oh-my-knowledge 0.18.0 → 0.20.0
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 +596 -326
- package/README.zh.md +917 -0
- package/dist/src/analysis/failure-clusterer.d.ts +96 -0
- package/dist/src/analysis/failure-clusterer.d.ts.map +1 -0
- package/dist/src/analysis/failure-clusterer.js +298 -0
- package/dist/src/analysis/failure-clusterer.js.map +1 -0
- package/dist/src/analysis/sample-diagnostics.d.ts +78 -0
- package/dist/src/analysis/sample-diagnostics.d.ts.map +1 -0
- package/dist/src/analysis/sample-diagnostics.js +259 -0
- package/dist/src/analysis/sample-diagnostics.js.map +1 -0
- package/dist/src/analysis/saturation.d.ts +85 -0
- package/dist/src/analysis/saturation.d.ts.map +1 -0
- package/dist/src/analysis/saturation.js +174 -0
- package/dist/src/analysis/saturation.js.map +1 -0
- package/dist/src/cli.js +772 -18
- package/dist/src/cli.js.map +1 -1
- package/dist/src/eval-core/bootstrap.d.ts +72 -0
- package/dist/src/eval-core/bootstrap.d.ts.map +1 -0
- package/dist/src/eval-core/bootstrap.js +174 -0
- package/dist/src/eval-core/bootstrap.js.map +1 -0
- package/dist/src/eval-core/evaluation-execution.d.ts +15 -1
- package/dist/src/eval-core/evaluation-execution.d.ts.map +1 -1
- package/dist/src/eval-core/evaluation-execution.js +37 -3
- package/dist/src/eval-core/evaluation-execution.js.map +1 -1
- package/dist/src/eval-core/evaluation-job.d.ts +10 -2
- package/dist/src/eval-core/evaluation-job.d.ts.map +1 -1
- package/dist/src/eval-core/evaluation-job.js +9 -1
- package/dist/src/eval-core/evaluation-job.js.map +1 -1
- package/dist/src/eval-core/evaluation-reporting.d.ts.map +1 -1
- package/dist/src/eval-core/evaluation-reporting.js +90 -0
- package/dist/src/eval-core/evaluation-reporting.js.map +1 -1
- package/dist/src/eval-core/schema.d.ts.map +1 -1
- package/dist/src/eval-core/schema.js +69 -0
- package/dist/src/eval-core/schema.js.map +1 -1
- package/dist/src/eval-core/verdict.d.ts +74 -0
- package/dist/src/eval-core/verdict.d.ts.map +1 -0
- package/dist/src/eval-core/verdict.js +283 -0
- package/dist/src/eval-core/verdict.js.map +1 -0
- package/dist/src/eval-workflows/each-evaluation-workflow.d.ts +16 -3
- package/dist/src/eval-workflows/each-evaluation-workflow.d.ts.map +1 -1
- package/dist/src/eval-workflows/each-evaluation-workflow.js +10 -2
- package/dist/src/eval-workflows/each-evaluation-workflow.js.map +1 -1
- package/dist/src/eval-workflows/evaluation-pipeline.d.ts +17 -1
- package/dist/src/eval-workflows/evaluation-pipeline.d.ts.map +1 -1
- package/dist/src/eval-workflows/evaluation-pipeline.js +45 -3
- package/dist/src/eval-workflows/evaluation-pipeline.js.map +1 -1
- package/dist/src/eval-workflows/run-evaluation.d.ts +23 -2
- package/dist/src/eval-workflows/run-evaluation.d.ts.map +1 -1
- package/dist/src/eval-workflows/run-evaluation.js +104 -4
- package/dist/src/eval-workflows/run-evaluation.js.map +1 -1
- package/dist/src/grading/assertions.d.ts +16 -0
- package/dist/src/grading/assertions.d.ts.map +1 -1
- package/dist/src/grading/assertions.js +385 -111
- package/dist/src/grading/assertions.js.map +1 -1
- package/dist/src/grading/debias-validate.d.ts +84 -0
- package/dist/src/grading/debias-validate.d.ts.map +1 -0
- package/dist/src/grading/debias-validate.js +173 -0
- package/dist/src/grading/debias-validate.js.map +1 -0
- package/dist/src/grading/gold-cli.d.ts +88 -0
- package/dist/src/grading/gold-cli.d.ts.map +1 -0
- package/dist/src/grading/gold-cli.js +251 -0
- package/dist/src/grading/gold-cli.js.map +1 -0
- package/dist/src/grading/gold-dataset.d.ts +73 -0
- package/dist/src/grading/gold-dataset.d.ts.map +1 -0
- package/dist/src/grading/gold-dataset.js +161 -0
- package/dist/src/grading/gold-dataset.js.map +1 -0
- package/dist/src/grading/human-gold.d.ts +102 -0
- package/dist/src/grading/human-gold.d.ts.map +1 -0
- package/dist/src/grading/human-gold.js +188 -0
- package/dist/src/grading/human-gold.js.map +1 -0
- package/dist/src/grading/index.d.ts +27 -2
- package/dist/src/grading/index.d.ts.map +1 -1
- package/dist/src/grading/index.js +36 -18
- package/dist/src/grading/index.js.map +1 -1
- package/dist/src/grading/judge.d.ts +65 -2
- package/dist/src/grading/judge.d.ts.map +1 -1
- package/dist/src/grading/judge.js +280 -23
- package/dist/src/grading/judge.js.map +1 -1
- package/dist/src/inputs/eval-config.js +19 -0
- package/dist/src/inputs/eval-config.js.map +1 -1
- package/dist/src/observability/{production-analyzer.d.ts → skill-health-analyzer.d.ts} +24 -2
- package/dist/src/observability/skill-health-analyzer.d.ts.map +1 -0
- package/dist/src/observability/{production-analyzer.js → skill-health-analyzer.js} +61 -6
- package/dist/src/observability/skill-health-analyzer.js.map +1 -0
- package/dist/src/observability/trace-adapter.d.ts.map +1 -1
- package/dist/src/observability/trace-adapter.js +27 -1
- package/dist/src/observability/trace-adapter.js.map +1 -1
- package/dist/src/renderer/html-renderer.d.ts.map +1 -1
- package/dist/src/renderer/html-renderer.js +40 -6
- package/dist/src/renderer/html-renderer.js.map +1 -1
- package/dist/src/renderer/layout.d.ts.map +1 -1
- package/dist/src/renderer/layout.js +138 -4
- package/dist/src/renderer/layout.js.map +1 -1
- package/dist/src/renderer/skill-health-renderer.d.ts +2 -2
- package/dist/src/renderer/skill-health-renderer.d.ts.map +1 -1
- package/dist/src/renderer/skill-health-renderer.js +39 -4
- package/dist/src/renderer/skill-health-renderer.js.map +1 -1
- package/dist/src/renderer/summary.d.ts +28 -1
- package/dist/src/renderer/summary.d.ts.map +1 -1
- package/dist/src/renderer/summary.js +322 -8
- package/dist/src/renderer/summary.js.map +1 -1
- package/dist/src/renderer/table.d.ts.map +1 -1
- package/dist/src/renderer/table.js +63 -2
- package/dist/src/renderer/table.js.map +1 -1
- package/dist/src/server/report-server.d.ts +2 -1
- package/dist/src/server/report-server.d.ts.map +1 -1
- package/dist/src/server/report-server.js +397 -2
- package/dist/src/server/report-server.js.map +1 -1
- package/dist/src/types.d.ts +247 -0
- package/dist/src/types.d.ts.map +1 -1
- package/package.json +24 -6
- package/dist/src/observability/production-analyzer.d.ts.map +0 -1
- package/dist/src/observability/production-analyzer.js.map +0 -1
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"evaluation-pipeline.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACnD,OAAO,EAAE,cAAc,EAAE,MAAM,mCAAmC,CAAC;AACnE,OAAO,EAAE,qBAAqB,EAAE,MAAM,kCAAkC,CAAC;AACzE,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,eAAe,EAAE,cAAc,EAAE,kBAAkB,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACzI,OAAO,EAAE,YAAY,EAAE,SAAS,EAAE,MAAM,sCAAsC,CAAC;AAC/E,OAAO,EAAE,qBAAqB,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AAEnG,OAAO,EACL,kBAAkB,EAClB,gBAAgB,GACjB,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC3D,OAAO,EACL,sBAAsB,EACtB,eAAe,EACf,mBAAmB,EACnB,eAAe,EACf,kBAAkB,EAClB,qBAAqB,EACrB,cAAc,EACd,iBAAiB,GAClB,MAAM,gCAAgC,CAAC;AA4BxC,KAAK,UAAU,4BAA4B,CAAC,EAC1C,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,WAAW,EACX,SAAS,EACT,OAAO,EACP,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAK,EACL,QAAQ,EACR,UAAU,
|
|
1
|
+
