oh-my-knowledge 0.30.0 → 0.32.0
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 +33 -672
- package/README.zh.md +45 -684
- package/dist/analysis/coverage-analyzer.d.ts.map +1 -0
- package/dist/analysis/coverage-analyzer.js.map +1 -0
- package/dist/analysis/failure-clusterer.d.ts.map +1 -0
- package/dist/analysis/failure-clusterer.js.map +1 -0
- package/dist/analysis/gap-analyzer.d.ts +121 -0
- package/dist/analysis/gap-analyzer.d.ts.map +1 -0
- package/dist/analysis/gap-analyzer.js +471 -0
- package/dist/analysis/gap-analyzer.js.map +1 -0
- package/dist/analysis/hedging-classifier.d.ts.map +1 -0
- package/dist/analysis/hedging-classifier.js.map +1 -0
- package/dist/analysis/report-diagnostics.d.ts +34 -0
- package/dist/analysis/report-diagnostics.d.ts.map +1 -0
- package/dist/analysis/report-diagnostics.js +753 -0
- package/dist/analysis/report-diagnostics.js.map +1 -0
- package/dist/analysis/sample-diagnostics.d.ts.map +1 -0
- package/dist/analysis/sample-diagnostics.js.map +1 -0
- package/dist/analysis/saturation.d.ts.map +1 -0
- package/dist/analysis/saturation.js.map +1 -0
- package/dist/authoring/evolver.d.ts +102 -0
- package/dist/authoring/evolver.d.ts.map +1 -0
- package/dist/authoring/evolver.js +681 -0
- package/dist/authoring/evolver.js.map +1 -0
- package/dist/authoring/generator.d.ts +38 -0
- package/dist/authoring/generator.d.ts.map +1 -0
- package/dist/authoring/generator.js +720 -0
- package/dist/authoring/generator.js.map +1 -0
- package/dist/authoring/sample-fixer.d.ts +51 -0
- package/dist/authoring/sample-fixer.d.ts.map +1 -0
- package/dist/authoring/sample-fixer.js +213 -0
- package/dist/authoring/sample-fixer.js.map +1 -0
- package/dist/cli/commands/doctor.d.ts +26 -0
- package/dist/cli/commands/doctor.d.ts.map +1 -0
- package/dist/cli/commands/doctor.js +302 -0
- package/dist/cli/commands/doctor.js.map +1 -0
- package/dist/cli/commands/eval/gold/compare.d.ts +17 -0
- package/dist/cli/commands/eval/gold/compare.d.ts.map +1 -0
- package/dist/cli/commands/eval/gold/compare.js +91 -0
- package/dist/cli/commands/eval/gold/compare.js.map +1 -0
- package/dist/cli/commands/eval/gold/index.d.ts +9 -0
- package/dist/cli/commands/eval/gold/index.d.ts.map +1 -0
- package/dist/cli/commands/eval/gold/index.js +34 -0
- package/dist/cli/commands/eval/gold/index.js.map +1 -0
- package/dist/cli/commands/eval/gold/init.d.ts +11 -0
- package/dist/cli/commands/eval/gold/init.d.ts.map +1 -0
- package/dist/cli/commands/eval/gold/init.js +51 -0
- package/dist/cli/commands/eval/gold/init.js.map +1 -0
- package/dist/cli/commands/eval/gold/validate.d.ts +12 -0
- package/dist/cli/commands/eval/gold/validate.d.ts.map +1 -0
- package/dist/cli/commands/eval/gold/validate.js +46 -0
- package/dist/cli/commands/eval/gold/validate.js.map +1 -0
- package/dist/cli/commands/eval/index.d.ts +54 -0
- package/dist/cli/commands/eval/index.d.ts.map +1 -0
- package/dist/cli/commands/eval/index.js +465 -0
- package/dist/cli/commands/eval/index.js.map +1 -0
- package/dist/cli/commands/evolve.d.ts +37 -0
- package/dist/cli/commands/evolve.d.ts.map +1 -0
- package/dist/cli/commands/evolve.js +283 -0
- package/dist/cli/commands/evolve.js.map +1 -0
- package/dist/cli/commands/init.d.ts +16 -0
- package/dist/cli/commands/init.d.ts.map +1 -0
- package/dist/cli/commands/init.js +152 -0
- package/dist/cli/commands/init.js.map +1 -0
- package/dist/cli/commands/observe/inbox.d.ts +23 -0
- package/dist/cli/commands/observe/inbox.d.ts.map +1 -0
- package/dist/cli/commands/observe/inbox.js +260 -0
- package/dist/cli/commands/observe/inbox.js.map +1 -0
- package/dist/cli/commands/observe/index.d.ts +22 -0
- package/dist/cli/commands/observe/index.d.ts.map +1 -0
- package/dist/cli/commands/observe/index.js +118 -0
- package/dist/cli/commands/observe/index.js.map +1 -0
- package/dist/cli/commands/observe/ingest.d.ts +13 -0
- package/dist/cli/commands/observe/ingest.d.ts.map +1 -0
- package/dist/cli/commands/observe/ingest.js +71 -0
- package/dist/cli/commands/observe/ingest.js.map +1 -0
- package/dist/cli/commands/observe/show.d.ts +13 -0
- package/dist/cli/commands/observe/show.d.ts.map +1 -0
- package/dist/cli/commands/observe/show.js +49 -0
- package/dist/cli/commands/observe/show.js.map +1 -0
- package/dist/cli/commands/sample.d.ts +38 -0
- package/dist/cli/commands/sample.d.ts.map +1 -0
- package/dist/cli/commands/sample.js +487 -0
- package/dist/cli/commands/sample.js.map +1 -0
- package/dist/cli/commands/studio.d.ts +23 -0
- package/dist/cli/commands/studio.d.ts.map +1 -0
- package/dist/cli/commands/studio.js +158 -0
- package/dist/cli/commands/studio.js.map +1 -0
- package/dist/cli/index.d.ts.map +1 -0
- package/dist/cli/index.js +29 -0
- package/dist/cli/index.js.map +1 -0
- package/dist/cli/lib/cli-exit.d.ts +16 -0
- package/dist/cli/lib/cli-exit.d.ts.map +1 -0
- package/dist/cli/lib/cli-exit.js +20 -0
- package/dist/cli/lib/cli-exit.js.map +1 -0
- package/dist/cli/lib/cmd-flags.d.ts +224 -0
- package/dist/cli/lib/cmd-flags.d.ts.map +1 -0
- package/dist/cli/lib/cmd-flags.js +46 -0
- package/dist/cli/lib/cmd-flags.js.map +1 -0
- package/dist/cli/lib/i18n-dict/common.d.ts +4 -0
- package/dist/cli/lib/i18n-dict/common.d.ts.map +1 -0
- package/dist/cli/lib/i18n-dict/common.js +67 -0
- package/dist/cli/lib/i18n-dict/common.js.map +1 -0
- package/dist/cli/lib/i18n-dict/evolve.d.ts +4 -0
- package/dist/cli/lib/i18n-dict/evolve.d.ts.map +1 -0
- package/dist/cli/lib/i18n-dict/evolve.js +43 -0
- package/dist/cli/lib/i18n-dict/evolve.js.map +1 -0
- package/dist/cli/lib/i18n-dict/gen.d.ts +4 -0
- package/dist/cli/lib/i18n-dict/gen.d.ts.map +1 -0
- package/dist/cli/lib/i18n-dict/gen.js +63 -0
- package/dist/cli/lib/i18n-dict/gen.js.map +1 -0
- package/dist/cli/lib/i18n-dict/help.d.ts +4 -0
- package/dist/cli/lib/i18n-dict/help.d.ts.map +1 -0
- package/dist/cli/lib/i18n-dict/help.js +281 -0
- package/dist/cli/lib/i18n-dict/help.js.map +1 -0
- package/dist/cli/lib/i18n-dict/init.d.ts +4 -0
- package/dist/cli/lib/i18n-dict/init.d.ts.map +1 -0
- package/dist/cli/lib/i18n-dict/init.js +27 -0
- package/dist/cli/lib/i18n-dict/init.js.map +1 -0
- package/dist/cli/lib/i18n-dict/run.d.ts +4 -0
- package/dist/cli/lib/i18n-dict/run.d.ts.map +1 -0
- package/dist/cli/lib/i18n-dict/run.js +143 -0
- package/dist/cli/lib/i18n-dict/run.js.map +1 -0
- package/dist/cli/lib/i18n-dict/types.d.ts +5 -0
- package/dist/cli/lib/i18n-dict/types.d.ts.map +1 -0
- package/dist/cli/lib/i18n-dict/types.js +2 -0
- package/dist/cli/lib/i18n-dict/types.js.map +1 -0
- package/dist/cli/lib/i18n-dict.d.ts +63 -0
- package/dist/cli/lib/i18n-dict.d.ts.map +1 -0
- package/dist/cli/lib/i18n-dict.js +67 -0
- package/dist/cli/lib/i18n-dict.js.map +1 -0
- package/dist/cli/lib/i18n.d.ts +29 -0
- package/dist/cli/lib/i18n.d.ts.map +1 -0
- package/dist/cli/lib/i18n.js +58 -0
- package/dist/cli/lib/i18n.js.map +1 -0
- package/dist/cli/lib/parse-run-config/judge-models.d.ts +24 -0
- package/dist/cli/lib/parse-run-config/judge-models.d.ts.map +1 -0
- package/dist/cli/lib/parse-run-config/judge-models.js +55 -0
- package/dist/cli/lib/parse-run-config/judge-models.js.map +1 -0
- package/dist/cli/lib/parse-run-config/samples-discovery.d.ts +19 -0
- package/dist/cli/lib/parse-run-config/samples-discovery.d.ts.map +1 -0
- package/dist/cli/lib/parse-run-config/samples-discovery.js +53 -0
- package/dist/cli/lib/parse-run-config/samples-discovery.js.map +1 -0
- package/dist/cli/lib/parse-run-config/variant-resolution.d.ts +18 -0
- package/dist/cli/lib/parse-run-config/variant-resolution.d.ts.map +1 -0
- package/dist/cli/lib/parse-run-config/variant-resolution.js +59 -0
- package/dist/cli/lib/parse-run-config/variant-resolution.js.map +1 -0
- package/dist/cli/lib/parse-run-config.d.ts +93 -0
- package/dist/cli/lib/parse-run-config.d.ts.map +1 -0
- package/dist/cli/lib/parse-run-config.js +157 -0
- package/dist/cli/lib/parse-run-config.js.map +1 -0
- package/dist/cli/lib/progress.d.ts.map +1 -0
- package/dist/cli/lib/progress.js.map +1 -0
- package/dist/cli/lib/resolve-skill-input.d.ts +8 -0
- package/dist/cli/lib/resolve-skill-input.d.ts.map +1 -0
- package/dist/cli/lib/resolve-skill-input.js +38 -0
- package/dist/cli/lib/resolve-skill-input.js.map +1 -0
- package/dist/cli/lib/run-tally.d.ts +18 -0
- package/dist/cli/lib/run-tally.d.ts.map +1 -0
- package/dist/cli/lib/run-tally.js.map +1 -0
- package/dist/cli/lib/shared.d.ts +12 -0
- package/dist/cli/lib/shared.d.ts.map +1 -0
- package/dist/cli/lib/shared.js +26 -0
- package/dist/cli/lib/shared.js.map +1 -0
- package/dist/cli/lib/update-check.d.ts +9 -0
- package/dist/cli/lib/update-check.d.ts.map +1 -0
- package/dist/cli/lib/update-check.js +113 -0
- package/dist/cli/lib/update-check.js.map +1 -0
- package/dist/cli/oclif/base-command.d.ts +8 -0
- package/dist/cli/oclif/base-command.d.ts.map +1 -0
- package/dist/cli/oclif/base-command.js +27 -0
- package/dist/cli/oclif/base-command.js.map +1 -0
- package/dist/cli/oclif/help.d.ts +16 -0
- package/dist/cli/oclif/help.d.ts.map +1 -0
- package/dist/cli/oclif/help.js +33 -0
- package/dist/cli/oclif/help.js.map +1 -0
- package/dist/cli/oclif/i18n.d.ts +16 -0
- package/dist/cli/oclif/i18n.d.ts.map +1 -0
- package/dist/cli/oclif/i18n.js +62 -0
- package/dist/cli/oclif/i18n.js.map +1 -0
- package/dist/cli/oclif/parsers.d.ts +10 -0
- package/dist/cli/oclif/parsers.d.ts.map +1 -0
- package/dist/cli/oclif/parsers.js +102 -0
- package/dist/cli/oclif/parsers.js.map +1 -0
- package/dist/cli/oclif/projection.d.ts +23 -0
- package/dist/cli/oclif/projection.d.ts.map +1 -0
- package/dist/cli/oclif/projection.js +52 -0
- package/dist/cli/oclif/projection.js.map +1 -0
- package/dist/cli/oclif/run.d.ts +2 -0
- package/dist/cli/oclif/run.d.ts.map +1 -0
- package/dist/cli/oclif/run.js +11 -0
- package/dist/cli/oclif/run.js.map +1 -0
- package/dist/diagnosis/observe-mapper.d.ts +68 -0
- package/dist/diagnosis/observe-mapper.d.ts.map +1 -0
- package/dist/diagnosis/observe-mapper.js +276 -0
- package/dist/diagnosis/observe-mapper.js.map +1 -0
- package/dist/diagnosis/observe-producer.d.ts +9 -0
- package/dist/diagnosis/observe-producer.d.ts.map +1 -0
- package/dist/diagnosis/observe-producer.js +199 -0
- package/dist/diagnosis/observe-producer.js.map +1 -0
- package/dist/diagnosis/studio-projection.d.ts +7 -0
- package/dist/diagnosis/studio-projection.d.ts.map +1 -0
- package/dist/diagnosis/studio-projection.js +84 -0
- package/dist/diagnosis/studio-projection.js.map +1 -0
- package/dist/diagnosis/types.d.ts +4 -0
- package/dist/diagnosis/types.d.ts.map +1 -0
- package/dist/diagnosis/types.js +20 -0
- package/dist/diagnosis/types.js.map +1 -0
- package/dist/doctor/fixer.d.ts +12 -0
- package/dist/doctor/fixer.d.ts.map +1 -0
- package/dist/doctor/fixer.js +326 -0
- package/dist/doctor/fixer.js.map +1 -0
- package/dist/doctor/health/builtin-dimensions.d.ts.map +1 -0
- package/dist/doctor/health/builtin-dimensions.js.map +1 -0
- package/dist/doctor/health/composer.d.ts.map +1 -0
- package/dist/doctor/health/composer.js +296 -0
- package/dist/doctor/health/composer.js.map +1 -0
- package/dist/doctor/health/dimension-registry.d.ts.map +1 -0
- package/dist/doctor/health/dimension-registry.js.map +1 -0
- package/dist/doctor/health/dimension-spec.d.ts +47 -0
- package/dist/doctor/health/dimension-spec.d.ts.map +1 -0
- package/dist/doctor/health/dimension-spec.js.map +1 -0
- package/dist/doctor/health/parser.d.ts +26 -0
- package/dist/doctor/health/parser.d.ts.map +1 -0
- package/dist/doctor/health/parser.js +228 -0
- package/dist/doctor/health/parser.js.map +1 -0
- package/dist/doctor/health/prompt-builder.d.ts +3 -0
- package/dist/doctor/health/prompt-builder.d.ts.map +1 -0
- package/dist/doctor/health/prompt-builder.js +2 -0
- package/dist/doctor/health/prompt-builder.js.map +1 -0
- package/dist/doctor/health/register.d.ts.map +1 -0
- package/dist/doctor/health/register.js.map +1 -0
- package/dist/doctor/index.d.ts.map +1 -0
- package/dist/doctor/index.js +246 -0
- package/dist/doctor/index.js.map +1 -0
- package/dist/doctor/messages.d.ts +21 -0
- package/dist/doctor/messages.d.ts.map +1 -0
- package/dist/doctor/messages.js +217 -0
- package/dist/doctor/messages.js.map +1 -0
- package/dist/doctor/preflight.d.ts.map +1 -0
- package/dist/doctor/preflight.js.map +1 -0
- package/dist/doctor/renderer.d.ts +17 -0
- package/dist/doctor/renderer.d.ts.map +1 -0
- package/dist/doctor/renderer.js +122 -0
- package/dist/doctor/renderer.js.map +1 -0
- package/dist/doctor/rules.d.ts.map +1 -0
- package/dist/doctor/rules.js +289 -0
- package/dist/doctor/rules.js.map +1 -0
- package/dist/eval-core/bootstrap.d.ts.map +1 -0
- package/dist/eval-core/bootstrap.js.map +1 -0
- package/dist/eval-core/cache.d.ts.map +1 -0
- package/dist/eval-core/cache.js.map +1 -0
- package/dist/eval-core/comparability.d.ts.map +1 -0
- package/dist/eval-core/comparability.js.map +1 -0
- package/dist/eval-core/dependency-checker.d.ts +37 -0
- package/dist/eval-core/dependency-checker.d.ts.map +1 -0
- package/dist/eval-core/dependency-checker.js.map +1 -0
- package/dist/eval-core/evaluation-execution.d.ts.map +1 -0
- package/dist/eval-core/evaluation-execution.js.map +1 -0
- package/dist/eval-core/evaluation-job.d.ts.map +1 -0
- package/dist/eval-core/evaluation-job.js.map +1 -0
- package/dist/eval-core/evaluation-reporting.d.ts.map +1 -0
- package/dist/eval-core/evaluation-reporting.js.map +1 -0
- package/dist/eval-core/execution-strategy.d.ts.map +1 -0
- package/dist/eval-core/execution-strategy.js +165 -0
- package/dist/eval-core/execution-strategy.js.map +1 -0
- package/dist/eval-core/fact-checker.d.ts.map +1 -0
- package/dist/eval-core/fact-checker.js.map +1 -0
- package/dist/eval-core/layer-gates.d.ts.map +1 -0
- package/dist/eval-core/layer-gates.js.map +1 -0
- package/dist/eval-core/mock-hook.cjs +215 -0
- package/dist/eval-core/mocks-runtime.d.ts +100 -0
- package/dist/eval-core/mocks-runtime.d.ts.map +1 -0
- package/dist/eval-core/mocks-runtime.js +559 -0
- package/dist/eval-core/mocks-runtime.js.map +1 -0
- package/dist/eval-core/schema.d.ts.map +1 -0
- package/dist/eval-core/schema.js.map +1 -0
- package/dist/eval-core/statistics.d.ts.map +1 -0
- package/dist/eval-core/statistics.js.map +1 -0
- package/dist/eval-core/task-planner.d.ts.map +1 -0
