
|
|
29
30
|
import type { Report } from '../types/index.js';
|
|
31
|
+
/**
|
|
32
|
+
* Below this sample count a non-significant diff is read as UNDERPOWERED
|
|
33
|
+
* (only large effects are detectable) rather than NOISE. Matches the
|
|
34
|
+
* pre-flight power band documented in `docs/specs/sample-design-spec.md`;
|
|
35
|
+
* guarded by `test/scripts/doc-constants-drift.test.ts`.
|
|
36
|
+
*/
|
|
37
|
+
export declare const UNDERPOWERED_MIN_SAMPLES = 20;
|
|
38
|
+
/**
|
|
39
|
+
* Inter-judge Pearson bands, shared with the report's ensemble-agreement audit
|
|
40
|
+
* badge (`src/renderer/summary.ts`) so the verdict gate and the UI read one scale.
|
|
41
|
+
* - `>= ENSEMBLE_STRONG_PEARSON` (0.7): judges agree on rank order (badge green).
|
|
42
|
+
* - `< ENSEMBLE_DISSENT_PEARSON` (0.4): strong disagreement (badge red). A
|
|
43
|
+
* would-be PROGRESS is downgraded to CAUTIOUS — the judge-layer signal driving
|
|
44
|
+
* the win is unreliable when the ensemble can't agree.
|
|
45
|
+
*/
|
|
46
|
+
export declare const ENSEMBLE_STRONG_PEARSON = 0.7;
|
|
47
|
+
export declare const ENSEMBLE_DISSENT_PEARSON = 0.4;
|
|
30
48
|
export type VerdictLevel = 'PROGRESS' | 'CAUTIOUS' | 'REGRESS' | 'NOISE' | 'UNDERPOWERED' | 'SOLO';
|
|
31
49
|
export interface VerdictResult {
|
|
32
50
|
level: VerdictLevel;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"verdict.d.ts","sourceRoot":"","sources":["../../src/eval-core/verdict.ts"],"names":[],"mappings":"AAAA
|
|
1
|
+
{"version":3,"file":"verdict.d.ts","sourceRoot":"","sources":["../../src/eval-core/verdict.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AAEH,OAAO,KAAK,EAAE,MAAM,EAAyC,MAAM,mBAAmB,CAAC;AAGvF;;;;;GAKG;AACH,eAAO,MAAM,wBAAwB,KAAK,CAAC;AAE3C;;;;;;;GAOG;AACH,eAAO,MAAM,uBAAuB,MAAM,CAAC;AAC3C,eAAO,MAAM,wBAAwB,MAAM,CAAC;AAE5C,MAAM,MAAM,YAAY,GACpB,UAAU,GACV,UAAU,GACV,SAAS,GACT,OAAO,GACP,cAAc,GACd,MAAM,CAAC;AAEX,MAAM,WAAW,aAAa;IAC5B,KAAK,EAAE,YAAY,CAAC;IACpB,6CAA6C;IAC7C,QAAQ,EAAE,MAAM,CAAC;IACjB,yEAAyE;IACzE,OAAO,CAAC,EAAE,KAAK,CAAC;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,YAAY,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC/F,yDAAyD;IACzD,SAAS,EAAE;QACT,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,UAAU,CAAC,EAAE,MAAM,CAAC;QACpB;6CACqC;QACrC,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,cAAc,CAAC,EAAE,MAAM,CAAC;QACxB,kBAAkB,CAAC,EAAE,MAAM,CAAC;KAC7B,CAAC;IACF,gEAAgE;IAChE,QAAQ,EAAE,MAAM,EAAE,CAAC;CACpB;AAED,MAAM,WAAW,cAAc;IAC7B,2EAA2E;IAC3E,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB;;;;OAIG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAC;CAC7B;AAED;;GAEG;AACH,wBAAgB,cAAc,CAAC,MAAM,EAAE,MAAM,EAAE,OAAO,GAAE,cAAmB,GAAG,aAAa,CAwE1F;AA0PD;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,aAAa,EAAE,OAAO,GAAE;IAAE,OAAO,CAAC,EAAE,OAAO,CAAA;CAAO,GAAG,MAAM,CAiBpG"}