{"version":3,"file":"evaluation-pipeline.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACnD,OAAO,EAAE,cAAc,EAAE,MAAM,mCAAmC,CAAC;AACnE,OAAO,EAAE,qBAAqB,EAAE,MAAM,kCAAkC,CAAC;AACzE,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,eAAe,EAAE,cAAc,EAAE,kBAAkB,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACzI,OAAO,EAAE,YAAY,EAAE,SAAS,EAAE,MAAM,sCAAsC,CAAC;AAC/E,OAAO,EAAE,qBAAqB,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AAEnG,OAAO,EACL,kBAAkB,EAClB,gBAAgB,GACjB,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC3D,OAAO,EACL,sBAAsB,EACtB,eAAe,EACf,mBAAmB,EACnB,eAAe,EACf,kBAAkB,EAClB,qBAAqB,EACrB,cAAc,EACd,iBAAiB,GAClB,MAAM,gCAAgC,CAAC;AA4BxC,KAAK,UAAU,4BAA4B,CAAC,EAC1C,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,WAAW,EACX,SAAS,EACT,OAAO,EACP,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAK,EACL,QAAQ,EACR,UAAU,EACV,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,GA4BP;IACC,MAAM,OAAO,GAAG,sBAAsB,CAAC;QACrC,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;QACvC,QAAQ,EAAE,YAAY;QACtB,aAAa,EAAE,iBAAiB;QAChC,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO;QACP,MAAM,EAAE,KAAK;QACb,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM;QACN,IAAI;QACJ,WAAW;QACX,WAAW;QACX,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;KACP,CAAC,CAAC;IACH,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,MAAM,EAAE,GAAG,EAAE,UAAU,EAAE,SAAS,EAAE,GAAG,mBAAmB,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC7E,MAAM,KAAK,GAAG,OAAO,KAAK,EAAE,CAAC;IAC7B,MAAM,gBAAgB,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,QAAQ,IAAI,kBAAkB,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAChG,MAAM,SAAS,GAAG,eAAe,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC,CAAC;IACjE,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IACpE,MAAM,UAAU,GAAG,cAAc,CAAC,SAAS,EAAE,KAAK,EAAE,SAAS,CAAC,CAAC;IAC/D,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,KAAK,EAAE,UAAU,CAAC,CAAC;IACrE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,UAAU,EAAE,UAAU,EAAE,gBAAgB,EAAE,CAAC;AACnG,CAAC;AAED,SAAS,qBAAqB,CAAC,KAAyB;IACtD,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,GAAG,GAAG,qBAAqB,CAAC,KAAK,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IAChE,MAAM,GAAG,GAAG,kBAAkB,CAAC;QAC7B,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,QAAQ,EAAE,KAAK,CAAC,KAAK;QACrB,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,UAAU;KACX,CAAC,CAAC;IACH,OAAO,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC;AACtB,CAAC;AAED,KAAK,UAAU,oBAAoB,CAAC,KAAyB,EAAE,GAAkB;IAC/E,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC3B,MAAM,KAAK,CAAC,gBAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IACtD,CAAC;AACH,CAAC;AAED,KAAK,UAAU,gBAAgB,CAAC,KAAyB,EAAE,GAAY;IACrE,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,SAAS,GAAG,eAAe,CAAC;QAChC,GAAG,EAAE,EAAE,GAAG,KAAK,CAAC,UAAU,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,UAAU,EAAE,SAAS,EAAE;QACnG,KAAK,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;QACvD,UAAU;KACX,CAAC,CAAC;IACH,KAAK,iBAAiB,CAAC,KAAK,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IACrD,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC3B,MAAM,KAAK,CAAC,gBAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC5D,CAAC;AACH,CAAC;AAED;;;;GAIG;AACH,SAAS,kBAAkB,CAAC,WAAmB;IAC7C,IAAI,CAAC,WAAW,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC;QAAE,OAAO,IAAI,CAAC;IAC1D,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;QACnD,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IACzE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,wBAAwB,CAAC,EAChC,MAAM,EACN,OAAO,EACP,SAAS,EACT,YAAY,EACZ,KAAK,EACL,WAAW,GAQZ;IACC,MAAM,CAAC,QAAQ,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC;IAEzC,MAAM,WAAW,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,aAAa,EAAE,EAAE,CAAC,CACjE,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,aAAa,EAAE,EAAE,CAAC,aAAa,CAAC,SAAS,IAAI,aAAa,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CACpH,CAAC,CAAC;IACH,IAAI,WAAW,EAAE,CAAC;QAChB,MAAM,gBAAgB,GAAG,MAAM,CAAC,WAAW,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC5G,MAAM,YAAY,GAAG,MAAM,CAAC,WAAW,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,GAAG,IAAI,IAAI,CAAC,CAAC,CAAC,CAAC;QAC5G,MAAM,QAAQ,GAAG,qBAAqB,CAAC,MAAM,EAAE,gBAAgB,EAAE,YAAY,CAAC,CAAC;QAC/E,IAAI,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACrC,MAAM,CAAC,QAAS,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACvC,CAAC;IACH,CAAC;IAED,6EAA6E;IAC7E,2EAA2E;IAC3E,gFAAgF;IAChF,MAAM,UAAU,GAAG,qBAAqB,CAAC,MAAM,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;IACvE,IAAI,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACvC,MAAM,WAAW,GAAG,kBAAkB,CAAC,WAAW,CAAC,CAAC;QACpD,KAAK,MAAM,OAAO,IAAI,YAAY,EAAE,CAAC;YACnC,MAAM,EAAE,GAAG,UAAU,CAAC,OAAO,CAAC,CAAC;YAC/B,IAAI,CAAC,EAAE;gBAAE,SAAS;YAClB,EAAE,CAAC,WAAW,GAAG,WAAW,CAAC;YAC7B,EAAE,CAAC,WAAW,GAAG,WAAW,CAAC;QAC/B,CAAC;QACD,MAAM,CAAC,QAAS,CAAC,UAAU,GAAG,UAAU,CAAC;IAC3C,CAAC;IAED,IAAI,KAAK,EAAE,CAAC;QACV,cAAc,CAAC,MAAM,EAAE,YAAY,EAAE,GAAG,YAAY,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,WAAW,EAAE,CAAC,CAAC;IACnF,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAkDD,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,EAC9C,WAAW,EACX,QAAQ,EACR,OAAO,EACP,KAAK,EACL,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,QAAQ,EACR,aAAa,EACb,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAK,GAAG,KAAK,EACb,WAAW,GAAG,CAAC,EACf,SAAS,EACT,OAAO,GAAG,KAAK,EACf,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,aAAa,GAAG,KAAK,EACrB,OAAO,GAAG,KAAK,EACf,KAAK,GAAG,CAAC,EACT,eAAe,EACf,QAAQ,EACR,YAAY,GAAG,KAAK,EACpB,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,GAAG,IAAI,EACnB,MAAM,GACoB;IAC1B,MAAM,YAAY,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IAChE,MAAM,QAAQ,GAAG,MAAM,4BAA4B,CAAC;QAClD,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,UAAU;QACV,OAAO;QACP,YAAY;QACZ,iBAAiB;QACjB,WAAW;QACX,SAAS;QACT,OAAO;QACP,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,KAAK,EAAE,aAAa,CAAC,YAAY,CAAC;QAClC,QAAQ;QACR,UAAU;QACV,MAAM;QACN,IAAI;QACJ,WAAW;QACX,WAAW;QACX,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;KACP,CAAC,CAAC;IAEH,IAAI,CAAC;QACH,IAAI,CAAC,aAAa,EAAE,CAAC;YACnB,IAAI,UAAU;gBAAE,UAAU,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC;YAC1E,MAAM,SAAS,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC;YACjC,IAAI,CAAC,OAAO;gBAAE,MAAM,SAAS,CAAC,aAAa,EAAE,UAAU,CAAC,CAAC;YAEzD,+EAA+E;YAC/E,MAAM,aAAa,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAC;YACxG,MAAM,GAAG,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,GAAG,IAAI,QAAQ,IAAI,OAAO,CAAC,GAAG,EAAE,CAAC;YAC3E,MAAM,SAAS,GAAG,MAAM,qBAAqB,CAAC,aAAa,EAAE,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,SAAS,CAAC,CAAC;YAChG,IAAI,CAAC,SAAS,CAAC,EAAE,EAAE,CAAC;gBAClB,MAAM,IAAI,KAAK,CAAC,sBAAsB,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC;YAC7D,CAAC;QACH,CAAC;QAED,8EAA8E;QAC9E,iFAAiF;QACjF,iFAAiF;QACjF,IAAI,cAAsD,CAAC;QAC3D,IAAI,WAAW,IAAI,WAAW,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;YAC3C,MAAM,EAAE,cAAc,EAAE,GAAG,MAAM,MAAM,CAAC,uBAAuB,CAAC,CAAC;YACjE,cAAc,GAAG,EAAE,CAAC;YACpB,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;YAC/B,KAAK,MAAM,EAAE,IAAI,WAAW,EAAE,CAAC;gBAC7B,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC;oBAC3B,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;oBACtB,cAAc,CAAC,EAAE,CAAC,QAAQ,CAAC,GAAG,cAAc,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;gBAC5D,CAAC;YACH,CAAC;QACH,CAAC;QAED,MAAM,EAAE,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,eAAe,EAAE,GAAG,MAAM,YAAY,CAAC;YAC7E,KAAK;YACL,QAAQ;YACR,aAAa;YACb,KAAK;YACL,UAAU;YACV,OAAO;YACP,WAAW;YACX,WAAW;YACX,SAAS;YACT,OAAO;YACP,OAAO;YACP,UAAU;YACV,KAAK;YACL,eAAe;YACf,WAAW;YACX,WAAW;YACX,cAAc;YACd,YAAY;YACZ,MAAM;SACP,CAAC,CAAC;QACH,IAAI,OAAO,GAAG,CAAC,IAAI,UAAU,EAAE,CAAC;YAC9B,UAAU,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,CAAC,MAAM,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,SAAS,EAAE,EAAE,EAAE,OAAO,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;QACzH,CAAC;QAED,MAAM,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,qBAAqB,CAAC,QAAQ,CAAC,CAAC;QACrD,MAAM,MAAM,GAAG,wBAAwB,CAAC;YACtC,MAAM,EAAE,eAAe,CAAC;gBACtB,KAAK,EAAE,QAAQ,CAAC,KAAK;gBACrB,QAAQ,EAAE,YAAY;gBACtB,KAAK;gBACL,UAAU;gBACV,OAAO;gBACP,YAAY;gBACZ,OAAO;gBACP,KAAK;gBACL,OAAO;gBACP,YAAY;gBACZ,SAAS;gBACT,OAAO,EAAE,QAAQ,CAAC,OAAO;gBACzB,GAAG;gBACH,GAAG;gBACH,YAAY;aACb,CAAC;YACF,OAAO;YACP,SAAS;YACT,YAAY;YACZ,KAAK;YACL,WAAW;SACZ,CAAC,CAAC;QACH,IAAI,eAAe,EAAE,CAAC;YACpB,MAAM,CAAC,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC;QACrC,CAAC;QACD,IAAI,MAAM,EAAE,CAAC;YACX,MAAM,CAAC,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QAC9B,CAAC;QACD,MAAM,QAAQ,GAAG,aAAa,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;QAClD,MAAM,oBAAoB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QAC1C,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC;IAC9B,CAAC;IAAC,OAAO,GAAY,EAAE,CAAC;QACtB,MAAM,gBAAgB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QACtC,MAAM,GAAG,CAAC;IACZ,CAAC;YAAS,CAAC;QACT,MAAM,cAAc,EAAE,CAAC;IACzB,CAAC;AACH,CAAC"}
|
|
@@ -24,6 +24,27 @@ interface CommonEvaluationOptions {
|
|
|
24
24
|
mcpConfig?: string;
|
|
25
25
|
verbose?: boolean;
|
|
26
26
|
layeredStats?: boolean;
|
|
27
|
+
/** --repeat N. 1 表示单次(默认); > 1 时在 runMultiple 层聚合 variance。
|
|
28
|
+
* 记入 report.meta.request.repeat 让 meta 如实反映用户输入。 */
|
|
29
|
+
repeat?: number;
|
|
30
|
+
/** --each 模式标记, true 表示当前评测是 each 批量流程(每个 skill 独立对比 baseline)。
|
|
31
|
+
* 记入 report.meta.request.each。 */
|
|
32
|
+
each?: boolean;
|
|
33
|
+
/** --judge-repeat N. 每条 sample × dimension 调 LLM judge N 次, 输出 stddev (judge 自一致性).
