@tangle-network/agent-eval 0.93.0 → 0.95.0
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/CHANGELOG.md +15 -0
- package/README.md +44 -30
- package/dist/action-policy.d.ts +24 -0
- package/dist/action-policy.d.ts.map +1 -0
- package/dist/action-policy.test.d.ts +2 -0
- package/dist/action-policy.test.d.ts.map +1 -0
- package/dist/active-learning.d.ts +41 -0
- package/dist/active-learning.d.ts.map +1 -0
- package/dist/adapters/http.d.ts +15 -21
- package/dist/adapters/http.d.ts.map +1 -0
- package/dist/adapters/http.js.map +1 -1
- package/dist/adapters/langchain.d.ts +7 -13
- package/dist/adapters/langchain.d.ts.map +1 -0
- package/dist/adapters/otel.d.ts +13 -24
- package/dist/adapters/otel.d.ts.map +1 -0
- package/dist/agent-profile-cell.d.ts +101 -0
- package/dist/agent-profile-cell.d.ts.map +1 -0
- package/dist/agent-profile.d.ts +27 -0
- package/dist/agent-profile.d.ts.map +1 -0
- package/dist/agent-profile.test.d.ts +2 -0
- package/dist/agent-profile.test.d.ts.map +1 -0
- package/dist/analyst/adapters.d.ts +62 -0
- package/dist/analyst/adapters.d.ts.map +1 -0
- package/dist/analyst/analyst.test.d.ts +2 -0
- package/dist/analyst/analyst.test.d.ts.map +1 -0
- package/dist/analyst/ax-service.d.ts +27 -0
- package/dist/analyst/ax-service.d.ts.map +1 -0
- package/dist/analyst/behavioral-analyst.d.ts +28 -0
- package/dist/analyst/behavioral-analyst.d.ts.map +1 -0
- package/dist/analyst/chat-client.d.ts +91 -0
- package/dist/analyst/chat-client.d.ts.map +1 -0
- package/dist/analyst/default-registry.d.ts +27 -0
- package/dist/analyst/default-registry.d.ts.map +1 -0
- package/dist/analyst/default-registry.test.d.ts +2 -0
- package/dist/analyst/default-registry.test.d.ts.map +1 -0
- package/dist/analyst/finding-signature.d.ts +48 -0
- package/dist/analyst/finding-signature.d.ts.map +1 -0
- package/dist/analyst/finding-subject.d.ts +146 -0
- package/dist/analyst/finding-subject.d.ts.map +1 -0
- package/dist/analyst/finding-subject.test.d.ts +2 -0
- package/dist/analyst/finding-subject.test.d.ts.map +1 -0
- package/dist/analyst/findings-store.d.ts +75 -0
- package/dist/analyst/findings-store.d.ts.map +1 -0
- package/dist/analyst/index.d.ts +28 -256
- package/dist/analyst/index.d.ts.map +1 -0
- package/dist/analyst/index.js +6 -66
- package/dist/analyst/index.js.map +1 -1
- package/dist/{kind-factory-5b7xXXOr.d.ts → analyst/kind-factory.d.ts} +11 -63
- package/dist/analyst/kind-factory.d.ts.map +1 -0
- package/dist/analyst/kinds/failure-mode.d.ts +19 -0
- package/dist/analyst/kinds/failure-mode.d.ts.map +1 -0
- package/dist/analyst/kinds/improvement.d.ts +23 -0
- package/dist/analyst/kinds/improvement.d.ts.map +1 -0
- package/dist/analyst/kinds/index.d.ts +22 -0
- package/dist/analyst/kinds/index.d.ts.map +1 -0
- package/dist/analyst/kinds/kinds.test.d.ts +2 -0
- package/dist/analyst/kinds/kinds.test.d.ts.map +1 -0
- package/dist/analyst/kinds/knowledge-gap.d.ts +28 -0
- package/dist/analyst/kinds/knowledge-gap.d.ts.map +1 -0
- package/dist/analyst/kinds/knowledge-poisoning.d.ts +22 -0
- package/dist/analyst/kinds/knowledge-poisoning.d.ts.map +1 -0
- package/dist/analyst/kinds/skill-usage.d.ts +84 -0
- package/dist/analyst/kinds/skill-usage.d.ts.map +1 -0
- package/dist/analyst/kinds/skill-usage.test.d.ts +2 -0
- package/dist/analyst/kinds/skill-usage.test.d.ts.map +1 -0
- package/dist/analyst/parse-tolerant.d.ts +26 -0
- package/dist/analyst/parse-tolerant.d.ts.map +1 -0
- package/dist/analyst/parse-tolerant.test.d.ts +2 -0
- package/dist/analyst/parse-tolerant.test.d.ts.map +1 -0
- package/dist/analyst/registry.budget.test.d.ts +2 -0
- package/dist/analyst/registry.budget.test.d.ts.map +1 -0
- package/dist/{default-registry-zoGHUQEH.d.ts → analyst/registry.d.ts} +8 -37
- package/dist/analyst/registry.d.ts.map +1 -0
- package/dist/analyst/steer-firewall.d.ts +35 -0
- package/dist/analyst/steer-firewall.d.ts.map +1 -0
- package/dist/analyst/steer-firewall.test.d.ts +2 -0
- package/dist/analyst/steer-firewall.test.d.ts.map +1 -0
- package/dist/analyst/structure-findings.d.ts +37 -0
- package/dist/analyst/structure-findings.d.ts.map +1 -0
- package/dist/analyst/structure-findings.test.d.ts +2 -0
- package/dist/analyst/structure-findings.test.d.ts.map +1 -0
- package/dist/analyst/tool-groups.d.ts +34 -0
- package/dist/analyst/tool-groups.d.ts.map +1 -0
- package/dist/{types-2VVIL04s.d.ts → analyst/types.d.ts} +19 -112
- package/dist/analyst/types.d.ts.map +1 -0
- package/dist/anti-slop.d.ts +59 -0
- package/dist/anti-slop.d.ts.map +1 -0
- package/dist/artifact-validator.d.ts +74 -0
- package/dist/artifact-validator.d.ts.map +1 -0
- package/dist/attestation.d.ts +63 -0
- package/dist/attestation.d.ts.map +1 -0
- package/dist/attestation.test.d.ts +2 -0
- package/dist/attestation.test.d.ts.map +1 -0
- package/dist/authenticity/index.d.ts +15 -16
- package/dist/authenticity/index.d.ts.map +1 -0
- package/dist/authenticity/index.test.d.ts +2 -0
- package/dist/authenticity/index.test.d.ts.map +1 -0
- package/dist/auto-pr.d.ts +120 -0
- package/dist/auto-pr.d.ts.map +1 -0
- package/dist/{baseline-DE36-Np7.d.ts → baseline.d.ts} +8 -44
- package/dist/baseline.d.ts.map +1 -0
- package/dist/behavior-dsl.d.ts +73 -0
- package/dist/behavior-dsl.d.ts.map +1 -0
- package/dist/belief-state/calibration.d.ts +10 -0
- package/dist/belief-state/calibration.d.ts.map +1 -0
- package/dist/belief-state/calibration.test.d.ts +2 -0
- package/dist/belief-state/calibration.test.d.ts.map +1 -0
- package/dist/belief-state/code-agent-corpus.d.ts +66 -0
- package/dist/belief-state/code-agent-corpus.d.ts.map +1 -0
- package/dist/belief-state/code-agent-corpus.test.d.ts +2 -0
- package/dist/belief-state/code-agent-corpus.test.d.ts.map +1 -0
- package/dist/belief-state/code-agent-evidence.d.ts +22 -0
- package/dist/belief-state/code-agent-evidence.d.ts.map +1 -0
- package/dist/belief-state/code-agent-evidence.test.d.ts +2 -0
- package/dist/belief-state/code-agent-evidence.test.d.ts.map +1 -0
- package/dist/belief-state/extract.d.ts +7 -0
- package/dist/belief-state/extract.d.ts.map +1 -0
- package/dist/belief-state/extract.test.d.ts +2 -0
- package/dist/belief-state/extract.test.d.ts.map +1 -0
- package/dist/belief-state/index.d.ts +14 -604
- package/dist/belief-state/index.d.ts.map +1 -0
- package/dist/belief-state/index.js +1 -1
- package/dist/belief-state/ope.d.ts +17 -0
- package/dist/belief-state/ope.d.ts.map +1 -0
- package/dist/belief-state/ope.test.d.ts +2 -0
- package/dist/belief-state/ope.test.d.ts.map +1 -0
- package/dist/belief-state/phase0-measurement.d.ts +55 -0
- package/dist/belief-state/phase0-measurement.d.ts.map +1 -0
- package/dist/belief-state/report.d.ts +17 -0
- package/dist/belief-state/report.d.ts.map +1 -0
- package/dist/belief-state/report.test.d.ts +2 -0
- package/dist/belief-state/report.test.d.ts.map +1 -0
- package/dist/belief-state/research-evidence.d.ts +23 -0
- package/dist/belief-state/research-evidence.d.ts.map +1 -0
- package/dist/belief-state/research-evidence.test.d.ts +2 -0
- package/dist/belief-state/research-evidence.test.d.ts.map +1 -0
- package/dist/belief-state/runtime-benchmark-corpus.d.ts +32 -0
- package/dist/belief-state/runtime-benchmark-corpus.d.ts.map +1 -0
- package/dist/belief-state/runtime-hooks.d.ts +87 -0
- package/dist/belief-state/runtime-hooks.d.ts.map +1 -0
- package/dist/belief-state/runtime-hooks.test.d.ts +2 -0
- package/dist/belief-state/runtime-hooks.test.d.ts.map +1 -0