- package/dist/eval-core/task-planner.js.map +1 -0
- package/dist/eval-core/verdict.d.ts.map +1 -0
- package/dist/eval-core/verdict.js.map +1 -0
- package/dist/eval-workflows/batch-evaluation-workflow.d.ts.map +1 -0
- package/dist/eval-workflows/batch-evaluation-workflow.js.map +1 -0
- package/dist/eval-workflows/evaluation-pipeline/preflight-warnings.d.ts +22 -0
- package/dist/eval-workflows/evaluation-pipeline/preflight-warnings.d.ts.map +1 -0
- package/dist/eval-workflows/evaluation-pipeline/preflight-warnings.js +67 -0
- package/dist/eval-workflows/evaluation-pipeline/preflight-warnings.js.map +1 -0
- package/dist/eval-workflows/evaluation-pipeline/report-finalize.d.ts +32 -0
- package/dist/eval-workflows/evaluation-pipeline/report-finalize.d.ts.map +1 -0
- package/dist/eval-workflows/evaluation-pipeline/report-finalize.js +57 -0
- package/dist/eval-workflows/evaluation-pipeline/report-finalize.js.map +1 -0
- package/dist/eval-workflows/evaluation-pipeline/run-state.d.ts +60 -0
- package/dist/eval-workflows/evaluation-pipeline/run-state.d.ts.map +1 -0
- package/dist/eval-workflows/evaluation-pipeline/run-state.js +88 -0
- package/dist/eval-workflows/evaluation-pipeline/run-state.js.map +1 -0
- package/dist/eval-workflows/evaluation-pipeline/test-set-hash.d.ts +21 -0
- package/dist/eval-workflows/evaluation-pipeline/test-set-hash.d.ts.map +1 -0
- package/dist/eval-workflows/evaluation-pipeline/test-set-hash.js +46 -0
- package/dist/eval-workflows/evaluation-pipeline/test-set-hash.js.map +1 -0
- package/dist/eval-workflows/evaluation-pipeline.d.ts +97 -0
- package/dist/eval-workflows/evaluation-pipeline.d.ts.map +1 -0
- package/dist/eval-workflows/evaluation-pipeline.js +179 -0
- package/dist/eval-workflows/evaluation-pipeline.js.map +1 -0
- package/dist/eval-workflows/evaluation-preparation.d.ts.map +1 -0
- package/dist/eval-workflows/evaluation-preparation.js.map +1 -0
- package/dist/eval-workflows/messages.d.ts +4 -0
- package/dist/eval-workflows/messages.d.ts.map +1 -0
- package/dist/eval-workflows/messages.js +30 -0
- package/dist/eval-workflows/messages.js.map +1 -0
- package/dist/eval-workflows/run-evaluation.d.ts.map +1 -0
- package/dist/eval-workflows/run-evaluation.js +528 -0
- package/dist/eval-workflows/run-evaluation.js.map +1 -0
- package/dist/executors/anthropic-api.d.ts.map +1 -0
- package/dist/executors/anthropic-api.js.map +1 -0
- package/dist/executors/claude-cli.d.ts.map +1 -0
- package/dist/executors/claude-cli.js +181 -0
- package/dist/executors/claude-cli.js.map +1 -0
- package/dist/executors/claude-sdk-trace.d.ts.map +1 -0
- package/dist/executors/claude-sdk-trace.js.map +1 -0
- package/dist/executors/claude-sdk.d.ts.map +1 -0
- package/dist/executors/claude-sdk.js.map +1 -0
- package/dist/executors/codex-cli-trace.d.ts.map +1 -0
- package/dist/executors/codex-cli-trace.js.map +1 -0
- package/dist/executors/codex-cli.d.ts.map +1 -0
- package/dist/executors/codex-cli.js.map +1 -0
- package/dist/executors/codex-sdk.d.ts.map +1 -0
- package/dist/executors/codex-sdk.js.map +1 -0
- package/dist/executors/gemini.d.ts.map +1 -0
- package/dist/executors/gemini.js.map +1 -0
- package/dist/executors/index.d.ts.map +1 -0
- package/dist/executors/index.js.map +1 -0
- package/dist/executors/openai-api.d.ts.map +1 -0
- package/dist/executors/openai-api.js.map +1 -0
- package/dist/executors/runtime-fingerprint.d.ts.map +1 -0
- package/dist/executors/runtime-fingerprint.js.map +1 -0
- package/dist/executors/script.d.ts.map +1 -0
- package/dist/executors/script.js.map +1 -0
- package/dist/executors/shared.d.ts +194 -0
- package/dist/executors/shared.d.ts.map +1 -0
- package/dist/executors/shared.js +253 -0
- package/dist/executors/shared.js.map +1 -0
- package/dist/grading/assertions.d.ts.map +1 -0
- package/dist/grading/assertions.js.map +1 -0
- package/dist/grading/debias-validate.d.ts.map +1 -0
- package/dist/grading/debias-validate.js.map +1 -0
- package/dist/grading/diagnostic.d.ts.map +1 -0
- package/dist/grading/diagnostic.js.map +1 -0
- package/dist/grading/gold-cli.d.ts.map +1 -0
- package/dist/grading/gold-cli.js.map +1 -0
- package/dist/grading/gold-dataset.d.ts.map +1 -0
- package/dist/grading/gold-dataset.js.map +1 -0
- package/dist/grading/human-gold.d.ts.map +1 -0
- package/dist/grading/human-gold.js.map +1 -0
- package/dist/grading/index.d.ts.map +1 -0
- package/dist/grading/index.js.map +1 -0
- package/dist/grading/judge.d.ts.map +1 -0
- package/dist/grading/judge.js.map +1 -0
- package/dist/grading/layered-scores.d.ts.map +1 -0
- package/dist/grading/layered-scores.js.map +1 -0
- package/dist/inputs/eval-config.d.ts.map +1 -0
- package/dist/inputs/eval-config.js.map +1 -0
- package/dist/inputs/load-samples.d.ts.map +1 -0
- package/dist/inputs/load-samples.js.map +1 -0
- package/dist/inputs/mcp-resolver.d.ts.map +1 -0
- package/dist/inputs/mcp-resolver.js.map +1 -0
- package/dist/inputs/skill-loader.d.ts.map +1 -0
- package/dist/inputs/skill-loader.js +261 -0
- package/dist/inputs/skill-loader.js.map +1 -0
- package/dist/inputs/url-fetcher.d.ts.map +1 -0
- package/dist/inputs/url-fetcher.js.map +1 -0
- package/dist/observability/experience-frontmatter.d.ts +37 -0
- package/dist/observability/experience-frontmatter.d.ts.map +1 -0
- package/dist/observability/experience-frontmatter.js +100 -0
- package/dist/observability/experience-frontmatter.js.map +1 -0
- package/dist/observability/experience.d.ts +23 -0
- package/dist/observability/experience.d.ts.map +1 -0
- package/dist/observability/experience.js +3438 -0
- package/dist/observability/experience.js.map +1 -0
- package/dist/observability/feedback-matchers.d.ts +37 -0
- package/dist/observability/feedback-matchers.d.ts.map +1 -0
- package/dist/observability/feedback-matchers.js +232 -0
- package/dist/observability/feedback-matchers.js.map +1 -0
- package/dist/observability/feedback-projection.d.ts +21 -0
- package/dist/observability/feedback-projection.d.ts.map +1 -0
- package/dist/observability/feedback-projection.js +20 -0
- package/dist/observability/feedback-projection.js.map +1 -0
- package/dist/observability/inbox-view-model.d.ts +43 -0
- package/dist/observability/inbox-view-model.d.ts.map +1 -0
- package/dist/observability/inbox-view-model.js +180 -0
- package/dist/observability/inbox-view-model.js.map +1 -0
- package/dist/observability/inbox.d.ts +28 -0
- package/dist/observability/inbox.d.ts.map +1 -0
- package/dist/observability/inbox.js +765 -0
- package/dist/observability/inbox.js.map +1 -0
- package/dist/observability/problem-patterns.d.ts +12 -0
- package/dist/observability/problem-patterns.d.ts.map +1 -0
- package/dist/observability/problem-patterns.js +244 -0
- package/dist/observability/problem-patterns.js.map +1 -0
- package/dist/observability/prompts/llm-enhanced-review.prompt.md +224 -0
- package/dist/observability/resolved-review.d.ts +37 -0
- package/dist/observability/resolved-review.d.ts.map +1 -0
- package/dist/observability/resolved-review.js +552 -0
- package/dist/observability/resolved-review.js.map +1 -0
- package/dist/observability/review-state.d.ts +20 -0
- package/dist/observability/review-state.d.ts.map +1 -0
- package/dist/observability/review-state.js +169 -0
- package/dist/observability/review-state.js.map +1 -0
- package/dist/observability/skill-chain-advisories.d.ts +16 -0
- package/dist/observability/skill-chain-advisories.d.ts.map +1 -0
- package/dist/observability/skill-chain-advisories.js +69 -0
- package/dist/observability/skill-chain-advisories.js.map +1 -0
- package/dist/observability/skill-chain.d.ts +7 -0
- package/dist/observability/skill-chain.d.ts.map +1 -0
- package/dist/observability/skill-chain.js +474 -0
- package/dist/observability/skill-chain.js.map +1 -0
- package/dist/observability/skill-health-analyzer.d.ts.map +1 -0
- package/dist/observability/skill-health-analyzer.js.map +1 -0
- package/dist/observability/soft-standards/constants.d.ts +5 -0
- package/dist/observability/soft-standards/constants.d.ts.map +1 -0
- package/dist/observability/soft-standards/constants.js +10 -0
- package/dist/observability/soft-standards/constants.js.map +1 -0
- package/dist/observability/soft-standards/index.d.ts +16 -0
- package/dist/observability/soft-standards/index.d.ts.map +1 -0
- package/dist/observability/soft-standards/index.js +15 -0
- package/dist/observability/soft-standards/index.js.map +1 -0
- package/dist/observability/soft-standards/llm-extractor.d.ts +6 -0
- package/dist/observability/soft-standards/llm-extractor.d.ts.map +1 -0
- package/dist/observability/soft-standards/llm-extractor.js +692 -0
- package/dist/observability/soft-standards/llm-extractor.js.map +1 -0
- package/dist/observability/soft-standards/runtime-evaluator.d.ts +12 -0
- package/dist/observability/soft-standards/runtime-evaluator.d.ts.map +1 -0
- package/dist/observability/soft-standards/runtime-evaluator.js +385 -0
- package/dist/observability/soft-standards/runtime-evaluator.js.map +1 -0
- package/dist/observability/soft-standards/skill-standards-store.d.ts +11 -0
- package/dist/observability/soft-standards/skill-standards-store.d.ts.map +1 -0
- package/dist/observability/soft-standards/skill-standards-store.js +158 -0
- package/dist/observability/soft-standards/skill-standards-store.js.map +1 -0
- package/dist/observability/soft-standards/types.d.ts +212 -0
- package/dist/observability/soft-standards/types.d.ts.map +1 -0
- package/dist/observability/soft-standards/types.js +2 -0
- package/dist/observability/soft-standards/types.js.map +1 -0
- package/dist/observability/text-signals.d.ts +14 -0
- package/dist/observability/text-signals.d.ts.map +1 -0
- package/dist/observability/text-signals.js +116 -0
- package/dist/observability/text-signals.js.map +1 -0
- package/dist/observability/trace-adapter.d.ts.map +1 -0
- package/dist/observability/trace-adapter.js.map +1 -0
- package/dist/observability/trace-attribution.d.ts +56 -0
- package/dist/observability/trace-attribution.d.ts.map +1 -0
- package/dist/observability/trace-attribution.js +200 -0
- package/dist/observability/trace-attribution.js.map +1 -0
- package/dist/observability/trace-segmenter.d.ts +52 -0
- package/dist/observability/trace-segmenter.d.ts.map +1 -0
- package/dist/observability/trace-segmenter.js +322 -0
- package/dist/observability/trace-segmenter.js.map +1 -0
- package/dist/observability/trace-source.d.ts +92 -0
- package/dist/observability/trace-source.d.ts.map +1 -0
- package/dist/observability/trace-source.js +502 -0
- package/dist/observability/trace-source.js.map +1 -0
- package/dist/renderer/html-renderer.d.ts.map +1 -0
- package/dist/renderer/html-renderer.js.map +1 -0
- package/dist/renderer/layout.d.ts.map +1 -0
- package/dist/renderer/layout.js.map +1 -0
- package/dist/renderer/observation-inbox/helpers.d.ts +14 -0
- package/dist/renderer/observation-inbox/helpers.d.ts.map +1 -0
- package/dist/renderer/observation-inbox/helpers.js +75 -0
- package/dist/renderer/observation-inbox/helpers.js.map +1 -0
- package/dist/renderer/observation-inbox/styles.d.ts +2 -0
- package/dist/renderer/observation-inbox/styles.d.ts.map +1 -0
- package/dist/renderer/observation-inbox/styles.js +5184 -0
- package/dist/renderer/observation-inbox/styles.js.map +1 -0
- package/dist/renderer/observation-inbox-renderer.d.ts +5 -0
- package/dist/renderer/observation-inbox-renderer.d.ts.map +1 -0
- package/dist/renderer/observation-inbox-renderer.js +7182 -0
- package/dist/renderer/observation-inbox-renderer.js.map +1 -0
- package/dist/renderer/skill-detail-renderer.d.ts +13 -0
- package/dist/renderer/skill-detail-renderer.d.ts.map +1 -0
- package/dist/renderer/skill-detail-renderer.js +1244 -0
- package/dist/renderer/skill-detail-renderer.js.map +1 -0
- package/dist/renderer/skill-health-renderer.d.ts.map +1 -0
- package/dist/renderer/skill-health-renderer.js.map +1 -0
- package/dist/renderer/skill-list-renderer.d.ts +3 -0
- package/dist/renderer/skill-list-renderer.d.ts.map +1 -0
- package/dist/renderer/skill-list-renderer.js +334 -0
- package/dist/renderer/skill-list-renderer.js.map +1 -0
- package/dist/renderer/summary.d.ts +68 -0
- package/dist/renderer/summary.d.ts.map +1 -0
- package/dist/renderer/summary.js +1896 -0
- package/dist/renderer/summary.js.map +1 -0
- package/dist/renderer/table.d.ts.map +1 -0
- package/dist/renderer/table.js.map +1 -0
- package/dist/renderer/test-view.d.ts.map +1 -0
- package/dist/renderer/test-view.js +905 -0
- package/dist/renderer/test-view.js.map +1 -0
- package/dist/renderer/trends.d.ts.map +1 -0
- package/dist/renderer/trends.js.map +1 -0
- package/dist/server/job-store.d.ts.map +1 -0
- package/dist/server/job-store.js.map +1 -0
- package/dist/server/report-server.d.ts.map +1 -0
- package/dist/server/report-server.js +892 -0
- package/dist/server/report-server.js.map +1 -0
- package/dist/server/report-store.d.ts.map +1 -0
- package/dist/server/report-store.js.map +1 -0
- package/dist/server/skill-index.d.ts +14 -0
- package/dist/server/skill-index.d.ts.map +1 -0
- package/dist/server/skill-index.js +381 -0
- package/dist/server/skill-index.js.map +1 -0
- package/dist/server/skill-insights.d.ts +32 -0
- package/dist/server/skill-insights.d.ts.map +1 -0
- package/dist/server/skill-insights.js +788 -0
- package/dist/server/skill-insights.js.map +1 -0
- package/dist/shared/hard-rules.d.ts +42 -0
- package/dist/shared/hard-rules.d.ts.map +1 -0
- package/dist/shared/hard-rules.js +242 -0
- package/dist/shared/hard-rules.js.map +1 -0
- package/dist/shared/llm-prompts/index.d.ts +14 -0
- package/dist/shared/llm-prompts/index.d.ts.map +1 -0
- package/dist/shared/llm-prompts/index.js +22 -0
- package/dist/shared/llm-prompts/index.js.map +1 -0
- package/dist/shared/llm-prompts/skill-health.d.ts +22 -0
- package/dist/shared/llm-prompts/skill-health.d.ts.map +1 -0
- package/dist/shared/llm-prompts/skill-health.js +170 -0
- package/dist/shared/llm-prompts/skill-health.js.map +1 -0
- package/dist/shared/time.d.ts.map +1 -0
- package/dist/shared/time.js.map +1 -0
- package/dist/shared/tool-search.d.ts.map +1 -0
- package/dist/shared/tool-search.js.map +1 -0
- package/dist/types/dependencies.d.ts +23 -0
- package/dist/types/dependencies.d.ts.map +1 -0
- package/dist/types/dependencies.js +2 -0
- package/dist/types/dependencies.js.map +1 -0
