@devflow-tools/benchmark 0.17.8 → 0.18.0
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/LICENSE +21 -0
- package/dist/accuracy-command-options.d.ts +3 -0
- package/dist/accuracy-command-options.d.ts.map +1 -0
- package/dist/accuracy-command-options.js +17 -0
- package/dist/accuracy-command-options.js.map +1 -0
- package/dist/context-symbol-quality-command.d.ts +59 -0
- package/dist/context-symbol-quality-command.d.ts.map +1 -0
- package/dist/context-symbol-quality-command.js +482 -0
- package/dist/context-symbol-quality-command.js.map +1 -0
- package/dist/evaluation-bundle-builder.d.ts +48 -0
- package/dist/evaluation-bundle-builder.d.ts.map +1 -0
- package/dist/evaluation-bundle-builder.js +496 -0
- package/dist/evaluation-bundle-builder.js.map +1 -0
- package/dist/evaluation-bundle-command.d.ts +2 -0
- package/dist/evaluation-bundle-command.d.ts.map +1 -0
- package/dist/evaluation-bundle-command.js +15 -0
- package/dist/evaluation-bundle-command.js.map +1 -0
- package/dist/evaluation-bundle.d.ts +55 -0
- package/dist/evaluation-bundle.d.ts.map +1 -0
- package/dist/evaluation-bundle.js +272 -0
- package/dist/evaluation-bundle.js.map +1 -0
- package/dist/evaluation-run-state.d.ts +23 -0
- package/dist/evaluation-run-state.d.ts.map +1 -0
- package/dist/evaluation-run-state.js +115 -0
- package/dist/evaluation-run-state.js.map +1 -0
- package/dist/evidence-quality-metrics.d.ts +15 -0
- package/dist/evidence-quality-metrics.d.ts.map +1 -1
- package/dist/evidence-quality-metrics.js +51 -1
- package/dist/evidence-quality-metrics.js.map +1 -1
- package/dist/graders/deterministic.d.ts.map +1 -1
- package/dist/graders/deterministic.js +5 -3
- package/dist/graders/deterministic.js.map +1 -1
- package/dist/index.d.ts +15 -2
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +12 -1
- package/dist/index.js.map +1 -1
- package/dist/release-quality-bundle-command.d.ts +3 -0
- package/dist/release-quality-bundle-command.d.ts.map +1 -0
- package/dist/release-quality-bundle-command.js +72 -0
- package/dist/release-quality-bundle-command.js.map +1 -0
- package/dist/release-quality-bundle-evaluator.d.ts +61 -0
- package/dist/release-quality-bundle-evaluator.d.ts.map +1 -0
- package/dist/release-quality-bundle-evaluator.js +321 -0
- package/dist/release-quality-bundle-evaluator.js.map +1 -0
- package/dist/release-quality-bundle-materializer.d.ts +79 -0
- package/dist/release-quality-bundle-materializer.d.ts.map +1 -0
- package/dist/release-quality-bundle-materializer.js +395 -0
- package/dist/release-quality-bundle-materializer.js.map +1 -0
- package/dist/release-quality-comparison.d.ts +34 -0
- package/dist/release-quality-comparison.d.ts.map +1 -0
- package/dist/release-quality-comparison.js +60 -0
- package/dist/release-quality-comparison.js.map +1 -0
- package/dist/release-quality-evaluation-command.d.ts +42 -0
- package/dist/release-quality-evaluation-command.d.ts.map +1 -0
- package/dist/release-quality-evaluation-command.js +91 -0
- package/dist/release-quality-evaluation-command.js.map +1 -0
- package/dist/release-quality-host-proof.d.ts +30 -0
- package/dist/release-quality-host-proof.d.ts.map +1 -0
- package/dist/release-quality-host-proof.js +109 -0
- package/dist/release-quality-host-proof.js.map +1 -0
- package/dist/release-quality-output-cleanup.d.ts +2 -0
- package/dist/release-quality-output-cleanup.d.ts.map +1 -0
- package/dist/release-quality-output-cleanup.js +18 -0
- package/dist/release-quality-output-cleanup.js.map +1 -0
- package/dist/release-quality-policy.d.ts +87 -0
- package/dist/release-quality-policy.d.ts.map +1 -0
- package/dist/release-quality-policy.js +179 -0
- package/dist/release-quality-policy.js.map +1 -0
- package/dist/release-quality-query-sampling.d.ts +36 -0
- package/dist/release-quality-query-sampling.d.ts.map +1 -0
- package/dist/release-quality-query-sampling.js +87 -0
- package/dist/release-quality-query-sampling.js.map +1 -0
- package/dist/release-quality-source-snapshot.d.ts +18 -0
- package/dist/release-quality-source-snapshot.d.ts.map +1 -0
- package/dist/release-quality-source-snapshot.js +46 -0
- package/dist/release-quality-source-snapshot.js.map +1 -0
- package/dist/retrieval-quality-corpus.d.ts +0 -1
- package/dist/retrieval-quality-corpus.d.ts.map +1 -1
- package/dist/retrieval-quality-corpus.js +1 -1
- package/dist/retrieval-quality-corpus.js.map +1 -1
- package/dist/retrieval-quality.d.ts.map +1 -1
- package/dist/retrieval-quality.js +29 -26
- package/dist/retrieval-quality.js.map +1 -1
- package/dist/run-accuracy-tests.js +82 -2
- package/dist/run-accuracy-tests.js.map +1 -1
- package/dist/suite-runner.js +1 -1
- package/dist/suite-runner.js.map +1 -1
- package/dist/unified-runtime-corpus.d.ts +64 -0
- package/dist/unified-runtime-corpus.d.ts.map +1 -0
- package/dist/unified-runtime-corpus.js +161 -0
- package/dist/unified-runtime-corpus.js.map +1 -0
- package/dist/unified-runtime-evaluation.d.ts +133 -0
- package/dist/unified-runtime-evaluation.d.ts.map +1 -0
- package/dist/unified-runtime-evaluation.js +388 -0
- package/dist/unified-runtime-evaluation.js.map +1 -0
- package/dist/unified-runtime-observation.d.ts +101 -0
- package/dist/unified-runtime-observation.d.ts.map +1 -0
- package/dist/unified-runtime-observation.js +2 -0
- package/dist/unified-runtime-observation.js.map +1 -0
- package/dist/unified-runtime-runner.d.ts +73 -0
- package/dist/unified-runtime-runner.d.ts.map +1 -0
- package/dist/unified-runtime-runner.js +620 -0
- package/dist/unified-runtime-runner.js.map +1 -0
- package/fixtures/evidence-runtime-corpus.json +0 -100
- package/fixtures/release-quality/bundles.json +1 -0
- package/fixtures/release-quality/calibration.json +602 -0
- package/fixtures/release-quality/ecc/corpus.json +257 -0
- package/fixtures/release-quality/ecc/input.json +62 -0
- package/fixtures/release-quality/llm-wiki/corpus.json +263 -0
- package/fixtures/release-quality/llm-wiki/input.json +72 -0
- package/fixtures/release-quality/mem0/corpus.json +288 -0
- package/fixtures/release-quality/mem0/input.json +68 -0
- package/fixtures/release-quality/projects.json +29 -0
- package/fixtures/release-quality/release-quality-policy.json +232 -0
- package/fixtures/release-quality/rsbuild-react/corpus.json +263 -0
- package/fixtures/release-quality/rsbuild-react/input.json +62 -0
- package/fixtures/retrieval-quality-corpus.json +7 -7
- package/fixtures/semantic-regex-inventory.json +3 -10
- package/fixtures/unified-runtime/baselines/2026-08-08.json +37 -0
- package/fixtures/unified-runtime/context.json +206 -0
- package/fixtures/unified-runtime/knowledge.json +224 -0
- package/fixtures/unified-runtime/memory.json +224 -0
- package/fixtures/unified-runtime/real-projects/rsbuild-react-2026-08-07/context.json +32 -0
- package/fixtures/unified-runtime/real-projects/rsbuild-react-2026-08-07/knowledge.json +32 -0
- package/fixtures/unified-runtime/real-projects/rsbuild-react-2026-08-07/memory.json +32 -0
- package/fixtures/unified-runtime/real-projects/rsbuild-react-2026-08-07/semantic.json +409 -0
- package/fixtures/unified-runtime/runs/rsbuild-react-2026-08-05.json +54 -0
- package/fixtures/unified-runtime/semantic.json +409 -0
- package/package.json +24 -6
|
@@ -153,10 +153,12 @@ function memoryOutcomeGrader() {
|
|
|
153
153
|
if (!isRecord(outcome))
|
|
154
154
|
return result('memory_outcome', 'unavailable', 'Memory outcome is unavailable.', startedAt, []);
|
|
155
155
|
const selected = Number(outcome.selected ?? 0);
|
|
156
|
-
const
|
|
156
|
+
const consumed = Number(outcome.consumed ?? 0);
|
|
157
|
+
const verified = Number(outcome.verified ?? 0);
|
|
157
158
|
const status = outcome.status;
|
|
158
|
-
const valid =
|
|
159
|
-
|
|
159
|
+
const valid = ['positive_outcome', 'negative_outcome', 'unknown_outcome'].includes(String(status))
|
|
160
|
+
&& selected >= consumed && consumed >= verified && verified >= 0;
|
|
161
|
+