- package/dist/belief-state/selective.d.ts +15 -0
- package/dist/belief-state/selective.d.ts.map +1 -0
- package/dist/belief-state/selective.test.d.ts +2 -0
- package/dist/belief-state/selective.test.d.ts.map +1 -0
- package/dist/belief-state/shadow-probe.d.ts +80 -0
- package/dist/belief-state/shadow-probe.d.ts.map +1 -0
- package/dist/belief-state/shadow-probe.test.d.ts +2 -0
- package/dist/belief-state/shadow-probe.test.d.ts.map +1 -0
- package/dist/belief-state/types.d.ts +195 -0
- package/dist/belief-state/types.d.ts.map +1 -0
- package/dist/belief-state/types.test.d.ts +2 -0
- package/dist/belief-state/types.test.d.ts.map +1 -0
- package/dist/benchmark.d.ts +14 -0
- package/dist/benchmark.d.ts.map +1 -0
- package/dist/benchmarks/index.d.ts +23 -4
- package/dist/benchmarks/index.d.ts.map +1 -0
- package/dist/benchmarks/routing/dataset.d.ts +34 -0
- package/dist/benchmarks/routing/dataset.d.ts.map +1 -0
- package/dist/benchmarks/routing/index.d.ts +34 -0
- package/dist/benchmarks/routing/index.d.ts.map +1 -0
- package/dist/benchmarks/types.d.ts +49 -0
- package/dist/benchmarks/types.d.ts.map +1 -0
- package/dist/bisector.d.ts +81 -0
- package/dist/bisector.d.ts.map +1 -0
- package/dist/budget-guard.d.ts +31 -0
- package/dist/budget-guard.d.ts.map +1 -0
- package/dist/builder-eval/builder-session.d.ts +111 -0
- package/dist/builder-eval/builder-session.d.ts.map +1 -0
- package/dist/builder-eval/correlation.d.ts +32 -0
- package/dist/builder-eval/correlation.d.ts.map +1 -0
- package/dist/builder-eval/index.d.ts +5 -250
- package/dist/builder-eval/index.d.ts.map +1 -0
- package/dist/builder-eval/index.js +2 -2
- package/dist/builder-eval/project-registry.d.ts +51 -0
- package/dist/builder-eval/project-registry.d.ts.map +1 -0
- package/dist/builder-eval/three-layer-eval.d.ts +55 -0
- package/dist/builder-eval/three-layer-eval.d.ts.map +1 -0
- package/dist/campaign/analyst-surface.d.ts +108 -0
- package/dist/campaign/analyst-surface.d.ts.map +1 -0
- package/dist/campaign/analyst-surface.test.d.ts +2 -0
- package/dist/campaign/analyst-surface.test.d.ts.map +1 -0
- package/dist/campaign/auto-pr.d.ts +46 -0
- package/dist/campaign/auto-pr.d.ts.map +1 -0
- package/dist/campaign/distillation/agreement-judge.d.ts +69 -0
- package/dist/campaign/distillation/agreement-judge.d.ts.map +1 -0
- package/dist/campaign/distillation/cli.d.ts +35 -0
- package/dist/campaign/distillation/cli.d.ts.map +1 -0
- package/dist/campaign/distillation/distillation.test.d.ts +2 -0
- package/dist/campaign/distillation/distillation.test.d.ts.map +1 -0
- package/dist/campaign/distillation/gold-scenarios.d.ts +54 -0
- package/dist/campaign/distillation/gold-scenarios.d.ts.map +1 -0
- package/dist/campaign/distillation/run-distillation.d.ts +119 -0
- package/dist/campaign/distillation/run-distillation.d.ts.map +1 -0
- package/dist/campaign/gates/compose.d.ts +12 -0
- package/dist/campaign/gates/compose.d.ts.map +1 -0
- package/dist/campaign/gates/default-production-gate.d.ts +58 -0
- package/dist/campaign/gates/default-production-gate.d.ts.map +1 -0
- package/dist/campaign/gates/heldout-gate.d.ts +12 -0
- package/dist/campaign/gates/heldout-gate.d.ts.map +1 -0
- package/dist/campaign/gates/promotion-policy.d.ts +125 -0
- package/dist/campaign/gates/promotion-policy.d.ts.map +1 -0
- package/dist/campaign/gates/promotion-policy.test.d.ts +2 -0
- package/dist/campaign/gates/promotion-policy.test.d.ts.map +1 -0
- package/dist/campaign/gates/sequential.d.ts +146 -0
- package/dist/campaign/gates/sequential.d.ts.map +1 -0
- package/dist/campaign/gates/sequential.test.d.ts +2 -0
- package/dist/campaign/gates/sequential.test.d.ts.map +1 -0
- package/dist/campaign/gates/statistical-heldout.d.ts +99 -0
- package/dist/campaign/gates/statistical-heldout.d.ts.map +1 -0
- package/dist/campaign/gates/statistical-heldout.test.d.ts +2 -0
- package/dist/campaign/gates/statistical-heldout.test.d.ts.map +1 -0
- package/dist/campaign/index.d.ts +38 -1341
- package/dist/campaign/index.d.ts.map +1 -0
- package/dist/campaign/index.js +1865 -1318
- package/dist/campaign/index.js.map +1 -1
- package/dist/campaign/labeled-store/fs-adapter.d.ts +59 -0
- package/dist/campaign/labeled-store/fs-adapter.d.ts.map +1 -0
- package/dist/campaign/presets/compare-proposers.d.ts +146 -0
- package/dist/campaign/presets/compare-proposers.d.ts.map +1 -0
- package/dist/campaign/presets/playback.d.ts +120 -0
- package/dist/campaign/presets/playback.d.ts.map +1 -0
- package/dist/campaign/presets/playback.test.d.ts +2 -0
- package/dist/campaign/presets/playback.test.d.ts.map +1 -0
- package/dist/campaign/presets/run-eval.d.ts +14 -0
- package/dist/campaign/presets/run-eval.d.ts.map +1 -0
- package/dist/campaign/presets/run-improvement-loop.d.ts +63 -0
- package/dist/campaign/presets/run-improvement-loop.d.ts.map +1 -0
- package/dist/campaign/presets/run-improvement-loop.test.d.ts +2 -0
- package/dist/campaign/presets/run-improvement-loop.test.d.ts.map +1 -0
- package/dist/campaign/presets/run-optimization.d.ts +92 -0
- package/dist/campaign/presets/run-optimization.d.ts.map +1 -0
- package/dist/campaign/presets/run-profile-matrix.d.ts +151 -0
- package/dist/campaign/presets/run-profile-matrix.d.ts.map +1 -0
- package/dist/campaign/presets/run-skill-opt.d.ts +96 -0
- package/dist/campaign/presets/run-skill-opt.d.ts.map +1 -0
- package/dist/campaign/proposers/_findings-text.d.ts +22 -0
- package/dist/campaign/proposers/_findings-text.d.ts.map +1 -0
- package/dist/campaign/proposers/ace.d.ts +33 -0
- package/dist/campaign/proposers/ace.d.ts.map +1 -0
- package/dist/campaign/proposers/ace.test.d.ts +2 -0
- package/dist/campaign/proposers/ace.test.d.ts.map +1 -0
- package/dist/campaign/proposers/analysis-edit.d.ts +32 -0
- package/dist/campaign/proposers/analysis-edit.d.ts.map +1 -0
- package/dist/campaign/proposers/evolutionary.d.ts +20 -0
- package/dist/campaign/proposers/evolutionary.d.ts.map +1 -0
- package/dist/campaign/proposers/fapo.d.ts +120 -0
- package/dist/campaign/proposers/fapo.d.ts.map +1 -0
- package/dist/campaign/proposers/gepa.d.ts +86 -0
- package/dist/campaign/proposers/gepa.d.ts.map +1 -0
- package/dist/campaign/proposers/halo.d.ts +44 -0
- package/dist/campaign/proposers/halo.d.ts.map +1 -0
- package/dist/campaign/proposers/halo.test.d.ts +2 -0
- package/dist/campaign/proposers/halo.test.d.ts.map +1 -0
- package/dist/campaign/proposers/memory.d.ts +47 -0
- package/dist/campaign/proposers/memory.d.ts.map +1 -0
- package/dist/campaign/proposers/memory.test.d.ts +2 -0
- package/dist/campaign/proposers/memory.test.d.ts.map +1 -0
- package/dist/campaign/proposers/skill-opt.d.ts +88 -0
- package/dist/campaign/proposers/skill-opt.d.ts.map +1 -0
- package/dist/campaign/proposers/trace-analyst.d.ts +48 -0
- package/dist/campaign/proposers/trace-analyst.d.ts.map +1 -0
- package/dist/campaign/proposers/trace-analyst.test.d.ts +2 -0
- package/dist/campaign/proposers/trace-analyst.test.d.ts.map +1 -0
- package/dist/campaign/provenance.d.ts +185 -0
- package/dist/campaign/provenance.d.ts.map +1 -0
- package/dist/campaign/run-campaign.d.ts +90 -0
- package/dist/campaign/run-campaign.d.ts.map +1 -0
- package/dist/campaign/score-utils.d.ts +26 -0
- package/dist/campaign/score-utils.d.ts.map +1 -0
- package/dist/campaign/skill-patch.d.ts +62 -0
- package/dist/campaign/skill-patch.d.ts.map +1 -0
- package/dist/campaign/storage.d.ts +38 -0