|
|
@@ -6,9 +6,10 @@
|
|
|
6
6
|
* v0.21 makes the data trustworthy. closes the last mile: senior
|
|
7
7
|
* engineers' #1 complaint is "the report has all the right numbers but I
|
|
8
8
|
* still need 30 minutes to read it before I can decide". This module
|
|
9
|
-
* aggregates the
|
|
10
|
-
*
|
|
11
|
-
* available) Krippendorff α against gold — and emits one
|
|
9
|
+
* aggregates the data sources omk already produces — Bootstrap CI on the
|
|
10
|
+
* pairwise diff, three-layer CI gate, saturation curve, inter-judge ensemble
|
|
11
|
+
* agreement, and (when available) Krippendorff α against gold — and emits one
|
|
12
|
+
* of six verdicts:
|
|
12
13
|
*
|
|
13
14
|
* - **PROGRESS** diff CI shows real positive shift, no layer regressed
|
|
14
15
|
* - **CAUTIOUS** positive shift but at least one layer broke its gate,
|
|
@@ -27,6 +28,23 @@
|
|
|
27
28
|
* empirical) is documented inline so users can audit and override.
|
|
28
29
|
*/
|
|
29
30
|
import { evaluateLayerGates } from './layer-gates.js';
|
|
31
|
+
/**
|
|
32
|
+
* Below this sample count a non-significant diff is read as UNDERPOWERED
|
|
33
|
+
* (only large effects are detectable) rather than NOISE. Matches the
|
|
34
|
+
* pre-flight power band documented in `docs/specs/sample-design-spec.md`;
|
|
35
|
+
* guarded by `test/scripts/doc-constants-drift.test.ts`.
|
|
36
|
+
*/
|
|
37
|
+
export const UNDERPOWERED_MIN_SAMPLES = 20;
|
|
38
|
+
/**
|
|
39
|
+
* Inter-judge Pearson bands, shared with the report's ensemble-agreement audit
|
|
40
|
+
* badge (`src/renderer/summary.ts`) so the verdict gate and the UI read one scale.
|
|
41
|
+
* - `>= ENSEMBLE_STRONG_PEARSON` (0.7): judges agree on rank order (badge green).
|
|
42
|
+
* - `< ENSEMBLE_DISSENT_PEARSON` (0.4): strong disagreement (badge red). A
|
|
43
|
+
* would-be PROGRESS is downgraded to CAUTIOUS — the judge-layer signal driving
|
|
44
|
+
* the win is unreliable when the ensemble can't agree.
|
|
45
|
+
*/
|
|
46
|
+
export const ENSEMBLE_STRONG_PEARSON = 0.7;
|
|
47
|
+
export const ENSEMBLE_DISSENT_PEARSON = 0.4;
|
|
30
48
|
/**
|
|
31
49
|
* Compute a verdict for a finished report. Pure function — no I/O.
|
|
32
50
|
*/
|
|
@@ -133,7 +151,7 @@ function verdictForPair(pair, summary, sampleCount, report, gateThreshold, trivi
|
|
|
133
151
|
// Diff CI contains 0. Distinguish "underpowered (saturation says: more samples needed)"
|
|
134
152
|
// from "noise (saturation says: we're saturated, the effect just isn't there)".