|
|
34
|
+
* 默认 1 (单次). 用于量化 LLM judge 在该 rubric 上的稳定性 — stddev 高 = 评分噪声大. */
|
|
35
|
+
judgeRepeat?: number;
|
|
36
|
+
/** --judge-models executor:model,executor:model,... — multi-judge ensemble. ≥ 2 个
|
|
37
|
+
* judge 时每条 sample × dimension 由所有 judge 各自打分, 输出 inter-judge agreement. */
|
|
38
|
+
judgeModels?: import('../types.js').JudgeConfig[];
|
|
39
|
+
/** --bootstrap. Distribution-free CI on each variant mean + pairwise diff. */
|
|
40
|
+
bootstrap?: boolean;
|
|
41
|
+
/** --bootstrap-samples N. Default 1000. */
|
|
42
|
+
bootstrapSamples?: number;
|
|
43
|
+
/** v0.21 Phase 3a length-debias toggle. Default true (judge prompt v3-cot-length).
|
|
44
|
+
* CLI passes false when --no-debias-length is set. */
|
|
45
|
+
lengthDebias?: boolean;
|
|
46
|
+
/** v0.22 — hard budget caps. */
|
|
47
|
+
budget?: import('../types.js').EvalBudget;
|
|
27
48
|
}
|
|
28
49
|
export interface RunEvaluationOptions extends CommonEvaluationOptions {
|
|
29
50
|
samplesPath: string;
|
|
@@ -76,7 +97,7 @@ export interface DryRunReport extends DryRunBase {
|
|
|
76
97
|
samplesPath: string;
|
|
77
98
|
tasks: DryRunTask[];
|
|
78
99
|
}
|
|
79
|
-
export declare function runEvaluation({ samplesPath, skillDir, variantSpecs, artifacts, model, judgeModel, outputDir, project, owner, tags, noJudge, dryRun, blind, concurrency, timeoutMs, noCache, executorName, judgeExecutorName, jobStore, persistJob, onProgress, skipPreflight, mcpConfig, verbose, retry, resume, layeredStats, }: RunEvaluationOptions): Promise<{
|
|
100
|
+
export declare function runEvaluation({ samplesPath, skillDir, variantSpecs, artifacts, model, judgeModel, outputDir, project, owner, tags, noJudge, dryRun, blind, concurrency, timeoutMs, noCache, executorName, judgeExecutorName, jobStore, persistJob, onProgress, skipPreflight, mcpConfig, verbose, retry, resume, layeredStats, repeat, each, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, }: RunEvaluationOptions): Promise<{
|
|
80
101
|
report: Report | DryRunReport;
|
|
81
102
|
filePath: string | null;
|
|
82
103
|
}>;
|
|
@@ -92,7 +113,7 @@ export interface DryRunEachReport extends DryRunBase {
|
|
|
92
113
|
artifacts: DryRunEachSkill[];
|
|
93
114
|
}
|
|
94
115
|
export declare function buildVarianceData(runs: Report[]): VarianceData | null;
|
|
95
|
-
export declare function runEachEvaluation({ skillDir, model, judgeModel, outputDir, project, owner, tags, noJudge, dryRun, concurrency, timeoutMs, executorName, judgeExecutorName, jobStore, persistJob, onProgress, onSkillProgress, skipPreflight, mcpConfig, verbose, }: RunEachEvaluationOptions): Promise<{
|
|
116
|
+
export declare function runEachEvaluation({ skillDir, model, judgeModel, outputDir, project, owner, tags, noJudge, dryRun, concurrency, timeoutMs, executorName, judgeExecutorName, jobStore, persistJob, onProgress, onSkillProgress, skipPreflight, mcpConfig, verbose, repeat, judgeRepeat, judgeModels, lengthDebias, }: RunEachEvaluationOptions): Promise<{
|
|
96
117
|
report: Report | DryRunEachReport;
|
|
97
118
|
filePath: string | null;
|
|
98
119
|
}>;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"run-evaluation.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/run-evaluation.ts"],"names":[],"mappings":"AAaA,OAAO,KAAK,EACV,QAAQ,EAER,QAAQ,EACR,gBAAgB,EAChB,MAAM,
|
|
1
|
+
{"version":3,"file":"run-evaluation.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/run-evaluation.ts"],"names":[],"mappings":"AAaA,OAAO,KAAK,EACV,QAAQ,EAER,QAAQ,EACR,gBAAgB,EAChB,MAAM,EAIN,YAAY,EAGZ,WAAW,EAGZ,MAAM,aAAa,CAAC;AAIrB,MAAM,WAAW,iBAAiB;IAChC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,UAAU,uBAAuB;IAC/B,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACrC,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAIlB,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB;yDACqD;IACrD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB;uCACmC;IACnC,IAAI,CAAC,EAAE,OAAO,CAAC;IACf;wEACoE;IACpE,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;iFAC6E;IAC7E,WAAW,CAAC,EAAE,OAAO,aAAa,EAAE,WAAW,EAAE,CAAC;IAClD,8EAA8E;IAC9E,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,2CAA2C;IAC3C,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B;2DACuD;IACvD,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,gCAAgC;IAChC,MAAM,CAAC,EAAE,OAAO,aAAa,EAAE,UAAU,CAAC;CAC3C;AAED,MAAM,WAAW,oBAAqB,SAAQ,uBAAuB;IACnE,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,CAAC,EAAE,WAAW,EAAE,CAAC;IAC7B,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,wBAAyB,SAAQ,uBAAuB;IACvE,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,eAAe,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE,iBAAiB,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;CAC9D;AAED,MAAM,WAAW,kBAAmB,SAAQ,oBAAoB;IAC9D,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,gBAAgB,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;CAC5E;AAED,UAAU,UAAU;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;IAC/B,cAAc,EAAE,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACnC,iBAAiB,EAAE,MAAM,CAAC;IAC1B,cAAc,EAAE,MAAM,CAAC;IACvB,cAAc,EAAE,MAAM,CAAC;IACvB,GAAG,EAAE,MAAM,GAAG,IAAI,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,OAAO,CAAC;IACnB,aAAa,EAAE,OAAO,CAAC;IACvB,aAAa,EAAE,OAAO,CAAC;IACvB,SAAS,EAAE,OAAO,CAAC;CACpB;AAED,UAAU,UAAU;IAClB,MAAM,EAAE,IAAI,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,YAAa,SAAQ,UAAU;IAC9C,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,UAAU,EAAE,CAAC;CACrB;AAED,wBAAsB,aAAa,CAAC,EAClC,WAAW,EACX,QAAQ,EACR,YAAiB,EACjB,SAAS,EACT,KAAqB,EACrB,UAAwB,EACxB,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,MAAc,EACd,KAAa,EACb,WAAe,EACf,SAAS,EACT,OAAe,EACf,YAAuB,EACvB,iBAAiB,EACjB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,aAAqB,EACrB,SAAS,EACT,OAAe,EACf,KAAS,EACT,MAAM,EACN,YAAoB,EACpB,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,GACP,EAAE,oBAAoB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,GAAG,YAAY,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAsF5F;AAED,UAAU,eAAe;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,gBAAiB,SAAQ,UAAU;IAClD,IAAI,EAAE,IAAI,CAAC;IACX,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,eAAe,EAAE,CAAC;CAC9B;AAyHD,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,YAAY,GAAG,IAAI,CA6ErE;AAED,wBAAsB,iBAAiB,CAAC,EACtC,QAAQ,EACR,KAAqB,EACrB,UAAwB,EACxB,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,MAAc,EACd,WAAe,EACf,SAAS,EACT,YAAuB,EACvB,iBAAiB,EACjB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,eAAsB,EACtB,aAAqB,EACrB,SAAS,EACT,OAAe,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,YAAY,GACb,EAAE,wBAAwB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,GAAG,gBAAgB,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CA0DpG;AAED,wBAAsB,WAAW,CAAC,EAAE,MAAU,EAAE,gBAAgB,EAAE,GAAG,MAAM,EAAE,EAAE,kBAAkB;;;;GAyBhG"}
|
|
@@ -6,7 +6,9 @@ import { confidenceInterval, tTest, effectSize } from '../eval-core/statistics.j
|
|
|
6
6
|
import { executeEachEvaluationRuns } from './each-evaluation-workflow.js';
|
|
7
7
|
import { buildDryRunEachArtifacts, buildDryRunTaskReport, prepareEvaluationRun, } from './evaluation-preparation.js';
|
|
8
8
|
import { executeEvaluationPipeline } from './evaluation-pipeline.js';
|
|
9
|
-
|
|
9
|
+
import { findSaturationPoint } from '../analysis/saturation.js';
|
|
10
|
+
import { bootstrapMeanCI } from '../eval-core/bootstrap.js';
|
|
11
|
+
export async function runEvaluation({ samplesPath, skillDir, variantSpecs = [], artifacts, model = DEFAULT_MODEL, judgeModel = JUDGE_MODEL, outputDir = DEFAULT_OUTPUT_DIR, project, owner, tags, noJudge = false, dryRun = false, blind = false, concurrency = 1, timeoutMs, noCache = false, executorName = 'claude', judgeExecutorName, jobStore = null, persistJob = true, onProgress = null, skipPreflight = false, mcpConfig, verbose = false, retry = 0, resume, layeredStats = false, repeat, each, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, }) {
|
|
10
12
|
const { samples, artifacts: resolvedArtifacts, tasks, variantNames, requires } = await prepareEvaluationRun({
|
|
11
13
|
samplesPath,
|
|
12
14
|
skillDir,
|
|
@@ -81,6 +83,14 @@ export async function runEvaluation({ samplesPath, skillDir, variantSpecs = [],
|
|
|
81
83
|
existingResults,
|
|
82
84
|
requires,
|
|
83
85
|
layeredStats,
|
|
86
|
+
repeat,
|
|
87
|
+
each,
|
|
88
|
+
judgeRepeat,
|
|
89
|
+
judgeModels,
|
|
90
|
+
bootstrap,
|
|
91
|
+
bootstrapSamples,
|
|
92
|
+
lengthDebias,
|
|
93
|
+
budget,
|
|
84
94
|
});
|
|
85
95
|
}
|
|
86
96
|
// Top-level (composite) extractor: feeds the legacy flat-field variance on
|
|
@@ -122,6 +132,81 @@ function buildComparisonMetric(scoresA, scoresB, meanA, meanB) {
|
|
|
122
132
|
effectSize: es,
|
|
123
133
|
};
|
|
124
134
|
}
|
|
135
|
+
/**
|
|
136
|
+
* Build saturation curve data from a sequence of runs.