- package/dist/types/diagnosis.d.ts +88 -0
- package/dist/types/diagnosis.d.ts.map +1 -0
- package/dist/types/diagnosis.js +2 -0
- package/dist/types/diagnosis.js.map +1 -0
- package/dist/types/doctor.d.ts +157 -0
- package/dist/types/doctor.d.ts.map +1 -0
- package/dist/types/doctor.js.map +1 -0
- package/dist/types/eval.d.ts +377 -0
- package/dist/types/eval.d.ts.map +1 -0
- package/dist/types/eval.js.map +1 -0
- package/dist/types/executor.d.ts.map +1 -0
- package/dist/types/executor.js.map +1 -0
- package/dist/types/index.d.ts +12 -0
- package/dist/types/index.d.ts.map +1 -0
- package/dist/types/index.js +12 -0
- package/dist/types/index.js.map +1 -0
- package/dist/types/judge.d.ts.map +1 -0
- package/dist/types/judge.js.map +1 -0
- package/dist/types/observability.d.ts +865 -0
- package/dist/types/observability.d.ts.map +1 -0
- package/dist/types/observability.js +14 -0
- package/dist/types/observability.js.map +1 -0
- package/dist/types/report.d.ts +560 -0
- package/dist/types/report.d.ts.map +1 -0
- package/dist/types/report.js.map +1 -0
- package/dist/types/shared.d.ts.map +1 -0
- package/dist/types/shared.js.map +1 -0
- package/dist/types/skill-index.d.ts +142 -0
- package/dist/types/skill-index.d.ts.map +1 -0
- package/dist/types/skill-index.js +2 -0
- package/dist/types/skill-index.js.map +1 -0
- package/dist/types/storage.d.ts.map +1 -0
- package/dist/types/storage.js.map +1 -0
- package/dist/util/safe-slice.d.ts +23 -0
- package/dist/util/safe-slice.d.ts.map +1 -0
- package/dist/util/safe-slice.js +33 -0
- package/dist/util/safe-slice.js.map +1 -0
- package/package.json +40 -8
- package/dist/src/analysis/coverage-analyzer.d.ts.map +0 -1
- package/dist/src/analysis/coverage-analyzer.js.map +0 -1
- package/dist/src/analysis/failure-clusterer.d.ts.map +0 -1
- package/dist/src/analysis/failure-clusterer.js.map +0 -1
- package/dist/src/analysis/gap-analyzer.d.ts +0 -121
- package/dist/src/analysis/gap-analyzer.d.ts.map +0 -1
- package/dist/src/analysis/gap-analyzer.js +0 -471
- package/dist/src/analysis/gap-analyzer.js.map +0 -1
- package/dist/src/analysis/hedging-classifier.d.ts.map +0 -1
- package/dist/src/analysis/hedging-classifier.js.map +0 -1
- package/dist/src/analysis/report-diagnostics.d.ts +0 -34
- package/dist/src/analysis/report-diagnostics.d.ts.map +0 -1
- package/dist/src/analysis/report-diagnostics.js +0 -753
- package/dist/src/analysis/report-diagnostics.js.map +0 -1
- package/dist/src/analysis/sample-diagnostics.d.ts.map +0 -1
- package/dist/src/analysis/sample-diagnostics.js.map +0 -1
- package/dist/src/analysis/saturation.d.ts.map +0 -1
- package/dist/src/analysis/saturation.js.map +0 -1
- package/dist/src/authoring/evolver.d.ts +0 -78
- package/dist/src/authoring/evolver.d.ts.map +0 -1
- package/dist/src/authoring/evolver.js +0 -320
- package/dist/src/authoring/evolver.js.map +0 -1
- package/dist/src/authoring/generator.d.ts +0 -35
- package/dist/src/authoring/generator.d.ts.map +0 -1
- package/dist/src/authoring/generator.js +0 -704
- package/dist/src/authoring/generator.js.map +0 -1
- package/dist/src/authoring/sample-fixer.d.ts +0 -45
- package/dist/src/authoring/sample-fixer.d.ts.map +0 -1
- package/dist/src/authoring/sample-fixer.js +0 -186
- package/dist/src/authoring/sample-fixer.js.map +0 -1
- package/dist/src/cli/cli-exit.d.ts +0 -15
- package/dist/src/cli/cli-exit.d.ts.map +0 -1
- package/dist/src/cli/cli-exit.js +0 -19
- package/dist/src/cli/cli-exit.js.map +0 -1
- package/dist/src/cli/commands/_shared.d.ts +0 -12
- package/dist/src/cli/commands/_shared.d.ts.map +0 -1
- package/dist/src/cli/commands/_shared.js +0 -26
- package/dist/src/cli/commands/_shared.js.map +0 -1
- package/dist/src/cli/commands/doctor.d.ts +0 -2
- package/dist/src/cli/commands/doctor.d.ts.map +0 -1
- package/dist/src/cli/commands/doctor.js +0 -174
- package/dist/src/cli/commands/doctor.js.map +0 -1
- package/dist/src/cli/commands/eval-gold.d.ts +0 -2
- package/dist/src/cli/commands/eval-gold.d.ts.map +0 -1
- package/dist/src/cli/commands/eval-gold.js +0 -137
- package/dist/src/cli/commands/eval-gold.js.map +0 -1
- package/dist/src/cli/commands/eval-runner.d.ts +0 -2
- package/dist/src/cli/commands/eval-runner.d.ts.map +0 -1
- package/dist/src/cli/commands/eval-runner.js +0 -299
- package/dist/src/cli/commands/eval-runner.js.map +0 -1
- package/dist/src/cli/commands/eval.d.ts +0 -2
- package/dist/src/cli/commands/eval.d.ts.map +0 -1
- package/dist/src/cli/commands/eval.js +0 -11
- package/dist/src/cli/commands/eval.js.map +0 -1
- package/dist/src/cli/commands/evolve.d.ts +0 -2
- package/dist/src/cli/commands/evolve.d.ts.map +0 -1
- package/dist/src/cli/commands/evolve.js +0 -132
- package/dist/src/cli/commands/evolve.js.map +0 -1
- package/dist/src/cli/commands/init.d.ts +0 -2
- package/dist/src/cli/commands/init.d.ts.map +0 -1
- package/dist/src/cli/commands/init.js +0 -110
- package/dist/src/cli/commands/init.js.map +0 -1
- package/dist/src/cli/commands/observe.d.ts +0 -2
- package/dist/src/cli/commands/observe.d.ts.map +0 -1
- package/dist/src/cli/commands/observe.js +0 -247
- package/dist/src/cli/commands/observe.js.map +0 -1
- package/dist/src/cli/commands/registry.d.ts +0 -11
- package/dist/src/cli/commands/registry.d.ts.map +0 -1
- package/dist/src/cli/commands/registry.js +0 -32
- package/dist/src/cli/commands/registry.js.map +0 -1
- package/dist/src/cli/commands/sample.d.ts +0 -15
- package/dist/src/cli/commands/sample.d.ts.map +0 -1
- package/dist/src/cli/commands/sample.js +0 -415
- package/dist/src/cli/commands/sample.js.map +0 -1
- package/dist/src/cli/commands/studio.d.ts +0 -2
- package/dist/src/cli/commands/studio.d.ts.map +0 -1
- package/dist/src/cli/commands/studio.js +0 -84
- package/dist/src/cli/commands/studio.js.map +0 -1
- package/dist/src/cli/i18n-dict.d.ts +0 -54
- package/dist/src/cli/i18n-dict.d.ts.map +0 -1
- package/dist/src/cli/i18n-dict.js +0 -1125
- package/dist/src/cli/i18n-dict.js.map +0 -1
- package/dist/src/cli/i18n.d.ts +0 -24
- package/dist/src/cli/i18n.d.ts.map +0 -1
- package/dist/src/cli/i18n.js +0 -53
- package/dist/src/cli/i18n.js.map +0 -1
- package/dist/src/cli/index.d.ts.map +0 -1
- package/dist/src/cli/index.js +0 -61
- package/dist/src/cli/index.js.map +0 -1
- package/dist/src/cli/parse-run-config.d.ts +0 -90
- package/dist/src/cli/parse-run-config.d.ts.map +0 -1
- package/dist/src/cli/parse-run-config.js +0 -302
- package/dist/src/cli/parse-run-config.js.map +0 -1
- package/dist/src/cli/parse-strict.d.ts +0 -7
- package/dist/src/cli/parse-strict.d.ts.map +0 -1
- package/dist/src/cli/parse-strict.js +0 -26
- package/dist/src/cli/parse-strict.js.map +0 -1
- package/dist/src/cli/progress.d.ts.map +0 -1
- package/dist/src/cli/progress.js.map +0 -1
- package/dist/src/cli/run-tally.d.ts +0 -18
- package/dist/src/cli/run-tally.d.ts.map +0 -1
- package/dist/src/cli/run-tally.js.map +0 -1
- package/dist/src/cli/update-check.d.ts +0 -3
- package/dist/src/cli/update-check.d.ts.map +0 -1
- package/dist/src/cli/update-check.js +0 -37
- package/dist/src/cli/update-check.js.map +0 -1
- package/dist/src/doctor/health/builtin-dimensions.d.ts.map +0 -1
- package/dist/src/doctor/health/builtin-dimensions.js.map +0 -1
- package/dist/src/doctor/health/composer.d.ts.map +0 -1
- package/dist/src/doctor/health/composer.js +0 -293
- package/dist/src/doctor/health/composer.js.map +0 -1
- package/dist/src/doctor/health/dimension-registry.d.ts.map +0 -1
- package/dist/src/doctor/health/dimension-registry.js.map +0 -1
- package/dist/src/doctor/health/dimension-spec.d.ts +0 -47
- package/dist/src/doctor/health/dimension-spec.d.ts.map +0 -1
- package/dist/src/doctor/health/dimension-spec.js.map +0 -1
- package/dist/src/doctor/health/parser.d.ts +0 -27
- package/dist/src/doctor/health/parser.d.ts.map +0 -1
- package/dist/src/doctor/health/parser.js +0 -191
- package/dist/src/doctor/health/parser.js.map +0 -1
- package/dist/src/doctor/health/prompt-builder.d.ts +0 -22
- package/dist/src/doctor/health/prompt-builder.d.ts.map +0 -1
- package/dist/src/doctor/health/prompt-builder.js +0 -170
- package/dist/src/doctor/health/prompt-builder.js.map +0 -1
- package/dist/src/doctor/health/register.d.ts.map +0 -1
- package/dist/src/doctor/health/register.js.map +0 -1
- package/dist/src/doctor/html-renderer.d.ts +0 -20
- package/dist/src/doctor/html-renderer.d.ts.map +0 -1
- package/dist/src/doctor/html-renderer.js +0 -376
- package/dist/src/doctor/html-renderer.js.map +0 -1
- package/dist/src/doctor/index.d.ts.map +0 -1
- package/dist/src/doctor/index.js +0 -245
- package/dist/src/doctor/index.js.map +0 -1
- package/dist/src/doctor/preflight.d.ts.map +0 -1
- package/dist/src/doctor/preflight.js.map +0 -1
- package/dist/src/doctor/renderer.d.ts +0 -17
- package/dist/src/doctor/renderer.d.ts.map +0 -1
- package/dist/src/doctor/renderer.js +0 -123
- package/dist/src/doctor/renderer.js.map +0 -1
- package/dist/src/doctor/rules.d.ts.map +0 -1
- package/dist/src/doctor/rules.js +0 -289
- package/dist/src/doctor/rules.js.map +0 -1
- package/dist/src/eval-core/bootstrap.d.ts.map +0 -1
- package/dist/src/eval-core/bootstrap.js.map +0 -1
- package/dist/src/eval-core/cache.d.ts.map +0 -1
- package/dist/src/eval-core/cache.js.map +0 -1
- package/dist/src/eval-core/comparability.d.ts.map +0 -1
- package/dist/src/eval-core/comparability.js.map +0 -1
- package/dist/src/eval-core/dependency-checker.d.ts +0 -58
- package/dist/src/eval-core/dependency-checker.d.ts.map +0 -1
- package/dist/src/eval-core/dependency-checker.js.map +0 -1
- package/dist/src/eval-core/evaluation-execution.d.ts.map +0 -1
- package/dist/src/eval-core/evaluation-execution.js.map +0 -1
- package/dist/src/eval-core/evaluation-job.d.ts.map +0 -1
- package/dist/src/eval-core/evaluation-job.js.map +0 -1
- package/dist/src/eval-core/evaluation-reporting.d.ts.map +0 -1
- package/dist/src/eval-core/evaluation-reporting.js.map +0 -1
- package/dist/src/eval-core/execution-strategy.d.ts.map +0 -1
- package/dist/src/eval-core/execution-strategy.js +0 -165
- package/dist/src/eval-core/execution-strategy.js.map +0 -1
- package/dist/src/eval-core/fact-checker.d.ts.map +0 -1
- package/dist/src/eval-core/fact-checker.js.map +0 -1
- package/dist/src/eval-core/layer-gates.d.ts.map +0 -1
- package/dist/src/eval-core/layer-gates.js.map +0 -1
- package/dist/src/eval-core/mock-hook.cjs +0 -203
- package/dist/src/eval-core/mocks-runtime.d.ts +0 -96
- package/dist/src/eval-core/mocks-runtime.d.ts.map +0 -1
- package/dist/src/eval-core/mocks-runtime.js +0 -323
- package/dist/src/eval-core/mocks-runtime.js.map +0 -1
- package/dist/src/eval-core/schema.d.ts.map +0 -1
- package/dist/src/eval-core/schema.js.map +0 -1
- package/dist/src/eval-core/statistics.d.ts.map +0 -1
- package/dist/src/eval-core/statistics.js.map +0 -1
- package/dist/src/eval-core/task-planner.d.ts.map +0 -1
- package/dist/src/eval-core/task-planner.js.map +0 -1
- package/dist/src/eval-core/verdict.d.ts.map +0 -1
- package/dist/src/eval-core/verdict.js.map +0 -1
- package/dist/src/eval-workflows/batch-evaluation-workflow.d.ts.map +0 -1
- package/dist/src/eval-workflows/batch-evaluation-workflow.js.map +0 -1
- package/dist/src/eval-workflows/evaluation-pipeline.d.ts +0 -109
- package/dist/src/eval-workflows/evaluation-pipeline.d.ts.map +0 -1
- package/dist/src/eval-workflows/evaluation-pipeline.js +0 -373
- package/dist/src/eval-workflows/evaluation-pipeline.js.map +0 -1
- package/dist/src/eval-workflows/evaluation-preparation.d.ts.map +0 -1
- package/dist/src/eval-workflows/evaluation-preparation.js.map +0 -1
- package/dist/src/eval-workflows/run-evaluation.d.ts.map +0 -1
- package/dist/src/eval-workflows/run-evaluation.js +0 -528
- package/dist/src/eval-workflows/run-evaluation.js.map +0 -1
- package/dist/src/executors/anthropic-api.d.ts.map +0 -1
- package/dist/src/executors/anthropic-api.js.map +0 -1
- package/dist/src/executors/claude-cli.d.ts.map +0 -1
- package/dist/src/executors/claude-cli.js +0 -179
- package/dist/src/executors/claude-cli.js.map +0 -1
- package/dist/src/executors/claude-sdk-trace.d.ts.map +0 -1
- package/dist/src/executors/claude-sdk-trace.js.map +0 -1
- package/dist/src/executors/claude-sdk.d.ts.map +0 -1
- package/dist/src/executors/claude-sdk.js.map +0 -1
- package/dist/src/executors/codex-cli-trace.d.ts.map +0 -1
- package/dist/src/executors/codex-cli-trace.js.map +0 -1
- package/dist/src/executors/codex-cli.d.ts.map +0 -1
- package/dist/src/executors/codex-cli.js.map +0 -1
- package/dist/src/executors/codex-sdk.d.ts.map +0 -1
- package/dist/src/executors/codex-sdk.js.map +0 -1
- package/dist/src/executors/gemini.d.ts.map +0 -1
- package/dist/src/executors/gemini.js.map +0 -1
- package/dist/src/executors/index.d.ts.map +0 -1
- package/dist/src/executors/index.js.map +0 -1
- package/dist/src/executors/openai-api.d.ts.map +0 -1
- package/dist/src/executors/openai-api.js.map +0 -1
- package/dist/src/executors/runtime-fingerprint.d.ts.map +0 -1
- package/dist/src/executors/runtime-fingerprint.js.map +0 -1
- package/dist/src/executors/script.d.ts.map +0 -1
- package/dist/src/executors/script.js.map +0 -1
- package/dist/src/executors/shared.d.ts +0 -194
- package/dist/src/executors/shared.d.ts.map +0 -1
- package/dist/src/executors/shared.js +0 -256
- package/dist/src/executors/shared.js.map +0 -1
- package/dist/src/grading/assertions.d.ts.map +0 -1
- package/dist/src/grading/assertions.js.map +0 -1
- package/dist/src/grading/debias-validate.d.ts.map +0 -1
- package/dist/src/grading/debias-validate.js.map +0 -1
- package/dist/src/grading/diagnostic.d.ts.map +0 -1
- package/dist/src/grading/diagnostic.js.map +0 -1
- package/dist/src/grading/gold-cli.d.ts.map +0 -1
- package/dist/src/grading/gold-cli.js.map +0 -1
- package/dist/src/grading/gold-dataset.d.ts.map +0 -1
- package/dist/src/grading/gold-dataset.js.map +0 -1
- package/dist/src/grading/human-gold.d.ts.map +0 -1
- package/dist/src/grading/human-gold.js.map +0 -1
- package/dist/src/grading/index.d.ts.map +0 -1
- package/dist/src/grading/index.js.map +0 -1
- package/dist/src/grading/judge.d.ts.map +0 -1
- package/dist/src/grading/judge.js.map +0 -1
- package/dist/src/grading/layered-scores.d.ts.map +0 -1
- package/dist/src/grading/layered-scores.js.map +0 -1
- package/dist/src/inputs/eval-config.d.ts.map +0 -1