|
|
135
153
|
const satVerdict = report.variance?.saturation?.verdicts?.[treatment];
|
|
136
|
-
const underpowered = sampleCount <
|
|
154
|
+
const underpowered = sampleCount < UNDERPOWERED_MIN_SAMPLES ||
|
|
137
155
|
(satVerdict && !satVerdict.saturated && satVerdict.confidence !== 'high');
|
|
138
156
|
if (underpowered) {
|
|
139
157
|
return {
|
|
@@ -167,6 +185,19 @@ function verdictForPair(pair, summary, sampleCount, report, gateThreshold, trivi
|
|
|
167
185
|
headline: `${headlineCore} · significant but practically tiny`,
|
|
168
186
|
};
|
|
169
187
|
}
|
|
188
|
+
// Ensemble dissent gate: a real, gate-passing gain is only PROGRESS if the
|
|
189
|
+
// judges that produced it actually agree. When inter-judge Pearson sits in the
|
|
190
|
+
// badge's red band (< ENSEMBLE_DISSENT_PEARSON) for either compared variant,
|
|
191
|
+
// the judge-layer signal behind the win is unreliable → downgrade to CAUTIOUS.
|
|
192
|
+
const dissent = ensembleDissent(pair, summary);
|
|
193
|
+
if (dissent) {
|
|
194
|
+
return {
|
|
195
|
+
control,
|
|
196
|
+
treatment,
|
|
197
|
+
level: 'CAUTIOUS',
|
|
198
|
+
headline: `${headlineCore} · gain real, but judges disagree on ${dissent.variant} (Pearson ${dissent.pearson} < ${ENSEMBLE_DISSENT_PEARSON})`,
|
|
199
|
+
};
|
|
200
|
+
}
|
|
170
201
|
// Did control break a gate? Then treatment winning isn't surprising — flag.
|
|
171
202
|
if (!cGate.allPass) {
|
|
172
203
|
return {
|
|
@@ -178,6 +209,25 @@ function verdictForPair(pair, summary, sampleCount, report, gateThreshold, trivi
|
|
|
178
209
|
}
|
|
179
210
|
return { control, treatment, level: 'PROGRESS', headline: `${headlineCore} · clean win` };
|
|
180
211
|
}
|
|
212
|
+
/**
|
|
213
|
+
* Strong inter-judge disagreement on the compared pair, if any. Reuses the
|
|
214
|
+
* persisted per-variant ensemble Pearson (`summary[v].judgeAgreement.pearson`).
|
|
215
|
+
* Returns the worst offender below the dissent threshold, or null when every
|
|
216
|
+
* variant agrees / has no ensemble signal (single- or constant-score judge ⇒
|
|
217
|
+
* pearson undefined, nothing to act on — its disagreement, if any, still shows
|
|
218
|
+
* in the report's MAD column).
|
|
219
|
+
*/
|
|
220
|
+
function ensembleDissent(pair, summary) {
|
|
221
|
+
let worst = null;
|
|
222
|
+
for (const variant of [pair.treatment, pair.control]) {
|
|
223
|
+
const pearson = summary[variant]?.judgeAgreement?.pearson;
|
|
224
|
+
if (pearson != null && pearson < ENSEMBLE_DISSENT_PEARSON) {
|
|
225
|
+
if (!worst || pearson < worst.pearson)
|
|
226
|
+
worst = { variant, pearson };
|
|
227
|
+
}
|
|
228
|
+
}
|
|
229
|
+
return worst;
|
|
230
|
+
}
|
|
181
231
|
function avgComposite(s) {
|
|
182
232
|
if (!s)
|
|
183
233
|
return 0;
|
|
@@ -1 +1 @@
|
|
|
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
|
+