|
|
137
|
+
*
|
|
138
|
+
* Each run contributes one checkpoint = "all samples up to and including this run".
|
|
139
|
+
* For each variant we compute (mean, CI) at that cumulative N. When repeat ≥ 5
|
|
140
|
+
* we additionally run findSaturationPoint to assess whether the curve has
|
|
141
|
+
* flattened. Below that threshold the data is recorded for plotting but no
|
|
142
|
+
* verdict is computed — the user still sees the curve, just without the auto
|
|
143
|
+
* "saturated at N=X" claim.
|
|
144
|
+
*/
|
|
145
|
+
function buildSaturationData(runs) {
|
|
146
|
+
if (runs.length < 2)
|
|
147
|
+
return undefined;
|
|
148
|
+
const variants = runs[0].meta.variants ?? [];
|
|
149
|
+
if (variants.length === 0)
|
|
150
|
+
return undefined;
|
|
151
|
+
// Per-variant: cumulative composite scores after each repeat.
|
|
152
|
+
const cumulativeByVariant = {};
|
|
153
|
+
const tracesByVariant = {};
|
|
154
|
+
const checkpointSampleCounts = [];
|
|
155
|
+
const acc = Object.fromEntries(variants.map((v) => [v, []]));
|
|
156
|
+
for (let runIdx = 0; runIdx < runs.length; runIdx++) {
|
|
157
|
+
const run = runs[runIdx];
|
|
158
|
+
for (const variant of variants) {
|
|
159
|
+
const newScores = [];
|
|
160
|
+
for (const entry of run.results ?? []) {
|
|
161
|
+
const v = entry.variants?.[variant];
|
|
162
|
+
if (!v || typeof v.compositeScore !== 'number' || v.compositeScore <= 0)
|
|
163
|
+
continue;
|
|
164
|
+
newScores.push(v.compositeScore);
|
|
165
|
+
}
|
|
166
|
+
acc[variant] = acc[variant].concat(newScores);
|
|
167
|
+
}
|
|
168
|
+
// Snapshot cumulative state for this checkpoint.
|
|
169
|
+
const checkpointN = acc[variants[0]]?.length ?? 0;
|
|
170
|
+
checkpointSampleCounts.push(checkpointN);
|
|
171
|
+
for (const variant of variants) {
|
|
172
|
+
if (!cumulativeByVariant[variant])
|
|
173
|
+
cumulativeByVariant[variant] = [];
|
|
174
|
+
cumulativeByVariant[variant].push([...acc[variant]]);
|
|
175
|
+
// Per-checkpoint trace: bootstrap CI on cumulative scores.
|
|
176
|
+
const ci = bootstrapMeanCI(acc[variant], 0.05, 1000);
|
|
177
|
+
if (!tracesByVariant[variant])
|
|
178
|
+
tracesByVariant[variant] = [];
|
|
179
|
+
tracesByVariant[variant].push({
|
|
180
|
+
n: acc[variant].length,
|
|
181
|
+
mean: ci.estimate,
|
|
182
|
+
ciLow: ci.low,
|
|
183
|
+
ciHigh: ci.high,
|
|
184
|
+
});
|
|
185
|
+
}
|
|
186
|
+
}
|
|
187
|
+
const verdicts = {};
|
|
188
|
+
if (runs.length >= 5) {
|
|
189
|
+
for (const variant of variants) {
|
|
190
|
+
const cumulative = cumulativeByVariant[variant];
|
|
191
|
+
if (!cumulative)
|
|
192
|
+
continue;
|
|
193
|
+
const r = findSaturationPoint(cumulative, 'bootstrap-ci-width');
|
|
194
|
+
verdicts[variant] = {
|
|
195
|
+
saturated: r.saturated,
|
|
196
|
+
atN: r.atN,
|
|
197
|
+
confidence: r.confidence,
|
|
198
|
+
method: r.method,
|
|
199
|
+
threshold: r.threshold,
|
|
200
|
+
reason: r.reason,
|
|
201
|
+
};
|
|
202
|
+
}
|
|
203
|
+
}
|
|
204
|
+
return {
|
|
205
|
+
checkpointSampleCounts,
|
|
206
|
+
perVariant: tracesByVariant,
|
|
207
|
+
...(Object.keys(verdicts).length > 0 ? { verdicts } : {}),
|
|
208
|
+
};
|
|
209
|
+
}
|
|
125
210
|
export function buildVarianceData(runs) {
|
|
126
211
|
if (runs.length <= 1) {
|
|
127
212
|
return null;
|
|
@@ -187,9 +272,15 @@ export function buildVarianceData(runs) {
|
|
|
187
272
|
});
|
|
188
273
|
}
|
|
189
274
|
}
|
|
190
|
-
|
|
275
|
+
const saturation = buildSaturationData(runs);
|
|
276
|
+
return {
|
|
277
|
+
runs: runs.length,
|
|
278
|
+
perVariant,
|
|
279
|
+
comparisons,
|
|
280
|
+
...(saturation ? { saturation } : {}),
|
|
281
|
+
};
|
|
191
282
|
}
|
|
192
|
-
export async function runEachEvaluation({ skillDir, model = DEFAULT_MODEL, judgeModel = JUDGE_MODEL, outputDir = DEFAULT_OUTPUT_DIR, project, owner, tags, noJudge = false, dryRun = false, concurrency = 1, timeoutMs, executorName = 'claude', judgeExecutorName, jobStore = null, persistJob = true, onProgress = null, onSkillProgress = null, skipPreflight = false, mcpConfig, verbose = false, }) {
|
|
283
|
+
export async function runEachEvaluation({ skillDir, model = DEFAULT_MODEL, judgeModel = JUDGE_MODEL, outputDir = DEFAULT_OUTPUT_DIR, project, owner, tags, noJudge = false, dryRun = false, concurrency = 1, timeoutMs, executorName = 'claude', judgeExecutorName, jobStore = null, persistJob = true, onProgress = null, onSkillProgress = null, skipPreflight = false, mcpConfig, verbose = false, repeat, judgeRepeat, judgeModels, lengthDebias, }) {
|
|
193
284
|
const skillEntries = discoverEachSkills(resolve(skillDir));
|
|
194
285
|
if (skillEntries.length === 0) {
|
|
195
286
|
throw new Error(`未发现带配对 eval-samples 的 skill:${skillDir}`);
|
|
@@ -232,8 +323,17 @@ export async function runEachEvaluation({ skillDir, model = DEFAULT_MODEL, judge
|
|
|
232
323
|
skipPreflight,
|
|
233
324
|
mcpConfig,
|
|
234
325
|
verbose,
|
|
326
|
+
repeat,
|
|
327
|
+
judgeRepeat,
|
|
328
|
+
judgeModels,
|
|
329
|
+
lengthDebias,
|
|
235
330
|
runSingleEvaluation: async (options) => {
|
|
236
|
-
|
|
331
|
+
// repeat > 1 时走 runMultiple 做 variance; each=true 标记让 meta.request 如实反映
|
|
332
|
+
if (repeat && repeat > 1) {
|
|
333
|
+
const multi = await runMultiple({ ...options, repeat, each: true, judgeRepeat, judgeModels, lengthDebias });
|
|
334
|
+
return { report: multi.report, filePath: multi.filePath };
|
|
335
|
+
}
|
|
336
|
+
const result = await runEvaluation({ ...options, each: true, judgeRepeat, judgeModels, lengthDebias });
|
|
237
337
|
return { report: result.report, filePath: result.filePath };
|
|
238
338
|
},
|
|
239
339
|
});