- package/dist/src/inputs/eval-config.js.map +0 -1
- package/dist/src/inputs/load-samples.d.ts.map +0 -1
- package/dist/src/inputs/load-samples.js.map +0 -1
- package/dist/src/inputs/mcp-resolver.d.ts.map +0 -1
- package/dist/src/inputs/mcp-resolver.js.map +0 -1
- package/dist/src/inputs/skill-loader.d.ts.map +0 -1
- package/dist/src/inputs/skill-loader.js +0 -253
- package/dist/src/inputs/skill-loader.js.map +0 -1
- package/dist/src/inputs/url-fetcher.d.ts.map +0 -1
- package/dist/src/inputs/url-fetcher.js.map +0 -1
- package/dist/src/observability/experience.d.ts +0 -260
- package/dist/src/observability/experience.d.ts.map +0 -1
- package/dist/src/observability/experience.js +0 -1233
- package/dist/src/observability/experience.js.map +0 -1
- package/dist/src/observability/inbox-view-model.d.ts +0 -27
- package/dist/src/observability/inbox-view-model.d.ts.map +0 -1
- package/dist/src/observability/inbox-view-model.js +0 -135
- package/dist/src/observability/inbox-view-model.js.map +0 -1
- package/dist/src/observability/inbox.d.ts +0 -125
- package/dist/src/observability/inbox.d.ts.map +0 -1
- package/dist/src/observability/inbox.js +0 -741
- package/dist/src/observability/inbox.js.map +0 -1
- package/dist/src/observability/problem-patterns.d.ts +0 -51
- package/dist/src/observability/problem-patterns.d.ts.map +0 -1
- package/dist/src/observability/problem-patterns.js +0 -205
- package/dist/src/observability/problem-patterns.js.map +0 -1
- package/dist/src/observability/review-state.d.ts +0 -54
- package/dist/src/observability/review-state.d.ts.map +0 -1
- package/dist/src/observability/review-state.js +0 -131
- package/dist/src/observability/review-state.js.map +0 -1
- package/dist/src/observability/skill-chain-advisories.d.ts +0 -25
- package/dist/src/observability/skill-chain-advisories.d.ts.map +0 -1
- package/dist/src/observability/skill-chain-advisories.js +0 -69
- package/dist/src/observability/skill-chain-advisories.js.map +0 -1
- package/dist/src/observability/skill-chain.d.ts +0 -61
- package/dist/src/observability/skill-chain.d.ts.map +0 -1
- package/dist/src/observability/skill-chain.js +0 -233
- package/dist/src/observability/skill-chain.js.map +0 -1
- package/dist/src/observability/skill-health-analyzer.d.ts.map +0 -1
- package/dist/src/observability/skill-health-analyzer.js.map +0 -1
- package/dist/src/observability/text-signals.d.ts +0 -7
- package/dist/src/observability/text-signals.d.ts.map +0 -1
- package/dist/src/observability/text-signals.js +0 -22
- package/dist/src/observability/text-signals.js.map +0 -1
- package/dist/src/observability/trace-adapter.d.ts.map +0 -1
- package/dist/src/observability/trace-adapter.js.map +0 -1
- package/dist/src/observability/trace-attribution.d.ts +0 -56
- package/dist/src/observability/trace-attribution.d.ts.map +0 -1
- package/dist/src/observability/trace-attribution.js +0 -200
- package/dist/src/observability/trace-attribution.js.map +0 -1
- package/dist/src/observability/trace-segmenter.d.ts +0 -52
- package/dist/src/observability/trace-segmenter.d.ts.map +0 -1
- package/dist/src/observability/trace-segmenter.js +0 -320
- package/dist/src/observability/trace-segmenter.js.map +0 -1
- package/dist/src/observability/trace-source.d.ts +0 -99
- package/dist/src/observability/trace-source.d.ts.map +0 -1
- package/dist/src/observability/trace-source.js +0 -492
- package/dist/src/observability/trace-source.js.map +0 -1
- package/dist/src/renderer/html-renderer.d.ts.map +0 -1
- package/dist/src/renderer/html-renderer.js.map +0 -1
- package/dist/src/renderer/layout.d.ts.map +0 -1
- package/dist/src/renderer/layout.js.map +0 -1
- package/dist/src/renderer/observation-inbox-renderer.d.ts +0 -4
- package/dist/src/renderer/observation-inbox-renderer.d.ts.map +0 -1
- package/dist/src/renderer/observation-inbox-renderer.js +0 -5376
- package/dist/src/renderer/observation-inbox-renderer.js.map +0 -1
- package/dist/src/renderer/skill-detail-renderer.d.ts +0 -15
- package/dist/src/renderer/skill-detail-renderer.d.ts.map +0 -1
- package/dist/src/renderer/skill-detail-renderer.js +0 -1234
- package/dist/src/renderer/skill-detail-renderer.js.map +0 -1
- package/dist/src/renderer/skill-health-renderer.d.ts.map +0 -1
- package/dist/src/renderer/skill-health-renderer.js.map +0 -1
- package/dist/src/renderer/skill-list-renderer.d.ts +0 -4
- package/dist/src/renderer/skill-list-renderer.d.ts.map +0 -1
- package/dist/src/renderer/skill-list-renderer.js +0 -327
- package/dist/src/renderer/skill-list-renderer.js.map +0 -1
- package/dist/src/renderer/summary.d.ts +0 -68
- package/dist/src/renderer/summary.d.ts.map +0 -1
- package/dist/src/renderer/summary.js +0 -1896
- package/dist/src/renderer/summary.js.map +0 -1
- package/dist/src/renderer/table.d.ts.map +0 -1
- package/dist/src/renderer/table.js.map +0 -1
- package/dist/src/renderer/test-view.d.ts.map +0 -1
- package/dist/src/renderer/test-view.js +0 -905
- package/dist/src/renderer/test-view.js.map +0 -1
- package/dist/src/renderer/trends.d.ts.map +0 -1
- package/dist/src/renderer/trends.js.map +0 -1
- package/dist/src/server/job-store.d.ts.map +0 -1
- package/dist/src/server/job-store.js.map +0 -1
- package/dist/src/server/report-server.d.ts.map +0 -1
- package/dist/src/server/report-server.js +0 -801
- package/dist/src/server/report-server.js.map +0 -1
- package/dist/src/server/report-store.d.ts.map +0 -1
- package/dist/src/server/report-store.js.map +0 -1
- package/dist/src/server/skill-index.d.ts +0 -77
- package/dist/src/server/skill-index.d.ts.map +0 -1
- package/dist/src/server/skill-index.js +0 -331
- package/dist/src/server/skill-index.js.map +0 -1
- package/dist/src/server/skill-insights.d.ts +0 -92
- package/dist/src/server/skill-insights.d.ts.map +0 -1
- package/dist/src/server/skill-insights.js +0 -676
- package/dist/src/server/skill-insights.js.map +0 -1
- package/dist/src/shared/hard-rules.d.ts +0 -40
- package/dist/src/shared/hard-rules.d.ts.map +0 -1
- package/dist/src/shared/hard-rules.js +0 -200
- package/dist/src/shared/hard-rules.js.map +0 -1
- package/dist/src/shared/time.d.ts.map +0 -1
- package/dist/src/shared/time.js.map +0 -1
- package/dist/src/shared/tool-search.d.ts.map +0 -1
- package/dist/src/shared/tool-search.js.map +0 -1
- package/dist/src/types/doctor.d.ts +0 -155
- package/dist/src/types/doctor.d.ts.map +0 -1
- package/dist/src/types/doctor.js.map +0 -1
- package/dist/src/types/eval.d.ts +0 -372
- package/dist/src/types/eval.d.ts.map +0 -1
- package/dist/src/types/eval.js.map +0 -1
- package/dist/src/types/executor.d.ts.map +0 -1
- package/dist/src/types/executor.js.map +0 -1
- package/dist/src/types/index.d.ts +0 -8
- package/dist/src/types/index.d.ts.map +0 -1
- package/dist/src/types/index.js +0 -8
- package/dist/src/types/index.js.map +0 -1
- package/dist/src/types/judge.d.ts.map +0 -1
- package/dist/src/types/judge.js.map +0 -1
- package/dist/src/types/report.d.ts +0 -551
- package/dist/src/types/report.d.ts.map +0 -1
- package/dist/src/types/report.js.map +0 -1
- package/dist/src/types/shared.d.ts.map +0 -1
- package/dist/src/types/shared.js.map +0 -1
- package/dist/src/types/storage.d.ts.map +0 -1
- package/dist/src/types/storage.js.map +0 -1
- package/dist/src/util/safe-slice.d.ts +0 -23
- package/dist/src/util/safe-slice.d.ts.map +0 -1
- package/dist/src/util/safe-slice.js +0 -33
- package/dist/src/util/safe-slice.js.map +0 -1
- /package/dist/{src/analysis → analysis}/coverage-analyzer.d.ts +0 -0
- /package/dist/{src/analysis → analysis}/coverage-analyzer.js +0 -0
- /package/dist/{src/analysis → analysis}/failure-clusterer.d.ts +0 -0
- /package/dist/{src/analysis → analysis}/failure-clusterer.js +0 -0
- /package/dist/{src/analysis → analysis}/hedging-classifier.d.ts +0 -0
- /package/dist/{src/analysis → analysis}/hedging-classifier.js +0 -0
- /package/dist/{src/analysis → analysis}/sample-diagnostics.d.ts +0 -0
- /package/dist/{src/analysis → analysis}/sample-diagnostics.js +0 -0
- /package/dist/{src/analysis → analysis}/saturation.d.ts +0 -0
- /package/dist/{src/analysis → analysis}/saturation.js +0 -0
- /package/dist/{src/cli → cli}/index.d.ts +0 -0
- /package/dist/{src/cli → cli/lib}/progress.d.ts +0 -0
- /package/dist/{src/cli → cli/lib}/progress.js +0 -0
- /package/dist/{src/cli → cli/lib}/run-tally.js +0 -0
- /package/dist/{src/doctor → doctor}/health/builtin-dimensions.d.ts +0 -0
- /package/dist/{src/doctor → doctor}/health/builtin-dimensions.js +0 -0
- /package/dist/{src/doctor → doctor}/health/composer.d.ts +0 -0
- /package/dist/{src/doctor → doctor}/health/dimension-registry.d.ts +0 -0
- /package/dist/{src/doctor → doctor}/health/dimension-registry.js +0 -0
- /package/dist/{src/doctor → doctor}/health/dimension-spec.js +0 -0
- /package/dist/{src/doctor → doctor}/health/register.d.ts +0 -0
- /package/dist/{src/doctor → doctor}/health/register.js +0 -0
- /package/dist/{src/doctor → doctor}/index.d.ts +0 -0
- /package/dist/{src/doctor → doctor}/preflight.d.ts +0 -0
- /package/dist/{src/doctor → doctor}/preflight.js +0 -0
- /package/dist/{src/doctor → doctor}/rules.d.ts +0 -0
- /package/dist/{src/eval-core → eval-core}/bootstrap.d.ts +0 -0
- /package/dist/{src/eval-core → eval-core}/bootstrap.js +0 -0
- /package/dist/{src/eval-core → eval-core}/cache.d.ts +0 -0
- /package/dist/{src/eval-core → eval-core}/cache.js +0 -0
- /package/dist/{src/eval-core → eval-core}/comparability.d.ts +0 -0
- /package/dist/{src/eval-core → eval-core}/comparability.js +0 -0
- /package/dist/{src/eval-core → eval-core}/dependency-checker.js +0 -0
- /package/dist/{src/eval-core → eval-core}/evaluation-execution.d.ts +0 -0
- /package/dist/{src/eval-core → eval-core}/evaluation-execution.js +0 -0
- /package/dist/{src/eval-core → eval-core}/evaluation-job.d.ts +0 -0
- /package/dist/{src/eval-core → eval-core}/evaluation-job.js +0 -0
- /package/dist/{src/eval-core → eval-core}/evaluation-reporting.d.ts +0 -0
- /package/dist/{src/eval-core → eval-core}/evaluation-reporting.js +0 -0
- /package/dist/{src/eval-core → eval-core}/execution-strategy.d.ts +0 -0
- /package/dist/{src/eval-core → eval-core}/fact-checker.d.ts +0 -0
- /package/dist/{src/eval-core → eval-core}/fact-checker.js +0 -0
- /package/dist/{src/eval-core → eval-core}/layer-gates.d.ts +0 -0
- /package/dist/{src/eval-core → eval-core}/layer-gates.js +0 -0
- /package/dist/{src/eval-core → eval-core}/schema.d.ts +0 -0
- /package/dist/{src/eval-core → eval-core}/schema.js +0 -0
- /package/dist/{src/eval-core → eval-core}/statistics.d.ts +0 -0
- /package/dist/{src/eval-core → eval-core}/statistics.js +0 -0
- /package/dist/{src/eval-core → eval-core}/task-planner.d.ts +0 -0
- /package/dist/{src/eval-core → eval-core}/task-planner.js +0 -0
- /package/dist/{src/eval-core → eval-core}/verdict.d.ts +0 -0
- /package/dist/{src/eval-core → eval-core}/verdict.js +0 -0
- /package/dist/{src/eval-workflows → eval-workflows}/batch-evaluation-workflow.d.ts +0 -0
- /package/dist/{src/eval-workflows → eval-workflows}/batch-evaluation-workflow.js +0 -0
- /package/dist/{src/eval-workflows → eval-workflows}/evaluation-preparation.d.ts +0 -0
- /package/dist/{src/eval-workflows → eval-workflows}/evaluation-preparation.js +0 -0
- /package/dist/{src/eval-workflows → eval-workflows}/run-evaluation.d.ts +0 -0
- /package/dist/{src/executors → executors}/anthropic-api.d.ts +0 -0
- /package/dist/{src/executors → executors}/anthropic-api.js +0 -0
- /package/dist/{src/executors → executors}/claude-cli.d.ts +0 -0
- /package/dist/{src/executors → executors}/claude-sdk-trace.d.ts +0 -0
- /package/dist/{src/executors → executors}/claude-sdk-trace.js +0 -0
- /package/dist/{src/executors → executors}/claude-sdk.d.ts +0 -0
- /package/dist/{src/executors → executors}/claude-sdk.js +0 -0
- /package/dist/{src/executors → executors}/codex-cli-trace.d.ts +0 -0
- /package/dist/{src/executors → executors}/codex-cli-trace.js +0 -0
- /package/dist/{src/executors → executors}/codex-cli.d.ts +0 -0
- /package/dist/{src/executors → executors}/codex-cli.js +0 -0
- /package/dist/{src/executors → executors}/codex-sdk.d.ts +0 -0
- /package/dist/{src/executors → executors}/codex-sdk.js +0 -0
- /package/dist/{src/executors → executors}/gemini.d.ts +0 -0
- /package/dist/{src/executors → executors}/gemini.js +0 -0
- /package/dist/{src/executors → executors}/index.d.ts +0 -0
- /package/dist/{src/executors → executors}/index.js +0 -0
- /package/dist/{src/executors → executors}/openai-api.d.ts +0 -0
- /package/dist/{src/executors → executors}/openai-api.js +0 -0
- /package/dist/{src/executors → executors}/runtime-fingerprint.d.ts +0 -0
- /package/dist/{src/executors → executors}/runtime-fingerprint.js +0 -0
- /package/dist/{src/executors → executors}/script.d.ts +0 -0
- /package/dist/{src/executors → executors}/script.js +0 -0
- /package/dist/{src/grading → grading}/assertions.d.ts +0 -0
- /package/dist/{src/grading → grading}/assertions.js +0 -0
- /package/dist/{src/grading → grading}/debias-validate.d.ts +0 -0
- /package/dist/{src/grading → grading}/debias-validate.js +0 -0
- /package/dist/{src/grading → grading}/diagnostic.d.ts +0 -0
- /package/dist/{src/grading → grading}/diagnostic.js +0 -0
- /package/dist/{src/grading → grading}/gold-cli.d.ts +0 -0
- /package/dist/{src/grading → grading}/gold-cli.js +0 -0
- /package/dist/{src/grading → grading}/gold-dataset.d.ts +0 -0
- /package/dist/{src/grading → grading}/gold-dataset.js +0 -0
- /package/dist/{src/grading → grading}/human-gold.d.ts +0 -0
- /package/dist/{src/grading → grading}/human-gold.js +0 -0
- /package/dist/{src/grading → grading}/index.d.ts +0 -0
- /package/dist/{src/grading → grading}/index.js +0 -0
- /package/dist/{src/grading → grading}/judge.d.ts +0 -0
- /package/dist/{src/grading → grading}/judge.js +0 -0
- /package/dist/{src/grading → grading}/layered-scores.d.ts +0 -0
- /package/dist/{src/grading → grading}/layered-scores.js +0 -0
- /package/dist/{src/inputs → inputs}/eval-config.d.ts +0 -0
- /package/dist/{src/inputs → inputs}/eval-config.js +0 -0
- /package/dist/{src/inputs → inputs}/load-samples.d.ts +0 -0
- /package/dist/{src/inputs → inputs}/load-samples.js +0 -0
- /package/dist/{src/inputs → inputs}/mcp-resolver.d.ts +0 -0
- /package/dist/{src/inputs → inputs}/mcp-resolver.js +0 -0