return result('memory_outcome', valid ? 'passed' : 'failed', valid ? `Memory outcome evaluated (${consumed}/${selected} consumed, ${verified} verified).` : 'Memory outcome was not verified.', startedAt, [{ type: 'memory_outcome', value: JSON.stringify({ status, selected, consumed, verified }) }], valid ? 1 : 0, 1);
|
|
160
162
|
},
|
|
161
163
|
};
|
|
162
164
|
}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"deterministic.js","sourceRoot":"","sources":["../../src/graders/deterministic.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAC9C,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAGzC,MAAM,wBAAwB,GAAG,EAAE,GAAG,IAAI,CAAC;AAC3C,MAAM,0BAA0B,GAAG,MAAM,CAAC;AAE1C,MAAM,CAAC,MAAM,qBAAqB,GAAuB;IACvD,qBAAqB,CAAC,mBAAmB,EAAE,iBAAiB,EAAE,CAAC,OAAO,EAAE,MAAM,EAAE,SAAS,CAAC,CAAC;IAC3F,qBAAqB,CAAC,qBAAqB,EAAE,mBAAmB,EAAE,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;IACxF,qBAAqB,CAAC,gBAAgB,EAAE,oBAAoB,EAAE,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;IACpF,aAAa,EAAE;IACf,qBAAqB,CAAC,kBAAkB,EAAE,sBAAsB,EAAE,CAAC,WAAW,EAAE,UAAU,EAAE,SAAS,CAAC,CAAC;IACvG,qBAAqB,CAAC,mBAAmB,EAAE,aAAa,EAAE,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;IAC9E,qBAAqB,CAAC,uBAAuB,EAAE,mBAAmB,EAAE,CAAC,SAAS,EAAE,YAAY,CAAC,CAAC;IAC9F,mBAAmB,EAAE;IACrB,qBAAqB,CAAC,mBAAmB,EAAE,kBAAkB,EAAE,CAAC,UAAU,EAAE,SAAS,CAAC,CAAC;IACvF,mBAAmB,EAAE;CACtB,CAAC;AAEF,MAAM,UAAU,uBAAuB;IACrC,OAAO;QACL,EAAE,EAAE,qBAAqB;QACzB,IAAI,EAAE,eAAe;QACrB,SAAS,EAAE,CAAC,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,CAAC;QACnD,gBAAgB,EAAE,CAAC,SAAS,CAAC;QAC7B,SAAS,EAAE,0BAA0B;QACrC,IAAI,EAAE,KAAK;QACX,QAAQ,EAAE,IAAI;QACd,KAAK,EAAE,KAAK,EAAC,OAAO,EAAC,EAAE;YACrB,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,MAAM,OAAO,GAAG,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;YACzC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,OAAO,OAAO,CAAC,IAAI,KAAK,QAAQ,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,EAAE,EAAE,CAAC;gBACnF,OAAO,MAAM,CAAC,qBAAqB,EAAE,aAAa,EAAE,6CAA6C,EAAE,SAAS,EAAE,EAAE,CAAC,CAAC;YACpH,CAAC;YACD,MAAM,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC;gBACtC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,IAAI,EAAkB,EAAE,CAAC,OAAO,IAAI,KAAK,QAAQ,CAAC;gBACzE,CAAC,CAAC,EAAE,CAAC;YACP,MAAM,SAAS,GAAG,aAAa,CAAC,OAAO,CAAC,SAAS,EAAE,GAAG,EAAE,0BAA0B,EAAE,0BAA0B,CAAC,CAAC;YAChH,MAAM,cAAc,GAAG,aAAa,CAAC,OAAO,CAAC,cAAc,EAAE,KAAK,EAAE,wBAAwB,EAAE,wBAAwB,CAAC,CAAC;YACxH,MAAM,SAAS,GAAG,MAAM,iBAAiB,CAAC,OAAO,CAAC,IAAI,EAAE,IAAI,EAAE;gBAC5D,GAAG,EAAE,OAAO,OAAO,CAAC,GAAG,KAAK,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW;gBACxE,SAAS;gBACT,cAAc;aACf,CAAC,CAAC;YACH,OAAO,MAAM,CACX,qBAAqB,EACrB,SAAS,CAAC,QAAQ,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,EACrE,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,4BAA4B,SAAS,CAAC,QAAQ,GAAG,EAC7F,SAAS,EACT;gBACE,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC,OAAO,CAAC,IAAI,EAAE,GAAG,IAAI,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE;gBAC7D,EAAE,IAAI,EAAE,aAAa,EAAE,KAAK,EAAE,SAAS,CAAC,UAAU,EAAE;aACrD,EACD,SAAS,CAAC,QAAQ,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EACvD,CAAC,CACF,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,IAAY,EACZ,IAAc,EACd,OAAqE;IAErE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,IAAI,0BAA0B,CAAC;IAClE,MAAM,cAAc,GAAG,OAAO,CAAC,cAAc,IAAI,wBAAwB,CAAC;IAC1E,OAAO,IAAI,OAAO,CAAC,OAAO,CAAC,EAAE;QAC3B,IAAI,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;QAC7B,IAAI,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;QAC7B,IAAI,QAAQ,GAAG,KAAK,CAAC;QACrB,MAAM,KAAK,GAAG,QAAQ,CAAC,IAAI,EAAE,IAAI,EAAE;YACjC,GAAG,EAAE,OAAO,CAAC,GAAG;YAChB,OAAO,EAAE,SAAS;YAClB,SAAS,EAAE,cAAc,GAAG,CAAC;YAC7B,QAAQ,EAAE,QAAQ;SACnB,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE,GAAG,EAAE,EAAE;YACrB,MAAM,GAAG,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,IAAI,EAAE,CAAC,CAAC;YAC7D,MAAM,GAAG,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,IAAI,EAAE,CAAC,CAAC;YAC7D,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;YACpE,MAAM,OAAO,GAAG,QAAQ,CAAC,QAAQ,CAAC,CAAC,EAAE,cAAc,CAAC,CAAC;YACrD,MAAM,IAAI,GAAG,OAAQ,KAAsC,EAAE,IAAI,KAAK,QAAQ;gBAC5E,CAAC,CAAC,MAAM,CAAE,KAA+B,CAAC,IAAI,CAAC;gBAC/C,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAClB,OAAO,CAAC;gBACN,QAAQ,EAAE,IAAI;gBACd,QAAQ;gBACR,UAAU,EAAE,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;gBAC9D,SAAS,EAAE,QAAQ,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM;aAC5C,CAAC,CAAC;QACL,CAAC,CAAC,CAAC;QACH,KAAK,CAAC,IAAI,CAAC,OAAO,EAAE,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;QAC9B,KAAK,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE;YACnC,IAAI,MAAM,KAAK,SAAS,IAAI,MAAM,KAAK,SAAS;gBAAE,QAAQ,GAAG,IAAI,CAAC;QACpE,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC;AAED,SAAS,qBAAqB,CAAC,EAAU,EAAE,GAAW,EAAE,SAAmB;IACzE,OAAO;QACL,EAAE;QACF,IAAI,EAAE,eAAe;QACrB,SAAS;QACT,gBAAgB,EAAE,CAAC,GAAG,CAAC;QACvB,SAAS,EAAE,KAAK;QAChB,IAAI,EAAE,MAAM;QACZ,QAAQ,EAAE,IAAI;QACd,KAAK,EAAE,OAAO,CAAC,EAAE;YACf,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,MAAM,KAAK,GAAG,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;YACpC,IAAI,OAAO,KAAK,KAAK,SAAS,EAAE,CAAC;gBAC/B,OAAO,MAAM,CAAC,EAAE,EAAE,aAAa,EAAE,qBAAqB,GAAG,kBAAkB,EAAE,SAAS,EAAE,EAAE,CAAC,CAAC;YAC9F,CAAC;YACD,OAAO,MAAM,CACX,EAAE,EACF,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,EAC3B,KAAK,CAAC,CAAC,CAAC,GAAG,GAAG,UAAU,CAAC,CAAC,CAAC,GAAG,GAAG,UAAU,EAC3C,SAAS,EACT,CAAC,EAAE,IAAI,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,EACrC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EACb,CAAC,CACF,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED,SAAS,aAAa;IACpB,OAAO;QACL,EAAE,EAAE,kBAAkB;QACtB,IAAI,EAAE,eAAe;QACrB,SAAS,EAAE,CAAC,QAAQ,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,CAAC;QACpD,gBAAgB,EAAE,CAAC,SAAS,CAAC;QAC7B,SAAS,EAAE,KAAK;QAChB,IAAI,EAAE,MAAM;QACZ,QAAQ,EAAE,IAAI;QACd,KAAK,EAAE,OAAO,CAAC,EAAE;YACf,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,MAAM,OAAO,GAAG,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;YACzC,IAAI,OAAO,OAAO,KAAK,QAAQ;gBAAE,OAAO,MAAM,CAAC,kBAAkB,EAAE,aAAa,EAAE,yBAAyB,EAAE,SAAS,EAAE,EAAE,CAAC,CAAC;YAC5H,MAAM,KAAK,GAAG,yCAAyC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;YACtE,OAAO,MAAM,CACX,kBAAkB,EAClB,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,EAC3B,KAAK,CAAC,CAAC,CAAC,yBAAyB,CAAC,CAAC,CAAC,2BAA2B,EAC/D,SAAS,EACT,CAAC,EAAE,IAAI,EAAE,cAAc,EAAE,KAAK,EAAE,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,EACrF,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EACb,CAAC,CACF,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED,SAAS,mBAAmB;IAC1B,MAAM,QAAQ,GAAG,IAAI,GAAG,CAAC,CAAC,WAAW,EAAE,QAAQ,EAAE,UAAU,EAAE,WAAW,EAAE,UAAU,EAAE,QAAQ,EAAE,0BAA0B,CAAC,CAAC,CAAC;IAC7H,OAAO;QACL,EAAE,EAAE,0BAA0B;QAC9B,IAAI,EAAE,eAAe;QACrB,SAAS,EAAE,CAAC,UAAU,EAAE,SAAS,EAAE,SAAS,CAAC;QAC7C,gBAAgB,EAAE,CAAC,gBAAgB,CAAC;QACpC,SAAS,EAAE,KAAK;QAChB,IAAI,EAAE,MAAM;QACZ,QAAQ,EAAE,IAAI;QACd,KAAK,EAAE,OAAO,CAAC,EAAE;YACf,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,MAAM,KAAK,GAAG,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC;YAC9C,IAAI,OAAO,KAAK,KAAK,QAAQ;gBAAE,OAAO,MAAM,CAAC,0BAA0B,EAAE,aAAa,EAAE,iCAAiC,EAAE,SAAS,EAAE,EAAE,CAAC,CAAC;YAC1I,MAAM,KAAK,GAAG,QAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YAClC,OAAO,MAAM,CACX,0BAA0B,EAC1B,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,EAC3B,KAAK,CAAC,CAAC,CAAC,qBAAqB,KAAK,GAAG,CAAC,CAAC,CAAC,qCAAqC,KAAK,GAAG,EACrF,SAAS,EACT,CAAC,EAAE,IAAI,EAAE,iBAAiB,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,EAC3C,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EACb,CAAC,CACF,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED,SAAS,mBAAmB;IAC1B,OAAO;QACL,EAAE,EAAE,gBAAgB;QACpB,IAAI,EAAE,eAAe;QACrB,SAAS,EAAE,CAAC,QAAQ,EAAE,SAAS,CAAC;QAChC,gBAAgB,EAAE,CAAC,eAAe,CAAC;QACnC,SAAS,EAAE,KAAK;QAChB,IAAI,EAAE,MAAM;QACZ,QAAQ,EAAE,IAAI;QACd,KAAK,EAAE,OAAO,CAAC,EAAE;YACf,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,MAAM,OAAO,GAAG,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAC;YAC/C,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC;gBAAE,OAAO,MAAM,CAAC,gBAAgB,EAAE,aAAa,EAAE,gCAAgC,EAAE,SAAS,EAAE,EAAE,CAAC,CAAC;YACxH,MAAM,QAAQ,GAAG,MAAM,CAAC,OAAO,CAAC,QAAQ,IAAI,CAAC,CAAC,CAAC;YAC/C,MAAM,