{"version":3,"file":"verdict.js","sourceRoot":"","sources":["../../src/eval-core/verdict.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AAGH,OAAO,EAAE,kBAAkB,EAAE,MAAM,kBAAkB,CAAC;AAEtD;;;;;GAKG;AACH,MAAM,CAAC,MAAM,wBAAwB,GAAG,EAAE,CAAC;AAE3C;;;;;;;GAOG;AACH,MAAM,CAAC,MAAM,uBAAuB,GAAG,GAAG,CAAC;AAC3C,MAAM,CAAC,MAAM,wBAAwB,GAAG,GAAG,CAAC;AA0C5C;;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,wBAAwB;YACtC,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,2EAA2E;IAC3E,+EAA+E;IAC/E,6EAA6E;IAC7E,+EAA+E;IAC/E,MAAM,OAAO,GAAG,eAAe,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IAC/C,IAAI,OAAO,EAAE,CAAC;QACZ,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,UAAU;YACjB,QAAQ,EAAE,GAAG,YAAY,wCAAwC,OAAO,CAAC,OAAO,aAAa,OAAO,CAAC,OAAO,MAAM,wBAAwB,GAAG;SAC9I,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;;;;;;;GAOG;AACH,SAAS,eAAe,CACtB,IAA4C,EAC5C,OAAuC;IAEvC,IAAI,KAAK,GAAgD,IAAI,CAAC;IAC9D,KAAK,MAAM,OAAO,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC;QACrD,MAAM,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,EAAE,cAAc,EAAE,OAAO,CAAC;QAC1D,IAAI,OAAO,IAAI,IAAI,IAAI,OAAO,GAAG,wBAAwB,EAAE,CAAC;YAC1D,IAAI,CAAC,KAAK,IAAI,OAAO,GAAG,KAAK,CAAC,OAAO;gBAAE,KAAK,GAAG,EAAE,OAAO,EAAE,OAAO,EAAE,CAAC;QACtE,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC;AACf,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 +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,EAAE,WAAW,EAAc,MAAM,EAAE,IAAI,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAEtG,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;IAClC;qDACiD;IACjD,cAAc,EAAE,MAAM,CAAC;IACvB,6DAA6D;IAC7D,kBAAkB,EAAE,MAAM,EAAE,CAAC;CAC9B;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,0DAA0D;IAC1D,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,oFAAoF;IACpF,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CACjD,GAAG,OAAO,CAAC,qBAAqB,CAAC,
|
|
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,EAAE,WAAW,EAAc,MAAM,EAAE,IAAI,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAEtG,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;IAClC;qDACiD;IACjD,cAAc,EAAE,MAAM,CAAC;IACvB,6DAA6D;IAC7D,kBAAkB,EAAE,MAAM,EAAE,CAAC;CAC9B;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,0DAA0D;IAC1D,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,oFAAoF;IACpF,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CACjD,GAAG,OAAO,CAAC,qBAAqB,CAAC,CAwEjC;AAED,wBAAgB,qBAAqB,CAAC,EACpC,KAAK,EACL,WAAW,EACX,YAAY,EACZ,WAAW,EACX,QAAQ,EACR,KAAK,EACL,YAAY,GACb,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,WAAW,EAAE,CAAC;IAC3B,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"}
|
|
@@ -9,17 +9,45 @@ export async function prepareEvaluationRun({ samplesPath, skillDir, variantSpecs
|
|
|
9
9
|
const { samples, requires, baseDir: samplesBaseDir, sourceFiles: samplesSourceFiles } = loadSamples(samplesPath);
|
|
10
10
|
// Build expressions from specs (preserving order) and resolve to artifacts.
|
|
11
11
|
const variantExpressions = variantSpecs.map((spec) => spec.expr);
|
|
12
|
-
|
|
13
|
-
//
|
|
14
|
-
|
|
15
|
-
|
|
16
|
-
|
|
17
|
-
|
|
18
|
-
|
|
19
|
-
|
|
20
|
-
|
|
21
|
-
|
|
22
|
-
|
|
12
|
+
// 不把 variantAllowedSkills 透进 resolveArtifacts:那里按解析出的 artifact 名查,
|
|
13
|
+
// 与 eval.yaml 的自定义 variant 名不一致时会漏绑。allowedSkills 统一在下面按 spec 绑定。
|
|
14
|
+
// resolveArtifacts 只保留 strictBaseline 默认(baseline → [])。
|
|
15
|
+
const resolvedArtifacts = artifacts || resolveArtifacts(resolve(skillDir), variantExpressions, { strictBaseline });
|
|
16
|
+
// Attach experimentRole to each artifact.