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"run-evaluation.js","sourceRoot":"","sources":["../../../src/eval-workflows/run-evaluation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,kBAAkB,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACzF,OAAO,EAAE,cAAc,EAAE,aAAa,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AACnF,OAAO,EAAE,kBAAkB,EAAE,MAAM,2BAA2B,CAAC;AAC/D,OAAO,EAAE,kBAAkB,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,4BAA4B,CAAC;AACnF,OAAO,EAAE,yBAAyB,EAAE,MAAM,+BAA+B,CAAC;AAC1E,OAAO,EACL,wBAAwB,EACxB,qBAAqB,EACrB,oBAAoB,GACrB,MAAM,6BAA6B,CAAC;AACrC,OAAO,EAAE,yBAAyB,EAAE,MAAM,0BAA0B,CAAC;AAuGrE,MAAM,CAAC,KAAK,UAAU,aAAa,CAAC,EAClC,WAAW,EACX,QAAQ,EACR,YAAY,GAAG,EAAE,EACjB,SAAS,EACT,KAAK,GAAG,aAAa,EACrB,UAAU,GAAG,WAAW,EACxB,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAO,GAAG,KAAK,EACf,MAAM,GAAG,KAAK,EACd,KAAK,GAAG,KAAK,EACb,WAAW,GAAG,CAAC,EACf,SAAS,EACT,OAAO,GAAG,KAAK,EACf,YAAY,GAAG,QAAQ,EACvB,iBAAiB,EACjB,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,aAAa,GAAG,KAAK,EACrB,SAAS,EACT,OAAO,GAAG,KAAK,EACf,KAAK,GAAG,CAAC,EACT,MAAM,EACN,YAAY,GAAG,KAAK,GACC;IACrB,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,iBAAiB,EAAE,KAAK,EAAE,YAAY,EAAE,QAAQ,EAAE,GAAG,MAAM,oBAAoB,CAAC;QAC1G,WAAW;QACX,QAAQ;QACR,YAAY;QACZ,SAAS;QACT,MAAM;QACN,SAAS;KACV,CAAC,CAAC;IAEH,IAAI,MAAM,EAAE,CAAC;QACX,OAAO;YACL,MAAM,EAAE,qBAAqB,CAAC;gBAC5B,KAAK;gBACL,UAAU;gBACV,YAAY;gBACZ,WAAW;gBACX,QAAQ;gBACR,KAAK;gBACL,YAAY;aACb,CAAC;YACF,QAAQ,EAAE,IAAI;SACf,CAAC;IACJ,CAAC;IAED,iEAAiE;IACjE,IAAI,eAAgG,CAAC;IACrG,IAAI,MAAM,EAAE,CAAC;QACX,MAAM,EAAE,eAAe,EAAE,GAAG,MAAM,MAAM,CAAC,2BAA2B,CAAC,CAAC;QACtE,MAAM,KAAK,GAAG,eAAe,CAAC,OAAO,CAAC,SAAS,IAAI,kBAAkB,CAAC,CAAC,CAAC;QACxE,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QACzC,IAAI,QAAQ,EAAE,CAAC;YACb,eAAe,GAAG,EAAE,CAAC;YACrB,KAAK,MAAM,KAAK,IAAI,QAAQ,CAAC,OAAO,IAAI,EAAE,EAAE,CAAC;gBAC3C,eAAe,CAAC,KAAK,CAAC,SAAS,CAAC,GAAG,KAAK,CAAC,QAAQ,CAAC;YACpD,CAAC;YACD,IAAI,UAAU,EAAE,CAAC;gBACf,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;gBACtH,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,YAAY,MAAM,QAAQ,KAAK,WAAW,CAAC,CAAC;YACnE,CAAC;QACH,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,YAAY,MAAM,cAAc,CAAC,CAAC;QACzD,CAAC;IACH,CAAC;IAED,MAAM,QAAQ,GAAe,cAAc,CAAC,YAAY,CAAC,CAAC;IAC1D,MAAM,aAAa,GAAe,cAAc,CAAC,iBAAiB,IAAI,YAAY,CAAC,CAAC;IACpF,OAAO,yBAAyB,CAAC;QAC/B,WAAW;QACX,QAAQ;QACR,OAAO;QACP,KAAK;QACL,SAAS,EAAE,iBAAiB;QAC5B,KAAK;QACL,UAAU;QACV,OAAO;QACP,YAAY;QACZ,iBAAiB,EAAE,iBAAiB,IAAI,YAAY;QACpD,QAAQ;QACR,aAAa;QACb,SAAS;QACT,OAAO;QACP,KAAK;QACL,IAAI;QACJ,KAAK;QACL,WAAW;QACX,SAAS;QACT,OAAO;QACP,QAAQ;QACR,UAAU;QACV,UAAU;QACV,aAAa;QACb,OAAO;QACP,KAAK;QACL,eAAe;QACf,QAAQ;QACR,YAAY;KACb,CAAC,CAAC;AACL,CAAC;AAeD,2EAA2E;AAC3E,0EAA0E;AAC1E,8EAA8E;AAC9E,wEAAwE;AACxE,MAAM,mBAAmB,GAAG,CAAC,CAA6B,EAAsB,EAAE,CAAC,CAAC,EAAE,iBAAiB,CAAC;AAExG,yEAAyE;AACzE,MAAM,iBAAiB,GAAyF;IAC9G,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,gBAAgB;IAChC,UAAU,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,aAAa;CACpC,CAAC;AAEF,4EAA4E;AAC5E,8EAA8E;AAC9E,kFAAkF;AAClF,iEAAiE;AACjE,MAAM,gBAAgB,GAAoF;IACxG,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,YAAY;IAC5B,QAAQ,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,gBAAgB;IACpC,KAAK,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,aAAa;CAC/B,CAAC;AAEF,SAAS,gBAAgB,CAAC,IAAc,EAAE,OAAe,EAAE,SAAgE;IACzH,MAAM,MAAM,GAAG,IAAI;SAChB,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;SAC/C,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAC;IACrD,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IACrC,OAAO,EAAE,MAAM,EAAE,GAAG,kBAAkB,CAAC,MAAM,CAAC,EAAE,CAAC;AACnD,CAAC;AAED,SAAS,qBAAqB,CAAC,OAAiB,EAAE,OAAiB,EAAE,KAAa,EAAE,KAAa;IAC/F,MAAM,CAAC,GAAG,KAAK,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;IAClC,MAAM,EAAE,GAAG,UAAU,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;IACxC,MAAM,QAAQ,GAAG,MAAM,CAAC,CAAC,KAAK,GAAG,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;IACpD,OAAO;QACL,QAAQ;QACR,UAAU,EAAE,CAAC,CAAC,UAAU;QACxB,EAAE,EAAE,CAAC,CAAC,EAAE;QACR,WAAW,EAAE,CAAC,CAAC,WAAW;QAC1B,UAAU,EAAE,EAAE;KACf,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,IAAc;IAC9C,IAAI,IAAI,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;QACrB,OAAO,IAAI,CAAC;IACd,CAAC;IAED,MAAM,QAAQ,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,IAAI,EAAE,CAAC;IAC7C,MAAM,UAAU,GAAoC,EAAE,CAAC;IACvD,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,6CAA6C;QAC7C,MAAM,SAAS,GAAG,gBAAgB,CAAC,IAAI,EAAE,OAAO,EAAE,mBAAmB,CAAC,CAAC;QACvE,IAAI,CAAC,SAAS;YAAE,SAAS;QAEzB,MAAM,QAAQ,GAA2D,EAAE,CAAC;QAC5E,MAAM,IAAI,GAAG,gBAAgB,CAAC,IAAI,EAAE,OAAO,EAAE,iBAAiB,CAAC,IAAI,CAAC,CAAC;QACrE,IAAI,IAAI;YAAE,QAAQ,CAAC,IAAI,GAAG,IAAI,CAAC;QAC/B,MAAM,UAAU,GAAG,gBAAgB,CAAC,IAAI,EAAE,OAAO,EAAE,iBAAiB,CAAC,UAAU,CAAC,CAAC;QACjF,IAAI,UAAU;YAAE,QAAQ,CAAC,UAAU,GAAG,UAAU,CAAC;QAEjD,uFAAuF;QACvF,MAAM,OAAO,GAAsD,EAAE,CAAC;QACtE,KAAK,MAAM,GAAG,IAAI,CAAC,MAAM,EAAE,UAAU,EAAE,OAAO,CAAU,EAAE,CAAC;YACzD,MAAM,UAAU,GAAG,gBAAgB,CAAC,IAAI,EAAE,OAAO,EAAE,gBAAgB,CAAC,GAAG,CAAC,CAAC,CAAC;YAC1E,IAAI,UAAU;gBAAE,OAAO,CAAC,GAAG,CAAC,GAAG,UAAU,CAAC;QAC5C,CAAC;QAED,UAAU,CAAC,OAAO,CAAC,GAAG;YACpB,GAAG,SAAS;YACZ,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACzD,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SACxD,CAAC;IACJ,CAAC;IAED,MAAM,WAAW,GAAyB,EAAE,CAAC;IAC7C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACzC,KAAK,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YAC7C,MAAM,EAAE,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;YACnC,MAAM,EAAE,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;YACnC,IAAI,CAAC,EAAE,IAAI,CAAC,EAAE;gBAAE,SAAS;YAEzB,MAAM,aAAa,GAAG,qBAAqB,CAAC,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC;YAEpF,MAAM,YAAY,GAAqE,EAAE,CAAC;YAC1F,KAAK,MAAM,GAAG,IAAI,CAAC,MAAM,EAAE,YAAY,CAAU,EAAE,CAAC;gBAClD,MAAM,EAAE,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,GAAG,CAAC,CAAC;gBAC9B,MAAM,EAAE,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,GAAG,CAAC,CAAC;gBAC9B,IAAI,EAAE,IAAI,EAAE,EAAE,CAAC;oBACb,YAAY,CAAC,GAAG,CAAC,GAAG,qBAAqB,CAAC,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC;gBACpF,CAAC;YACH,CAAC;YAED,qFAAqF;YACrF,MAAM,WAAW,GAAgE,EAAE,CAAC;YACpF,KAAK,MAAM,GAAG,IAAI,CAAC,MAAM,EAAE,UAAU,EAAE,OAAO,CAAU,EAAE,CAAC;gBACzD,MAAM,EAAE,GAAG,EAAE,CAAC,OAAO,EAAE,CAAC,GAAG,CAAC,CAAC;gBAC7B,MAAM,EAAE,GAAG,EAAE,CAAC,OAAO,EAAE,CAAC,GAAG,CAAC,CAAC;gBAC7B,IAAI,EAAE,IAAI,EAAE,EAAE,CAAC;oBACb,WAAW,CAAC,GAAG,CAAC,GAAG,qBAAqB,CAAC,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC;gBACnF,CAAC;YACH,CAAC;YAED,WAAW,CAAC,IAAI,CAAC;gBACf,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC;gBACd,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC;gBACd,GAAG,aAAa;gBAChB,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC3E,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aACzE,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,OAAO,EAAE,IAAI,EAAE,IAAI,CAAC,MAAM,EAAE,UAAU,EAAE,WAAW,EAAE,CAAC;AACxD,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,iBAAiB,CAAC,EACtC,QAAQ,EACR,KAAK,GAAG,aAAa,EACrB,UAAU,GAAG,WAAW,EACxB,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAO,GAAG,KAAK,EACf,MAAM,GAAG,KAAK,EACd,WAAW,GAAG,CAAC,EACf,SAAS,EACT,YAAY,GAAG,QAAQ,EACvB,iBAAiB,EACjB,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,eAAe,GAAG,IAAI,EACtB,aAAa,GAAG,KAAK,EACrB,SAAS,EACT,OAAO,GAAG,KAAK,GACU;IACzB,MAAM,YAAY,GAAG,kBAAkB,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC;IAC3D,IAAI,YAAY,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC9B,MAAM,IAAI,KAAK,CAAC,+BAA+B,QAAQ,EAAE,CAAC,CAAC;IAC7D,CAAC;IAED,IAAI,MAAM,EAAE,CAAC;QACX,MAAM,EAAE,SAAS,EAAE,YAAY,EAAE,UAAU,EAAE,GAAG,wBAAwB,CAAC,YAAY,CAAC,CAAC;QACvF,OAAO;YACL,MAAM,EAAE;gBACN,MAAM,EAAE,IAAI;gBACZ,IAAI,EAAE,IAAI;gBACV,KAAK;gBACL,UAAU;gBACV,QAAQ,EAAE,YAAY;gBACtB,QAAQ;gBACR,cAAc,EAAE,YAAY,CAAC,MAAM;gBACnC,UAAU;gBACV,SAAS,EAAE,YAAY;aACxB;YACD,QAAQ,EAAE,IAAI;SACf,CAAC;IACJ,CAAC;IACD,OAAO,yBAAyB,CAAC;QAC/B,QAAQ;QACR,YAAY;QACZ,KAAK;QACL,UAAU;QACV,SAAS;QACT,OAAO;QACP,KAAK;QACL,IAAI;QACJ,OAAO;QACP,WAAW;QACX,SAAS;QACT,YAAY;QACZ,iBAAiB;QACjB,QAAQ;QACR,UAAU;QACV,UAAU;QACV,eAAe;QACf,aAAa;QACb,SAAS;QACT,OAAO;QACP,mBAAmB,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE;YACrC,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,OAAO,CAAC,CAAC;YAC5C,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,MAAgB,EAAE,QAAQ,EAAE,MAAM,CAAC,QAAQ,EAAE,CAAC;QACxE,CAAC;KACF,CAAC,CAAC;AACL,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,WAAW,CAAC,EAAE,MAAM,GAAG,CAAC,EAAE,gBAAgB,EAAE,GAAG,MAAM,EAAsB;IAC/F,MAAM,IAAI,GAAa,EAAE,CAAC;IAC1B,MAAM,cAAc,GAAG,MAAM,CAAC,SAAS,CAAC;IAExC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAChC,gBAAgB,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAClD,MAAM,MAAM,GAAG,CAAC,KAAK,MAAM,GAAG,CAAC,CAAC;QAChC,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM,aAAa,CAAC;YACrC,GAAG,MAAM;YACT,SAAS,EAAE,MAAM,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,CAAC,IAAI;YACzC,UAAU,EAAE,MAAM;SACnB,CAAC,CAAC;QACH,IAAI,CAAC,IAAI,CAAC,MAAgB,CAAC,CAAC;IAC9B,CAAC;IAED,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACrC,MAAM,UAAU,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC;IAC3C,IAAI,QAAQ,GAAkB,IAAI,CAAC;IACnC,IAAI,UAAU,EAAE,CAAC;QACf,MAAM,CAAC,QAAQ,GAAG,UAAU,CAAC;QAC7B,gDAAgD;QAChD,QAAQ,GAAG,aAAa,CAAC,MAAM,EAAE,cAAc,IAAI,kBAAkB,CAAC,CAAC;IACzE,CAAC;IAED,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;AAC1C,CAAC"}
|
|
1
|
+
{"version":3,"file":"run-evaluation.js","sourceRoot":"","sources":["../../../src/eval-workflows/run-evaluation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,kBAAkB,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACzF,OAAO,EAAE,cAAc,EAAE,aAAa,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AACnF,OAAO,EAAE,kBAAkB,EAAE,MAAM,2BAA2B,CAAC;AAC/D,OAAO,EAAE,kBAAkB,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,4BAA4B,CAAC;AACnF,OAAO,EAAE,yBAAyB,EAAE,MAAM,+BAA+B,CAAC;AAC1E,OAAO,EACL,wBAAwB,EACxB,qBAAqB,EACrB,oBAAoB,GACrB,MAAM,6BAA6B,CAAC;AACrC,OAAO,EAAE,yBAAyB,EAAE,MAAM,0BAA0B,CAAC;AAkBrE,OAAO,EAAE,mBAAmB,EAAE,MAAM,2BAA2B,CAAC;AAChE,OAAO,EAAE,eAAe,EAAE,MAAM,2BAA2B,CAAC;AA4G5D,MAAM,CAAC,KAAK,UAAU,aAAa,CAAC,EAClC,WAAW,EACX,QAAQ,EACR,YAAY,GAAG,EAAE,EACjB,SAAS,EACT,KAAK,GAAG,aAAa,EACrB,UAAU,GAAG,WAAW,EACxB,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAO,GAAG,KAAK,EACf,MAAM,GAAG,KAAK,EACd,KAAK,GAAG,KAAK,EACb,WAAW,GAAG,CAAC,EACf,SAAS,EACT,OAAO,GAAG,KAAK,EACf,YAAY,GAAG,QAAQ,EACvB,iBAAiB,EACjB,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,aAAa,GAAG,KAAK,EACrB,SAAS,EACT,OAAO,GAAG,KAAK,EACf,KAAK,GAAG,CAAC,EACT,MAAM,EACN,YAAY,GAAG,KAAK,EACpB,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,GACe;IACrB,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,iBAAiB,EAAE,KAAK,EAAE,YAAY,EAAE,QAAQ,EAAE,GAAG,MAAM,oBAAoB,CAAC;QAC1G,WAAW;QACX,QAAQ;QACR,YAAY;QACZ,SAAS;QACT,MAAM;QACN,SAAS;KACV,CAAC,CAAC;IAEH,IAAI,MAAM,EAAE,CAAC;QACX,OAAO;YACL,MAAM,EAAE,qBAAqB,CAAC;gBAC5B,KAAK;gBACL,UAAU;gBACV,YAAY;gBACZ,WAAW;gBACX,QAAQ;gBACR,KAAK;gBACL,YAAY;aACb,CAAC;YACF,QAAQ,EAAE,IAAI;SACf,CAAC;IACJ,CAAC;IAED,iEAAiE;IACjE,IAAI,eAAgG,CAAC;IACrG,IAAI,MAAM,EAAE,CAAC;QACX,MAAM,EAAE,eAAe,EAAE,GAAG,MAAM,MAAM,CAAC,2BAA2B,CAAC,CAAC;QACtE,MAAM,KAAK,GAAG,eAAe,CAAC,OAAO,CAAC,SAAS,IAAI,kBAAkB,CAAC,CAAC,CAAC;QACxE,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QACzC,IAAI,QAAQ,EAAE,CAAC;YACb,eAAe,GAAG,EAAE,CAAC;YACrB,KAAK,MAAM,KAAK,IAAI,QAAQ,CAAC,OAAO,IAAI,EAAE,EAAE,CAAC;gBAC3C,eAAe,CAAC,KAAK,CAAC,SAAS,CAAC,GAAG,KAAK,CAAC,QAAQ,CAAC;YACpD,CAAC;YACD,IAAI,UAAU,EAAE,CAAC;gBACf,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;gBACtH,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,YAAY,MAAM,QAAQ,KAAK,WAAW,CAAC,CAAC;YACnE,CAAC;QACH,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,YAAY,MAAM,cAAc,CAAC,CAAC;QACzD,CAAC;IACH,CAAC;IAED,MAAM,QAAQ,GAAe,cAAc,CAAC,YAAY,CAAC,CAAC;IAC1D,MAAM,aAAa,GAAe,cAAc,CAAC,iBAAiB,IAAI,YAAY,CAAC,CAAC;IACpF,OAAO,yBAAyB,CAAC;QAC/B,WAAW;QACX,QAAQ;QACR,OAAO;QACP,KAAK;QACL,SAAS,EAAE,iBAAiB;QAC5B,KAAK;QACL,UAAU;QACV,OAAO;QACP,YAAY;QACZ,iBAAiB,EAAE,iBAAiB,IAAI,YAAY;QACpD,QAAQ;QACR,aAAa;QACb,SAAS;QACT,OAAO;QACP,KAAK;QACL,IAAI;QACJ,KAAK;QACL,WAAW;QACX,SAAS;QACT,OAAO;QACP,QAAQ;QACR,UAAU;QACV,UAAU;QACV,aAAa;QACb,OAAO;QACP,KAAK;QACL,eAAe;QACf,QAAQ;QACR,YAAY;QACZ,MAAM;QACN,IAAI;QACJ,WAAW;QACX,WAAW;QACX,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;KACP,CAAC,CAAC;AACL,CAAC;AAeD,2EAA2E;AAC3E,0EAA0E;AAC1E,8EAA8E;AAC9E,wEAAwE;AACxE,MAAM,mBAAmB,GAAG,CAAC,CAA6B,EAAsB,EAAE,CAAC,CAAC,EAAE,iBAAiB,CAAC;AAExG,yEAAyE;AACzE,MAAM,iBAAiB,GAAyF;IAC9G,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,gBAAgB;IAChC,UAAU,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,aAAa;CACpC,CAAC;AAEF,4EAA4E;AAC5E,8EAA8E;AAC9E,kFAAkF;AAClF,iEAAiE;AACjE,MAAM,gBAAgB,GAAoF;IACxG,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,YAAY;IAC5B,QAAQ,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,gBAAgB;IACpC,KAAK,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