- package/dist/campaign/storage.d.ts.map +1 -0
- package/dist/{types-BU-7W85F.d.ts → campaign/types.d.ts} +98 -99
- package/dist/campaign/types.d.ts.map +1 -0
- package/dist/campaign/worktree/index.d.ts +53 -0
- package/dist/campaign/worktree/index.d.ts.map +1 -0
- package/dist/canary.d.ts +101 -0
- package/dist/canary.d.ts.map +1 -0
- package/dist/causal-attribution.d.ts +45 -0
- package/dist/causal-attribution.d.ts.map +1 -0
- package/dist/{chunk-TWS7AZEY.js → chunk-2T4EZACH.js} +2 -2
- package/dist/chunk-2T4EZACH.js.map +1 -0
- package/dist/{chunk-LMZQ2Z4U.js → chunk-56GC6VYO.js} +126 -1
- package/dist/chunk-56GC6VYO.js.map +1 -0
- package/dist/{chunk-QG2OVF2D.js → chunk-77T4STFI.js} +154 -112
- package/dist/chunk-77T4STFI.js.map +1 -0
- package/dist/{chunk-ZFIBGEOL.js → chunk-7QTQKIDD.js} +179 -178
- package/dist/chunk-7QTQKIDD.js.map +1 -0
- package/dist/{chunk-UMMZHCPB.js → chunk-AQ5WQAIV.js} +26 -9
- package/dist/chunk-AQ5WQAIV.js.map +1 -0
- package/dist/{chunk-CVVHBFGN.js → chunk-CWNP4DV4.js} +53 -5
- package/dist/chunk-CWNP4DV4.js.map +1 -0
- package/dist/{chunk-QS3RBQPI.js → chunk-D5KBVULY.js} +2 -2
- package/dist/chunk-DJWX3GVS.js +81 -0
- package/dist/chunk-DJWX3GVS.js.map +1 -0
- package/dist/{chunk-S6OZEZQK.js → chunk-HMA63UEO.js} +37 -9
- package/dist/{chunk-S6OZEZQK.js.map → chunk-HMA63UEO.js.map} +1 -1
- package/dist/{chunk-UHMJT4T7.js → chunk-IZCEK2HR.js} +2 -2
- package/dist/{chunk-6SOJM3VR.js → chunk-KKWJD5E6.js} +21 -21
- package/dist/chunk-KKWJD5E6.js.map +1 -0
- package/dist/{chunk-KWRRMR3J.js → chunk-LO6IOIJ2.js} +10 -10
- package/dist/chunk-LO6IOIJ2.js.map +1 -0
- package/dist/{chunk-L3JOU6XM.js → chunk-NACM5RS7.js} +3 -3
- package/dist/{chunk-E4GH6USR.js → chunk-NZEQVRH5.js} +2 -2
- package/dist/chunk-NZEQVRH5.js.map +1 -0
- package/dist/{chunk-XY4DDNEG.js → chunk-PSWWQXHF.js} +14 -89
- package/dist/chunk-PSWWQXHF.js.map +1 -0
- package/dist/{chunk-4FBZZIYD.js → chunk-QTMYW64F.js} +2 -2
- package/dist/{chunk-D3V5B42D.js → chunk-S4SYLDFX.js} +7 -4
- package/dist/chunk-S4SYLDFX.js.map +1 -0
- package/dist/{chunk-47X6LRCE.js → chunk-UFSG7ACU.js} +10 -7
- package/dist/chunk-UFSG7ACU.js.map +1 -0
- package/dist/{chunk-CY6U5S3X.js → chunk-URWVKRCS.js} +2 -2
- package/dist/{chunk-GSH6QNNS.js → chunk-X74V6ESX.js} +102 -79
- package/dist/chunk-X74V6ESX.js.map +1 -0
- package/dist/{chunk-QAY5UIJO.js → chunk-YBIGNSCZ.js} +178 -59
- package/dist/chunk-YBIGNSCZ.js.map +1 -0
- package/dist/{chunk-Y47J2LJ3.js → chunk-Z6L6YSU6.js} +5 -5
- package/dist/{chunk-T375SUOZ.js → chunk-ZOPLCJSY.js} +86 -31
- package/dist/chunk-ZOPLCJSY.js.map +1 -0
- package/dist/ci-gate.d.ts +44 -0
- package/dist/ci-gate.d.ts.map +1 -0
- package/dist/cli.d.ts +2 -0
- package/dist/cli.d.ts.map +1 -0
- package/dist/cli.js +2 -2
- package/dist/client.d.ts +77 -0
- package/dist/client.d.ts.map +1 -0
- package/dist/client.test.d.ts +2 -0
- package/dist/client.test.d.ts.map +1 -0
- package/dist/command-runner.d.ts +74 -0
- package/dist/command-runner.d.ts.map +1 -0
- package/dist/command-runner.test.d.ts +2 -0
- package/dist/command-runner.test.d.ts.map +1 -0
- package/dist/completion-verifier.d.ts +147 -0
- package/dist/completion-verifier.d.ts.map +1 -0
- package/dist/completion-verifier.test.d.ts +9 -0
- package/dist/completion-verifier.test.d.ts.map +1 -0
- package/dist/concurrency.d.ts +23 -0
- package/dist/concurrency.d.ts.map +1 -0
- package/dist/contamination-guard.d.ts +81 -0
- package/dist/contamination-guard.d.ts.map +1 -0
- package/dist/{analyze-runs-DwCEkpO_.d.ts → contract/analyze-runs.d.ts} +9 -10
- package/dist/contract/analyze-runs.d.ts.map +1 -0
- package/dist/contract/define-agent-eval.d.ts +52 -0
- package/dist/contract/define-agent-eval.d.ts.map +1 -0
- package/dist/contract/diff.d.ts +114 -0
- package/dist/contract/diff.d.ts.map +1 -0
- package/dist/contract/index.d.ts +106 -640
- package/dist/contract/index.d.ts.map +1 -0
- package/dist/contract/index.js +131 -21
- package/dist/contract/index.js.map +1 -1
- package/dist/{insight-report-BBwvOh6x.d.ts → contract/insight-report.d.ts} +16 -19
- package/dist/contract/insight-report.d.ts.map +1 -0
- package/dist/contract/insight-types-fwd.d.ts +7 -0
- package/dist/contract/insight-types-fwd.d.ts.map +1 -0
- package/dist/contract/intake/agent-trace.d.ts +97 -0
- package/dist/contract/intake/agent-trace.d.ts.map +1 -0
- package/dist/{code-agent-session-BO8nCnv3.d.ts → contract/intake/code-agent-session.d.ts} +15 -17
- package/dist/contract/intake/code-agent-session.d.ts.map +1 -0
- package/dist/contract/intake/feedback-table.d.ts +87 -0
- package/dist/contract/intake/feedback-table.d.ts.map +1 -0
- package/dist/contract/intake/index.d.ts +22 -0
- package/dist/contract/intake/index.d.ts.map +1 -0
- package/dist/contract/intake/otel-spans.d.ts +33 -0
- package/dist/contract/intake/otel-spans.d.ts.map +1 -0
- package/dist/contract/self-improve.d.ts +284 -0
- package/dist/contract/self-improve.d.ts.map +1 -0
- package/dist/{control-runtime-DuFBYg7A.d.ts → control-runtime.d.ts} +23 -26
- package/dist/control-runtime.d.ts.map +1 -0
- package/dist/control-runtime.test.d.ts +2 -0
- package/dist/control-runtime.test.d.ts.map +1 -0
- package/dist/control.d.ts +11 -9
- package/dist/control.d.ts.map +1 -0
- package/dist/control.js +2 -2
- package/dist/convergence.d.ts +29 -0
- package/dist/convergence.d.ts.map +1 -0
- package/dist/{cost-ledger-DuSqlw5B.d.ts → cost-ledger.d.ts} +10 -11
- package/dist/cost-ledger.d.ts.map +1 -0
- package/dist/cost-ledger.test.d.ts +2 -0
- package/dist/cost-ledger.test.d.ts.map +1 -0
- package/dist/cost-report.d.ts +43 -0
- package/dist/cost-report.d.ts.map +1 -0
- package/dist/cost-report.test.d.ts +2 -0
- package/dist/cost-report.test.d.ts.map +1 -0
- package/dist/cost-tracker.d.ts +76 -0
- package/dist/cost-tracker.d.ts.map +1 -0
- package/dist/{counterfactual-Dwibr5IW.d.ts → counterfactual.d.ts} +12 -13
- package/dist/counterfactual.d.ts.map +1 -0
- package/dist/cross-trace-diff.d.ts +56 -0
- package/dist/cross-trace-diff.d.ts.map +1 -0
- package/dist/{dataset-BbGkaN2I.d.ts → dataset.d.ts} +11 -14
- package/dist/dataset.d.ts.map +1 -0
- package/dist/deploy-gate-layer.d.ts +125 -0
- package/dist/deploy-gate-layer.d.ts.map +1 -0
- package/dist/deploy-gate-layer.test.d.ts +2 -0
- package/dist/deploy-gate-layer.test.d.ts.map +1 -0
- package/dist/description-length-gate.d.ts +119 -0
- package/dist/description-length-gate.d.ts.map +1 -0
- package/dist/detectors/edge.test.d.ts +2 -0
- package/dist/detectors/edge.test.d.ts.map +1 -0
- package/dist/detectors/index.d.ts +81 -0
- package/dist/detectors/index.d.ts.map +1 -0
- package/dist/detectors/index.test.d.ts +2 -0
- package/dist/detectors/index.test.d.ts.map +1 -0
- package/dist/diagnose/causal-sweep.d.ts +100 -0
- package/dist/diagnose/causal-sweep.d.ts.map +1 -0
- package/dist/diagnose/index.d.ts +36 -0
- package/dist/diagnose/index.d.ts.map +1 -0
- package/dist/diagnose/remediation.d.ts +68 -0
- package/dist/diagnose/remediation.d.ts.map +1 -0
- package/dist/diagnose/repair.d.ts +77 -0
- package/dist/diagnose/repair.d.ts.map +1 -0
- package/dist/diagnose.d.ts +1 -251
- package/dist/diagnose.js +2 -2
- package/dist/discover-personas.d.ts +35 -0
- package/dist/discover-personas.d.ts.map +1 -0
- package/dist/driver.d.ts +95 -0
- package/dist/driver.d.ts.map +1 -0
- package/dist/driver.test.d.ts +8 -0
- package/dist/driver.test.d.ts.map +1 -0