|
|
17
|
+
// - 当 artifacts 由本函数从 variantSpecs.map(spec => spec.expr) 解析时,两者顺序一一对应,
|
|
18
|
+
// 按 index 绑定 role,并把 spec.name 同步成消歧后的最终唯一名。不能按 spec.name 匹配
|
|
19
|
+
// artifact.name:同 basename 的 variant 被 ensureUniqueVariantNames 消歧后(v1/greeter /
|
|
20
|
+
// v2/greeter)就和 spec 的派生短名(greeter)对不上,会丢 role。
|
|
21
|
+
// - 当 artifacts 由外部传入(如 batch workflow,role 已预置)时,index 无法对齐,
|
|
22
|
+
// 退回按 name 匹配,且 if-already-set 守卫会保留预置 role。
|
|
23
|
+
if (!artifacts && variantSpecs.length === resolvedArtifacts.length) {
|
|
24
|
+
variantSpecs.forEach((spec, i) => {
|
|
25
|
+
const artifact = resolvedArtifacts[i];
|
|
26
|
+
if (!artifact.experimentRole)
|
|
27
|
+
artifact.experimentRole = spec.role;
|
|
28
|
+
// allowedSkills 按 spec 身份绑定(eval 主路径的唯一隔离绑定点,显式声明优先于
|
|
29
|
+
// strictBaseline 默认)。首选 spec.allowedSkills(eval.yaml 经 configVariantsToSpecs 带上);
|
|
30
|
+
// 旧的按变量名 map 仅作 legacy fallback,兜住「CLI roles 覆盖 config 时 spec 不带
|
|
31
|
+
// allowedSkills」等边角,行为与重构前一致。
|
|
32
|
+
// TODO: 待所有路径都按 spec 携带 allowedSkills 后,移除 variantAllowedSkills 这条 fallback,
|
|
33
|
+
// 与 resolveArtifacts 里的按名绑定一并收成单一来源。
|
|
34
|
+
const explicit = spec.allowedSkills ?? variantAllowedSkills?.[spec.name];
|
|
35
|
+
if (explicit !== undefined)
|
|
36
|
+
artifact.allowedSkills = explicit;
|
|
37
|
+
spec.name = artifact.name; // 同步唯一名,让 spec 与 report / variantNames 一致(放最后,前面还要用旧名查)
|
|
38
|
+
});
|
|
39
|
+
}
|
|
40
|
+
else {
|
|
41
|
+
const roleByName = {};
|
|
42
|
+
for (const spec of variantSpecs)
|
|
43
|
+
roleByName[spec.name] = spec.role;
|
|
44
|
+
for (const artifact of resolvedArtifacts) {
|
|
45
|
+
if (artifact.experimentRole)
|
|
46
|
+
continue; // preserve if already set (e.g. batch workflow)
|
|
47
|
+
const role = roleByName[artifact.name];
|
|
48
|
+
if (role)
|
|
49
|
+
artifact.experimentRole = role;
|
|
50
|
+
}
|
|
23
51
|
}
|
|
24
52
|
if (!dryRun) {
|
|
25
53
|
const mcpServers = loadMcpConfig(mcpConfig);
|
|
@@ -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;AAiBvD,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,OAAO,EAAE,cAAc,EAAE,WAAW,EAAE,kBAAkB,EAAE,GAAG,WAAW,CAAC,WAAW,CAAC,CAAC;IAEjH,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,
|
|