,aAAa;CAC/B,CAAC;AAEF,SAAS,gBAAgB,CAAC,IAAc,EAAE,OAAe,EAAE,SAAgE;IACzH,MAAM,MAAM,GAAG,IAAI;SAChB,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;SAC/C,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAC;IACrD,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IACrC,OAAO,EAAE,MAAM,EAAE,GAAG,kBAAkB,CAAC,MAAM,CAAC,EAAE,CAAC;AACnD,CAAC;AAED,SAAS,qBAAqB,CAAC,OAAiB,EAAE,OAAiB,EAAE,KAAa,EAAE,KAAa;IAC/F,MAAM,CAAC,GAAG,KAAK,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;IAClC,MAAM,EAAE,GAAG,UAAU,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;IACxC,MAAM,QAAQ,GAAG,MAAM,CAAC,CAAC,KAAK,GAAG,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;IACpD,OAAO;QACL,QAAQ;QACR,UAAU,EAAE,CAAC,CAAC,UAAU;QACxB,EAAE,EAAE,CAAC,CAAC,EAAE;QACR,WAAW,EAAE,CAAC,CAAC,WAAW;QAC1B,UAAU,EAAE,EAAE;KACf,CAAC;AACJ,CAAC;AAED;;;;;;;;;GASG;AACH,SAAS,mBAAmB,CAAC,IAAc;IACzC,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,SAAS,CAAC;IACtC,MAAM,QAAQ,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,IAAI,EAAE,CAAC;IAC7C,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAE5C,8DAA8D;IAC9D,MAAM,mBAAmB,GAA+B,EAAE,CAAC;IAC3D,MAAM,eAAe,GAAsF,EAAE,CAAC;IAE9G,MAAM,sBAAsB,GAAa,EAAE,CAAC;IAC5C,MAAM,GAAG,GAA6B,MAAM,CAAC,WAAW,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;IAEvF,KAAK,IAAI,MAAM,GAAG,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,CAAC;QACpD,MAAM,GAAG,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC;QACzB,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;YAC/B,MAAM,SAAS,GAAa,EAAE,CAAC;YAC/B,KAAK,MAAM,KAAK,IAAI,GAAG,CAAC,OAAO,IAAI,EAAE,EAAE,CAAC;gBACtC,MAAM,CAAC,GAAG,KAAK,CAAC,QAAQ,EAAE,CAAC,OAAO,CAAC,CAAC;gBACpC,IAAI,CAAC,CAAC,IAAI,OAAO,CAAC,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,CAAC,cAAc,IAAI,CAAC;oBAAE,SAAS;gBAClF,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC;YACnC,CAAC;YACD,GAAG,CAAC,OAAO,CAAC,GAAG,GAAG,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;QAChD,CAAC;QACD,iDAAiD;QACjD,MAAM,WAAW,GAAG,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE,MAAM,IAAI,CAAC,CAAC;QAClD,sBAAsB,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QACzC,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;YAC/B,IAAI,CAAC,mBAAmB,CAAC,OAAO,CAAC;gBAAE,mBAAmB,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC;YACrE,mBAAmB,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YAErD,2DAA2D;YAC3D,MAAM,EAAE,GAAG,eAAe,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;YACrD,IAAI,CAAC,eAAe,CAAC,OAAO,CAAC;gBAAE,eAAe,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC;YAC7D,eAAe,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC;gBAC5B,CAAC,EAAE,GAAG,CAAC,OAAO,CAAC,CAAC,MAAM;gBACtB,IAAI,EAAE,EAAE,CAAC,QAAQ;gBACjB,KAAK,EAAE,EAAE,CAAC,GAAG;gBACb,MAAM,EAAE,EAAE,CAAC,IAAI;aAChB,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,MAAM,QAAQ,GAA+B,EAAE,CAAC;IAChD,IAAI,IAAI,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;QACrB,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;YAC/B,MAAM,UAAU,GAAG,mBAAmB,CAAC,OAAO,CAAC,CAAC;YAChD,IAAI,CAAC,UAAU;gBAAE,SAAS;YAC1B,MAAM,CAAC,GAAG,mBAAmB,CAAC,UAAU,EAAE,oBAAoB,CAAC,CAAC;YAChE,QAAQ,CAAC,OAAO,CAAC,GAAG;gBAClB,SAAS,EAAE,CAAC,CAAC,SAAS;gBACtB,GAAG,EAAE,CAAC,CAAC,GAAG;gBACV,UAAU,EAAE,CAAC,CAAC,UAAU;gBACxB,MAAM,EAAE,CAAC,CAAC,MAAM;gBAChB,SAAS,EAAE,CAAC,CAAC,SAAS;gBACtB,MAAM,EAAE,CAAC,CAAC,MAAM;aACjB,CAAC;QACJ,CAAC;IACH,CAAC;IAED,OAAO;QACL,sBAAsB;QACtB,UAAU,EAAE,eAAe;QAC3B,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC1D,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,IAAc;IAC9C,IAAI,IAAI,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;QACrB,OAAO,IAAI,CAAC;IACd,CAAC;IAED,MAAM,QAAQ,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,IAAI,EAAE,CAAC;IAC7C,MAAM,UAAU,GAAoC,EAAE,CAAC;IACvD,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,6CAA6C;QAC7C,MAAM,SAAS,GAAG,gBAAgB,CAAC,IAAI,EAAE,OAAO,EAAE,mBAAmB,CAAC,CAAC;QACvE,IAAI,CAAC,SAAS;YAAE,SAAS;QAEzB,MAAM,QAAQ,GAA2D,EAAE,CAAC;QAC5E,MAAM,IAAI,GAAG,gBAAgB,CAAC,IAAI,EAAE,OAAO,EAAE,iBAAiB,CAAC,IAAI,CAAC,CAAC;QACrE,IAAI,IAAI;YAAE,QAAQ,CAAC,IAAI,GAAG,IAAI,CAAC;QAC/B,MAAM,UAAU,GAAG,gBAAgB,CAAC,IAAI,EAAE,OAAO,EAAE,iBAAiB,CAAC,UAAU,CAAC,CAAC;QACjF,IAAI,UAAU;YAAE,QAAQ,CAAC,UAAU,GAAG,UAAU,CAAC;QAEjD,uFAAuF;QACvF,MAAM,OAAO,GAAsD,EAAE,CAAC;QACtE,KAAK,MAAM,GAAG,IAAI,CAAC,MAAM,EAAE,UAAU,EAAE,OAAO,CAAU,EAAE,CAAC;YACzD,MAAM,UAAU,GAAG,gBAAgB,CAAC,IAAI,EAAE,OAAO,EAAE,gBAAgB,CAAC,GAAG,CAAC,CAAC,CAAC;YAC1E,IAAI,UAAU;gBAAE,OAAO,CAAC,GAAG,CAAC,GAAG,UAAU,CAAC;QAC5C,CAAC;QAED,UAAU,CAAC,OAAO,CAAC,GAAG;YACpB,GAAG,SAAS;YACZ,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACzD,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SACxD,CAAC;IACJ,CAAC;IAED,MAAM,WAAW,GAAyB,EAAE,CAAC;IAC7C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACzC,KAAK,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YAC7C,MAAM,EAAE,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;YACnC,MAAM,EAAE,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;YACnC,IAAI,CAAC,EAAE,IAAI,CAAC,EAAE;gBAAE,SAAS;YAEzB,MAAM,aAAa,GAAG,qBAAqB,CAAC,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC;YAEpF,MAAM,YAAY,GAAqE,EAAE,CAAC;YAC1F,KAAK,MAAM,GAAG,IAAI,CAAC,MAAM,EAAE,YAAY,CAAU,EAAE,CAAC;gBAClD,MAAM,EAAE,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,GAAG,CAAC,CAAC;gBAC9B,MAAM,EAAE,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,GAAG,CAAC,CAAC;gBAC9B,IAAI,EAAE,IAAI,EAAE,EAAE,CAAC;oBACb,YAAY,CAAC,GAAG,CAAC,GAAG,qBAAqB,CAAC,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC;gBACpF,CAAC;YACH,CAAC;YAED,qFAAqF;YACrF,MAAM,WAAW,GAAgE,EAAE,CAAC;YACpF,KAAK,MAAM,GAAG,IAAI,CAAC,MAAM,EAAE,UAAU,EAAE,OAAO,CAAU,EAAE,CAAC;gBACzD,MAAM,EAAE,GAAG,EAAE,CAAC,OAAO,EAAE,CAAC,GAAG,CAAC,CAAC;gBAC7B,MAAM,EAAE,GAAG,EAAE,CAAC,OAAO,EAAE,CAAC,GAAG,CAAC,CAAC;gBAC7B,IAAI,EAAE,IAAI,EAAE,EAAE,CAAC;oBACb,WAAW,CAAC,GAAG,CAAC,GAAG,qBAAqB,CAAC,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC;gBACnF,CAAC;YACH,CAAC;YAED,WAAW,CAAC,IAAI,CAAC;gBACf,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC;gBACd,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC;gBACd,GAAG,aAAa;gBAChB,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC3E,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