|
|
1
|
+
{"version":3,"file":"deterministic.js","sourceRoot":"","sources":["../../src/graders/deterministic.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAC9C,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAGzC,MAAM,wBAAwB,GAAG,EAAE,GAAG,IAAI,CAAC;AAC3C,MAAM,0BAA0B,GAAG,MAAM,CAAC;AAE1C,MAAM,CAAC,MAAM,qBAAqB,GAAuB;IACvD,qBAAqB,CAAC,mBAAmB,EAAE,iBAAiB,EAAE,CAAC,OAAO,EAAE,MAAM,EAAE,SAAS,CAAC,CAAC;IAC3F,qBAAqB,CAAC,qBAAqB,EAAE,mBAAmB,EAAE,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;IACxF,qBAAqB,CAAC,gBAAgB,EAAE,oBAAoB,EAAE,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;IACpF,aAAa,EAAE;IACf,qBAAqB,CAAC,kBAAkB,EAAE,sBAAsB,EAAE,CAAC,WAAW,EAAE,UAAU,EAAE,SAAS,CAAC,CAAC;IACvG,qBAAqB,CAAC,mBAAmB,EAAE,aAAa,EAAE,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;IAC9E,qBAAqB,CAAC,uBAAuB,EAAE,mBAAmB,EAAE,CAAC,SAAS,EAAE,YAAY,CAAC,CAAC;IAC9F,mBAAmB,EAAE;IACrB,qBAAqB,CAAC,mBAAmB,EAAE,kBAAkB,EAAE,CAAC,UAAU,EAAE,SAAS,CAAC,CAAC;IACvF,mBAAmB,EAAE;CACtB,CAAC;AAEF,MAAM,UAAU,uBAAuB;IACrC,OAAO;QACL,EAAE,EAAE,qBAAqB;QACzB,IAAI,EAAE,eAAe;QACrB,SAAS,EAAE,CAAC,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,CAAC;QACnD,gBAAgB,EAAE,CAAC,SAAS,CAAC;QAC7B,SAAS,EAAE,0BAA0B;QACrC,IAAI,EAAE,KAAK;QACX,QAAQ,EAAE,IAAI;QACd,KAAK,EAAE,KAAK,EAAC,OAAO,EAAC,EAAE;YACrB,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,MAAM,OAAO,GAAG,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;YACzC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,OAAO,OAAO,CAAC,IAAI,KAAK,QAAQ,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,EAAE,EAAE,CAAC;gBACnF,OAAO,MAAM,CAAC,qBAAqB,EAAE,aAAa,EAAE,6CAA6C,EAAE,SAAS,EAAE,EAAE,CAAC,CAAC;YACpH,CAAC;YACD,MAAM,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC;gBACtC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,IAAI,EAAkB,EAAE,CAAC,OAAO,IAAI,KAAK,QAAQ,CAAC;gBACzE,CAAC,CAAC,EAAE,CAAC;YACP,MAAM,SAAS,GAAG,aAAa,CAAC,OAAO,CAAC,SAAS,EAAE,GAAG,EAAE,0BAA0B,EAAE,0BAA0B,CAAC,CAAC;YAChH,MAAM,cAAc,GAAG,aAAa,CAAC,OAAO,CAAC,cAAc,EAAE,KAAK,EAAE,wBAAwB,EAAE,wBAAwB,CAAC,CAAC;YACxH,MAAM,SAAS,GAAG,MAAM,iBAAiB,CAAC,OAAO,CAAC,IAAI,EAAE,IAAI,EAAE;gBAC5D,GAAG,EAAE,OAAO,OAAO,CAAC,GAAG,KAAK,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW;gBACxE,SAAS;gBACT,cAAc;aACf,CAAC,CAAC;YACH,OAAO,MAAM,CACX,qBAAqB,EACrB,SAAS,CAAC,QAAQ,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,EACrE,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,4BAA4B,SAAS,CAAC,QAAQ,GAAG,EAC7F,SAAS,EACT;gBACE,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC,OAAO,CAAC,IAAI,EAAE,GAAG,IAAI,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE;gBAC7D,EAAE,IAAI,EAAE,aAAa,EAAE,KAAK,EAAE,SAAS,CAAC,UAAU,EAAE;aACrD,EACD,SAAS,CAAC,QAAQ,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EACvD,CAAC,CACF,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,IAAY,EACZ,IAAc,EACd,OAAqE;IAErE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,IAAI,0BAA0B,CAAC;IAClE,MAAM,cAAc,GAAG,OAAO,CAAC,cAAc,IAAI,wBAAwB,CAAC;IAC1E,OAAO,IAAI,OAAO,CAAC,OAAO,CAAC,EAAE;QAC3B,IAAI,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;QAC7B,IAAI,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;QAC7B,IAAI,QAAQ,GAAG,KAAK,CAAC;QACrB,MAAM,KAAK,GAAG,QAAQ,CAAC,IAAI,EAAE,IAAI,EAAE;YACjC,GAAG,EAAE,OAAO,CAAC,GAAG;YAChB,OAAO,EAAE,SAAS;YAClB,SAAS,EAAE,cAAc,GAAG,CAAC;YAC7B,QAAQ,EAAE,QAAQ;SACnB,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE,GAAG,EAAE,EAAE;YACrB,MAAM,GAAG,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,IAAI,EAAE,CAAC,CAAC;YAC7D,MAAM,GAAG,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,IAAI,EAAE,CAAC,CAAC;YAC7D,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;YACpE,MAAM,OAAO,GAAG,QAAQ,CAAC,QAAQ,CAAC,CAAC,EAAE,cAAc,CAAC,CAAC;YACrD,MAAM,IAAI,GAAG,OAAQ,KAAsC,EAAE,IAAI,KAAK,QAAQ;gBAC5E,CAAC,CAAC,MAAM,CAAE,KAA+B,CAAC,IAAI,CAAC;gBAC/C,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAClB,OAAO,CAAC;gBACN,QAAQ,EAAE,IAAI;gBACd,QAAQ;gBACR,UAAU,EAAE,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;gBAC9D,SAAS,EAAE,QAAQ,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM;aAC5C,CAAC,CAAC;QACL,CAAC,CAAC,CAAC;QACH,KAAK,CAAC,IAAI,CAAC,OAAO,EAAE,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;QAC9B,KAAK,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE;YACnC,IAAI,MAAM,KAAK,SAAS,IAAI,MAAM,KAAK,SAAS;gBAAE,QAAQ,GAAG,IAAI,CAAC;QACpE,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC;AAED,SAAS,qBAAqB,CAAC,EAAU,EAAE,GAAW,EAAE,SAAmB;IACzE,OAAO;QACL,EAAE;QACF,IAAI,EAAE,eAAe;QACrB,SAAS;QACT,gBAAgB,EAAE,CAAC,GAAG,CAAC;QACvB,SAAS,EAAE,KAAK;QAChB,IAAI,EAAE,MAAM;QACZ,QAAQ,EAAE,IAAI;QACd,KAAK,EAAE,OAAO,CAAC,EAAE;YACf,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,MAAM,KAAK,GAAG,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;YACpC,IAAI,OAAO,KAAK,KAAK,SAAS,EAAE,CAAC;gBAC/B,OAAO,MAAM,CAAC,EAAE,EAAE,aAAa,EAAE,qBAAqB,GAAG,kBAAkB,EAAE,SAAS,EAAE,EAAE,CAAC,CAAC;YAC9F,CAAC;YACD,OAAO,MAAM,CACX,EAAE,EACF,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,EAC3B,KAAK,CAAC,CAAC,CAAC,GAAG,GAAG,UAAU,CAAC,CAAC,CAAC,GAAG,GAAG,UAAU,EAC3C,SAAS,EACT,CAAC,EAAE,IAAI,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,EACrC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EACb,CAAC,CACF,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED,SAAS,aAAa;IACpB,OAAO;QACL,EAAE,EAAE,kBAAkB;QACtB,IAAI,EAAE,eAAe;QACrB,SAAS,EAAE,CAAC,QAAQ,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,CAAC;QACpD,gBAAgB,EAAE,CAAC,SAAS,CAAC;QAC7B,SAAS,EAAE,KAAK;QAChB,IAAI,EAAE,MAAM;QACZ,QAAQ,EAAE,IAAI;QACd,KAAK,EAAE,OAAO,CAAC,EAAE;YACf,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,MAAM,OAAO,GAAG,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;YACzC,IAAI,OAAO,OAAO,KAAK,QAAQ;gBAAE,OAAO,MAAM,CAAC,kBAAkB,EAAE,aAAa,EAAE,yBAAyB,EAAE,SAAS,EAAE,EAAE,CAAC,CAAC;YAC5H,MAAM,KAAK,GAAG,yCAAyC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;YACtE,OAAO,MAAM,CACX,kBAAkB,EAClB,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,EAC3B,KAAK,CAAC,CAAC,CAAC,yBAAyB,CAAC,CAAC,CAAC,2BAA2B,EAC/D,SAAS,EACT,CAAC,EAAE,IAAI,EAAE,cAAc,EAAE,KAAK,EAAE,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,EACrF,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EACb,CAAC,CACF,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED,SAAS,mBAAmB;IAC1B,MAAM,QAAQ,GAAG,IAAI,GAAG,CAAC,CAAC,WAAW,EAAE,QAAQ,EAAE,UAAU,EAAE,WAAW,EAAE,UAAU,EAAE,QAAQ,EAAE,0BAA0B,CAAC,CAAC,CAAC;IAC7H,OAAO;QACL,EAAE,EAAE,0BAA0B;QAC9B,IAAI,EAAE,eAAe;QACrB,SAAS,EAAE,CAAC,UAAU,EAAE,SAAS,EAAE,SAAS,CAAC;QAC7C,gBAAgB,EAAE,CAAC,gBAAgB,CAAC;QACpC,SAAS,EAAE,KAAK;QAChB,IAAI,EAAE,MAAM;QACZ,QAAQ,EAAE,IAAI;QACd,KAAK,EAAE,OAAO,CAAC,EAAE;YACf,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,MAAM,KAAK,GAAG,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC;YAC9C,IAAI,OAAO,KAAK,KAAK,QAAQ;gBAAE,OAAO,MAAM,CAAC,0BAA0B,EAAE,aAAa,EAAE,iCAAiC,EAAE,SAAS,EAAE,EAAE,CAAC,CAAC;YAC1I,MAAM,KAAK,GAAG,QAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YAClC,OAAO,MAAM,CACX,0BAA0B,EAC1B,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,EAC3B,KAAK,CAAC,CAAC,CAAC,qBAAqB,KAAK,GAAG,CAAC,CAAC,CAAC,qCAAqC,KAAK,GAAG,EACrF,SAAS,EACT,CAAC,EAAE,IAAI,EAAE,iBAAiB,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,EAC3C,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EACb,CAAC,CACF,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED,SAAS,mBAAmB;IAC1B,OAAO;QACL,EAAE,EAAE,gBAAgB;QACpB,IAAI,EAAE,eAAe;QACrB,SAAS,EAAE,CAAC,QAAQ,EAAE,SAAS,CAAC;QAChC,gBAAgB,EAAE,CAAC,eAAe,CAAC;QACnC,SAAS,EAAE,KAAK;QAChB,IAAI,EAAE,MAAM;QACZ,QAAQ,EAAE,IAAI;QACd,KAAK,EAAE,OAAO,CAAC,EAAE;YACf,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,MAAM,OAAO,GAAG,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAC;YAC/C,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC;gBAAE,OAAO,MAAM,CAAC,gBAAgB,EAAE,aAAa,EAAE,gCAAgC,EAAE,SAAS,EAAE,EAAE,CAAC,CAAC;YACxH,MAAM,QAAQ,GAAG,MAAM,CAAC,OAAO,CAAC,QAAQ,IAAI,CAAC,CAAC,CAAC;YAC/C,MAAM,QAAQ,GAAG,MAAM,CAAC,OAAO,CAAC,QAAQ,IAAI,CAAC,CAAC,CAAC;YAC/C,MAAM,QAAQ,GAAG,MAAM,CAAC,OAAO,CAAC,QAAQ,IAAI,