- package/dist/dual-agent-bench.d.ts +81 -0
- package/dist/dual-agent-bench.d.ts.map +1 -0
- package/dist/error-count-extractor.d.ts +47 -0
- package/dist/error-count-extractor.d.ts.map +1 -0
- package/dist/error-count-extractor.test.d.ts +2 -0
- package/dist/error-count-extractor.test.d.ts.map +1 -0
- package/dist/{errors-CzMUYo7b.d.ts → errors.d.ts} +10 -11
- package/dist/errors.d.ts.map +1 -0
- package/dist/{researcher-DE6Gpnb4.d.ts → eval-campaign.d.ts} +34 -156
- package/dist/eval-campaign.d.ts.map +1 -0
- package/dist/eval-campaign.test.d.ts +2 -0
- package/dist/eval-campaign.test.d.ts.map +1 -0
- package/dist/eval-tools.d.ts +55 -0
- package/dist/eval-tools.d.ts.map +1 -0
- package/dist/eval-trace-store.d.ts +107 -0
- package/dist/eval-trace-store.d.ts.map +1 -0
- package/dist/eval-trace-store.test.d.ts +2 -0
- package/dist/eval-trace-store.test.d.ts.map +1 -0
- package/dist/executor.d.ts +38 -0
- package/dist/executor.d.ts.map +1 -0
- package/dist/executor.test.d.ts +10 -0
- package/dist/executor.test.d.ts.map +1 -0
- package/dist/experiment-tracker.d.ts +178 -0
- package/dist/experiment-tracker.d.ts.map +1 -0
- package/dist/experiment-tracker.test.d.ts +2 -0
- package/dist/experiment-tracker.test.d.ts.map +1 -0
- package/dist/failure-taxonomy.d.ts +38 -0
- package/dist/failure-taxonomy.d.ts.map +1 -0
- package/dist/{feedback-trajectory-D9OVLrg9.d.ts → feedback-trajectory.d.ts} +36 -38
- package/dist/feedback-trajectory.d.ts.map +1 -0
- package/dist/feedback-trajectory.test.d.ts +2 -0
- package/dist/feedback-trajectory.test.d.ts.map +1 -0
- package/dist/flow-layer.d.ts +90 -0
- package/dist/flow-layer.d.ts.map +1 -0
- package/dist/flow-layer.test.d.ts +2 -0
- package/dist/flow-layer.test.d.ts.map +1 -0
- package/dist/fuzz/capsule.d.ts +46 -0
- package/dist/fuzz/capsule.d.ts.map +1 -0
- package/dist/fuzz/cube.d.ts +36 -0
- package/dist/fuzz/cube.d.ts.map +1 -0
- package/dist/fuzz/explorer-cost.test.d.ts +2 -0
- package/dist/fuzz/explorer-cost.test.d.ts.map +1 -0
- package/dist/fuzz/explorer.d.ts +64 -0
- package/dist/fuzz/explorer.d.ts.map +1 -0
- package/dist/fuzz/fuzz-agent.d.ts +16 -0
- package/dist/fuzz/fuzz-agent.d.ts.map +1 -0
- package/dist/fuzz/fuzz-agent.test.d.ts +2 -0
- package/dist/fuzz/fuzz-agent.test.d.ts.map +1 -0
- package/dist/fuzz/gates.d.ts +33 -0
- package/dist/fuzz/gates.d.ts.map +1 -0
- package/dist/fuzz/index.d.ts +26 -0
- package/dist/fuzz/index.d.ts.map +1 -0
- package/dist/fuzz/policies.d.ts +28 -0
- package/dist/fuzz/policies.d.ts.map +1 -0
- package/dist/fuzz/tools.d.ts +20 -0
- package/dist/fuzz/tools.d.ts.map +1 -0
- package/dist/fuzz/types.d.ts +307 -0
- package/dist/fuzz/types.d.ts.map +1 -0
- package/dist/fuzz.d.ts +1 -547
- package/dist/golden-matcher.d.ts +71 -0
- package/dist/golden-matcher.d.ts.map +1 -0
- package/dist/governance/eu-ai-act.d.ts +37 -0
- package/dist/governance/eu-ai-act.d.ts.map +1 -0
- package/dist/governance/index.d.ts +5 -135
- package/dist/governance/index.d.ts.map +1 -0
- package/dist/governance/nist-ai-rmf.d.ts +15 -0
- package/dist/governance/nist-ai-rmf.d.ts.map +1 -0
- package/dist/governance/soc2.d.ts +12 -0
- package/dist/governance/soc2.d.ts.map +1 -0
- package/dist/governance/types.d.ts +66 -0
- package/dist/governance/types.d.ts.map +1 -0
- package/dist/harness-optimizer.d.ts +82 -0
- package/dist/harness-optimizer.d.ts.map +1 -0
- package/dist/held-out-gate.d.ts +135 -0
- package/dist/held-out-gate.d.ts.map +1 -0
- package/dist/hosted/client.d.ts +73 -0
- package/dist/hosted/client.d.ts.map +1 -0
- package/dist/hosted/from-env.test.d.ts +8 -0
- package/dist/hosted/from-env.test.d.ts.map +1 -0
- package/dist/hosted/index.d.ts +10 -238
- package/dist/hosted/index.d.ts.map +1 -0
- package/dist/hosted/types.d.ts +159 -0
- package/dist/hosted/types.d.ts.map +1 -0
- package/dist/index.d.ts +277 -5643
- package/dist/index.d.ts.map +1 -0
- package/dist/index.js +317 -108
- package/dist/index.js.map +1 -1
- package/dist/integrity/backend-integrity.d.ts +71 -0
- package/dist/integrity/backend-integrity.d.ts.map +1 -0
- package/dist/integrity/preflight.d.ts +72 -0
- package/dist/integrity/preflight.d.ts.map +1 -0
- package/dist/integrity/preflight.test.d.ts +2 -0
- package/dist/integrity/preflight.test.d.ts.map +1 -0
- package/dist/integrity/single-backend.d.ts +67 -0
- package/dist/integrity/single-backend.d.ts.map +1 -0
- package/dist/intent-match-judge.d.ts +69 -0
- package/dist/intent-match-judge.d.ts.map +1 -0
- package/dist/intent-match-judge.test.d.ts +2 -0
- package/dist/intent-match-judge.test.d.ts.map +1 -0
- package/dist/{judge-calibration-0p2QcWNE.d.ts → judge-calibration.d.ts} +16 -17
- package/dist/judge-calibration.d.ts.map +1 -0
- package/dist/judge-ensemble.d.ts +66 -0
- package/dist/judge-ensemble.d.ts.map +1 -0
- package/dist/judge-ensemble.test.d.ts +8 -0
- package/dist/judge-ensemble.test.d.ts.map +1 -0
- package/dist/judge-families.d.ts +38 -0
- package/dist/judge-families.d.ts.map +1 -0
- package/dist/judge-panel.d.ts +65 -0
- package/dist/judge-panel.d.ts.map +1 -0
- package/dist/judge-retry.d.ts +70 -0
- package/dist/judge-retry.d.ts.map +1 -0
- package/dist/judge-runner.d.ts +36 -0
- package/dist/judge-runner.d.ts.map +1 -0
- package/dist/judge-runner.test.d.ts +2 -0
- package/dist/judge-runner.test.d.ts.map +1 -0
- package/dist/judges.d.ts +74 -0
- package/dist/judges.d.ts.map +1 -0
- package/dist/keyword-coverage-judge.d.ts +89 -0
- package/dist/keyword-coverage-judge.d.ts.map +1 -0
- package/dist/keyword-coverage-judge.test.d.ts +2 -0
- package/dist/keyword-coverage-judge.test.d.ts.map +1 -0
- package/dist/knowledge/index.d.ts +3 -103
- package/dist/knowledge/index.d.ts.map +1 -0
- package/dist/knowledge/readiness.d.ts +26 -0
- package/dist/knowledge/readiness.d.ts.map +1 -0
- package/dist/knowledge/types.d.ts +75 -0
- package/dist/knowledge/types.d.ts.map +1 -0
- package/dist/live-proof.d.ts +62 -0
- package/dist/live-proof.d.ts.map +1 -0
- package/dist/{llm-client-BeEcAokY.d.ts → llm-client.d.ts} +52 -23
- package/dist/llm-client.d.ts.map +1 -0
- package/dist/llm-client.test.d.ts +2 -0
- package/dist/llm-client.test.d.ts.map +1 -0
- package/dist/locked-jsonl-appender.d.ts +19 -0
- package/dist/locked-jsonl-appender.d.ts.map +1 -0
- package/dist/matrix/aggregation.d.ts +16 -0
- package/dist/matrix/aggregation.d.ts.map +1 -0
- package/dist/matrix/index.d.ts +12 -30
- package/dist/matrix/index.d.ts.map +1 -0
- package/dist/matrix/runner.d.ts +15 -0
- package/dist/matrix/runner.d.ts.map +1 -0
- package/dist/{types-mn5Aqk7x.d.ts → matrix/types.d.ts} +11 -15
- package/dist/matrix/types.d.ts.map +1 -0
- package/dist/meta-eval/calibration.d.ts +47 -0
- package/dist/meta-eval/calibration.d.ts.map +1 -0
- package/dist/meta-eval/correlation-study.d.ts +53 -0
- package/dist/meta-eval/correlation-study.d.ts.map +1 -0
- package/dist/meta-eval/index.d.ts +6 -181
- package/dist/meta-eval/index.d.ts.map +1 -0
- package/dist/meta-eval/index.js +1 -1
- package/dist/{outcome-store-rnXLEqSn.d.ts → meta-eval/outcome-store.d.ts} +7 -8
- package/dist/meta-eval/outcome-store.d.ts.map +1 -0
- package/dist/{rubric-predictive-validity-Cy_W-hWZ.d.ts → meta-eval/rubric-predictive-validity.d.ts} +8 -11
- package/dist/meta-eval/rubric-predictive-validity.d.ts.map +1 -0
- package/dist/meta-eval/sentinel.d.ts +169 -0
- package/dist/meta-eval/sentinel.d.ts.map +1 -0
- package/dist/metrics.d.ts +63 -0
- package/dist/metrics.d.ts.map +1 -0