aACzE,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,MAAM,UAAU,GAAG,mBAAmB,CAAC,IAAI,CAAC,CAAC;IAC7C,OAAO;QACL,IAAI,EAAE,IAAI,CAAC,MAAM;QACjB,UAAU;QACV,WAAW;QACX,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACtC,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,iBAAiB,CAAC,EACtC,QAAQ,EACR,KAAK,GAAG,aAAa,EACrB,UAAU,GAAG,WAAW,EACxB,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAO,GAAG,KAAK,EACf,MAAM,GAAG,KAAK,EACd,WAAW,GAAG,CAAC,EACf,SAAS,EACT,YAAY,GAAG,QAAQ,EACvB,iBAAiB,EACjB,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,eAAe,GAAG,IAAI,EACtB,aAAa,GAAG,KAAK,EACrB,SAAS,EACT,OAAO,GAAG,KAAK,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,YAAY,GACa;IACzB,MAAM,YAAY,GAAG,kBAAkB,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC;IAC3D,IAAI,YAAY,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC9B,MAAM,IAAI,KAAK,CAAC,+BAA+B,QAAQ,EAAE,CAAC,CAAC;IAC7D,CAAC;IAED,IAAI,MAAM,EAAE,CAAC;QACX,MAAM,EAAE,SAAS,EAAE,YAAY,EAAE,UAAU,EAAE,GAAG,wBAAwB,CAAC,YAAY,CAAC,CAAC;QACvF,OAAO;YACL,MAAM,EAAE;gBACN,MAAM,EAAE,IAAI;gBACZ,IAAI,EAAE,IAAI;gBACV,KAAK;gBACL,UAAU;gBACV,QAAQ,EAAE,YAAY;gBACtB,QAAQ;gBACR,cAAc,EAAE,YAAY,CAAC,MAAM;gBACnC,UAAU;gBACV,SAAS,EAAE,YAAY;aACxB;YACD,QAAQ,EAAE,IAAI;SACf,CAAC;IACJ,CAAC;IACD,OAAO,yBAAyB,CAAC;QAC/B,QAAQ;QACR,YAAY;QACZ,KAAK;QACL,UAAU;QACV,SAAS;QACT,OAAO;QACP,KAAK;QACL,IAAI;QACJ,OAAO;QACP,WAAW;QACX,SAAS;QACT,YAAY;QACZ,iBAAiB;QACjB,QAAQ;QACR,UAAU;QACV,UAAU;QACV,eAAe;QACf,aAAa;QACb,SAAS;QACT,OAAO;QACP,MAAM;QACN,WAAW;QACX,WAAW;QACX,YAAY;QACZ,mBAAmB,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE;YACrC,wEAAwE;YACxE,IAAI,MAAM,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;gBACzB,MAAM,KAAK,GAAG,MAAM,WAAW,CAAC,EAAE,GAAG,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,WAAW,EAAE,YAAY,EAAE,CAAC,CAAC;gBAC5G,OAAO,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,CAAC;YAC5D,CAAC;YACD,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,EAAE,GAAG,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,WAAW,EAAE,YAAY,EAAE,CAAC,CAAC;YACvG,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,MAAgB,EAAE,QAAQ,EAAE,MAAM,CAAC,QAAQ,EAAE,CAAC;QACxE,CAAC;KACF,CAAC,CAAC;AACL,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,WAAW,CAAC,EAAE,MAAM,GAAG,CAAC,EAAE,gBAAgB,EAAE,GAAG,MAAM,EAAsB;IAC/F,MAAM,IAAI,GAAa,EAAE,CAAC;IAC1B,MAAM,cAAc,GAAG,MAAM,CAAC,SAAS,CAAC;IAExC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAChC,gBAAgB,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAClD,MAAM,MAAM,GAAG,CAAC,KAAK,MAAM,GAAG,CAAC,CAAC;QAChC,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM,aAAa,CAAC;YACrC,GAAG,MAAM;YACT,SAAS,EAAE,MAAM,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,CAAC,IAAI;YACzC,UAAU,EAAE,MAAM;SACnB,CAAC,CAAC;QACH,IAAI,CAAC,IAAI,CAAC,MAAgB,CAAC,CAAC;IAC9B,CAAC;IAED,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACrC,MAAM,UAAU,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC;IAC3C,IAAI,QAAQ,GAAkB,IAAI,CAAC;IACnC,IAAI,UAAU,EAAE,CAAC;QACf,MAAM,CAAC,QAAQ,GAAG,UAAU,CAAC;QAC7B,gDAAgD;QAChD,QAAQ,GAAG,aAAa,CAAC,MAAM,EAAE,cAAc,IAAI,kBAAkB,CAAC,CAAC;IACzE,CAAC;IAED,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;AAC1C,CAAC"}
|
|
@@ -7,6 +7,22 @@ export interface AsyncAssertionContext {
|
|
|
7
7
|
samplesDir: string;
|
|
8
8
|
}
|
|
9
9
|
export declare function ratioToScore(ratio: number): number;
|
|
10
|
+
/**
|
|
11
|
+
* ROUGE-N (recall-oriented). N-gram overlap divided by the reference's
|
|
12
|
+
* n-gram count. Multi-set form per Lin 2004 — repeated n-grams are clipped
|
|
13
|
+
* to their reference count, so a candidate that repeats a single matching
|
|
14
|
+
* n-gram many times can't game the score above 1.
|
|
15
|
+
*/
|
|
16
|
+
export declare function rougeN(candidate: string, reference: string, n: number): number;
|
|
17
|
+
/** Standard Levenshtein edit distance with O(min(m,n)) memory. */
|
|
18
|
+
export declare function levenshtein(a: string, b: string): number;
|
|
19
|
+
/**
|
|
20
|
+
* BLEU score (corpus-level form, single reference). Geometric mean of
|
|
21
|
+
* precision at n=1..maxN, multiplied by the brevity penalty for short
|
|
22
|
+
* candidates. Unsmoothed — returns 0 if any n-gram precision is 0, which
|
|
23
|
+
* is a known sharpness of the original BLEU; threshold accordingly.
|
|
24
|
+
*/
|
|
25
|
+
export declare function bleu(candidate: string, reference: string, maxN?: number): number;
|
|
10
26
|
export declare function validateJsonSchema(data: unknown, schema: Record<string, unknown>): boolean;
|
|
11
27
|
export declare function runAssertions(output: string, assertions: Assertion[], context?: {
|
|
12
28
|
costUSD?: number;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"assertions.d.ts","sourceRoot":"","sources":["../../../src/grading/assertions.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,SAAS,EAAmB,gBAAgB,EAAE,UAAU,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAMlH,eAAO,MAAM,qBAAqB,
|
|
1
|
+
{"version":3,"file":"assertions.d.ts","sourceRoot":"","sources":["../../../src/grading/assertions.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,SAAS,EAAmB,gBAAgB,EAAE,UAAU,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAMlH,eAAO,MAAM,qBAAqB,aAQhC,CAAC;AAEH,MAAM,WAAW,qBAAqB;IACpC,QAAQ,EAAE,UAAU,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;CACpB;AAwBD,wBAAgB,YAAY,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAElD;AAkCD;;;;;GAKG;AACH,wBAAgB,MAAM,CAAC,SAAS,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,GAAG,MAAM,CAK9E;AAED,kEAAkE;AAClE,wBAAgB,WAAW,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,GAAG,MAAM,CAmBxD;AAED;;;;;GAKG;AACH,wBAAgB,IAAI,CAAC,SAAS,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,IAAI,SAAI,GAAG,MAAM,CAgB3E;AAED,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,OAAO,CAQ1F;AAoHD,wBAAgB,aAAa,CAC3B,MAAM,EAAE,MAAM,EACd,UAAU,EAAE,SAAS,EAAE,EACvB,OAAO,GAAE;IAAE,OAAO,CAAC,EAAE,MAAM,CAAC;IAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,YAAY,EAAE,CAAA;CAAO,GACrG,gBAAgB,CA8BlB;AAED,wBAAsB,kBAAkB,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,EAAE,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,EAAE,EAAE,qBAAqB,GAAG,OAAO,CAAC,gBAAgB,CAAC,CA2FhL"}
|