- /package/dist/{src/inputs → inputs}/skill-loader.d.ts +0 -0
- /package/dist/{src/inputs → inputs}/url-fetcher.d.ts +0 -0
- /package/dist/{src/inputs → inputs}/url-fetcher.js +0 -0
- /package/dist/{src/observability → observability}/skill-health-analyzer.d.ts +0 -0
- /package/dist/{src/observability → observability}/skill-health-analyzer.js +0 -0
- /package/dist/{src/observability → observability}/trace-adapter.d.ts +0 -0
- /package/dist/{src/observability → observability}/trace-adapter.js +0 -0
- /package/dist/{src/renderer → renderer}/html-renderer.d.ts +0 -0
- /package/dist/{src/renderer → renderer}/html-renderer.js +0 -0
- /package/dist/{src/renderer → renderer}/layout.d.ts +0 -0
- /package/dist/{src/renderer → renderer}/layout.js +0 -0
- /package/dist/{src/renderer → renderer}/skill-health-renderer.d.ts +0 -0
- /package/dist/{src/renderer → renderer}/skill-health-renderer.js +0 -0
- /package/dist/{src/renderer → renderer}/table.d.ts +0 -0
- /package/dist/{src/renderer → renderer}/table.js +0 -0
- /package/dist/{src/renderer → renderer}/test-view.d.ts +0 -0
- /package/dist/{src/renderer → renderer}/trends.d.ts +0 -0
- /package/dist/{src/renderer → renderer}/trends.js +0 -0
- /package/dist/{src/server → server}/job-store.d.ts +0 -0
- /package/dist/{src/server → server}/job-store.js +0 -0
- /package/dist/{src/server → server}/report-server.d.ts +0 -0
- /package/dist/{src/server → server}/report-store.d.ts +0 -0
- /package/dist/{src/server → server}/report-store.js +0 -0
- /package/dist/{src/shared → shared}/time.d.ts +0 -0
- /package/dist/{src/shared → shared}/time.js +0 -0
- /package/dist/{src/shared → shared}/tool-search.d.ts +0 -0
- /package/dist/{src/shared → shared}/tool-search.js +0 -0
- /package/dist/{src/types → types}/doctor.js +0 -0
- /package/dist/{src/types → types}/eval.js +0 -0
- /package/dist/{src/types → types}/executor.d.ts +0 -0
- /package/dist/{src/types → types}/executor.js +0 -0
- /package/dist/{src/types → types}/judge.d.ts +0 -0
- /package/dist/{src/types → types}/judge.js +0 -0
- /package/dist/{src/types → types}/report.js +0 -0
- /package/dist/{src/types → types}/shared.d.ts +0 -0
- /package/dist/{src/types → types}/shared.js +0 -0
- /package/dist/{src/types → types}/storage.d.ts +0 -0
- /package/dist/{src/types → types}/storage.js +0 -0
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"task-planner.js","sourceRoot":"","sources":["../../src/eval-core/task-planner.ts"],"names":[],"mappings":"AAEA;;;;;;GAMG;AACH,MAAM,UAAU,wBAAwB,CAAC,GAAkC;IACzE,IAAI,CAAC,GAAG;QAAE,OAAO,IAAI,CAAC;IACtB,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,IAAI,GAAG,CAAC,aAAa,IAAI,GAAG,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACtD,KAAK,CAAC,IAAI,CAAC,2DAA2D,CAAC,CAAC;QACxE,KAAK,MAAM,CAAC,IAAI,GAAG,CAAC,aAAa;YAAE,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC;IAChE,CAAC;IACD,IAAI,GAAG,CAAC,eAAe,IAAI,GAAG,CAAC,eAAe,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC1D,KAAK,CAAC,IAAI,CAAC,2CAA2C,CAAC,CAAC;QACxD,KAAK,MAAM,CAAC,IAAI,GAAG,CAAC,eAAe;YAAE,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC;IAClE,CAAC;IACD,IAAI,GAAG,CAAC,KAAK,IAAI,GAAG,CAAC,KAAK,CAAC,IAAI,EAAE,EAAE,CAAC;QAClC,KAAK,CAAC,IAAI,CAAC,QAAQ,GAAG,CAAC,KAAK,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;IACzC,CAAC;IACD,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IACpC,OAAO;QACL,qBAAqB;QACrB,EAAE;QACF,GAAG,KAAK;QACR,EAAE;QACF,6EAA6E;KAC9E,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACf,CAAC;AAED,MAAM,UAAU,UAAU,CAAC,OAAiB,EAAE,QAAkB,EAAE,MAAqC;IACrG,MAAM,SAAS,GAAe,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACvD,IAAI,EAAE,OAAO;QACb,IAAI,EAAE,OAAO,KAAK,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,OAAO;QACnD,MAAM,EAAE,OAAO,KAAK,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,QAAQ;QACtD,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,IAAI,IAAI;KACjC,CAAC,CAAC,CAAC;IACJ,OAAO,uBAAuB,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;AACrD,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,OAAiB,EAAE,SAAqB;IAC9E,MAAM,KAAK,GAAW,EAAE,CAAC;IAEzB,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;YACjC,sEAAsE;YACtE,MAAM,UAAU,GAAG,wBAAwB,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC;YAChE,MAAM,cAAc,GAAG,MAAM,CAAC,OAAO;gBACnC,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,eAAe,MAAM,CAAC,OAAO,UAAU;gBACzD,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC;YAClB,MAAM,UAAU,GAAG,UAAU;gBAC3B,CAAC,CAAC,GAAG,UAAU,cAAc,cAAc,EAAE;gBAC7C,CAAC,CAAC,cAAc,CAAC;YAEnB,KAAK,CAAC,IAAI,CAAC;gBACT,SAAS,EAAE,MAAM,CAAC,SAAS;gBAC3B,OAAO,EAAE,QAAQ,CAAC,IAAI;gBACtB,QAAQ;gBACR,MAAM,EAAE,UAAU;gBAClB,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,IAAI;gBAC7B,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;gBACrC,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;gBACrC,eAAe,EAAE,QAAQ,CAAC,OAAO;gBACjC,GAAG,EAAE,QAAQ,CAAC,GAAG,IAAI,QAAQ,CAAC,SAAS,IAAI,MAAM,CAAC,GAAG,IAAI,IAAI;gBAC7D,OAAO,EAAE,MAAM;aAChB,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,OAAO,KAAK,CAAC;AACf,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"verdict.d.ts","sourceRoot":"","sources":["../../src/eval-core/verdict.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAEH,OAAO,KAAK,EAAE,MAAM,EAAyC,MAAM,mBAAmB,CAAC;AAGvF,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;AAsND;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,aAAa,EAAE,OAAO,GAAE;IAAE,OAAO,CAAC,EAAE,OAAO,CAAA;CAAO,GAAG,MAAM,CAiBpG"}
|
|
@@ -0,0 +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"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"batch-evaluation-workflow.d.ts","sourceRoot":"","sources":["../../src/eval-workflows/batch-evaluation-workflow.ts"],"names":[],"mappings":"AAMA,OAAO,KAAK,EACV,QAAQ,EACR,qBAAqB,EAErB,gBAAgB,EAEhB,QAAQ,EACR,gBAAgB,EAEjB,MAAM,mBAAmB,CAAC;AAE3B,UAAU,0BAA0B;IAClC,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,OAAO,EAAE,OAAO,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,EAAE,IAAI,CAAC;IACf,UAAU,EAAE,KAAK,CAAC;IAClB,UAAU,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACpC,gBAAgB,EAAE,OAAO,CAAC;IAC1B,yEAAyE;IACzE,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,IAAI,CAAC,EAAE,IAAI,GAAG,IAAI,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,kFAAkF;IAClF,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,8EAA8E;IAC9E,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,0CAA0C;IAC1C,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,8DAA8D;IAC9D,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,6BAA6B;IAC7B,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,uDAAuD;IACvD,MAAM,CAAC,EAAE,OAAO,mBAAmB,EAAE,UAAU,CAAC;CACjD;AAED,UAAU,sBAAsB;IAC9B,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,MAAM,EAAE,gBAAgB,CAAC;IACzB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;CACzB;AA+DD,wBAAgB,0BAA0B,CAAC,EACzC,UAAU,EACV,QAAQ,EACR,YAAY,EACZ,YAAY,EACZ,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,OAAO,EACP,KAAK,EACL,IAAI,EACJ,WAAW,EACX,SAAS,EACT,YAAY,EACZ,MAAM,EACN,WAAW,EACX,OAAO,EACP,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,GACP,EAAE;IACD,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC9E,YAAY,EAAE,sBAAsB,EAAE,CAAC;IACvC,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,OAAO,EAAE,OAAO,CAAC;IACjB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,MAAM,CAAC,EAAE,OAAO,mBAAmB,EAAE,UAAU,CAAC;CACjD,GAAG;IAAE,MAAM,EAAE,qBAAqB,CAAC;IAAC,GAAG,EAAE,OAAO,mBAAmB,EAAE,aAAa,CAAA;CAAE,CA4FpF;AAED,wBAAsB,0BAA0B,CAAC,EAC/C,QAAQ,EACR,YAAY,EACZ,KAAK,EACL,UAAU,EACV,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,WAAe,EACf,SAAS,EACT,YAAY,EACZ,iBAAiB,EACjB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,eAAsB,EACtB,gBAAwB,EACxB,UAAkB,EAClB,IAAW,EACX,SAAS,EACT,OAAe,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,YAAY,EACZ,SAAS,EACT,gBAAgB,EAChB,MAAM,EACN,OAAe,EACf,cAAc,EACd,oBAAoB,EACpB,mBAAmB,GACpB,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC9E,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACrC,eAAe,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;IAC5G,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B,kCAAkC;IAClC,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,IAAI,CAAC,EAAE,IAAI,GAAG,IAAI,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,MAAM,CAAC,EAAE,OAAO,mBAAmB,EAAE,UAAU,CAAC;IAChD,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,wEAAwE;IACxE,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,mDAAmD;IACnD,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;IAChD,mBAAmB,EAAE,CAAC,OAAO,EAAE,0BAA0B,KAAK,OAAO,CAAC;QAAE,MAAM,EAAE,gBAAgB,CAAC;QAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;KAAE,CAAC,CAAC;CAC9H,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,qBAAqB,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CA8FtE"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"batch-evaluation-workflow.js","sourceRoot":"","sources":["../../src/eval-workflows/batch-evaluation-workflow.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAC7C,OAAO,EAAE,kBAAkB,EAAE,aAAa,EAAE,aAAa,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACnI,OAAO,EAAE,sBAAsB,EAAE,mBAAmB,EAAE,kBAAkB,EAAE,qBAAqB,EAAE,MAAM,gCAAgC,CAAC;AACxI,OAAO,EAAE,6BAA6B,EAAE,MAAM,qCAAqC,CAAC;AACpF,OAAO,EAAE,kBAAkB,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC9E,OAAO,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AAyD7D,SAAS,aAAa,CAAC,QAAoD;IACzE,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,aAAa,CAAC,KAAa;IAClC,OAAO,KAAK;SACT,UAAU,CAAC,SAAS,EAAE,GAAG,CAAC;SAC1B,UAAU,CAAC,mBAAmB,EAAE,GAAG,CAAC;SACpC,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC;WACrB,OAAO,CAAC;AACf,CAAC;AAED,SAAS,qBAAqB,CAAC,MAAwB,EAAE,gBAAwB;IAC/E,OAAO;QACL,QAAQ,EAAE,MAAM,CAAC,OAAO,CAAC,QAAQ,IAAK,EAAqB;QAC3D,CAAC,gBAAgB,CAAC,EAAE,MAAM,CAAC,OAAO,CAAC,gBAAgB,CAAC,IAAK,EAAqB;KAC/E,CAAC;AACJ,CAAC;AAED,SAAS,eAAe,CAAC,YAAsC;IAC7D,OAAO,YAAY,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,WAAW,EAAE,MAAM,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC;QAC/E,IAAI;QACJ,SAAS;QACT,WAAW;QACX,QAAQ,EAAE,MAAM,CAAC,EAAE;QACnB,UAAU,EAAE,QAAQ;QACpB,MAAM,EAAE,WAAW;QACnB,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,WAAW;QACpC,YAAY,EAAE,MAAM,CAAC,IAAI,CAAC,YAAY;QACtC,YAAY,EAAE,MAAM,CAAC,IAAI,CAAC,cAAc,EAAE,CAAC,IAAI,CAAC,IAAI,IAAI;QACxD,OAAO,EAAE,qBAAqB,CAAC,MAAM,EAAE,IAAI,CAAC;QAC5C,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC1D,CAAC,CAAC,CAAC;AACN,CAAC;AAED,SAAS,4BAA4B,CAAC,EACpC,QAAQ,EACR,YAAY,EACZ,YAAY,EACZ,KAAK,GAMN;IACC,MAAM,gBAAgB,GAA+C,EAAE,CAAC;IACxE,KAAK,MAAM,KAAK,IAAI,YAAY,EAAE,CAAC;QACjC,gBAAgB,CAAC,KAAK,CAAC,IAAI,CAAC;YAC1B,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,gBAAgB,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC;mBAC7C,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,eAAe;mBACjC,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE;oBACpD,QAAQ,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,QAAQ;iBAChE,CAAC,CAAC;IACP,CAAC;IACD,OAAO,gBAAgB,CAAC;AAC1B,CAAC;AAED,MAAM,UAAU,0BAA0B,CAAC,EACzC,UAAU,EACV,QAAQ,EACR,YAAY,EACZ,YAAY,EACZ,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,OAAO,EACP,KAAK,EACL,IAAI,EACJ,WAAW,EACX,SAAS,EACT,YAAY,EACZ,MAAM,EACN,WAAW,EACX,OAAO,EACP,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,GAwBP;IACC,MAAM,oBAAoB,GAAG,WAAW,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC;QAChE,CAAC,CAAC,WAAW;QACb,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,iBAAiB,IAAI,YAAY,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;IACzE,MAAM,OAAO,GAAG,sBAAsB,CAAC;QACrC,WAAW,EAAE,EAAE;QACf,QAAQ;QACR,SAAS,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;YACtC,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,IAAI,EAAE,OAAO;YACb,MAAM,EAAE,WAAW;YACnB,OAAO,EAAE,IAAI;YACb,OAAO,EAAE,KAAK,CAAC,SAAS;SACzB,CAAC,CAAC;QACH,KAAK;QACL,QAAQ,EAAE,YAAY;QACtB,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO;QACP,MAAM,EAAE,KAAK;QACb,KAAK,EAAE,KAAK;QACZ,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM;QACN,WAAW,EAAE,oBAAoB;QACjC,KAAK,EAAE,IAAI;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,UAAU,EAAE,SAAS,CAAC,CAAC;IAClF,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,GAAG,GAAG,qBAAqB,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IAC1D,MAAM,GAAG,GAAG,kBAAkB,CAAC;QAC7B,KAAK,EAAE,OAAO,UAAU,EAAE;QAC1B,KAAK,EAAE,UAAU;QACjB,QAAQ,EAAE,UAAU;QACpB,OAAO;QACP,SAAS;QACT,SAAS;QACT,UAAU;KACX,CAAC,CAAC;IACH,MAAM,WAAW,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;IAChG,MAAM,gBAAgB,GAAG,4BAA4B,CAAC,EAAE,QAAQ,EAAE,YAAY,EAAE,YAAY,EAAE,KAAK,EAAE,CAAC,CAAC;IACvG,MAAM,eAAe,GAAG,aAAa,CAAC,gBAAgB,CAAC;WAClD,MAAM,CAAC,MAAM,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC;WAClC,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE,EAAE,QAAQ,EAAE,CAAC,CAAC;IACtE,MAAM,iBAAiB,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE,CACrD,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,iBAAiB,KAAK,KAAK;WAC1C,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CAC7D,OAAO,CAAC,gBAAgB,KAAK,KAAK,IAAI,OAAO,CAAC,iBAAiB,KAAK,KAAK,CAAC,CAAC,CAAC;IAEhF,MAAM,eAAe,GAAoD,oBAAoB,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC;QACzG,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,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC;KACpG,CAAC,CAAC,CAAC;IAEJ,MAAM,MAAM,GAA0B;QACpC,IAAI,EAAE,kBAAkB;QACxB,EAAE,EAAE,UAAU;QACd,IAAI,EAAE,OAAO;QACb,IAAI,EAAE;YACJ,IAAI,EAAE,OAAO;YACb,KAAK;YACL,QAAQ,EAAE,YAAY;YACtB,QAAQ;YACR,WAAW;YACX,SAAS,EAAE,WAAW,GAAG,CAAC;YAC1B,cAAc,EAAE,YAAY,CAAC,MAAM;YACnC,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,UAAU;YACrB,UAAU,EAAE,aAAa,EAAE;YAC3B,WAAW,EAAE,OAAO,CAAC,OAAO;YAC5B,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,OAAO;YACP,GAAG;YACH,GAAG;YACH,OAAO,EAAE,UAAU,EAAE;SACtB;QACD,KAAK,EAAE,eAAe,CAAC,YAAY,CAAC;KACrC,CAAC;IAEF,OAAO,EAAE,MAAM,EAAE,GAAG,EAAE,CAAC;AACzB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,0BAA0B,CAAC,EAC/C,QAAQ,EACR,YAAY,EACZ,KAAK,EACL,UAAU,EACV,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAO,GAAG,KAAK,EACf,WAAW,GAAG,CAAC,EACf,SAAS,EACT,YAAY,EACZ,iBAAiB,EACjB,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,eAAe,GAAG,IAAI,EACtB,gBAAgB,GAAG,KAAK,EACxB,UAAU,GAAG,KAAK,EAClB,IAAI,GAAG,IAAI,EACX,SAAS,EACT,OAAO,GAAG,KAAK,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,YAAY,EACZ,SAAS,EACT,gBAAgB,EAChB,MAAM,EACN