- package/dist/model-seats.d.ts +71 -0
- package/dist/model-seats.d.ts.map +1 -0
- package/dist/model-seats.test.d.ts +2 -0
- package/dist/model-seats.test.d.ts.map +1 -0
- package/dist/muffled-gate-scanner.d.ts +102 -0
- package/dist/muffled-gate-scanner.d.ts.map +1 -0
- package/dist/{multi-layer-verifier-DUZXrPDA.d.ts → multi-layer-verifier.d.ts} +12 -15
- package/dist/multi-layer-verifier.d.ts.map +1 -0
- package/dist/multi-layer-verifier.test.d.ts +2 -0
- package/dist/multi-layer-verifier.test.d.ts.map +1 -0
- package/dist/multi-toolchain-layer.d.ts +80 -0
- package/dist/multi-toolchain-layer.d.ts.map +1 -0
- package/dist/multi-toolchain-layer.test.d.ts +2 -0
- package/dist/multi-toolchain-layer.test.d.ts.map +1 -0
- package/dist/multishot/default-tools.d.ts +34 -0
- package/dist/multishot/default-tools.d.ts.map +1 -0
- package/dist/multishot/index.d.ts +7 -290
- package/dist/multishot/index.d.ts.map +1 -0
- package/dist/multishot/index.js.map +1 -1
- package/dist/multishot/judges.d.ts +32 -0
- package/dist/multishot/judges.d.ts.map +1 -0
- package/dist/multishot/matrix.d.ts +107 -0
- package/dist/multishot/matrix.d.ts.map +1 -0
- package/dist/multishot/multishot.d.ts +23 -0
- package/dist/multishot/multishot.d.ts.map +1 -0
- package/dist/multishot/router.d.ts +37 -0
- package/dist/multishot/router.d.ts.map +1 -0
- package/dist/multishot/types.d.ts +60 -0
- package/dist/multishot/types.d.ts.map +1 -0
- package/dist/observability.d.ts +71 -0
- package/dist/observability.d.ts.map +1 -0
- package/dist/openapi.json +1 -1
- package/dist/oracle.d.ts +55 -0
- package/dist/oracle.d.ts.map +1 -0
- package/dist/orthogonality.d.ts +35 -0
- package/dist/orthogonality.d.ts.map +1 -0
- package/dist/otel-pipeline.d.ts +31 -0
- package/dist/otel-pipeline.d.ts.map +1 -0
- package/dist/paraphrase.d.ts +107 -0
- package/dist/paraphrase.d.ts.map +1 -0
- package/dist/{pareto-E-pembql.d.ts → pareto.d.ts} +9 -10
- package/dist/pareto.d.ts.map +1 -0
- package/dist/partition-held-out.d.ts +70 -0
- package/dist/partition-held-out.d.ts.map +1 -0
- package/dist/partition-held-out.test.d.ts +2 -0
- package/dist/partition-held-out.test.d.ts.map +1 -0
- package/dist/perf/index.d.ts +13 -119
- package/dist/perf/index.d.ts.map +1 -0
- package/dist/perf/integrity.d.ts +30 -0
- package/dist/perf/integrity.d.ts.map +1 -0
- package/dist/perf/journey.d.ts +45 -0
- package/dist/perf/journey.d.ts.map +1 -0
- package/dist/perf/ratchet.d.ts +47 -0
- package/dist/perf/ratchet.d.ts.map +1 -0
- package/dist/pipelines/budget-breach.d.ts +31 -0
- package/dist/pipelines/budget-breach.d.ts.map +1 -0
- package/dist/pipelines/budget-breach.test.d.ts +2 -0
- package/dist/pipelines/budget-breach.test.d.ts.map +1 -0
- package/dist/pipelines/failure-cluster.d.ts +38 -0
- package/dist/pipelines/failure-cluster.d.ts.map +1 -0
- package/dist/pipelines/failure-cluster.test.d.ts +2 -0
- package/dist/pipelines/failure-cluster.test.d.ts.map +1 -0
- package/dist/pipelines/first-divergence.d.ts +26 -0
- package/dist/pipelines/first-divergence.d.ts.map +1 -0
- package/dist/pipelines/first-divergence.test.d.ts +2 -0
- package/dist/pipelines/first-divergence.test.d.ts.map +1 -0
- package/dist/pipelines/index.d.ts +8 -173
- package/dist/pipelines/index.d.ts.map +1 -0
- package/dist/pipelines/index.js +40 -15
- package/dist/pipelines/index.js.map +1 -1
- package/dist/pipelines/judge-agreement.d.ts +26 -0
- package/dist/pipelines/judge-agreement.d.ts.map +1 -0
- package/dist/pipelines/judge-agreement.test.d.ts +2 -0
- package/dist/pipelines/judge-agreement.test.d.ts.map +1 -0
- package/dist/pipelines/regression.d.ts +23 -0
- package/dist/pipelines/regression.d.ts.map +1 -0
- package/dist/pipelines/regression.test.d.ts +2 -0
- package/dist/pipelines/regression.test.d.ts.map +1 -0
- package/dist/pipelines/stuck-loop.d.ts +32 -0
- package/dist/pipelines/stuck-loop.d.ts.map +1 -0
- package/dist/pipelines/stuck-loop.test.d.ts +2 -0
- package/dist/pipelines/stuck-loop.test.d.ts.map +1 -0
- package/dist/pipelines/tool-waste.d.ts +34 -0
- package/dist/pipelines/tool-waste.d.ts.map +1 -0
- package/dist/pipelines/tool-waste.test.d.ts +2 -0
- package/dist/pipelines/tool-waste.test.d.ts.map +1 -0
- package/dist/playbook.d.ts +16 -0
- package/dist/playbook.d.ts.map +1 -0
- package/dist/pr-review-benchmark.d.ts +88 -0
- package/dist/pr-review-benchmark.d.ts.map +1 -0
- package/dist/pr-review-benchmark.test.d.ts +2 -0
- package/dist/pr-review-benchmark.test.d.ts.map +1 -0
- package/dist/pre-registration.d.ts +125 -0
- package/dist/pre-registration.d.ts.map +1 -0
- package/dist/prm/builtin-rubrics.d.ts +33 -0
- package/dist/prm/builtin-rubrics.d.ts.map +1 -0
- package/dist/prm/index.d.ts +5 -100
- package/dist/prm/index.d.ts.map +1 -0
- package/dist/prm/index.js +55 -12
- package/dist/prm/index.js.map +1 -1
- package/dist/prm/inference.d.ts +29 -0
- package/dist/prm/inference.d.ts.map +1 -0
- package/dist/prm/inference.test.d.ts +2 -0
- package/dist/prm/inference.test.d.ts.map +1 -0
- package/dist/{rubric-BOfxn4ja.d.ts → prm/rubric.d.ts} +10 -13
- package/dist/prm/rubric.d.ts.map +1 -0
- package/dist/prm/training-export.d.ts +38 -0
- package/dist/prm/training-export.d.ts.map +1 -0
- package/dist/produced-state.d.ts +63 -0
- package/dist/produced-state.d.ts.map +1 -0
- package/dist/produced-state.test.d.ts +8 -0
- package/dist/produced-state.test.d.ts.map +1 -0
- package/dist/profile/baselines.d.ts +37 -0
- package/dist/profile/baselines.d.ts.map +1 -0
- package/dist/profile/index.d.ts +105 -0
- package/dist/profile/index.d.ts.map +1 -0
- package/dist/promotion-gate.d.ts +93 -0
- package/dist/promotion-gate.d.ts.map +1 -0
- package/dist/prompt-registry.d.ts +41 -0
- package/dist/prompt-registry.d.ts.map +1 -0
- package/dist/propose-review-control.d.ts +49 -0
- package/dist/propose-review-control.d.ts.map +1 -0
- package/dist/propose-review-control.test.d.ts +2 -0
- package/dist/propose-review-control.test.d.ts.map +1 -0
- package/dist/propose-review.d.ts +155 -0
- package/dist/propose-review.d.ts.map +1 -0
- package/dist/{red-team-BXHil6c8.d.ts → red-team.d.ts} +14 -16
- package/dist/red-team.d.ts.map +1 -0
- package/dist/reference-replay-steering.d.ts +11 -0
- package/dist/reference-replay-steering.d.ts.map +1 -0
- package/dist/reference-replay.d.ts +176 -0
- package/dist/reference-replay.d.ts.map +1 -0
- package/dist/reflective-mutation.d.ts +79 -0
- package/dist/reflective-mutation.d.ts.map +1 -0
- package/dist/registry.d.ts +31 -0
- package/dist/registry.d.ts.map +1 -0
- package/dist/release-confidence.d.ts +128 -0
- package/dist/release-confidence.d.ts.map +1 -0
- package/dist/release-report.d.ts +11 -0
- package/dist/release-report.d.ts.map +1 -0
- package/dist/replay.d.ts +120 -0
- package/dist/replay.d.ts.map +1 -0
- package/dist/reporter.d.ts +14 -0
- package/dist/reporter.d.ts.map +1 -0
- package/dist/reporting.d.ts +15 -15
- package/dist/reporting.d.ts.map +1 -0
- package/dist/reporting.js +3 -3
- package/dist/researcher.d.ts +140 -0
- package/dist/researcher.d.ts.map +1 -0
- package/dist/reviewer.d.ts +118 -0
- package/dist/reviewer.d.ts.map +1 -0
- package/dist/reviewer.test.d.ts +2 -0
- package/dist/reviewer.test.d.ts.map +1 -0
- package/dist/reward-model-export.d.ts +60 -0
- package/dist/reward-model-export.d.ts.map +1 -0
- package/dist/rl/active-curriculum.d.ts +110 -0
- package/dist/rl/active-curriculum.d.ts.map +1 -0