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;AAiBvD,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,OAAO,EAAE,cAAc,EAAE,WAAW,EAAE,kBAAkB,EAAE,GAAG,WAAW,CAAC,WAAW,CAAC,CAAC;IAEjH,4EAA4E;IAC5E,MAAM,kBAAkB,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACjE,mEAAmE;IACnE,kEAAkE;IAClE,yDAAyD;IACzD,MAAM,iBAAiB,GAAG,SAAS,IAAI,gBAAgB,CACrD,OAAO,CAAC,QAAQ,CAAC,EACjB,kBAAkB,EAClB,EAAE,cAAc,EAAE,CACnB,CAAC;IAEF,0CAA0C;IAC1C,0EAA0E;IAC1E,iEAAiE;IACjE,qFAAqF;IACrF,oDAAoD;IACpD,gEAAgE;IAChE,iDAAiD;IACjD,IAAI,CAAC,SAAS,IAAI,YAAY,CAAC,MAAM,KAAK,iBAAiB,CAAC,MAAM,EAAE,CAAC;QACnE,YAAY,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC,EAAE,EAAE;YAC/B,MAAM,QAAQ,GAAG,iBAAiB,CAAC,CAAC,CAAC,CAAC;YACtC,IAAI,CAAC,QAAQ,CAAC,cAAc;gBAAE,QAAQ,CAAC,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC;YAClE,qDAAqD;YACrD,kFAAkF;YAClF,iEAAiE;YACjE,8BAA8B;YAC9B,6EAA6E;YAC7E,2CAA2C;YAC3C,MAAM,QAAQ,GAAG,IAAI,CAAC,aAAa,IAAI,oBAAoB,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACzE,IAAI,QAAQ,KAAK,SAAS;gBAAE,QAAQ,CAAC,aAAa,GAAG,QAAQ,CAAC;YAC9D,IAAI,CAAC,IAAI,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAC,wDAAwD;QACrF,CAAC,CAAC,CAAC;IACL,CAAC;SAAM,CAAC;QACN,MAAM,UAAU,GAAwC,EAAE,CAAC;QAC3D,KAAK,MAAM,IAAI,IAAI,YAAY;YAAE,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC;QACnE,KAAK,MAAM,QAAQ,IAAI,iBAAiB,EAAE,CAAC;YACzC,IAAI,QAAQ,CAAC,cAAc;gBAAE,SAAS,CAAE,gDAAgD;YACxF,MAAM,IAAI,GAAG,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;YACvC,IAAI,IAAI;gBAAE,QAAQ,CAAC,cAAc,GAAG,IAAI,CAAC;QAC3C,CAAC;IACH,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;QACR,cAAc;QACd,kBAAkB;KACnB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,EACpC,KAAK,EACL,WAAW,EACX,YAAY,EACZ,WAAW,EACX,QAAQ,EACR,KAAK,EACL,YAAY,GASb;IACC,OAAO;QACL,MAAM,EAAE,IAAa;QACrB,KAAK;QACL,WAAW;QACX,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"}
|
|
@@ -137,7 +137,7 @@ export interface DryRunBatchReport extends DryRunBase {
|
|
|
137
137
|
totalArtifacts: number;
|
|
138
138
|
artifacts: DryRunBatchSkill[];
|
|
139
139
|
}
|
|
140
|
-
export declare function buildVarianceData(runs: Report[]): VarianceData | null;
|
|
140
|
+
export declare function buildVarianceData(runs: Report[], bootstrapSamples?: number, seed?: number): VarianceData | null;
|
|
141
141
|
export declare function runBatchEvaluation({ skillDir, model, outputDir, project, owner, tags, noJudge, dryRun, concurrency, timeoutMs, executorName, jobStore, persistJob, onProgress, onSkillProgress, skipConnectivity, skipDoctor, lang, mcpConfig, verbose, repeat, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, noCache, strictBaseline, variantAllowedSkills, }: RunBatchEvaluationOptions): Promise<{
|
|
142
142
|
report: BatchEvaluationReport | DryRunBatchReport;
|
|
143
143
|
filePath: string | null;
|