CAAC,CAAC,CAAC;YAC/C,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;YAC9B,MAAM,KAAK,GAAG,CAAC,kBAAkB,EAAE,kBAAkB,EAAE,iBAAiB,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;mBAC7F,QAAQ,IAAI,QAAQ,IAAI,QAAQ,IAAI,QAAQ,IAAI,QAAQ,IAAI,CAAC,CAAC;YACnE,OAAO,MAAM,CACX,gBAAgB,EAChB,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,EAC3B,KAAK,CAAC,CAAC,CAAC,6BAA6B,QAAQ,IAAI,QAAQ,cAAc,QAAQ,aAAa,CAAC,CAAC,CAAC,kCAAkC,EACjI,SAAS,EACT,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,KAAK,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,MAAM,EAAE,QAAQ,EAAE,QAAQ,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,EAC7F,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EACb,CAAC,CACF,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,MAAM,CACpB,QAAgB,EAChB,MAA8B,EAC9B,MAAc,EACd,SAAiB,EACjB,QAAkC,EAClC,KAAc,EACd,SAAkB;IAElB,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC,CAAC;IACvD,MAAM,OAAO,GAAG,UAAU,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC;QACnE,QAAQ,EAAE,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS;KACrD,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,CAAC;IACjC,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,eAAe,EAAE,MAAM,EAAE,KAAK,EAAE,SAAS,EAAE,QAAQ,EAAE,MAAM,EAAE,UAAU,EAAE,OAAO,EAAE,CAAC;AAC9G,CAAC;AAED,SAAS,QAAQ,CAAC,KAAc;IAC9B,OAAO,OAAO,CAAC,KAAK,CAAC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;AAC9E,CAAC;AAED,SAAS,aAAa,CAAC,KAAc,EAAE,OAAe,EAAE,OAAe,EAAE,QAAgB;IACvF,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC;QACxD,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC;QACzD,CAAC,CAAC,QAAQ,CAAC;AACf,CAAC"}
|
package/dist/index.d.ts
CHANGED
|
@@ -16,12 +16,25 @@ export type { RetrievalQualityCase, RetrievalQualityFixture, RetrievalQualityRep
|
|
|
16
16
|
export { runRetrievalQualityCorpus } from './retrieval-quality-corpus.js';
|
|
17
17
|
export type { EvidenceRuntimeCorpusFixture, EvidenceRuntimeQualityReport, QualityCorpusReport, RetrievalQualityCorpusFixture, RetrievalQualityCorpusProject, RetrievalQualityCorpusReport, } from './retrieval-quality-corpus.js';
|
|
18
18
|
export { runEvidenceRuntimeQualityCorpus } from './retrieval-quality-corpus.js';
|
|
19
|
-
export
|
|
20
|
-
export
|
|
19
|
+
export * from './evidence-quality-metrics.js';
|
|
20
|
+
export * from './release-quality-comparison.js';
|
|
21
|
+
export * from './release-quality-policy.js';
|
|
21
22
|
export * from './graders/index.js';
|
|
22
23
|
export { aggregateRepeatedTrials, createTrialReceipt, renderRepeatedTrialReport, } from './repeated-trials.js';
|
|
23
24
|
export type { RepeatedTrial, RepeatedTrialReport, RepeatedTrialTaskReport, } from './repeated-trials.js';
|
|
24
25
|
export { evaluateSemanticFixtures } from './semantic-intent-quality.js';
|
|
25
26
|
export { evaluateSemanticCalibration, evaluateSemanticReleaseGates, wilsonInterval95 } from './semantic-intent-quality.js';
|
|
26
27
|
export type { SemanticIntentCorpus, SemanticIntentExpectedFrame, SemanticIntentFixture, SemanticMutationPermission, SemanticQualityReport, SemanticCalibrationReport, SemanticCalibrationSample, SemanticReleaseGates, } from './semantic-intent-quality.js';
|
|
28
|
+
export { loadUnifiedRuntimeCorpus } from './unified-runtime-corpus.js';
|
|
29
|
+
export type { UnifiedAvailability, UnifiedCaseClass, UnifiedCaseSlice, UnifiedChannel, UnifiedLanguage, UnifiedRetrievalCase, UnifiedRuntimeCorpus, UnifiedSemanticCase, UnifiedVariant, } from './unified-runtime-corpus.js';
|
|
30
|
+
export { evaluateUnifiedRuntimeCorpus } from './unified-runtime-evaluation.js';
|
|
31
|
+
export { runUnifiedRuntimeGold } from './unified-runtime-runner.js';
|
|
32
|
+
export type * from './unified-runtime-observation.js';
|
|
33
|
+
export type { EvaluationBundleEvidenceV1, UnifiedEvaluationSlice, UnifiedLatencyMetric, UnifiedRerankerMetric, UnifiedRuntimeEvaluationReport, } from './unified-runtime-evaluation.js';
|
|
34
|
+
export * from './evaluation-bundle.js';
|
|
35
|
+
export * from './evaluation-bundle-builder.js';
|
|
36
|
+
export * from './evaluation-run-state.js';
|
|
37
|
+
export * from './release-quality-bundle-materializer.js';
|
|
38
|
+
export * from './release-quality-bundle-evaluator.js';
|
|
39
|
+
export * from './release-quality-evaluation-command.js';
|
|
27
40
|
//# sourceMappingURL=index.d.ts.map
|
package/dist/index.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxD,YAAY,EAAE,gBAAgB,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC;AAC1E,OAAO,EAAE,uBAAuB,EAAE,iBAAiB,EAAE,MAAM,oBAAoB,CAAC;AAChF,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AACvD,YAAY,EAAE,gBAAgB,EAAE,MAAM,uBAAuB,CAAC;AAC9D,OAAO,EAAE,sBAAsB,EAAE,MAAM,qBAAqB,CAAC;AAC7D,YAAY,EAAE,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AACpE,OAAO,EAAE,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AACtD,YAAY,EAAE,wBAAwB,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AACnF,OAAO,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAC7D,YAAY,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAClE,OAAO,EAAE,4BAA4B,EAAE,MAAM,wBAAwB,CAAC;AACtE,OAAO,EAAE,4BAA4B,EAAE,MAAM,6BAA6B,CAAC;AAC3E,YAAY,EAAE,sBAAsB,EAAE,MAAM,6BAA6B,CAAC;AAC1E,YAAY,EACV,oBAAoB,EACpB,uBAAuB,EACvB,sBAAsB,EACtB,0BAA0B,GAC3B,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAE,yBAAyB,EAAE,MAAM,+BAA+B,CAAC;AAC1E,YAAY,EACV,4BAA4B,EAC5B,4BAA4B,EAC5B,mBAAmB,EACnB,6BAA6B,EAC7B,6BAA6B,EAC7B,4BAA4B,GAC7B,MAAM,+BAA+B,CAAC;AACvC,OAAO,EAAE,+BAA+B,EAAE,MAAM,+BAA+B,CAAC;AAChF,
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxD,YAAY,EAAE,gBAAgB,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC;AAC1E,OAAO,EAAE,uBAAuB,EAAE,iBAAiB,EAAE,MAAM,oBAAoB,CAAC;AAChF,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AACvD,YAAY,EAAE,gBAAgB,EAAE,MAAM,uBAAuB,CAAC;AAC9D,OAAO,EAAE,sBAAsB,EAAE,MAAM,qBAAqB,CAAC;AAC7D,YAAY,EAAE,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AACpE,OAAO,EAAE,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AACtD,YAAY,EAAE,wBAAwB,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AACnF,OAAO,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAC7D,YAAY,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAClE,OAAO,EAAE,4BAA4B,EAAE,MAAM,wBAAwB,CAAC;AACtE,OAAO,EAAE,4BAA4B,EAAE,MAAM,6BAA6B,CAAC;AAC3E,YAAY,EAAE,sBAAsB,EAAE,MAAM,6BAA6B,CAAC;AAC1E,YAAY,EACV,oBAAoB,EACpB,uBAAuB,EACvB,sBAAsB,EACtB,0BAA0B,GAC3B,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAE,yBAAyB,EAAE,MAAM,+BAA+B,CAAC;AAC1E,YAAY,EACV,4BAA4B,EAC5B,4BAA4B,EAC5B,mBAAmB,EACnB,6BAA6B,EAC7B,6BAA6B,EAC7B,4BAA4B,GAC7B,MAAM,+BAA+B,CAAC;AACvC,OAAO,EAAE,+BAA+B,EAAE,MAAM,+BAA+B,CAAC;AAChF,cAAc,+BAA+B,CAAC;AAC9C,cAAc,iCAAiC,CAAC;AAChD,cAAc,6BAA6B,CAAC;AAC5C,cAAc,oBAAoB,CAAC;AACnC,OAAO,EACL,uBAAuB,EACvB,kBAAkB,EAClB,yBAAyB,GAC1B,MAAM,sBAAsB,CAAC;AAC9B,YAAY,EACV,aAAa,EACb,mBAAmB,EACnB,uBAAuB,GACxB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAAE,wBAAwB,EAAE,MAAM,8BAA8B,CAAC;AACxE,OAAO,EAAE,2BAA2B,EAAE,4BAA4B,EAAE,gBAAgB,EAAE,MAAM,8BAA8B,CAAC;AAC3H,YAAY,EACV,oBAAoB,EACpB,2BAA2B,EAC3B,qBAAqB,EACrB,0BAA0B,EAC1B,qBAAqB,EACrB,yBAAyB,EACzB,yBAAyB,EACzB,oBAAoB,GACrB,MAAM,8BAA8B,CAAC;AACtC,OAAO,EAAE,wBAAwB,EAAE,MAAM,6BAA6B,CAAC;AACvE,YAAY,EACV,mBAAmB,EACnB,gBAAgB,EAChB,gBAAgB,EAChB,cAAc,EACd,eAAe,EACf,oBAAoB,EACpB,oBAAoB,EACpB,mBAAmB,EACnB,cAAc,GACf,MAAM,6BAA6B,CAAC;AACrC,OAAO,EAAE,4BAA4B,EAAE,MAAM,iCAAiC,CAAC;AAC/E,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,mBAAmB,kCAAkC,CAAC;AACtD,YAAY,EACV,0BAA0B,EAC1B,sBAAsB,EACtB,oBAAoB,EACpB,qBAAqB,EACrB,8BAA8B,GAC/B,MAAM,iCAAiC,CAAC;AACzC,cAAc,wBAAwB,CAAC;AACvC,cAAc,gCAAgC,CAAC;AAC/C,cAAc,2BAA2B,CAAC;AAC1C,cAAc,0CAA0C,CAAC;AACzD,cAAc,uCAAuC,CAAC;AACtD,cAAc,yCAAyC,CAAC"}
|
package/dist/index.js
CHANGED
|
@@ -8,9 +8,20 @@ export { runRetrievalQualityBenchmark } from './retrieval-quality.js';
|
|
|
8
8
|
export { createIsolatedQualityProject } from './isolated-quality-store.js';
|
|
9
9
|
export { runRetrievalQualityCorpus } from './retrieval-quality-corpus.js';
|
|
10
10
|
export { runEvidenceRuntimeQualityCorpus } from './retrieval-quality-corpus.js';
|
|
11