- package/dist/rl/adaptation-eval.d.ts +109 -0
- package/dist/rl/adaptation-eval.d.ts.map +1 -0
- package/dist/{adversarial-DIVcDoI_.d.ts → rl/adversarial.d.ts} +6 -7
- package/dist/rl/adversarial.d.ts.map +1 -0
- package/dist/rl/compute-curves.d.ts +127 -0
- package/dist/rl/compute-curves.d.ts.map +1 -0
- package/dist/rl/contamination.d.ts +117 -0
- package/dist/rl/contamination.d.ts.map +1 -0
- package/dist/rl/corpus.d.ts +55 -0
- package/dist/rl/corpus.d.ts.map +1 -0
- package/dist/rl/corpus.test.d.ts +2 -0
- package/dist/rl/corpus.test.d.ts.map +1 -0
- package/dist/rl/dataset.d.ts +102 -0
- package/dist/rl/dataset.d.ts.map +1 -0
- package/dist/rl/dataset.test.d.ts +2 -0
- package/dist/rl/dataset.test.d.ts.map +1 -0
- package/dist/rl/exporters.d.ts +141 -0
- package/dist/rl/exporters.d.ts.map +1 -0
- package/dist/rl/index.d.ts +49 -0
- package/dist/rl/index.d.ts.map +1 -0
- package/dist/{off-policy-DiwuKKg7.d.ts → rl/off-policy.d.ts} +8 -9
- package/dist/rl/off-policy.d.ts.map +1 -0
- package/dist/rl/predictive-validity-researcher.d.ts +69 -0
- package/dist/rl/predictive-validity-researcher.d.ts.map +1 -0
- package/dist/rl/preferences.d.ts +141 -0
- package/dist/rl/preferences.d.ts.map +1 -0
- package/dist/rl/process-reward.d.ts +122 -0
- package/dist/rl/process-reward.d.ts.map +1 -0
- package/dist/rl/reward-hacking.d.ts +104 -0
- package/dist/rl/reward-hacking.d.ts.map +1 -0
- package/dist/rl/rl-campaign.d.ts +85 -0
- package/dist/rl/rl-campaign.d.ts.map +1 -0
- package/dist/rl/run-record-adapters.d.ts +56 -0
- package/dist/rl/run-record-adapters.d.ts.map +1 -0
- package/dist/rl/sim-fidelity.d.ts +166 -0
- package/dist/rl/sim-fidelity.d.ts.map +1 -0
- package/dist/rl/sim-fidelity.test.d.ts +2 -0
- package/dist/rl/sim-fidelity.test.d.ts.map +1 -0
- package/dist/rl/tournament.d.ts +115 -0
- package/dist/rl/tournament.d.ts.map +1 -0
- package/dist/rl/verifiable-reward.d.ts +124 -0
- package/dist/rl/verifiable-reward.d.ts.map +1 -0
- package/dist/rl.d.ts +1 -1192
- package/dist/rl.js +615 -615
- package/dist/rl.js.map +1 -1
- package/dist/{run-campaign-RDGAM5KJ.js → run-campaign-WXY7KI67.js} +3 -3
- package/dist/run-critic.d.ts +23 -0
- package/dist/run-critic.d.ts.map +1 -0
- package/dist/run-evidence.d.ts +32 -0
- package/dist/run-evidence.d.ts.map +1 -0
- package/dist/{run-record-e7vj1uZQ.d.ts → run-record.d.ts} +16 -122
- package/dist/run-record.d.ts.map +1 -0
- package/dist/run-record.test.d.ts +2 -0
- package/dist/run-record.test.d.ts.map +1 -0
- package/dist/run-score.d.ts +31 -0
- package/dist/run-score.d.ts.map +1 -0
- package/dist/runtime-trajectory.d.ts +47 -0
- package/dist/runtime-trajectory.d.ts.map +1 -0
- package/dist/{test-graded-scenario-BdVaPyHT.d.ts → sandbox-harness.d.ts} +34 -60
- package/dist/sandbox-harness.d.ts.map +1 -0
- package/dist/sandbox-harness.test.d.ts +2 -0
- package/dist/sandbox-harness.test.d.ts.map +1 -0
- package/dist/sandbox-pool.d.ts +74 -0
- package/dist/sandbox-pool.d.ts.map +1 -0
- package/dist/sandbox-pool.test.d.ts +2 -0
- package/dist/sandbox-pool.test.d.ts.map +1 -0
- package/dist/scorecard.d.ts +133 -0
- package/dist/scorecard.d.ts.map +1 -0
- package/dist/scorecard.test.d.ts +2 -0
- package/dist/scorecard.test.d.ts.map +1 -0
- package/dist/self-play.d.ts +69 -0
- package/dist/self-play.d.ts.map +1 -0
- package/dist/semantic-concept-judge.d.ts +135 -0
- package/dist/semantic-concept-judge.d.ts.map +1 -0
- package/dist/semantic-concept-judge.test.d.ts +2 -0
- package/dist/semantic-concept-judge.test.d.ts.map +1 -0
- package/dist/{sequential-5iSVfzl2.d.ts → sequential.d.ts} +9 -10
- package/dist/sequential.d.ts.map +1 -0
- package/dist/{series-convergence-D5OWMBg6.d.ts → series-convergence.d.ts} +4 -5
- package/dist/series-convergence.d.ts.map +1 -0
- package/dist/slo.d.ts +48 -0
- package/dist/slo.d.ts.map +1 -0
- package/dist/state-continuity.d.ts +47 -0
- package/dist/state-continuity.d.ts.map +1 -0
- package/dist/{statistics-C7PozGrZ.d.ts → statistics.d.ts} +152 -39
- package/dist/statistics.d.ts.map +1 -0
- package/dist/statistics.test.d.ts +2 -0
- package/dist/statistics.test.d.ts.map +1 -0
- package/dist/steering-optimizer.d.ts +58 -0
- package/dist/steering-optimizer.d.ts.map +1 -0
- package/dist/steering.d.ts +24 -0
- package/dist/steering.d.ts.map +1 -0
- package/dist/storyboard/code-edit.d.ts +64 -0
- package/dist/storyboard/code-edit.d.ts.map +1 -0
- package/dist/storyboard/code-edit.test.d.ts +2 -0
- package/dist/storyboard/code-edit.test.d.ts.map +1 -0
- package/dist/storyboard/index.d.ts +16 -81
- package/dist/storyboard/index.d.ts.map +1 -0
- package/dist/storyboard/index.test.d.ts +2 -0
- package/dist/storyboard/index.test.d.ts.map +1 -0
- package/dist/{summary-report-DGmUucwQ.d.ts → summary-report.d.ts} +23 -160
- package/dist/summary-report.d.ts.map +1 -0
- package/dist/telemetry/client.d.ts +35 -0
- package/dist/telemetry/client.d.ts.map +1 -0
- package/dist/telemetry/index.d.ts +17 -35
- package/dist/telemetry/index.d.ts.map +1 -0
- package/dist/telemetry/schema.d.ts +61 -0
- package/dist/telemetry/schema.d.ts.map +1 -0
- package/dist/telemetry/sink-fetch.d.ts +39 -0
- package/dist/telemetry/sink-fetch.d.ts.map +1 -0
- package/dist/telemetry/{file.d.ts → sink-file.d.ts} +5 -7
- package/dist/telemetry/sink-file.d.ts.map +1 -0
- package/dist/test-graded-scenario.d.ts +42 -0
- package/dist/test-graded-scenario.d.ts.map +1 -0
- package/dist/testing.d.ts +5 -0
- package/dist/testing.d.ts.map +1 -0
- package/dist/testing.js +8 -0
- package/dist/testing.js.map +1 -0
- package/dist/tool-use-metrics.d.ts +35 -0
- package/dist/tool-use-metrics.d.ts.map +1 -0
- package/dist/trace/capture-fetch.d.ts +48 -0
- package/dist/trace/capture-fetch.d.ts.map +1 -0
- package/dist/trace/capture-fetch.test.d.ts +2 -0
- package/dist/trace/capture-fetch.test.d.ts.map +1 -0
- package/dist/{emitter-DEZwY14K.d.ts → trace/emitter.d.ts} +9 -12
- package/dist/trace/emitter.d.ts.map +1 -0
- package/dist/trace/extract-usage.d.ts +46 -0
- package/dist/trace/extract-usage.d.ts.map +1 -0
- package/dist/trace/extract-usage.test.d.ts +2 -0
- package/dist/trace/extract-usage.test.d.ts.map +1 -0
- package/dist/trace/index.d.ts +15 -0
- package/dist/trace/index.d.ts.map +1 -0
- package/dist/{integrity-VJ9A7aST.d.ts → trace/integrity.d.ts} +11 -14
- package/dist/trace/integrity.d.ts.map +1 -0
- package/dist/trace/otel-bridge.d.ts +29 -0
- package/dist/trace/otel-bridge.d.ts.map +1 -0
- package/dist/trace/otel-export.d.ts +52 -0
- package/dist/trace/otel-export.d.ts.map +1 -0
- package/dist/trace/otel.d.ts +57 -0
- package/dist/trace/otel.d.ts.map +1 -0
- package/dist/trace/otlp-attributes.d.ts +17 -0
- package/dist/trace/otlp-attributes.d.ts.map +1 -0
- package/dist/trace/query.d.ts +29 -0
- package/dist/trace/query.d.ts.map +1 -0
- package/dist/trace/query.test.d.ts +2 -0
- package/dist/trace/query.test.d.ts.map +1 -0
- package/dist/{raw-provider-sink-C46HDghv.d.ts → trace/raw-provider-sink.d.ts} +13 -14
- package/dist/trace/raw-provider-sink.d.ts.map +1 -0
- package/dist/{redact-B40YG2M_.d.ts → trace/redact.d.ts} +7 -8
- package/dist/trace/redact.d.ts.map +1 -0
- package/dist/{schema-m0gsnbt3.d.ts → trace/schema.d.ts} +29 -30
- package/dist/trace/schema.d.ts.map +1 -0
- package/dist/trace/store-to-otlp.d.ts +72 -0
- package/dist/trace/store-to-otlp.d.ts.map +1 -0
- package/dist/trace/store-to-otlp.test.d.ts +2 -0
- package/dist/trace/store-to-otlp.test.d.ts.map +1 -0