,OAAO,GAAG,KAAK,EACf,cAAc,EACd,oBAAoB,EACpB,mBAAmB,GAsCpB;IACC,MAAM,UAAU,GAAG,aAAa,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC;IAC5C,MAAM,YAAY,GAA6B,EAAE,CAAC;IAClD,IAAI,YAAY,GAAG,CAAC,CAAC;IAErB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,YAAY,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC7C,MAAM,KAAK,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC;QAC9B,eAAe,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,YAAY,CAAC,MAAM,EAAE,CAAC,CAAC;QAErG,8DAA8D;QAC9D,MAAM,qBAAqB,GAAG,oBAAoB,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,SAAS;YAC5E,CAAC,CAAC,EAAE,GAAG,oBAAoB,EAAE,CAAC,KAAK,CAAC,SAAS,CAAC,EAAE,oBAAoB,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE;YAClF,CAAC,CAAC,oBAAoB,CAAC;QACzB,MAAM,cAAc,GAAG,gBAAgB,CACrC,OAAO,CAAC,QAAQ,CAAC,EACjB,CAAC,UAAU,EAAE,KAAK,CAAC,SAAS,CAAC,EAC7B,EAAE,cAAc,EAAE,oBAAoB,EAAE,qBAAqB,EAAE,CAChE,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE;YACjB,IAAI,QAAQ,CAAC,IAAI,KAAK,KAAK,CAAC,SAAS,EAAE,CAAC;gBACtC,OAAO,EAAE,GAAG,QAAQ,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,cAAc,EAAE,WAAoB,EAAE,CAAC;YACjF,CAAC;YACD,OAAO,EAAE,GAAG,QAAQ,EAAE,cAAc,EAAE,SAAkB,EAAE,CAAC;QAC7D,CAAC,CAAC,CAAC;QACH,MAAM,UAAU,GAAG,GAAG,UAAU,IAAI,MAAM,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,IAAI,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;QAClG,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,GAAG,MAAM,mBAAmB,CAAC;YACrD,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,QAAQ;YACR,SAAS,EAAE,cAAc;YACzB,KAAK;YACL,UAAU;YACV,SAAS;YACT,OAAO;YACP,WAAW;YACX,SAAS;YACT,OAAO;YACP,YAAY;YACZ,iBAAiB;YACjB,QAAQ,EAAE,IAAI;YACd,UAAU,EAAE,KAAK;YACjB,UAAU;YACV,gBAAgB,EAAE,gBAAgB,IAAI,CAAC,GAAG,CAAC;YAC3C,UAAU;YACV,IAAI;YACJ,SAAS;YACT,OAAO;YACP,KAAK,EAAE,UAAU;YACjB,WAAW;YACX,WAAW;YACX,YAAY;YACZ,SAAS;YACT,gBAAgB;YAChB,MAAM;SACP,CAAC,CAAC;QAEH,YAAY,CAAC,IAAI,CAAC;YAChB,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,MAAM;YACN,QAAQ;SACT,CAAC,CAAC;QAEH,YAAY,IAAI,MAAM,CAAC,IAAI,CAAC,YAAY,CAAC;QACzC,eAAe,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,YAAY,CAAC,MAAM,EAAE,CAAC,CAAC;IACtG,CAAC;IAED,MAAM,EAAE,MAAM,EAAE,WAAW,EAAE,GAAG,EAAE,GAAG,0BAA0B,CAAC;QAC9D,UAAU;QACV,QAAQ;QACR,YAAY;QACZ,YAAY;QACZ,KAAK;QACL,UAAU;QACV,OAAO;QACP,YAAY;QACZ,iBAAiB;QACjB,OAAO;QACP,KAAK;QACL,IAAI;QACJ,WAAW;QACX,SAAS;QACT,YAAY;QACZ,MAAM;QACN,WAAW;QACX,OAAO;QACP,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;KACP,CAAC,CAAC;IACH,MAAM,QAAQ,GAAG,aAAa,CAAC,WAAW,EAAE,SAAS,CAAC,CAAC;IACvD,MAAM,gBAAgB,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,QAAQ,IAAI,kBAAkB,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAChG,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IAClE,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,QAAQ,EAAE,CAAC;AAC3C,CAAC"}
|
|
@@ -0,0 +1,22 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* preflight 阶段的结构性 / 隔离性警告。
|
|
3
|
+
*
|
|
4
|
+
* 两组互不相关的 warning,共用「执行前 stderr 提示,不是 verdict gate」的语义:
|
|
5
|
+
* - power: 基于 sample 数 / repeat 的硬底线(n<5 不可靠, n<20 仅能测大效应,
|
|
6
|
+
* repeat<2 无法测稳定性)。**不是 MDE / power analysis 预测** —— run 之前
|
|
7
|
+
* 还没有 σ,预测「CI 半宽 ~ ±0.4」是手挥;真正的 power 判定在 post-hoc
|
|
8
|
+
* verdict(UNDERPOWERED state + saturation curve)
|
|
9
|
+
* - isolation: 用户显式 `--no-strict-baseline` 且存在 baseline-kind variant 且
|
|
10
|
+
* `~/.claude/skills/` 里有内容 —— baseline 会被 SDK 全发现污染,verdict / Δ
|
|
11
|
+
* 不可信。默认 strict 时不出 warn
|
|
12
|
+
*
|
|
13
|
+
* `buildXxxWarnings` 是纯函数,export 给单测断言文案语义;`emitXxxWarnings` 是
|
|
14
|
+
* 写 stderr 的薄包装,只给 orchestrator 用。
|
|
15
|
+
*/
|
|
16
|
+
import type { Artifact } from '../../types/index.js';
|
|
17
|
+
import type { Lang } from '../../types/shared.js';
|
|
18
|
+
export declare function buildPowerWarnings(sampleCount: number, repeat: number, lang?: Lang): string[];
|
|
19
|
+
export declare function emitPowerWarnings(sampleCount: number, repeat: number, lang: Lang): void;
|
|
20
|
+
export declare function buildIsolationWarnings(artifacts: Artifact[], strictBaseline: boolean | undefined): string[];
|
|
21
|
+
export declare function emitIsolationWarnings(artifacts: Artifact[], strictBaseline: boolean | undefined): void;
|
|
22
|
+
//# sourceMappingURL=preflight-warnings.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"preflight-warnings.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline/preflight-warnings.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAKH,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,sBAAsB,CAAC;AACrD,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,uBAAuB,CAAC;AAGlD,wBAAgB,kBAAkB,CAAC,WAAW,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,GAAE,IAAW,GAAG,MAAM,EAAE,CAWnG;AAED,wBAAgB,iBAAiB,CAAC,WAAW,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,GAAG,IAAI,CAIvF;AAED,wBAAgB,sBAAsB,CACpC,SAAS,EAAE,QAAQ,EAAE,EACrB,cAAc,EAAE,OAAO,GAAG,SAAS,GAClC,MAAM,EAAE,CAsBV;AAED,wBAAgB,qBAAqB,CAAC,SAAS,EAAE,QAAQ,EAAE,EAAE,cAAc,EAAE,OAAO,GAAG,SAAS,GAAG,IAAI,CAItG"}
|
|
@@ -0,0 +1,67 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* preflight 阶段的结构性 / 隔离性警告。
|
|
3
|
+
*
|
|
4
|
+
* 两组互不相关的 warning,共用「执行前 stderr 提示,不是 verdict gate」的语义:
|
|
5
|
+
* - power: 基于 sample 数 / repeat 的硬底线(n<5 不可靠, n<20 仅能测大效应,
|
|
6
|
+
* repeat<2 无法测稳定性)。**不是 MDE / power analysis 预测** —— run 之前
|
|
7
|
+
* 还没有 σ,预测「CI 半宽 ~ ±0.4」是手挥;真正的 power 判定在 post-hoc
|
|
8
|
+
* verdict(UNDERPOWERED state + saturation curve)
|
|
9
|
+
* - isolation: 用户显式 `--no-strict-baseline` 且存在 baseline-kind variant 且
|
|
10
|
+
* `~/.claude/skills/` 里有内容 —— baseline 会被 SDK 全发现污染,verdict / Δ
|
|
11
|
+
* 不可信。默认 strict 时不出 warn
|
|
12
|
+
*
|
|
13
|
+
* `buildXxxWarnings` 是纯函数,export 给单测断言文案语义;`emitXxxWarnings` 是
|
|
14
|
+
* 写 stderr 的薄包装,只给 orchestrator 用。
|
|
15
|
+
*/
|
|
16
|
+
import { existsSync, readdirSync } from 'node:fs';
|
|
17
|
+
import { homedir } from 'node:os';
|
|
18
|
+
import { join } from 'node:path';
|
|
19
|
+
import { tEvalWorkflowMessage } from '../messages.js';
|
|
20
|
+
export function buildPowerWarnings(sampleCount, repeat, lang = 'zh') {
|
|
21
|
+
const warnings = [];
|
|
22
|
+
if (sampleCount < 5) {
|
|
23
|
+
warnings.push(tEvalWorkflowMessage('power_warning_tiny_n', lang, { n: sampleCount }));
|
|
24
|
+
}
|
|
25
|
+
else if (sampleCount < 20) {
|
|
26
|
+
warnings.push(tEvalWorkflowMessage('power_warning_small_n', lang, { n: sampleCount }));
|
|
27
|
+
}
|
|
28
|
+
if (repeat < 2) {
|
|
29
|
+
warnings.push(tEvalWorkflowMessage('power_warning_repeat_one', lang));
|
|
30
|
+
}
|
|
31
|
+
return warnings;
|
|
32
|
+
}
|
|
33
|
+
export function emitPowerWarnings(sampleCount, repeat, lang) {
|
|
34
|
+
for (const w of buildPowerWarnings(sampleCount, repeat, lang)) {
|
|
35
|
+
process.stderr.write(`${w}\n`);
|
|
36
|
+
}
|
|
37
|
+
}
|
|
38
|
+
export function buildIsolationWarnings(artifacts, strictBaseline) {
|
|
39
|
+
// Only warn when user explicitly disabled isolation.
|
|
40
|
+
if (strictBaseline !== false)
|
|
41
|
+
return [];
|
|
42
|
+
const hasBaselineKind = artifacts.some((a) => a.kind === 'baseline');
|
|
43
|
+
if (!hasBaselineKind)
|
|
44
|
+
return [];
|
|
45
|
+
// Check ~/.claude/skills/ for content (avoid hard-coding home — read at runtime).
|
|
46
|
+
const skillsDir = join(homedir(), '.claude', 'skills');
|
|
47
|
+
if (!existsSync(skillsDir))
|
|
48
|
+
return [];
|
|
49
|
+
let skillCount = 0;
|
|
50
|
+
try {
|
|
51
|
+
skillCount = readdirSync(skillsDir).filter((entry) => !entry.startsWith('.')).length;
|
|
52
|
+
}
|
|
53
|
+
catch {
|
|
54
|
+
return [];
|
|
55
|
+
}
|
|
56
|
+
if (skillCount === 0)
|
|
57
|
+
return [];
|
|
58
|
+
return [
|
|
59
|
+
`⚠ baseline 隔离已关闭(--no-strict-baseline)。检测到 ~/.claude/skills/ 内有 ${skillCount} 个 skill, baseline variant 可能被 auto-discovery 污染。除非你确认要这种比较,建议恢复默认 strict 模式。`,
|
|
60
|
+
];
|
|
61
|
+
}
|
|
62
|
+
export function emitIsolationWarnings(artifacts, strictBaseline) {
|
|
63
|
+
for (const w of buildIsolationWarnings(artifacts, strictBaseline)) {
|
|
64
|
+
process.stderr.write(`${w}\n`);
|
|
65
|
+
}
|
|
66
|
+
}
|
|
67
|
+
//# sourceMappingURL=preflight-warnings.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"preflight-warnings.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline/preflight-warnings.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,EAAE,UAAU,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAClD,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAGjC,OAAO,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAC;AAEtD,MAAM,UAAU,kBAAkB,CAAC,WAAmB,EAAE,MAAc,EAAE,OAAa,IAAI;IACvF,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;QACpB,QAAQ,CAAC,IAAI,CAAC,oBAAoB,CAAC,sBAAsB,EAAE,IAAI,EAAE,EAAE,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC;IACxF,CAAC;SAAM,IAAI,WAAW,GAAG,EAAE,EAAE,CAAC;QAC5B,QAAQ,CAAC,IAAI,CAAC,oBAAoB,CAAC,uBAAuB,EAAE,IAAI,EAAE,EAAE,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC;IACzF,CAAC;IACD,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;QACf,QAAQ,CAAC,IAAI,CAAC,oBAAoB,CAAC,0BAA0B,EAAE,IAAI,CAAC,CAAC,CAAC;IACxE,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,WAAmB,EAAE,MAAc,EAAE,IAAU;IAC/E,KAAK,MAAM,CAAC,IAAI,kBAAkB,CAAC,WAAW,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,CAAC;QAC9D,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACjC,CAAC;AACH,CAAC;AAED,MAAM,UAAU,sBAAsB,CACpC,SAAqB,EACrB,cAAmC;IAEnC,qDAAqD;IACrD,IAAI,cAAc,KAAK,KAAK;QAAE,OAAO,EAAE,CAAC;IAExC,MAAM,eAAe,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,UAAU,CAAC,CAAC;IACrE,IAAI,CAAC,eAAe;QAAE,OAAO,EAAE,CAAC;IAEhC,kFAAkF;IAClF,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC;IACvD,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC;QAAE,OAAO,EAAE,CAAC;IAEtC,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,CAAC;QACH,UAAU,GAAG,WAAW,CAAC,SAAS,CAAC,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC;IACvF,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;IACD,IAAI,UAAU,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAEhC,OAAO;QACL,mEAAmE,UAAU,+EAA+E;KAC7J,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,SAAqB,EAAE,cAAmC;IAC9F,KAAK,MAAM,CAAC,IAAI,sBAAsB,CAAC,SAAS,EAAE,cAAc,CAAC,EAAE,CAAC;QAClE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACjC,CAAC;AACH,CAAC"}
|
|
@@ -0,0 +1,32 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Report 最终化 —— task execution 结果 → 完整 Report。
|
|
3
|
+
*
|
|
4
|
+
* `executeTasks` 出来的是裸结果(results / totalCostUSD),`aggregateReport` 把
|
|
5
|
+
* 它们包成 Report 骨架。`finalizeEvaluationReport` 在这之上挂三层后置分析:
|
|
6
|
+
* - `report.analysis` 主体: diagnostics + sampleQuality(传入 samples 用于
|
|
7
|
+
* capability / difficulty / construct / provenance 覆盖率聚合)
|
|
8
|
+
* - 可选 coverage: 仅当结果里含 tool trace 数据时计算
|
|
9
|
+
* - 可选 gapReports: 文本信号(markers / hedging)始终适用,与 tool trace 无关;
|
|
10
|
+
* 带上 testSetHash 水印(spec §7.1 强制要求)
|
|
11
|
+
*
|
|
12
|
+
* 最后一步 `applyBlindMode` 是 blind 模式下的字段脱敏,放在所有 analysis 之后,
|
|
13
|
+
* 避免脱敏后字段被分析逻辑读取。
|
|
14
|
+
*
|
|
15
|
+
* 仅被 orchestrator 调用;独立拆出主要为让 orchestrator 的 try-finally 主干
|
|
16
|
+
* 看起来纯粹是「执行→收尾」时序。
|
|
17
|
+
*/
|
|
18
|
+
import type { Artifact, Report, Sample, VariantResult } from '../../types/index.js';
|
|
19
|
+
type EvaluationResults = Record<string, Record<string, VariantResult>>;
|
|
20
|
+
export declare function finalizeEvaluationReport({ report, results, artifacts, variantNames, blind, samplesPath, samplesSourceFiles, samples, }: {
|
|
21
|
+
report: Report;
|
|
22
|
+
results: EvaluationResults;
|
|
23
|
+
artifacts: Artifact[];
|
|
24
|
+
variantNames: string[];
|
|
25
|
+
blind: boolean;
|
|
26
|
+
samplesPath: string;
|
|
27
|
+
/** 目录模式下,bundle 内所有源文件;单文件模式下 [samplesPath]。computeTestSetHash 用。 */
|
|
28
|
+
samplesSourceFiles?: string[];
|
|
29
|
+
samples: Sample[];
|
|
30
|
+
}): Report;
|
|
31
|
+
export {};
|
|
32
|
+
//# sourceMappingURL=report-finalize.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"report-finalize.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline/report-finalize.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAMH,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,EAAE,MAAM,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AAGpF,KAAK,iBAAiB,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC,CAAC;AAEvE,wBAAgB,wBAAwB,CAAC,EACvC,MAAM,EACN,OAAO,EACP,SAAS,EACT,YAAY,EACZ,KAAK,EACL,WAAW,EACX,kBAAkB,EAClB,OAAO,GACR,EAAE;IACD,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,iBAAiB,CAAC;IAC3B,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,KAAK,EAAE,OAAO,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;IACpB,qEAAqE;IACrE,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC9B,OAAO,EAAE,MAAM,EAAE,CAAC;CACnB,GAAG,MAAM,CAsCT"}
|
|
@@ -0,0 +1,57 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Report 最终化 —— task execution 结果 → 完整 Report。
|
|
3
|
+
*
|
|
4
|
+
* `executeTasks` 出来的是裸结果(results / totalCostUSD),`aggregateReport` 把
|
|
5
|
+
* 它们包成 Report 骨架。`finalizeEvaluationReport` 在这之上挂三层后置分析:
|
|
6
|
+
* - `report.analysis` 主体: diagnostics + sampleQuality(传入 samples 用于
|
|
7
|
+
* capability / difficulty / construct / provenance 覆盖率聚合)
|
|
8
|
+
* - 可选 coverage: 仅当结果里含 tool trace 数据时计算
|
|
9
|
+
* - 可选 gapReports: 文本信号(markers / hedging)始终适用,与 tool trace 无关;
|
|
10
|
+
* 带上 testSetHash 水印(spec §7.1 强制要求)
|
|
11
|
+
*
|
|
12
|
+
* 最后一步 `applyBlindMode` 是 blind 模式下的字段脱敏,放在所有 analysis 之后,
|
|
13
|
+
* 避免脱敏后字段被分析逻辑读取。
|
|
14
|
+
*
|
|
15
|
+
* 仅被 orchestrator 调用;独立拆出主要为让 orchestrator 的 try-finally 主干
|
|
16
|
+
* 看起来纯粹是「执行→收尾」时序。
|
|
17
|
+
*/
|
|
18
|
+
import { analyzeResults } from '../../analysis/report-diagnostics.js';
|
|
19
|
+
import { computeReportCoverage } from '../../analysis/coverage-analyzer.js';
|
|
20
|
+
import { computeReportGapRates } from '../../analysis/gap-analyzer.js';
|
|
21
|
+
import { applyBlindMode } from '../../eval-core/evaluation-reporting.js';
|
|
22
|
+
import { computeTestSetHash } from './test-set-hash.js';
|
|
23
|
+
export function finalizeEvaluationReport({ report, results, artifacts, variantNames, blind, samplesPath, samplesSourceFiles, samples, }) {
|
|
24
|
+
// pass samples so analyzeResults can populate analysis.sampleQuality
|
|
25
|
+
// (capability/difficulty/construct/provenance coverage aggregate). Without
|
|
26
|
+
// samples, analysis.sampleQuality is omitted (老报告读取仍可工作).