|
-
export
|
|
11
|
+
export * from './evidence-quality-metrics.js';
|
|
12
|
+
export * from './release-quality-comparison.js';
|
|
13
|
+
export * from './release-quality-policy.js';
|
|
12
14
|
export * from './graders/index.js';
|
|
13
15
|
export { aggregateRepeatedTrials, createTrialReceipt, renderRepeatedTrialReport, } from './repeated-trials.js';
|
|
14
16
|
export { evaluateSemanticFixtures } from './semantic-intent-quality.js';
|
|
15
17
|
export { evaluateSemanticCalibration, evaluateSemanticReleaseGates, wilsonInterval95 } from './semantic-intent-quality.js';
|
|
18
|
+
export { loadUnifiedRuntimeCorpus } from './unified-runtime-corpus.js';
|
|
19
|
+
export { evaluateUnifiedRuntimeCorpus } from './unified-runtime-evaluation.js';
|
|
20
|
+
export { runUnifiedRuntimeGold } from './unified-runtime-runner.js';
|
|
21
|
+
export * from './evaluation-bundle.js';
|
|
22
|
+
export * from './evaluation-bundle-builder.js';
|
|
23
|
+
export * from './evaluation-run-state.js';
|
|
24
|
+
export * from './release-quality-bundle-materializer.js';
|
|
25
|
+
export * from './release-quality-bundle-evaluator.js';
|
|
26
|
+
export * from './release-quality-evaluation-command.js';
|
|
16
27
|
//# sourceMappingURL=index.js.map
|
package/dist/index.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AAExD,OAAO,EAAE,uBAAuB,EAAE,iBAAiB,EAAE,MAAM,oBAAoB,CAAC;AAChF,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AAEvD,OAAO,EAAE,sBAAsB,EAAE,MAAM,qBAAqB,CAAC;AAE7D,OAAO,EAAE,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AAEtD,OAAO,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAE7D,OAAO,EAAE,4BAA4B,EAAE,MAAM,wBAAwB,CAAC;AACtE,OAAO,EAAE,4BAA4B,EAAE,MAAM,6BAA6B,CAAC;AAQ3E,OAAO,EAAE,yBAAyB,EAAE,MAAM,+BAA+B,CAAC;AAS1E,OAAO,EAAE,+BAA+B,EAAE,MAAM,+BAA+B,CAAC;AAChF,
|
|
1
|
+
{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AAExD,OAAO,EAAE,uBAAuB,EAAE,iBAAiB,EAAE,MAAM,oBAAoB,CAAC;AAChF,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AAEvD,OAAO,EAAE,sBAAsB,EAAE,MAAM,qBAAqB,CAAC;AAE7D,OAAO,EAAE,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AAEtD,OAAO,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAE7D,OAAO,EAAE,4BAA4B,EAAE,MAAM,wBAAwB,CAAC;AACtE,OAAO,EAAE,4BAA4B,EAAE,MAAM,6BAA6B,CAAC;AAQ3E,OAAO,EAAE,yBAAyB,EAAE,MAAM,+BAA+B,CAAC;AAS1E,OAAO,EAAE,+BAA+B,EAAE,MAAM,+BAA+B,CAAC;AAChF,cAAc,+BAA+B,CAAC;AAC9C,cAAc,iCAAiC,CAAC;AAChD,cAAc,6BAA6B,CAAC;AAC5C,cAAc,oBAAoB,CAAC;AACnC,OAAO,EACL,uBAAuB,EACvB,kBAAkB,EAClB,yBAAyB,GAC1B,MAAM,sBAAsB,CAAC;AAM9B,OAAO,EAAE,wBAAwB,EAAE,MAAM,8BAA8B,CAAC;AACxE,OAAO,EAAE,2BAA2B,EAAE,4BAA4B,EAAE,gBAAgB,EAAE,MAAM,8BAA8B,CAAC;AAW3H,OAAO,EAAE,wBAAwB,EAAE,MAAM,6BAA6B,CAAC;AAYvE,OAAO,EAAE,4BAA4B,EAAE,MAAM,iCAAiC,CAAC;AAC/E,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AASpE,cAAc,wBAAwB,CAAC;AACvC,cAAc,gCAAgC,CAAC;AAC/C,cAAc,2BAA2B,CAAC;AAC1C,cAAc,0CAA0C,CAAC;AACzD,cAAc,uCAAuC,CAAC;AACtD,cAAc,yCAAyC,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"release-quality-bundle-command.d.ts","sourceRoot":"","sources":["../src/release-quality-bundle-command.ts"],"names":[],"mappings":"AAgCA,wBAAsB,8BAA8B,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CAsClF;AAED,wBAAgB,gCAAgC,CAAC,IAAI,EAAE,MAAM,GAAG,IAAI,CAEnE"}
|
|
@@ -0,0 +1,72 @@
|
|
|
1
|
+
import { execFileSync } from 'node:child_process';
|
|
2
|
+
import { existsSync, mkdirSync, realpathSync, writeFileSync, } from 'node:fs';
|
|
3
|
+
import { join, resolve } from 'node:path';
|
|
4
|
+
import { fileURLToPath } from 'node:url';
|
|
5
|
+
import { stableRuntimeJson } from '@devflow-tools/sdk';
|
|
6
|
+
import { loadReleaseQualityMaterializationTargets, materializeReleaseQualityBundle, } from './release-quality-bundle-materializer.js';
|
|
7
|
+
import { removeReleaseQualityOutput } from './release-quality-output-cleanup.js';
|
|
8
|
+
export async function runReleaseQualityBundleCommand(argv) {
|
|
9
|
+
const manifestPath = requiredOption(argv, '--manifest');
|
|
10
|
+
const sourcesRoot = requiredOption(argv, '--sources');
|
|
11
|
+
const outputRoot = resolve(requiredOption(argv, '--output'));
|
|
12
|
+
const builderCommit = requiredCommit(argv, '--builder-commit');
|
|
13
|
+
const builderIdentity = `devflow:${builderCommit}`;
|
|
14
|
+
if (existsSync(outputRoot))
|
|
15
|
+
throw new Error('RELEASE_QUALITY_OUTPUT_EXISTS');
|
|
16
|
+
mkdirSync(outputRoot, { recursive: true });
|
|
17
|
+
try {
|
|
18
|
+
const targets = loadReleaseQualityMaterializationTargets({
|
|
19
|
+
manifestPath,
|
|
20
|
+
sourcesRoot,
|
|
21
|
+
outputRoot,
|
|
22
|
+
});
|
|
23
|
+
const bundles = [];
|
|
24
|
+
for (const target of targets) {
|
|
25
|
+
const result = await materializeReleaseQualityBundle(target, builderIdentity);
|
|
26
|
+
bundles.push({
|
|
27
|
+
projectId: target.input.project.projectId,
|
|
28
|
+
datasetDirectory: target.outputPath.slice(outputRoot.length + 1),
|
|
29
|
+
bundleId: result.manifest.bundleId,
|
|
30
|
+
projectSnapshotId: result.manifest.projectSnapshotId,
|
|
31
|
+
channelSnapshotHashes: result.manifest.channelSnapshotHashes,
|
|
32
|
+
builderIdentity: result.manifest.builderIdentity,
|
|
33
|
+
});
|
|
34
|
+
}
|
|
35
|
+
const mapping = {
|
|
36
|
+
schemaVersion: 'devflow.release-quality-bundles.v1',
|
|
37
|
+
bundles,
|
|
38
|
+
};
|
|
39
|
+
const mappingPath = join(outputRoot, 'bundles.json');
|
|
40
|
+
writeFileSync(mappingPath, `${stableRuntimeJson(mapping)}\n`);
|
|
41
|
+
process.stdout.write(`${stableRuntimeJson(mapping)}\n`);
|
|
42
|
+
}
|
|
43
|
+
catch (error) {
|
|
44
|
+
removeReleaseQualityOutput(outputRoot);
|
|
45
|
+
throw error;
|
|
46
|
+
}
|
|
47
|
+
}
|
|
48
|
+
export function removeReleaseQualityBundleOutput(path) {
|
|
49
|
+
removeReleaseQualityOutput(path);
|
|
50
|
+
}
|
|
51
|
+
function requiredOption(argv, name) {
|
|
52
|
+
const index = argv.indexOf(name);
|
|
53
|
+
const value = index >= 0 ? argv[index + 1] : undefined;
|
|
54
|
+
if (!value)
|
|
55
|
+
throw new Error(`${name} requires a path`);
|
|
56
|
+
return resolve(value);
|
|
57
|
+
}
|
|
58
|
+
function requiredCommit(argv, name) {
|
|
59
|
+
const index = argv.indexOf(name);
|
|
60
|
+
const value = index >= 0 ? argv[index + 1] : undefined;
|
|
61
|
+
if (!value || !/^[0-9a-f]{40}$/u.test(value))
|
|
62
|
+
throw new Error(`${name} requires a Git commit`);
|
|
63
|
+
const head = execFileSync('git', ['rev-parse', 'HEAD'], { encoding: 'utf8' }).trim();
|
|
64
|
+
if (value !== head)
|
|
65
|
+
throw new Error('RELEASE_QUALITY_BUILDER_COMMIT_MISMATCH');
|
|
66
|
+
return value;
|
|
67
|
+
}
|
|
68
|
+
const commandPath = process.argv[1] ? realpathSync(process.argv[1]) : '';
|
|
69
|
+
if (commandPath === realpathSync(fileURLToPath(import.meta.url)) || commandPath.endsWith('/release-quality-bundle-command.ts')) {
|
|
70
|
+
await runReleaseQualityBundleCommand(process.argv.slice(2));
|
|
71
|
+
}
|
|
72
|
+