- package/dist/{store-CKUAgsJz.d.ts → trace/store.d.ts} +37 -13
- package/dist/trace/store.d.ts.map +1 -0
- package/dist/trace/store.test.d.ts +2 -0
- package/dist/trace/store.test.d.ts.map +1 -0
- package/dist/{analyst-C8HHvfJp.d.ts → trace-analyst/analyst.d.ts} +12 -14
- package/dist/trace-analyst/analyst.d.ts.map +1 -0
- package/dist/trace-analyst/analyst.test.d.ts +2 -0
- package/dist/trace-analyst/analyst.test.d.ts.map +1 -0
- package/dist/trace-analyst/behavioral-metrics.d.ts +40 -0
- package/dist/trace-analyst/behavioral-metrics.d.ts.map +1 -0
- package/dist/trace-analyst/behavioral-metrics.test.d.ts +2 -0
- package/dist/trace-analyst/behavioral-metrics.test.d.ts.map +1 -0
- package/dist/trace-analyst/hook.d.ts +55 -0
- package/dist/trace-analyst/hook.d.ts.map +1 -0
- package/dist/trace-analyst/index.d.ts +18 -0
- package/dist/trace-analyst/index.d.ts.map +1 -0
- package/dist/trace-analyst/insights.d.ts +71 -0
- package/dist/trace-analyst/insights.d.ts.map +1 -0
- package/dist/trace-analyst/insights.test.d.ts +2 -0
- package/dist/trace-analyst/insights.test.d.ts.map +1 -0
- package/dist/trace-analyst/otlp-flatten.d.ts +42 -0
- package/dist/trace-analyst/otlp-flatten.d.ts.map +1 -0
- package/dist/trace-analyst/otlp-span.d.ts +85 -0
- package/dist/trace-analyst/otlp-span.d.ts.map +1 -0
- package/dist/trace-analyst/otlp-span.test.d.ts +2 -0
- package/dist/trace-analyst/otlp-span.test.d.ts.map +1 -0
- package/dist/trace-analyst/otlp-to-run-records.d.ts +115 -0
- package/dist/trace-analyst/otlp-to-run-records.d.ts.map +1 -0
- package/dist/trace-analyst/otlp-to-run-records.test.d.ts +2 -0
- package/dist/trace-analyst/otlp-to-run-records.test.d.ts.map +1 -0
- package/dist/trace-analyst/otlp-to-run-records.timestamps.test.d.ts +2 -0
- package/dist/trace-analyst/otlp-to-run-records.timestamps.test.d.ts.map +1 -0
- package/dist/trace-analyst/prompts.d.ts +6 -0
- package/dist/trace-analyst/prompts.d.ts.map +1 -0
- package/dist/trace-analyst/store-otlp.d.ts +126 -0
- package/dist/trace-analyst/store-otlp.d.ts.map +1 -0
- package/dist/trace-analyst/store-otlp.test.d.ts +8 -0
- package/dist/trace-analyst/store-otlp.test.d.ts.map +1 -0
- package/dist/trace-analyst/store-otlp.timestamps.test.d.ts +2 -0
- package/dist/trace-analyst/store-otlp.timestamps.test.d.ts.map +1 -0
- package/dist/trace-analyst/store.d.ts +63 -0
- package/dist/trace-analyst/store.d.ts.map +1 -0
- package/dist/trace-analyst/tools.d.ts +44 -0
- package/dist/trace-analyst/tools.d.ts.map +1 -0
- package/dist/trace-analyst/tools.test.d.ts +10 -0
- package/dist/trace-analyst/tools.test.d.ts.map +1 -0
- package/dist/{store-C1YxJDEK.d.ts → trace-analyst/types.d.ts} +18 -74
- package/dist/trace-analyst/types.d.ts.map +1 -0
- package/dist/trace-contracts.d.ts +180 -0
- package/dist/trace-contracts.d.ts.map +1 -0
- package/dist/traced-analyst.d.ts +26 -0
- package/dist/traced-analyst.d.ts.map +1 -0
- package/dist/traced-judges.d.ts +27 -0
- package/dist/traced-judges.d.ts.map +1 -0
- package/dist/traces.d.ts +4 -947
- package/dist/traces.d.ts.map +1 -0
- package/dist/traces.js +42 -12
- package/dist/{trajectory-GEdXJCL5.d.ts → trajectory.d.ts} +8 -9
- package/dist/trajectory.d.ts.map +1 -0
- package/dist/{types-Croy5h7V.d.ts → types.d.ts} +33 -33
- package/dist/types.d.ts.map +1 -0
- package/dist/ui-finding.d.ts +104 -0
- package/dist/ui-finding.d.ts.map +1 -0
- package/dist/verdict-cache.d.ts +78 -0
- package/dist/verdict-cache.d.ts.map +1 -0
- package/dist/verdict-cache.test.d.ts +2 -0
- package/dist/verdict-cache.test.d.ts.map +1 -0
- package/dist/{verdict-C9MlYujm.d.ts → verdict.d.ts} +2 -3
- package/dist/verdict.d.ts.map +1 -0
- package/dist/visual-diff.d.ts +32 -0
- package/dist/visual-diff.d.ts.map +1 -0
- package/dist/wire/handlers.d.ts +54 -0
- package/dist/wire/handlers.d.ts.map +1 -0
- package/dist/wire/index.d.ts +13 -570
- package/dist/wire/index.d.ts.map +1 -0
- package/dist/wire/index.js +2 -2
- package/dist/wire/openapi.d.ts +3 -0
- package/dist/wire/openapi.d.ts.map +1 -0
- package/dist/wire/rpc.d.ts +21 -0
- package/dist/wire/rpc.d.ts.map +1 -0
- package/dist/wire/rubrics.d.ts +34 -0
- package/dist/wire/rubrics.d.ts.map +1 -0
- package/dist/wire/schemas.d.ts +410 -0
- package/dist/wire/schemas.d.ts.map +1 -0
- package/dist/wire/server.d.ts +60 -0
- package/dist/wire/server.d.ts.map +1 -0
- package/dist/workflow/event-schema.d.ts +5 -0
- package/dist/workflow/event-schema.d.ts.map +1 -0
- package/dist/workflow/feedback-pack.d.ts +99 -0
- package/dist/workflow/feedback-pack.d.ts.map +1 -0
- package/dist/workflow/index.d.ts +22 -495
- package/dist/workflow/index.d.ts.map +1 -0
- package/dist/workflow/index.js +2 -2
- package/dist/workflow/intelligence-export.d.ts +62 -0
- package/dist/workflow/intelligence-export.d.ts.map +1 -0
- package/dist/workflow/partner-report.d.ts +49 -0
- package/dist/workflow/partner-report.d.ts.map +1 -0
- package/dist/workflow/phase-graph.d.ts +43 -0
- package/dist/workflow/phase-graph.d.ts.map +1 -0
- package/dist/workflow/promotion-gate.d.ts +61 -0
- package/dist/workflow/promotion-gate.d.ts.map +1 -0
- package/dist/workflow/run-record.d.ts +12 -0
- package/dist/workflow/run-record.d.ts.map +1 -0
- package/dist/workflow/runtime-adapter.d.ts +20 -0
- package/dist/workflow/runtime-adapter.d.ts.map +1 -0
- package/dist/workflow/sanitize.d.ts +21 -0
- package/dist/workflow/sanitize.d.ts.map +1 -0
- package/dist/workflow/schema.d.ts +5 -0
- package/dist/workflow/schema.d.ts.map +1 -0
- package/dist/workflow/summary.d.ts +43 -0
- package/dist/workflow/summary.d.ts.map +1 -0
- package/dist/workflow/trace-event-fields.d.ts +6 -0
- package/dist/workflow/trace-event-fields.d.ts.map +1 -0
- package/dist/workflow/trajectory.d.ts +15 -0
- package/dist/workflow/trajectory.d.ts.map +1 -0
- package/dist/workflow/types.d.ts +68 -0
- package/dist/workflow/types.d.ts.map +1 -0
- package/dist/workspace-inspector.d.ts +67 -0
- package/dist/workspace-inspector.d.ts.map +1 -0
- package/dist/wrangler-deploy-runner.test.d.ts +2 -0
- package/dist/wrangler-deploy-runner.test.d.ts.map +1 -0
- package/docs/campaign-proposers.md +170 -0
- package/docs/concepts.md +8 -4
- package/docs/customer-journeys.md +15 -13
- package/docs/design/loop-taxonomy.md +34 -66
- package/docs/distributed-driver.md +14 -14
- package/docs/feature-guide.md +1 -1
- package/docs/hosted-ingest-spec.md +2 -3
- package/docs/multi-shot-optimization.md +8 -8
- package/docs/product-eval-adoption.md +1 -1
- package/docs/self-improvement-map.md +33 -29
- package/package.json +9 -15
- package/dist/calibration-Cpr3WaX3.d.ts +0 -101
- package/dist/chunk-47X6LRCE.js.map +0 -1
- package/dist/chunk-6SOJM3VR.js.map +0 -1
- package/dist/chunk-CVVHBFGN.js.map +0 -1
- package/dist/chunk-D3V5B42D.js.map +0 -1
- package/dist/chunk-E4GH6USR.js.map +0 -1
- package/dist/chunk-GSH6QNNS.js.map +0 -1
- package/dist/chunk-KWRRMR3J.js.map +0 -1
- package/dist/chunk-LMZQ2Z4U.js.map +0 -1
- package/dist/chunk-QAY5UIJO.js.map +0 -1
- package/dist/chunk-QG2OVF2D.js.map +0 -1
- package/dist/chunk-T375SUOZ.js.map +0 -1
- package/dist/chunk-TWS7AZEY.js.map +0 -1
- package/dist/chunk-UMMZHCPB.js.map +0 -1
- package/dist/chunk-XY4DDNEG.js.map +0 -1
- package/dist/chunk-ZFIBGEOL.js.map +0 -1
- package/dist/control-_Qb7skHX.d.ts +0 -259
- package/dist/corpus-BoR-041R.d.ts +0 -560
- package/dist/failure-cluster-CL7IVgkJ.d.ts +0 -76