|
|
27
|
+
report.analysis = analyzeResults(report, { samples });
|
|
28
|
+
const hasToolData = Object.values(results).some((sampleResults) => (Object.values(sampleResults).some((variantResult) => variantResult.toolCalls && variantResult.toolCalls.length > 0)));
|
|
29
|
+
if (hasToolData) {
|
|
30
|
+
const artifactContents = Object.fromEntries(artifacts.map((artifact) => [artifact.name, artifact.content]));
|
|
31
|
+
const artifactCwds = Object.fromEntries(artifacts.map((artifact) => [artifact.name, artifact.cwd || null]));
|
|
32
|
+
const coverage = computeReportCoverage(report, artifactContents, artifactCwds);
|
|
33
|
+
if (Object.keys(coverage).length > 0) {
|
|
34
|
+
report.analysis.coverage = coverage;
|
|
35
|
+
}
|
|
36
|
+
}
|
|
37
|
+
// Gap rate computation runs on every successful report regardless of whether
|
|
38
|
+
// tool trace data is present — text-based signals (markers, hedging) still
|
|
39
|
+
// apply. The samples-file SHA is the mandatory watermark required by spec §7.1.
|
|
40
|
+
const gapReports = computeReportGapRates(report.results, variantNames);
|
|
41
|
+
if (Object.keys(gapReports).length > 0) {
|
|
42
|
+
const testSetHash = computeTestSetHash(samplesPath, samplesSourceFiles);
|
|
43
|
+
for (const variant of variantNames) {
|
|
44
|
+
const gr = gapReports[variant];
|
|
45
|
+
if (!gr)
|
|
46
|
+
continue;
|
|
47
|
+
gr.testSetPath = samplesPath;
|
|
48
|
+
gr.testSetHash = testSetHash;
|
|
49
|
+
}
|
|
50
|
+
report.analysis.gapReports = gapReports;
|
|
51
|
+
}
|
|
52
|
+
if (blind) {
|
|
53
|
+
applyBlindMode(report, variantNames, `${variantNames.join(',')}:${samplesPath}`);
|
|
54
|
+
}
|
|
55
|
+
return report;
|
|
56
|
+
}
|
|
57
|
+
//# sourceMappingURL=report-finalize.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"report-finalize.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline/report-finalize.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,EAAE,cAAc,EAAE,MAAM,sCAAsC,CAAC;AACtE,OAAO,EAAE,qBAAqB,EAAE,MAAM,qCAAqC,CAAC;AAC5E,OAAO,EAAE,qBAAqB,EAAE,MAAM,gCAAgC,CAAC;AACvE,OAAO,EAAE,cAAc,EAAE,MAAM,yCAAyC,CAAC;AAEzE,OAAO,EAAE,kBAAkB,EAAE,MAAM,oBAAoB,CAAC;AAIxD,MAAM,UAAU,wBAAwB,CAAC,EACvC,MAAM,EACN,OAAO,EACP,SAAS,EACT,YAAY,EACZ,KAAK,EACL,WAAW,EACX,kBAAkB,EAClB,OAAO,GAWR;IACC,qEAAqE;IACrE,2EAA2E;IAC3E,0DAA0D;IAC1D,MAAM,CAAC,QAAQ,GAAG,cAAc,CAAC,MAAM,EAAE,EAAE,OAAO,EAAE,CAAC,CAAC;IAEtD,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,EAAE,kBAAkB,CAAC,CAAC;QACxE,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"}
|
|
@@ -0,0 +1,60 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* evaluation pipeline 的运行状态生命周期。
|
|
3
|
+
*
|
|
4
|
+
* 把 `EvaluationRunState` + 4 个 helper(initialize / finalizeSuccessfulRun /
|
|
5
|
+
* persistSuccessfulJob / persistFailedJob)从 evaluation-pipeline.ts 拆出来,
|
|
6
|
+
* 让 orchestrator 主体回归「编排」单职责。这些 helper 之间是线性时序耦合
|
|
7
|
+
* (init → preflight → executeTasks → finalize → persist),向上只被
|
|
8
|
+
* `executeEvaluationPipeline` 调用,向下只依赖 `eval-core/evaluation-job` 与
|
|
9
|
+
* `server/job-store` 的纯 job-store 抽象,与 task execution / report assembly
|
|
10
|
+
* 解耦。
|
|
11
|
+
*
|
|
12
|
+
* 不再单独 export 给测试 —— job 生命周期通过 orchestrator 间接验证,无独立单测。
|
|
13
|
+
*/
|
|
14
|
+
import type { Artifact, EvaluationJob, EvaluationRequest, EvaluationRun, JobStore } from '../../types/index.js';
|
|
15
|
+
export interface EvaluationRunState {
|
|
16
|
+
request: EvaluationRequest;
|
|
17
|
+
runId: string;
|
|
18
|
+
jobId: string;
|
|
19
|
+
createdAt: string;
|
|
20
|
+
startedAt: string;
|
|
21
|
+
initialRun: EvaluationRun;
|
|
22
|
+
runningJob: EvaluationJob;
|
|
23
|
+
resolvedJobStore: JobStore | null;
|
|
24
|
+
}
|
|
25
|
+
export declare function initializeEvaluationRunState({ samplesPath, skillDir, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, concurrency, timeoutMs, noCache, blind, project, owner, tags, runId, jobStore, persistJob, repeat, batch, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, effort, }: {
|
|
26
|
+
samplesPath: string;
|
|
27
|
+
skillDir: string;
|
|
28
|
+
artifacts: Artifact[];
|
|
29
|
+
model: string;
|
|
30
|
+
judgeModel: string;
|
|
31
|
+
noJudge: boolean;
|
|
32
|
+
executorName: string;
|
|
33
|
+
judgeExecutorName: string;
|
|
34
|
+
concurrency: number;
|
|
35
|
+
timeoutMs?: number;
|
|
36
|
+
noCache: boolean;
|
|
37
|
+
blind: boolean;
|
|
38
|
+
project?: string;
|
|
39
|
+
owner?: string;
|
|
40
|
+
tags?: string[];
|
|
41
|
+
runId: string;
|
|
42
|
+
jobStore?: JobStore | null;
|
|
43
|
+
persistJob?: boolean;
|
|
44
|
+
repeat?: number;
|
|
45
|
+
batch?: boolean;
|
|
46
|
+
judgeRepeat?: number;
|
|
47
|
+
judgeModels?: import('../../types/index.js').JudgeConfig[];
|
|
48
|
+
bootstrap?: boolean;
|
|
49
|
+
bootstrapSamples?: number;
|
|
50
|
+
lengthDebias?: boolean;
|
|
51
|
+
budget?: import('../../types/index.js').EvalBudget;
|
|
52
|
+
effort?: 'low' | 'medium' | 'high' | 'xhigh' | 'max';
|
|
53
|
+
}): Promise<EvaluationRunState>;
|
|
54
|
+
export declare function finalizeSuccessfulRun(state: EvaluationRunState): {
|
|
55
|
+
run: EvaluationRun;
|
|
56
|
+
job: EvaluationJob;
|
|
57
|
+
};
|
|
58
|
+
export declare function persistSuccessfulJob(state: EvaluationRunState, job: EvaluationJob): Promise<void>;
|
|
59
|
+
export declare function persistFailedJob(state: EvaluationRunState, err: unknown): Promise<void>;
|
|
60
|
+
//# sourceMappingURL=run-state.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"run-state.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline/run-state.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAgBH,OAAO,KAAK,EACV,QAAQ,EACR,aAAa,EACb,iBAAiB,EACjB,aAAa,EACb,QAAQ,EACT,MAAM,sBAAsB,CAAC;AAE9B,MAAM,WAAW,kBAAkB;IACjC,OAAO,EAAE,iBAAiB,CAAC;IAC3B,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,aAAa,CAAC;IAC1B,UAAU,EAAE,aAAa,CAAC;IAC1B,gBAAgB,EAAE,QAAQ,GAAG,IAAI,CAAC;CACnC;AAED,wBAAsB,4BAA4B,CAAC,EACjD,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,KAAK,EACL,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,EACN,MAAM,GACP,EAAE;IACD,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,KAAK,EAAE,OAAO,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,OAAO,sBAAsB,EAAE,WAAW,EAAE,CAAC;IAC3D,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,MAAM,CAAC,EAAE,OAAO,sBAAsB,EAAE,UAAU,CAAC;IACnD,MAAM,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,GAAG,KAAK,CAAC;CACtD,GAAG,OAAO,CAAC,kBAAkB,CAAC,CAsC9B;AAED,wBAAgB,qBAAqB,CAAC,KAAK,EAAE,kBAAkB;;;EAa9D;AAED,wBAAsB,oBAAoB,CAAC,KAAK,EAAE,kBAAkB,EAAE,GAAG,EAAE,aAAa,GAAG,OAAO,CAAC,IAAI,CAAC,CAIvG;AAED,wBAAsB,gBAAgB,CAAC,KAAK,EAAE,kBAAkB,EAAE,GAAG,EAAE,OAAO,GAAG,OAAO,CAAC,IAAI,CAAC,CAW7F"}
|
|
@@ -0,0 +1,88 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* evaluation pipeline 的运行状态生命周期。
|
|
3
|
+
*
|
|
4
|
+
* 把 `EvaluationRunState` + 4 个 helper(initialize / finalizeSuccessfulRun /
|
|
5
|
+
* persistSuccessfulJob / persistFailedJob)从 evaluation-pipeline.ts 拆出来,
|
|
6
|
+
* 让 orchestrator 主体回归「编排」单职责。这些 helper 之间是线性时序耦合
|
|
7
|
+
* (init → preflight → executeTasks → finalize → persist),向上只被
|
|
8
|
+
* `executeEvaluationPipeline` 调用,向下只依赖 `eval-core/evaluation-job` 与
|
|
9
|
+
* `server/job-store` 的纯 job-store 抽象,与 task execution / report assembly
|
|
10
|
+
* 解耦。
|
|
11
|
+
*
|
|
12
|
+
* 不再单独 export 给测试 —— job 生命周期通过 orchestrator 间接验证,无独立单测。
|
|
13
|
+
*/
|
|
14
|
+
import { buildEvaluationRequest, createFailedJob, createEvaluationRun, createQueuedJob, createSucceededJob, finalizeEvaluationRun, markJobRunning, failEvaluationRun, } from '../../eval-core/evaluation-job.js';
|
|
15
|
+
import { createFileJobStore, DEFAULT_JOBS_DIR, } from '../../server/job-store.js';
|
|
16
|
+
export async function initializeEvaluationRunState({ samplesPath, skillDir, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, concurrency, timeoutMs, noCache, blind, project, owner, tags, runId, jobStore, persistJob, repeat, batch, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, effort, }) {
|
|
17
|
+
const effectiveJudges = judgeModels && judgeModels.length > 0
|
|
18
|
+
? judgeModels
|
|
19
|
+
: [{ executor: judgeExecutorName, model: judgeModel }];
|
|
20
|
+
const request = buildEvaluationRequest({
|
|
21
|
+
samplesPath,
|
|
22
|
+
skillDir,
|
|
23
|
+
artifacts,
|
|
24
|
+
model,
|
|
25
|
+
executor: executorName,
|
|
26
|
+
noJudge,
|
|
27
|
+
concurrency,
|
|
28
|
+
timeoutMs,
|
|
29
|
+
noCache,
|
|
30
|
+
dryRun: false,
|
|
31
|
+
blind,
|
|
32
|
+
project,
|
|
33
|
+
owner,
|
|
34
|
+
tags,
|
|
35
|
+
repeat,
|
|
36
|
+
batch,
|
|
37
|
+
judgeRepeat,
|
|
38
|
+
judgeModels: effectiveJudges,
|
|
39
|
+
bootstrap,
|
|
40
|
+
bootstrapSamples,
|
|
41
|
+
lengthDebias,
|
|
42
|
+
budget,
|
|
43
|
+
effort,
|
|
44
|
+
});
|
|
45
|
+
const createdAt = new Date().toISOString();
|
|
46
|
+
const { run: initialRun, startedAt } = createEvaluationRun(runId, createdAt);
|
|
47
|
+
const jobId = `job-${runId}`;
|
|
48
|
+
const resolvedJobStore = persistJob ? (jobStore ?? createFileJobStore(DEFAULT_JOBS_DIR)) : null;
|
|
49
|
+
const queuedJob = createQueuedJob({ jobId, request, createdAt });
|
|
50
|
+
if (resolvedJobStore)
|
|
51
|
+
await resolvedJobStore.save(jobId, queuedJob);
|
|
52
|
+
const runningJob = markJobRunning(queuedJob, runId, startedAt);
|
|
53
|
+
if (resolvedJobStore)
|
|
54
|
+
await resolvedJobStore.save(jobId, runningJob);
|
|
55
|
+
return { request, runId, jobId, createdAt, startedAt, initialRun, runningJob, resolvedJobStore };
|
|
56
|
+
}
|
|
57
|
+
export function finalizeSuccessfulRun(state) {
|
|
58
|
+
const finishedAt = new Date().toISOString();
|
|
59
|
+
const run = finalizeEvaluationRun(state.initialRun, finishedAt);
|
|
60
|
+
const job = createSucceededJob({
|
|
61
|
+
jobId: state.jobId,
|
|
62
|
+
runId: state.runId,
|
|
63
|
+
reportId: state.runId,
|
|
64
|
+
request: state.request,
|
|
65
|
+
createdAt: state.createdAt,
|
|
66
|
+
startedAt: state.startedAt,
|
|
67
|
+
finishedAt,
|
|
68
|
+
});
|
|
69
|
+
return { run, job };
|
|
70
|
+
}
|
|
71
|
+
export async function persistSuccessfulJob(state, job) {
|
|
72
|
+
if (state.resolvedJobStore) {
|
|
73
|
+
await state.resolvedJobStore.save(state.jobId, job);
|
|
74
|
+
}
|
|
75
|
+
}
|
|
76
|
+
export async function persistFailedJob(state, err) {
|
|
77
|
+
const finishedAt = new Date().toISOString();
|
|
78
|
+
const failedJob = createFailedJob({
|
|
79
|
+
job: { ...state.runningJob, runId: state.runId, startedAt: state.startedAt, finishedAt: undefined },
|
|
80
|
+
error: err instanceof Error ? err.message : String(err),
|
|
81
|
+
finishedAt,
|
|
82
|
+
});
|
|
83
|
+
void failEvaluationRun(state.initialRun, finishedAt);
|
|
84
|
+
if (state.resolvedJobStore) {
|
|
85
|
+
await state.resolvedJobStore.save(state.jobId, failedJob);
|
|
86
|
+
}
|
|
87
|
+
}
|
|
88
|
+
//# sourceMappingURL=run-state.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"run-state.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline/run-state.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EACL,sBAAsB,EACtB,eAAe,EACf,mBAAmB,EACnB,eAAe,EACf,kBAAkB,EAClB,qBAAqB,EACrB,cAAc,EACd,iBAAiB,GAClB,MAAM,mCAAmC,CAAC;AAC3C,OAAO,EACL,kBAAkB,EAClB,gBAAgB,GACjB,MAAM,2BAA2B,CAAC;AAoBnC,MAAM,CAAC,KAAK,UAAU,4BAA4B,CAAC,EACjD,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,KAAK,EACL,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,EACN,MAAM,GA6BP;IACC,MAAM,eAAe,GAAiD,WAAW,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC;QACzG,CAAC,CAAC,WAAW;QACb,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,iBAAiB,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;IACzD,MAAM,OAAO,GAAG,sBAAsB,CAAC;QACrC,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,QAAQ,EAAE,YAAY;QACtB,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO;QACP,MAAM,EAAE,KAAK;QACb,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM;QACN,KAAK;QACL,WAAW;QACX,WAAW,EAAE,eAAe;QAC5B,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;QACN,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,MAAM,UAAU,qBAAqB,CAAC,KAAyB;IAC7D,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,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,KAAyB,EAAE,GAAkB;IACtF,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,MAAM,CAAC,KAAK,UAAU,gBAAgB,CAAC,KAAyB,EAAE,GAAY;IAC5E,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"}
|
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* 测试集水印 hash 计算(spec §7.1)。
|
|
3
|
+
*
|
|
4
|
+
* 规则:
|
|
5
|
+
* - 单文件 / sourceFiles 单元素:hash 该文件内容
|
|
6
|
+