//# sourceMappingURL=release-quality-bundle-command.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"release-quality-bundle-command.js","sourceRoot":"","sources":["../src/release-quality-bundle-command.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EACL,UAAU,EACV,SAAS,EACT,YAAY,EACZ,aAAa,GACd,MAAM,SAAS,CAAC;AACjB,OAAO,EAAW,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACnD,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AACzC,OAAO,EAAE,iBAAiB,EAAE,MAAM,oBAAoB,CAAC;AACvD,OAAO,EACL,wCAAwC,EACxC,+BAA+B,GAChC,MAAM,0CAA0C,CAAC;AAClD,OAAO,EAAE,0BAA0B,EAAE,MAAM,qCAAqC,CAAC;AAkBjF,MAAM,CAAC,KAAK,UAAU,8BAA8B,CAAC,IAAc;IACjE,MAAM,YAAY,GAAG,cAAc,CAAC,IAAI,EAAE,YAAY,CAAC,CAAC;IACxD,MAAM,WAAW,GAAG,cAAc,CAAC,IAAI,EAAE,WAAW,CAAC,CAAC;IACtD,MAAM,UAAU,GAAG,OAAO,CAAC,cAAc,CAAC,IAAI,EAAE,UAAU,CAAC,CAAC,CAAC;IAC7D,MAAM,aAAa,GAAG,cAAc,CAAC,IAAI,EAAE,kBAAkB,CAAC,CAAC;IAC/D,MAAM,eAAe,GAAG,WAAW,aAAa,EAAE,CAAC;IACnD,IAAI,UAAU,CAAC,UAAU,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,+BAA+B,CAAC,CAAC;IAE7E,SAAS,CAAC,UAAU,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC3C,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,wCAAwC,CAAC;YACvD,YAAY;YACZ,WAAW;YACX,UAAU;SACX,CAAC,CAAC;QACH,MAAM,OAAO,GAA6C,EAAE,CAAC;QAC7D,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;YAC7B,MAAM,MAAM,GAAG,MAAM,+BAA+B,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;YAC9E,OAAO,CAAC,IAAI,CAAC;gBACX,SAAS,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,SAAS;gBACzC,gBAAgB,EAAE,MAAM,CAAC,UAAU,CAAC,KAAK,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC;gBAChE,QAAQ,EAAE,MAAM,CAAC,QAAQ,CAAC,QAAQ;gBAClC,iBAAiB,EAAE,MAAM,CAAC,QAAQ,CAAC,iBAAiB;gBACpD,qBAAqB,EAAE,MAAM,CAAC,QAAQ,CAAC,qBAAqB;gBAC5D,eAAe,EAAE,MAAM,CAAC,QAAQ,CAAC,eAAe;aACjD,CAAC,CAAC;QACL,CAAC;QACD,MAAM,OAAO,GAAkC;YAC7C,aAAa,EAAE,oCAAoC;YACnD,OAAO;SACR,CAAC;QACF,MAAM,WAAW,GAAG,IAAI,CAAC,UAAU,EAAE,cAAc,CAAC,CAAC;QACrD,aAAa,CAAC,WAAW,EAAE,GAAG,iBAAiB,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QAC9D,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,iBAAiB,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IAC1D,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,0BAA0B,CAAC,UAAU,CAAC,CAAC;QACvC,MAAM,KAAK,CAAC;IACd,CAAC;AACH,CAAC;AAED,MAAM,UAAU,gCAAgC,CAAC,IAAY;IAC3D,0BAA0B,CAAC,IAAI,CAAC,CAAC;AACnC,CAAC;AAED,SAAS,cAAc,CAAC,IAAc,EAAE,IAAY;IAClD,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IACjC,MAAM,KAAK,GAAG,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACvD,IAAI,CAAC,KAAK;QAAE,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,kBAAkB,CAAC,CAAC;IACvD,OAAO,OAAO,CAAC,KAAK,CAAC,CAAC;AACxB,CAAC;AAED,SAAS,cAAc,CAAC,IAAc,EAAE,IAAY;IAClD,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IACjC,MAAM,KAAK,GAAG,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACvD,IAAI,CAAC,KAAK,IAAI,CAAC,iBAAiB,CAAC,IAAI,CAAC,KAAK,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,wBAAwB,CAAC,CAAC;IAC/F,MAAM,IAAI,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,WAAW,EAAE,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IACrF,IAAI,KAAK,KAAK,IAAI;QAAE,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAC;IAC/E,OAAO,KAAK,CAAC;AACf,CAAC;AAED,MAAM,WAAW,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AACzE,IAAI,WAAW,KAAK,YAAY,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,WAAW,CAAC,QAAQ,CAAC,oCAAoC,CAAC,EAAE,CAAC;IAC/H,MAAM,8BAA8B,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;AAC9D,CAAC"}
|
|
@@ -0,0 +1,61 @@
|
|
|
1
|
+
import { type EvidenceRerankerPort, type EvaluationChannel, type ProviderExecutionStatus } from '@devflow-tools/sdk';
|
|
2
|
+
import { type AggregatedEvidenceQualityMetric, type EvidenceQualityMetric } from './evidence-quality-metrics.js';
|
|
3
|
+
import { type ReleaseQualityQuerySamplingEvidence } from './release-quality-query-sampling.js';
|
|
4
|
+
export interface ReleaseQualityBundleCaseResult {
|
|
5
|
+
caseId: string;
|
|
6
|
+
channel: EvaluationChannel;
|
|
7
|
+
caseClass: 'positive' | 'lifecycle';
|
|
8
|
+
expectedLifecycle?: 'explicit_save_durable' | 'latest_episode_selected';
|
|
9
|
+
candidateRefs: string[];
|
|
10
|
+
selectedRefs: string[];
|
|
11
|
+
candidateRecall: EvidenceQualityMetric;
|
|
12
|
+
recallAt5: EvidenceQualityMetric;
|
|
13
|
+
precisionAt5: EvidenceQualityMetric;
|
|
14
|
+
forbiddenContamination: EvidenceQualityMetric;
|
|
15
|
+
provenanceCompleteness: EvidenceQualityMetric;
|
|
16
|
+
sourceVersionCorrectness: EvidenceQualityMetric;
|
|
17
|
+
querySampling: ReleaseQualityQuerySamplingEvidence;
|
|
18
|
+
reranker: {
|
|
19
|
+
profileId: string;
|
|
20
|
+
profileVersion: string;
|
|
21
|
+
status: ProviderExecutionStatus;
|
|
22
|
+
reasonCodes: string[];
|
|
23
|
+
};
|
|
24
|
+
durationMs: number;
|
|
25
|
+
}
|
|
26
|
+
interface ReleaseQualityChannelMetrics {
|
|
27
|
+
candidateRecall: AggregatedEvidenceQualityMetric;
|
|
28
|
+
recallAt5: AggregatedEvidenceQualityMetric;
|
|
29
|
+
precisionAt5: AggregatedEvidenceQualityMetric;
|
|
30
|
+
forbiddenContamination: AggregatedEvidenceQualityMetric;
|
|
31
|
+
provenanceCompleteness: AggregatedEvidenceQualityMetric;
|
|
32
|
+
}
|
|
33
|
+
export interface ReleaseQualityBundleReport {
|
|
34
|
+
schemaVersion: 'devflow.release-quality-bundle-report.v1';
|
|
35
|
+
bundleId: string;
|
|
36
|
+
projectSnapshotId: string;
|
|
37
|
+
builderIdentity: string;
|
|
38
|
+
corpusHash: string;
|
|
39
|
+
labelHashes: Record<EvaluationChannel, string>;
|
|
40
|
+
channelSnapshotHashes: Record<EvaluationChannel, string>;
|
|
41
|
+
cases: ReleaseQualityBundleCaseResult[];
|
|
42
|
+
channels: Record<EvaluationChannel, ReleaseQualityChannelMetrics>;
|
|
43
|
+
memoryDurability: EvidenceQualityMetric;
|
|
44
|
+
memoryTemporalCorrectness: EvidenceQualityMetric;
|
|
45
|
+
knowledgeVersionCorrectness: EvidenceQualityMetric;
|
|
46
|
+
consumption: {
|
|
47
|
+
status: 'unavailable';
|
|
48
|
+
reasonCode: 'host_ledger_required';
|
|
49
|
+
};
|
|
50
|
+
verification: {
|
|
51
|
+
status: 'unavailable';
|
|
52
|
+
reasonCode: 'verification_receipt_required';
|
|
53
|
+
};
|
|
54
|
+
bundleUnchanged: boolean;
|
|
55
|
+
}
|
|
56
|
+
export declare function evaluateReleaseQualityBundle(input: {
|
|
57
|
+
bundlePath: string;
|
|
58
|
+
reranker: EvidenceRerankerPort;
|
|
59
|
+
}): Promise<ReleaseQualityBundleReport>;
|
|
60
|
+
export {};
|
|
61
|
+
//# sourceMappingURL=release-quality-bundle-evaluator.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"release-quality-bundle-evaluator.d.ts","sourceRoot":"","sources":["../src/release-quality-bundle-evaluator.ts"],"names":[],"mappings":"AAIA,OAAO,EAML,KAAK,oBAAoB,EACzB,KAAK,iBAAiB,EACtB,KAAK,uBAAuB,EAC7B,MAAM,oBAAoB,CAAC;AAM5B,OAAO,EAQL,KAAK,+BAA+B,EACpC,KAAK,qBAAqB,EAC3B,MAAM,+BAA+B,CAAC;AAiBvC,OAAO,EAIL,KAAK,mCAAmC,EACzC,MAAM,qCAAqC,CAAC;AAW7C,MAAM,WAAW,8BAA8B;IAC7C,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,iBAAiB,CAAC;IAC3B,SAAS,EAAE,UAAU,GAAG,WAAW,CAAC;IACpC,iBAAiB,CAAC,EAAE,uBAAuB,GAAG,yBAAyB,CAAC;IACxE,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,eAAe,EAAE,qBAAqB,CAAC;IACvC,SAAS,EAAE,qBAAqB,CAAC;IACjC,YAAY,EAAE,qBAAqB,CAAC;IACpC,sBAAsB,EAAE,qBAAqB,CAAC;IAC9C,sBAAsB,EAAE,qBAAqB,CAAC;IAC9C,wBAAwB,EAAE,qBAAqB,CAAC;IAChD,aAAa,EAAE,mCAAmC,CAAC;IACnD,QAAQ,EAAE;QACR,SAAS,EAAE,MAAM,CAAC;QAClB,cAAc,EAAE,MAAM,CAAC;QACvB,MAAM,EAAE,uBAAuB,CAAC;QAChC,WAAW,EAAE,MAAM,EAAE,CAAC;KACvB,CAAC;IACF,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,UAAU,4BAA4B;IACpC,eAAe,EAAE,+BAA+B,CAAC;IACjD,SAAS,EAAE,+BAA+B,CAAC;IAC3C,YAAY,EAAE,+BAA+B,CAAC;IAC9C,sBAAsB,EAAE,+BAA+B,CAAC;IACxD,sBAAsB,EAAE,+BAA+B,CAAC;CACzD;AAED,MAAM,WAAW,0BAA0B;IACzC,aAAa,EAAE,0CAA0C,CAAC;IAC1D,QAAQ,EAAE,MAAM,CAAC;IACjB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,eAAe,EAAE,MAAM,CAAC;IACxB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC,iBAAiB,EAAE,MAAM,CAAC,CAAC;IAC/C,qBAAqB,EAAE,MAAM,CAAC,iBAAiB,EAAE,MAAM,CAAC,CAAC;IACzD,KAAK,EAAE,8BAA8B,EAAE,CAAC;IACxC,QAAQ,EAAE,MAAM,CAAC,iBAAiB,EAAE,4BAA4B,CAAC,CAAC;IAClE,gBAAgB,EAAE,qBAAqB,CAAC;IACxC,yBAAyB,EAAE,qBAAqB,CAAC;IACjD,2BAA2B,EAAE,qBAAqB,CAAC;IACnD,WAAW,EAAE;QAAE,MAAM,EAAE,aAAa,CAAC;QAAC,UAAU,EAAE,sBAAsB,CAAA;KAAE,CAAC;IAC3E,YAAY,EAAE;QAAE,MAAM,EAAE,aAAa,CAAC;QAAC,UAAU,EAAE,+BAA+B,CAAA;KAAE,CAAC;IACrF,eAAe,EAAE,OAAO,CAAC;CAC1B;AAED,wBAAsB,4BAA4B,CAAC,KAAK,EAAE;IACxD,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,oBAAoB,CAAC;CAChC,GAAG,OAAO,CAAC,0BAA0B,CAAC,CAmFtC"}
|
|
@@ -0,0 +1,321 @@
|
|
|
1
|
+
import { readFileSync } from 'node:fs';
|
|
2
|
+
import { join } from 'node:path';
|
|
3
|
+
import Database from 'better-sqlite3';
|
|
4
|
+
import { DevFlowDatabase } from '@devflow-tools/database';
|
|
5
|
+
import { evaluationContextStableRef, evaluationKnowledgeStableRef, evaluationMemoryStableRef, createRuntimeActivationSnapshotV1, } from '@devflow-tools/sdk';
|
|
6
|
+
import { createEngines } from '@devflow-tools/mcp-server/composition';
|
|
7
|
+
import { RetrievalCoordinator, RetrievalSessionRepository } from '@devflow-tools/retrieval-engine';
|
|
8
|
+
import { resolveTaskSemanticContext, } from '@devflow-tools/semantic-engine';
|
|
9
|
+
import { aggregateEvidenceQualityMetrics, candidateRecallAt20, evidenceQualityMetric, forbiddenContaminationAt5, precisionAt5, recallAt5, selectBestGoldSet, } from './evidence-quality-metrics.js';
|
|
10
|
+
import { createEvaluationRunState } from './evaluation-run-state.js';
|
|
11
|
+
import { captureProducer, identityFor, noOpTaskRuntime, rankedEvidenceRefs, restrictBenchmarkRetrievalChannel, stageCapture, } from './unified-runtime-runner.js';
|
|
12
|
+
import { validateEvaluationBundle, } from './evaluation-bundle.js';
|
|
13
|
+
import { createReleaseQualityQuerySamplingReplay, parseReleaseQualityQuerySampling, } from './release-quality-query-sampling.js';
|
|
14
|
+
export async function evaluateReleaseQualityBundle(input) {
|
|
15
|
+
const bundlePath = input.bundlePath;
|
|
16
|
+
const validation = validateEvaluationBundle(bundlePath);
|
|
17
|
+
if (validation.status !== 'valid')
|
|
18
|
+
throw new Error(`EVALUATION_BUNDLE_INVALID:${validation.reasonCode}`);
|
|
19
|
+
const { manifest } = validation;
|
|
20
|
+
const corpus = readJson(join(bundlePath, 'corpus.json'));
|
|
21
|
+
const profiles = readJson(join(bundlePath, 'profiles.json'));
|
|
22
|
+
const labels = readLabels(bundlePath);
|
|
23
|
+
const runState = createEvaluationRunState({
|
|
24
|
+
bundlePath,
|
|
25
|
+
kind: 'retrieval',
|
|
26
|
+
caseId: manifest.bundleId.slice('bundle:'.length, 'bundle:'.length + 16),
|
|
27
|
+
});
|
|
28
|
+
const engines = await createEngines(runState.dataSources.sourceRoot, {
|
|
29
|
+
warmSymbolVectors: false,
|
|
30
|
+
dataSources: runState.dataSources,
|
|
31
|
+
});
|
|
32
|
+
const memoryDatabase = new Database(runState.dataSources.memoryDatabasePath, { readonly: true });
|
|
33
|
+
const knowledgeDatabase = new Database(runState.dataSources.knowledgeDatabasePath, { readonly: true });
|
|
34
|
+
const operationalDatabase = new DevFlowDatabase(runState.dataSources.sourceRoot, {
|
|
35
|
+
dbPath: runState.dataSources.operationalDatabasePath,
|
|
36
|
+
});
|
|
37
|
+
try {
|
|
38
|
+
engines.knowledge.activatePersistedLocalSources();
|
|
39
|
+
const cases = [];
|
|
40
|
+
for (const rawCase of corpus.cases) {
|
|
41
|
+
const item = releaseQualityCase(rawCase);
|
|
42
|
+
const label = labels[item.channel].labels.find(candidate => candidate.caseId === item.caseId);
|
|
43
|
+
if (!label)
|
|
44
|
+
throw new Error(`EVALUATION_LABEL_CASE_MISSING:${item.caseId}`);
|
|
45
|
+
cases.push(await evaluateCase({
|
|
46
|
+
item,
|
|
47
|
+
label,
|
|
48
|
+
manifest,
|
|
49
|
+
engines,
|
|
50
|
+
memoryDatabase,
|
|
51
|
+
knowledgeDatabase,
|
|
52
|
+
operationalDatabase,
|
|
53
|
+
profiles,
|
|
54
|
+
reranker: input.reranker,
|
|
55
|
+
sourceRoot: runState.dataSources.sourceRoot,
|
|
56
|
+
}));
|
|
57
|
+
}
|
|
58
|
+
await runState.verifyBundleUnchanged();
|
|
59
|
+
const positive = cases.filter(item => item.caseClass === 'positive');
|
|
60
|
+
const channels = Object.fromEntries(['context', 'memory', 'knowledge'].map(channel => {
|
|
61
|
+
const channelCases = positive.filter(item => item.channel === channel);
|
|
62
|
+
return [channel, aggregateChannel(channelCases)];
|
|
63
|
+
}));
|
|
64
|
+
const lifecycle = cases.filter(item => item.channel === 'memory' && item.caseClass === 'lifecycle');
|
|
65
|
+
return {
|
|
66
|
+
schemaVersion: 'devflow.release-quality-bundle-report.v1',
|
|
67
|
+
bundleId: manifest.bundleId,
|
|
68
|
+
projectSnapshotId: manifest.projectSnapshotId,
|
|
69
|
+
builderIdentity: manifest.builderIdentity,
|
|
70
|
+
corpusHash: manifest.corpusHash,
|
|
71
|
+
labelHashes: manifest.labelHashes,
|
|
72
|
+
channelSnapshotHashes: manifest.channelSnapshotHashes,
|
|
73
|
+
cases,
|
|
74
|
+
channels,
|
|
75
|
+
memoryDurability: evidenceQualityMetric(lifecycle.filter(item => item.recallAt5.numerator === item.recallAt5.denominator).length, lifecycle.length),
|
|
76
|
+
memoryTemporalCorrectness: evidenceQualityMetric(lifecycle.filter(item => item.expectedLifecycle === 'latest_episode_selected'
|
|
77
|
+
&& item.recallAt5.numerator === item.recallAt5.denominator
|
|
78
|
+
&& item.forbiddenContamination.numerator === 0).length, lifecycle.filter(item => item.expectedLifecycle === 'latest_episode_selected').length),
|
|
79
|
+
knowledgeVersionCorrectness: combine(cases
|
|
80
|
+
.filter(item => item.channel === 'knowledge' && item.caseClass === 'positive')
|
|
81
|
+
.map(item => item.sourceVersionCorrectness)),
|
|
82
|
+
consumption: { status: 'unavailable', reasonCode: 'host_ledger_required' },
|
|
83
|
+
verification: { status: 'unavailable', reasonCode: 'verification_receipt_required' },
|
|
84
|
+
bundleUnchanged: true,
|
|
85
|
+
};
|
|
86
|
+
}
|
|
87
|
+
finally {
|
|
88
|
+
memoryDatabase.close();
|
|
89
|
+
knowledgeDatabase.close();
|
|
90
|
+
operationalDatabase.close();
|
|
91
|
+
closeEngines(engines);
|
|
92
|
+
runState.dispose();
|
|
93
|
+
}
|
|
94
|
+
}
|
|
95
|
+
async function evaluateCase(input) {
|
|
96
|
+
const startedAt = performance.now();
|
|
97
|
+
const resolved = await canonicalRefs(input);
|
|
98
|
+
const candidateRefs = resolved.candidateRefs.slice(0, 20);
|
|
99
|
+
const selectedRefs = resolved.selectedRefs.slice(0, 5);
|
|
100
|
+
const gold = new Set(selectBestGoldSet({
|
|
101
|
+
goldSets: input.label.goldSets,
|
|
102
|
+
candidateRefs,
|
|
103
|
+
selectedRefs,
|
|
104
|
+
}));
|
|
105
|
+
return {
|
|
106
|
+
caseId: input.item.caseId,
|
|
107
|
+
channel: input.item.channel,
|
|
108
|
+
caseClass: input.item.caseClass,
|
|
109
|
+
...(input.item.expectedLifecycle ? { expectedLifecycle: input.item.expectedLifecycle } : {}),
|
|
110
|
+
candidateRefs,
|
|
111
|
+
selectedRefs,
|
|
112
|
+
candidateRecall: candidateRecallAt20(candidateRefs, gold),
|
|
113
|
+
recallAt5: recallAt5(selectedRefs, gold),
|
|
114
|
+
precisionAt5: precisionAt5(selectedRefs, gold),
|
|
115
|
+
forbiddenContamination: forbiddenContaminationAt5(selectedRefs, new Set(input.label.forbiddenRefs)),
|
|
116
|
+
provenanceCompleteness: evidenceQualityMetric(resolved.resolved, resolved.returned),
|
|
117
|
+
sourceVersionCorrectness: input.item.channel === 'knowledge'
|
|
118
|
+
? evidenceQualityMetric(resolved.versionCorrect, resolved.returned)
|
|
119
|
+
: evidenceQualityMetric(0, 0),
|
|
120
|
+
querySampling: resolved.querySampling,
|
|
121
|
+
reranker: resolved.reranker,
|
|
122
|
+
durationMs: Math.max(0, Math.round((performance.now() - startedAt) * 1000) / 1000),
|
|
123
|
+
};
|
|
124
|
+
}
|
|
125
|
+
async function canonicalRefs(input) {
|
|
126
|
+
const identity = identityFor(input.item.caseId, 'benchmark:release-quality', input.sourceRoot, input.manifest.projectId);
|
|
127
|
+
const querySampling = createReleaseQualityQuerySamplingReplay(input.item.querySampling);
|
|
128
|
+
const semantic = await resolveTaskSemanticContext({
|
|
129
|
+
normalizedPrompt: input.item.prompt,
|
|
130
|
+
identity,
|
|
131
|
+
revision: 1,
|
|
132
|
+
activation: createRuntimeActivationSnapshotV1({
|
|
133
|
+
semantic: 'authoritative',
|
|
134
|
+
retrieval: 'authoritative',
|
|
135
|
+
memoryExtraction: 'deferred',
|
|
136
|
+
knowledgeCompilation: 'deferred',
|
|
137
|
+
embeddingProfiles: {
|
|
138
|
+
code_retrieval: input.profiles.embeddingProfiles.code_retrieval.profileHash,
|
|
139
|
+
},
|
|
140
|
+
rolloutPolicyVersion: 'release-quality-bundle.v1',
|
|
141
|
+
createdAt: 1,
|
|
142
|
+
}),
|
|
143
|
+
router: input.engines.semanticRouter,
|
|
144
|
+
retrievalQuerySampling: {
|
|
145
|
+
sample: querySampling.sample,
|
|
146
|
+
},
|
|
147
|
+
policy: {
|
|
148
|
+
policyHash: 'release-quality-bundle.v1',
|
|
149
|
+
budgets: {
|
|
150
|
+
hookDeadlineMs: 3_000,
|
|
151
|
+
contextTokens: 8_000,
|
|
152
|
+
responseReserveTokens: 2_000,
|
|
153
|
+
perChannelMs: { code: 1_500, memory: 800, knowledge: 1_500 },
|
|
154
|
+
},
|
|
155
|
+
},
|
|
156
|
+
now: 1,
|
|
157
|
+
});
|
|
158
|
+
const querySamplingEvidence = querySampling.evidence();
|
|
159
|
+
const context = restrictBenchmarkRetrievalChannel(semantic.context, input.item.channel, input.item.prompt.normalize('NFKC'));
|
|
160
|
+
const stage = stageCapture();
|
|
161
|
+