* - 多文件(目录模式):按文件名排序后,把每个文件 basename + sha256 拼成
|
|
7
|
+
* manifest,对 manifest 再 sha256,取前 12 hex chars。这样:
|
|
8
|
+
* 1. 加 / 删 sample 文件 → hash 变
|
|
9
|
+
* 2. 任一文件内容改 → hash 变
|
|
10
|
+
* 3. 同样文件不同顺序 → hash 不变(排序后稳定)
|
|
11
|
+
*
|
|
12
|
+
* 历史问题: 之前对目录 readFileSync() 抛 EISDIR → 返回 null,gap report 水印丢失。
|
|
13
|
+
*
|
|
14
|
+
* 导出:
|
|
15
|
+
* - `computeTestSetHash`: pipeline 内部使用
|
|
16
|
+
* - `_computeTestSetHashForTest`: 单测专用别名,保留以兼容历史 import surface
|
|
17
|
+
* (`test/eval-workflows/test-set-hash.test.ts`)
|
|
18
|
+
*/
|
|
19
|
+
export declare function computeTestSetHash(samplesPath: string, sourceFiles?: string[]): string | null;
|
|
20
|
+
export declare function _computeTestSetHashForTest(samplesPath: string, sourceFiles?: string[]): string | null;
|
|
21
|
+
//# sourceMappingURL=test-set-hash.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"test-set-hash.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline/test-set-hash.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAKH,wBAAgB,kBAAkB,CAAC,WAAW,EAAE,MAAM,EAAE,WAAW,CAAC,EAAE,MAAM,EAAE,GAAG,MAAM,GAAG,IAAI,CAmB7F;AAED,wBAAgB,0BAA0B,CAAC,WAAW,EAAE,MAAM,EAAE,WAAW,CAAC,EAAE,MAAM,EAAE,GAAG,MAAM,GAAG,IAAI,CAErG"}
|
|
@@ -0,0 +1,46 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* 测试集水印 hash 计算(spec §7.1)。
|
|
3
|
+
*
|
|
4
|
+
* 规则:
|
|
5
|
+
* - 单文件 / sourceFiles 单元素:hash 该文件内容
|
|
6
|
+
* - 多文件(目录模式):按文件名排序后,把每个文件 basename + sha256 拼成
|
|
7
|
+
* manifest,对 manifest 再 sha256,取前 12 hex chars。这样:
|
|
8
|
+
* 1. 加 / 删 sample 文件 → hash 变
|
|
9
|
+
* 2. 任一文件内容改 → hash 变
|
|
10
|
+
* 3. 同样文件不同顺序 → hash 不变(排序后稳定)
|
|
11
|
+
*
|
|
12
|
+
* 历史问题: 之前对目录 readFileSync() 抛 EISDIR → 返回 null,gap report 水印丢失。
|
|
13
|
+
*
|
|
14
|
+
* 导出:
|
|
15
|
+
* - `computeTestSetHash`: pipeline 内部使用
|
|
16
|
+
* - `_computeTestSetHashForTest`: 单测专用别名,保留以兼容历史 import surface
|
|
17
|
+
* (`test/eval-workflows/test-set-hash.test.ts`)
|
|
18
|
+
*/
|
|
19
|
+
import { createHash } from 'node:crypto';
|
|
20
|
+
import { existsSync, readFileSync } from 'node:fs';
|
|
21
|
+
export function computeTestSetHash(samplesPath, sourceFiles) {
|
|
22
|
+
// 优先 sourceFiles(目录模式可靠),fallback samplesPath 兼容老调用
|
|
23
|
+
const files = sourceFiles && sourceFiles.length > 0
|
|
24
|
+
? [...sourceFiles].sort()
|
|
25
|
+
: (samplesPath && existsSync(samplesPath) ? [samplesPath] : []);
|
|
26
|
+
if (files.length === 0)
|
|
27
|
+
return null;
|
|
28
|
+
try {
|
|
29
|
+
if (files.length === 1) {
|
|
30
|
+
return createHash('sha256').update(readFileSync(files[0], 'utf-8')).digest('hex').slice(0, 12);
|
|
31
|
+
}
|
|
32
|
+
const manifest = files.map((f) => {
|
|
33
|
+
const base = f.split(/[/\\]/).pop() ?? f;
|
|
34
|
+
const inner = createHash('sha256').update(readFileSync(f, 'utf-8')).digest('hex');
|
|
35
|
+
return `${base}:${inner}`;
|
|
36
|
+
}).join('\n');
|
|
37
|
+
return createHash('sha256').update(manifest).digest('hex').slice(0, 12);
|
|
38
|
+
}
|
|
39
|
+
catch {
|
|
40
|
+
return null;
|
|
41
|
+
}
|
|
42
|
+
}
|
|
43
|
+
export function _computeTestSetHashForTest(samplesPath, sourceFiles) {
|
|
44
|
+
return computeTestSetHash(samplesPath, sourceFiles);
|
|
45
|
+
}
|
|
46
|
+
//# sourceMappingURL=test-set-hash.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"test-set-hash.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline/test-set-hash.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AAEnD,MAAM,UAAU,kBAAkB,CAAC,WAAmB,EAAE,WAAsB;IAC5E,oDAAoD;IACpD,MAAM,KAAK,GAAG,WAAW,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC;QACjD,CAAC,CAAC,CAAC,GAAG,WAAW,CAAC,CAAC,IAAI,EAAE;QACzB,CAAC,CAAC,CAAC,WAAW,IAAI,UAAU,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAClE,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IACpC,IAAI,CAAC;QACH,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACvB,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QACjG,CAAC;QACD,MAAM,QAAQ,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;YAC/B,MAAM,IAAI,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;YACzC,MAAM,KAAK,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,YAAY,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YAClF,OAAO,GAAG,IAAI,IAAI,KAAK,EAAE,CAAC;QAC5B,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACd,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IAC1E,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,MAAM,UAAU,0BAA0B,CAAC,WAAmB,EAAE,WAAsB;IACpF,OAAO,kBAAkB,CAAC,WAAW,EAAE,WAAW,CAAC,CAAC;AACtD,CAAC"}
|
|
@@ -0,0 +1,97 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* evaluation pipeline orchestrator —— 单次 evaluation run 的「执行+收尾」编排。
|
|
3
|
+
*
|
|
4
|
+
* 主入口 `executeEvaluationPipeline` 把以下几段串成线性时序:
|
|
5
|
+
* 1. `initializeEvaluationRunState` 建 run / job 元数据并写 jobStore
|
|
6
|
+
* 2. resolve judge models + executor map(必须先于 preflight,见函数内注释)
|
|
7
|
+
* 3. `preflight` + `preflightAllJudges` LLM 连通性
|
|
8
|
+
* 4. `emitPowerWarnings` + `emitIsolationWarnings` 结构性 / 隔离性预警
|
|
9
|
+
* 5. `executeTasks` 真正跑用例
|
|
10
|
+
* 6. `finalizeSuccessfulRun` + `finalizeEvaluationReport` 收尾 + 后置分析
|
|
11
|
+
* 7. `persistReport` + `persistSuccessfulJob` 落盘
|
|
12
|
+
* 失败路径走 `persistFailedJob`, finally 关 MCP server。
|
|
13
|
+
*
|
|
14
|
+
* 4 块被拆出去的 helper 都在 `./evaluation-pipeline/` 目录下,与本文件 1:1 协作:
|
|
15
|
+
* - run-state.ts: EvaluationRunState 生命周期(init / finalizeSuccess / persistSuccess / persistFailed)
|
|
16
|
+
* - test-set-hash.ts: 测试集水印 hash(spec §7.1)
|
|
17
|
+
* - preflight-warnings.ts: power + isolation 预警
|
|
18
|
+
* - report-finalize.ts: report 后置分析装配
|
|
19
|
+
*
|
|
20
|
+
* 兼容 re-export: `buildPowerWarnings` / `buildIsolationWarnings` /
|
|
21
|
+
* `_computeTestSetHashForTest` 保留在本模块的 export surface,避免 test 与
|
|
22
|
+
* `run-evaluation.ts:275` 动态 import 路径改动。
|
|
23
|
+
*/
|
|
24
|
+
import type { DependencyRequirements } from '../eval-core/dependency-checker.js';
|
|
25
|
+
import type { Artifact, ExecutorFn, JobStore, ProgressCallback, Report, Sample, Task } from '../types/index.js';
|
|
26
|
+
import type { Lang } from '../types/shared.js';
|
|
27
|
+
export { buildPowerWarnings, buildIsolationWarnings } from './evaluation-pipeline/preflight-warnings.js';
|
|
28
|
+
export { _computeTestSetHashForTest } from './evaluation-pipeline/test-set-hash.js';
|
|
29
|
+
export interface EvaluationPipelineOptions {
|
|
30
|
+
samplesPath: string;
|
|
31
|
+
/** Sample bundle 根目录。loadSamples 输出,缺省时 fallback dirname(samplesPath)。 */
|
|
32
|
+
samplesBaseDir?: string;
|
|
33
|
+
/** Sample bundle 内合并的源文件绝对路径列表。目录模式下用于 computeTestSetHash 遍历。 */
|
|
34
|
+
samplesSourceFiles?: string[];
|
|
35
|
+
skillDir: string;
|
|
36
|
+
samples: Sample[];
|
|
37
|
+
tasks: Task[];
|
|
38
|
+
artifacts: Artifact[];
|
|
39
|
+
model: string;
|
|
40
|
+
judgeModel: string;
|
|
41
|
+
noJudge: boolean;
|
|
42
|
+
executorName: string;
|
|
43
|
+
judgeExecutorName: string;
|
|
44
|
+
executor: ExecutorFn;
|
|
45
|
+
judgeExecutor: ExecutorFn;
|
|
46
|
+
outputDir?: string | null;
|
|
47
|
+
project?: string;
|
|
48
|
+
owner?: string;
|
|
49
|
+
tags?: string[];
|
|
50
|
+
blind?: boolean;
|
|
51
|
+
concurrency?: number;
|
|
52
|
+
timeoutMs?: number;
|
|
53
|
+
noCache?: boolean;
|
|
54
|
+
jobStore?: JobStore | null;
|
|
55
|
+
persistJob?: boolean;
|
|
56
|
+
onProgress?: ProgressCallback | null;
|
|
57
|
+
skipConnectivity?: boolean;
|
|
58
|
+
verbose?: boolean;
|
|
59
|
+
retry?: number;
|
|
60
|
+
existingResults?: Record<string, Record<string, VariantResult>>;
|
|
61
|
+
requires?: DependencyRequirements;
|
|
62
|
+
layeredStats?: boolean;
|
|
63
|
+
/** 透传到 meta.request.repeat */
|
|
64
|
+
repeat?: number;
|
|
65
|
+
/** 透传到 meta.request.batch */
|
|
66
|
+
batch?: boolean;
|
|
67
|
+
/** 透传到 meta.request.judgeRepeat 与 grade(),每条 sample × dimension judge N 次 */
|
|
68
|
+
judgeRepeat?: number;
|
|
69
|
+
/** Multi-judge ensemble configs (≥ 2 entries triggers ensemble mode). */
|
|
70
|
+
judgeModels?: import('../types/index.js').JudgeConfig[];
|
|
71
|
+
/** --bootstrap. */
|
|
72
|
+
bootstrap?: boolean;
|
|
73
|
+
/** --bootstrap-samples N. Default 1000. */
|
|
74
|
+
bootstrapSamples?: number;
|
|
75
|
+
/** v0.21 length-debias toggle. Default true; --no-debias-length flips to false. */
|
|
76
|
+
lengthDebias?: boolean;
|
|
77
|
+
/** hard budget caps. */
|
|
78
|
+
budget?: import('../types/index.js').EvalBudget;
|
|
79
|
+
/** strict-baseline default state (only used to decide whether to emit
|
|
80
|
+
* isolation-disabled pre-flight warnings). True/undefined = default behavior
|
|
81
|
+
* (no warning); false = user explicitly disabled, warn if ~/.claude/skills/ has content. */
|
|
82
|
+
strictBaseline?: boolean;
|
|
83
|
+
/** Explicit persisted run id. Used by batch workflows that need stable child ids. */
|
|
84
|
+
runId?: string;
|
|
85
|
+
/** CLI output language for warnings emitted by the pipeline. */
|
|
86
|
+
lang?: Lang;
|
|
87
|
+
/** Reasoning effort 透传到 executor。默认 'low'(由 RunConfig 兜底)。 */
|
|
88
|
+
effort?: 'low' | 'medium' | 'high' | 'xhigh' | 'max';
|
|
89
|
+
/** 关闭 diagnostic。Default false。 */
|
|
90
|
+
noDiagnostic?: boolean;
|
|
91
|
+
}
|
|
92
|
+
type VariantResult = import('../types/index.js').VariantResult;
|
|
93
|
+
export declare function executeEvaluationPipeline({ samplesPath, samplesBaseDir, samplesSourceFiles, skillDir, samples, tasks, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, executor, judgeExecutor, outputDir, project, owner, tags, blind, concurrency, timeoutMs, noCache, jobStore, persistJob, onProgress, skipConnectivity, verbose, retry, existingResults, requires: _requires, layeredStats, repeat, batch, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, strictBaseline, runId, lang, effort, noDiagnostic, }: EvaluationPipelineOptions): Promise<{
|
|
94
|
+
report: Report;
|
|
95
|
+
filePath: string | null;
|
|
96
|
+
}>;
|
|
97
|
+
//# sourceMappingURL=evaluation-pipeline.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluation-pipeline.d.ts","sourceRoot":"","sources":["../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;GAsBG;AAIH,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AAEjF,OAAO,KAAK,EACV,QAAQ,EACR,UAAU,EACV,QAAQ,EACR,gBAAgB,EAChB,MAAM,EACN,MAAM,EACN,IAAI,EACL,MAAM,mBAAmB,CAAC;AAC3B,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,oBAAoB,CAAC;AAW/C,OAAO,EAAE,kBAAkB,EAAE,sBAAsB,EAAE,MAAM,6CAA6C,CAAC;AACzG,OAAO,EAAE,0BAA0B,EAAE,MAAM,wCAAwC,CAAC;AAEpF,MAAM,WAAW,yBAAyB;IACxC,WAAW,EAAE,MAAM,CAAC;IACpB,0EAA0E;IAC1E,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,iEAAiE;IACjE,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC9B,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,QAAQ,EAAE,UAAU,CAAC;IACrB,aAAa,EAAE,UAAU,CAAC;IAC1B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACrC,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,eAAe,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC,CAAC;IAChE,QAAQ,CAAC,EAAE,sBAAsB,CAAC;IAClC,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,8BAA8B;IAC9B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,6BAA6B;IAC7B,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,6EAA6E;IAC7E,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,yEAAyE;IACzE,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,mBAAmB;IACnB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,2CAA2C;IAC3C,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,mFAAmF;IACnF,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,wBAAwB;IACxB,MAAM,CAAC,EAAE,OAAO,mBAAmB,EAAE,UAAU,CAAC;IAChD;;iGAE6F;IAC7F,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,qFAAqF;IACrF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,gEAAgE;IAChE,IAAI,CAAC,EAAE,IAAI,CAAC;IACZ,8DAA8D;IAC9D,MAAM,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,GAAG,KAAK,CAAC;IACrD,mCAAmC;IACnC,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAED,KAAK,aAAa,GAAG,OAAO,mBAAmB,EAAE,aAAa,CAAC;AAE/D,wBAAsB,yBAAyB,CAAC,EAC9C,WAAW,EACX,cAAc,EACd,kBAAkB,EAClB,QAAQ,EACR,OAAO,EACP,KAAK,EACL,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,QAAQ,EACR,aAAa,EACb,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAa,EACb,WAAe,EACf,SAAS,EACT,OAAe,EACf,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,gBAAwB,EACxB,OAAe,EACf,KAAS,EACT,eAAe,EAGf,QAAQ,EAAE,SAAS,EACnB,YAAoB,EACpB,MAAM,EACN,KAAK,EACL,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAmB,EACnB,MAAM,EACN,cAAc,EACd,KAAK,EACL,IAAW,EACX,MAAM,EACN,YAAY,GACb,EAAE,yBAAyB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CA+IlF"}
|