const result = await new RetrievalCoordinator({
|
|
162
|
+
code: captureProducer(input.engines.codeEvidence, stage),
|
|
163
|
+
memory: captureProducer(input.engines.memoryQuery, stage),
|
|
164
|
+
knowledge: captureProducer(input.engines.knowledgeQuery, stage),
|
|
165
|
+
sessions: new RetrievalSessionRepository(input.operationalDatabase),
|
|
166
|
+
reranker: input.reranker,
|
|
167
|
+
taskRuntime: noOpTaskRuntime(),
|
|
168
|
+
}).retrieve({ context, taskRevision: 1, tokenBudget: 8_000 });
|
|
169
|
+
const channel = input.item.channel === 'context' ? 'code' : input.item.channel;
|
|
170
|
+
const reranker = result.receipt.reranker?.providerReceipt;
|
|
171
|
+
if (!reranker || reranker.status !== 'ready') {
|
|
172
|
+
const legs = (result.receipt.legs ?? [])
|
|
173
|
+
.filter(leg => leg.channel === channel)
|
|
174
|
+
.map(leg => `${leg.status}:${leg.candidateCount}:${leg.reasonCode ?? 'none'}`)
|
|
175
|
+
.join(',');
|
|
176
|
+
throw new Error(`RELEASE_QUALITY_RERANKER_UNAVAILABLE:${input.item.caseId}:${channel}:`
|
|
177
|
+
+ `${reranker?.status ?? 'missing'}:${reranker?.reasonCodes.join(',') ?? 'none'}`
|
|
178
|
+
+ `:candidates=${stage.candidates.length}:legs=${legs || 'none'}`);
|
|
179
|
+
}
|
|
180
|
+
const candidates = stage.candidates.filter(candidate => candidate.channel === channel);
|
|
181
|
+
const mapped = new Map();
|
|
182
|
+
let resolved = 0;
|
|
183
|
+
let versionCorrect = 0;
|
|
184
|
+
for (const candidate of candidates) {
|
|
185
|
+
const reference = evaluationRef(input, candidate);
|
|
186
|
+
if (!reference)
|
|
187
|
+
continue;
|
|
188
|
+
mapped.set(candidate.stableRef, reference.ref);
|
|
189
|
+
resolved += 1;
|
|
190
|
+
versionCorrect += reference.versionCorrect;
|
|
191
|
+
}
|
|
192
|
+
return {
|
|
193
|
+
candidateRefs: unique(rankedEvidenceRefs(result.receipt, input.item.channel)
|
|
194
|
+
.flatMap(stableRef => mapped.get(stableRef) ?? [])),
|
|
195
|
+
selectedRefs: unique((result.receipt.selectedEvidence ?? [])
|
|
196
|
+
.filter(candidate => candidate.channel === channel)
|
|
197
|
+
.flatMap(candidate => mapped.get(candidate.stableRef) ?? [])),
|
|
198
|
+
returned: candidates.length,
|
|
199
|
+
resolved,
|
|
200
|
+
versionCorrect,
|
|
201
|
+
querySampling: querySamplingEvidence,
|
|
202
|
+
reranker: {
|
|
203
|
+
profileId: result.receipt.reranker.profileId,
|
|
204
|
+
profileVersion: result.receipt.reranker.profileVersion,
|
|
205
|
+
status: reranker.status,
|
|
206
|
+
reasonCodes: reranker.reasonCodes,
|
|
207
|
+
},
|
|
208
|
+
};
|
|
209
|
+
}
|
|
210
|
+
function evaluationRef(input, candidate) {
|
|
211
|
+
if (candidate.channel === 'code') {
|
|
212
|
+
const path = stringPayload(candidate, 'path');
|
|
213
|
+
const symbolId = stringPayload(candidate, 'symbolId');
|
|
214
|
+
if (!path || !symbolId)
|
|
215
|
+
return undefined;
|
|
216
|
+
return { ref: evaluationContextStableRef({
|
|
217
|
+
projectSnapshotId: input.manifest.projectSnapshotId,
|
|
218
|
+
normalizedPath: path,
|
|
219
|
+
symbolId,
|
|
220
|
+
range: [integerPayload(candidate, 'startLine'), integerPayload(candidate, 'endLine')],
|
|
221
|
+
}), versionCorrect: 0 };
|
|
222
|
+
}
|
|
223
|
+
if (candidate.channel === 'memory') {
|
|
224
|
+
const id = stringPayload(candidate, 'id');
|
|
225
|
+
if (!id)
|
|
226
|
+
return undefined;
|
|
227
|
+
if (candidate.payload.type === 'cold_episode') {
|
|
228
|
+
const row = input.memoryDatabase.prepare('SELECT episode_hash FROM memory_episodes_v1 WHERE episode_id = ?')
|
|
229
|
+
.get(id);
|
|
230
|
+
return row ? { ref: evaluationMemoryStableRef({
|
|
231
|
+
channelSnapshotHash: input.manifest.channelSnapshotHashes.memory,
|
|
232
|
+
kind: 'episode',
|
|
233
|
+
immutableId: id,
|
|
234
|
+
version: row.episode_hash,
|
|
235
|
+
}), versionCorrect: 0 } : undefined;
|
|
236
|
+
}
|
|
237
|
+
const row = input.memoryDatabase.prepare('SELECT version FROM memories WHERE id = ?')
|
|
238
|
+
.get(id);
|
|
239
|
+
return row ? { ref: evaluationMemoryStableRef({
|
|
240
|
+
channelSnapshotHash: input.manifest.channelSnapshotHashes.memory,
|
|
241
|
+
kind: 'fact',
|
|
242
|
+
immutableId: id,
|
|
243
|
+
version: row.version,
|
|
244
|
+
}), versionCorrect: 0 } : undefined;
|
|
245
|
+
}
|
|
246
|
+
const id = stringPayload(candidate, 'id');
|
|
247
|
+
if (!id)
|
|
248
|
+
return undefined;
|
|
249
|
+
const row = input.knowledgeDatabase.prepare(`SELECT c.content_hash, d.source, d.version
|
|
250
|
+
FROM knowledge_chunks c JOIN knowledge_docs d ON d.id = c.doc_id WHERE c.id = ?`);
|
|
251
|
+
const knowledge = row.get(id);
|
|
252
|
+
if (!knowledge)
|
|
253
|
+
return undefined;
|
|
254
|
+
return { ref: evaluationKnowledgeStableRef({
|
|
255
|
+
sourceId: knowledge.source,
|
|
256
|
+
sourceVersion: knowledge.version,
|
|
257
|
+
contentHash: knowledge.content_hash,
|
|
258
|
+
chunkId: id,
|
|
259
|
+
}), versionCorrect: candidate.payload.version === knowledge.version ? 1 : 0 };
|
|
260
|
+
}
|
|
261
|
+
function stringPayload(candidate, key) {
|
|
262
|
+
const value = candidate.payload[key];
|
|
263
|
+
return typeof value === 'string' && value.length > 0 ? value : undefined;
|
|
264
|
+
}
|
|
265
|
+
function integerPayload(candidate, key) {
|
|
266
|
+
const value = candidate.payload[key];
|
|
267
|
+
return Number.isSafeInteger(value) && value >= 0 ? value : 0;
|
|
268
|
+
}
|
|
269
|
+
function unique(refs) {
|
|
270
|
+
return [...new Set(refs)];
|
|
271
|
+
}
|
|
272
|
+
function aggregateChannel(cases) {
|
|
273
|
+
return {
|
|
274
|
+
candidateRecall: aggregateEvidenceQualityMetrics(cases.map(item => item.candidateRecall)),
|
|
275
|
+
recallAt5: aggregateEvidenceQualityMetrics(cases.map(item => item.recallAt5)),
|
|
276
|
+
precisionAt5: aggregateEvidenceQualityMetrics(cases.map(item => item.precisionAt5)),
|
|
277
|
+
forbiddenContamination: aggregateEvidenceQualityMetrics(cases.map(item => item.forbiddenContamination)),
|
|
278
|
+
provenanceCompleteness: aggregateEvidenceQualityMetrics(cases.map(item => item.provenanceCompleteness)),
|
|
279
|
+
};
|
|
280
|
+
}
|
|
281
|
+
function combine(metrics) {
|
|
282
|
+
return evidenceQualityMetric(metrics.reduce((total, item) => total + item.numerator, 0), metrics.reduce((total, item) => total + item.denominator, 0));
|
|
283
|
+
}
|
|
284
|
+
function readLabels(bundlePath) {
|
|
285
|
+
const entries = ['context', 'memory', 'knowledge'].map(channel => [
|
|
286
|
+
channel,
|
|
287
|
+
readJson(join(bundlePath, `labels/${channel}.json`)),
|
|
288
|
+
]);
|
|
289
|
+
return Object.fromEntries(entries);
|
|
290
|
+
}
|
|
291
|
+
function readJson(path) {
|
|
292
|
+
return JSON.parse(readFileSync(path, 'utf8'));
|
|
293
|
+
}
|
|
294
|
+
function releaseQualityCase(value) {
|
|
295
|
+
if (typeof value.caseId !== 'string'
|
|
296
|
+
|| !['context', 'memory', 'knowledge'].includes(value.channel)
|
|
297
|
+
|| (value.caseClass !== 'positive' && value.caseClass !== 'lifecycle')
|
|
298
|
+
|| typeof value.prompt !== 'string' || !value.prompt.trim()
|
|
299
|
+
|| (value.expectedLifecycle !== undefined
|
|
300
|
+
&& value.expectedLifecycle !== 'explicit_save_durable'
|
|
301
|
+
&& value.expectedLifecycle !== 'latest_episode_selected')) {
|
|
302
|
+
throw new Error(`EVALUATION_CORPUS_CASE_INVALID:${String(value.caseId)}`);
|
|
303
|
+
}
|
|
304
|
+
return {
|
|
305
|
+
caseId: value.caseId,
|
|
306
|
+
channel: value.channel,
|
|
307
|
+
caseClass: value.caseClass,
|
|
308
|
+
prompt: value.prompt,
|
|
309
|
+
querySampling: parseReleaseQualityQuerySampling(value.querySampling),
|
|
310
|
+
...(value.expectedLifecycle ? { expectedLifecycle: value.expectedLifecycle } : {}),
|
|
311
|
+
};
|
|
312
|
+
}
|
|
313
|
+
function closeEngines(engines) {
|
|
314
|
+
engines.context.close();
|
|
315
|
+
engines.memory.close();
|
|
316
|
+
engines.knowledge.close();
|
|
317
|
+
engines.telemetry.close();
|
|
318
|
+
engines.codegraph.close();
|
|
319
|
+
engines.data.close?.();
|
|
320
|
+
}
|
|
321
|
+
//# sourceMappingURL=release-quality-bundle-evaluator.js.map
|