- package/dist/harness-optimizer-EnEnQPsr.d.ts +0 -106
- package/dist/index-Bx3gZ8xl.d.ts +0 -159
- package/dist/pre-registration-mAnCugl9.d.ts +0 -523
- package/dist/provenance-LnqRT0sS.d.ts +0 -441
- package/dist/query-CqTxMwDw.d.ts +0 -31
- package/dist/release-report-euXIV_Sk.d.ts +0 -233
- package/dist/run-critic-BAIjX99r.d.ts +0 -56
- package/dist/run-improvement-loop-5z_l5zDz.d.ts +0 -427
- package/dist/runtime-trajectory-BDgfGZSr.d.ts +0 -49
- package/dist/semantic-concept-judge-Dn8Z6KEG.d.ts +0 -624
- package/dist/sink-fetch-B1Yg4Til.d.ts +0 -101
- package/docs/design/external-agent-wedge.md +0 -89
- package/docs/design/phase-d-rfc.md +0 -125
- package/docs/design/phase4-consumer-migration.md +0 -70
- package/docs/design/primitives-integration-spec.md +0 -393
- package/docs/design/product-self-improvement-loop.md +0 -146
- package/docs/design/self-improvement-engine.md +0 -140
- package/docs/design/self-improvement-protocol.md +0 -223
- package/docs/design/self-improvement-roadmap.md +0 -106
- package/docs/design/substrate-gaps.md +0 -118
- package/docs/phase-b-pairing-kit.md +0 -188
- package/docs/phase-b-runbook.md +0 -176
- package/docs/pilot/README.md +0 -62
- package/docs/pilot/customer-checklist.md +0 -90
- package/docs/pilot/integration-foreign-stack.md +0 -296
- package/docs/pilot/integration-tangle-stack.md +0 -248
- package/docs/pilot/one-pager.md +0 -161
- package/docs/pilot/sample-insight-report.json +0 -172
- package/docs/quickstart-external.md +0 -229
- package/docs/research/belief-state-agent-eval-roadmap.md +0 -593
- package/docs/research/research-roadmap.md +0 -205
- package/docs/specs/driver-honest-spec.md +0 -251
- package/docs/specs/hermes-self-improvement-audit.md +0 -93
- package/docs/specs/profile-versioning.md +0 -291
- package/docs/three-package-architecture.md +0 -168
- /package/dist/{chunk-QS3RBQPI.js.map → chunk-D5KBVULY.js.map} +0 -0
- /package/dist/{chunk-UHMJT4T7.js.map → chunk-IZCEK2HR.js.map} +0 -0
- /package/dist/{chunk-L3JOU6XM.js.map → chunk-NACM5RS7.js.map} +0 -0
- /package/dist/{chunk-4FBZZIYD.js.map → chunk-QTMYW64F.js.map} +0 -0
- /package/dist/{chunk-CY6U5S3X.js.map → chunk-URWVKRCS.js.map} +0 -0
- /package/dist/{chunk-Y47J2LJ3.js.map → chunk-Z6L6YSU6.js.map} +0 -0
- /package/dist/{run-campaign-RDGAM5KJ.js.map → run-campaign-WXY7KI67.js.map} +0 -0
|
@@ -0,0 +1,176 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Reference replay — score an agent against withheld historical outcomes.
|
|
3
|
+
*
|
|
4
|
+
* This is the generic version of the public-audit replay pattern:
|
|
5
|
+
* run a candidate system on an old task, keep the reference answers hidden
|
|
6
|
+
* until after execution, then score recall/precision and gate promotion
|
|
7
|
+
* across train/dev/test/holdout splits.
|
|
8
|
+
*/
|
|
9
|
+
export type ReferenceReplaySplit = 'train' | 'dev' | 'test' | 'holdout';
|
|
10
|
+
export type ReferenceReplayMatchStrategy = 'reference-order' | 'global-greedy';
|
|
11
|
+
export interface ReferenceReplayItem {
|
|
12
|
+
id: string;
|
|
13
|
+
title: string;
|
|
14
|
+
description?: string;
|
|
15
|
+
severity?: string;
|
|
16
|
+
tags?: string[];
|
|
17
|
+
weight?: number;
|
|
18
|
+
}
|
|
19
|
+
export interface ReferenceReplayCandidate {
|
|
20
|
+
id: string;
|
|
21
|
+
title: string;
|
|
22
|
+
description?: string;
|
|
23
|
+
severity?: string;
|
|
24
|
+
tags?: string[];
|
|
25
|
+
metadata?: Record<string, unknown>;
|
|
26
|
+
}
|
|
27
|
+
export interface ReferenceReplayScenario {
|
|
28
|
+
id: string;
|
|
29
|
+
split?: ReferenceReplaySplit;
|
|
30
|
+
references: ReferenceReplayItem[];
|
|
31
|
+
candidates: ReferenceReplayCandidate[];
|
|
32
|
+
metadata?: Record<string, unknown>;
|
|
33
|
+
}
|
|
34
|
+
export interface ReferenceReplayCase<Input = unknown> {
|
|
35
|
+
id: string;
|
|
36
|
+
split?: ReferenceReplaySplit;
|
|
37
|
+
input: Input;
|
|
38
|
+
references: ReferenceReplayItem[];
|
|
39
|
+
metadata?: Record<string, unknown>;
|
|
40
|
+
}
|
|
41
|
+
export interface ReferenceReplayExecutionScenario<Input = unknown> {
|
|
42
|
+
id: string;
|
|
43
|
+
split: ReferenceReplaySplit;
|
|
44
|
+
input: Input;
|
|
45
|
+
metadata?: Record<string, unknown>;
|
|
46
|
+
}
|
|
47
|
+
export interface ReferenceReplayRunContext {
|
|
48
|
+
runId: string;
|
|
49
|
+
caseIndex: number;
|
|
50
|
+
abortSignal?: AbortSignal;
|
|
51
|
+
}
|
|
52
|
+
export interface ReferenceReplayAdapter<Input = unknown> {
|
|
53
|
+
run(scenario: ReferenceReplayExecutionScenario<Input>, context: ReferenceReplayRunContext): Promise<ReferenceReplayCandidate[]>;
|
|
54
|
+
}
|
|
55
|
+
export type ReferenceReplayAdapterFn<Input = unknown> = (scenario: ReferenceReplayExecutionScenario<Input>, context: ReferenceReplayRunContext) => Promise<ReferenceReplayCandidate[]>;
|
|
56
|
+
export type ReferenceReplayAdapterLike<Input = unknown> = ReferenceReplayAdapter<Input> | ReferenceReplayAdapterFn<Input>;
|
|
57
|
+
export interface ReferenceReplayMatch {
|
|
58
|
+
scenarioId: string;
|
|
59
|
+
referenceId: string;
|
|
60
|
+
candidateId: string | null;
|
|
61
|
+
score: number;
|
|
62
|
+
matched: boolean;
|
|
63
|
+
weight: number;
|
|
64
|
+
reason: string;
|
|
65
|
+
}
|
|
66
|
+
export interface ReferenceReplayScenarioScore {
|
|
67
|
+
scenarioId: string;
|
|
68
|
+
split: ReferenceReplaySplit;
|
|
69
|
+
matched: number;
|
|
70
|
+
total: number;
|
|
71
|
+
falsePositives: number;
|
|
72
|
+
matchedWeight: number;
|
|
73
|
+
totalWeight: number;
|
|
74
|
+
precision: number;
|
|
75
|
+
recall: number;
|
|
76
|
+
f1: number;
|
|
77
|
+
matches: ReferenceReplayMatch[];
|
|
78
|
+
}
|
|
79
|
+
export interface ReferenceReplayAggregate {
|
|
80
|
+
matched: number;
|
|
81
|
+
total: number;
|
|
82
|
+
falsePositives: number;
|
|
83
|
+
matchedWeight: number;
|
|
84
|
+
totalWeight: number;
|
|
85
|
+
precision: number;
|
|
86
|
+
recall: number;
|
|
87
|
+
f1: number;
|
|
88
|
+
weightedRecall: number;
|
|
89
|
+
}
|
|
90
|
+
export interface ReferenceReplayScore {
|
|
91
|
+
scenarios: ReferenceReplayScenarioScore[];
|
|
92
|
+
aggregate: ReferenceReplayAggregate;
|
|
93
|
+
bySplit: Partial<Record<ReferenceReplaySplit, ReferenceReplayAggregate>>;
|
|
94
|
+
}
|
|
95
|
+
export interface ReferenceMatchResult {
|
|
96
|
+
score: number;
|
|
97
|
+
reason?: string;
|
|
98
|
+
}
|
|
99
|
+
export type ReferenceReplayMatcher = (reference: ReferenceReplayItem, candidate: ReferenceReplayCandidate, scenario: ReferenceReplayScenario) => ReferenceMatchResult;
|
|
100
|
+
export interface ReferenceReplayScoreOptions {
|
|
101
|
+
matcher?: ReferenceReplayMatcher;
|
|
102
|
+
matchThreshold?: number;
|
|
103
|
+
matchStrategy?: ReferenceReplayMatchStrategy;
|
|
104
|
+
includeHoldout?: boolean;
|
|
105
|
+
splits?: ReferenceReplaySplit[];
|
|
106
|
+
}
|
|
107
|
+
export interface ReferenceReplayPromotionPolicy {
|
|
108
|
+
/** Splits that must improve or stay flat. Default: ['dev', 'test']. */
|
|
109
|
+
requiredSplits?: ReferenceReplaySplit[];
|
|
110
|
+
/** Minimum aggregate F1 lift required on required splits. Default 0. */
|
|
111
|
+
minF1Delta?: number;
|
|
112
|
+
/** Maximum F1 drop allowed on any compared split. Default 0. */
|
|
113
|
+
maxRegression?: number;
|
|
114
|
+
/** If true, holdout must be present and must not regress. Default true. */
|
|
115
|
+
requireHoldoutNonRegression?: boolean;
|
|
116
|
+
}
|
|
117
|
+
export interface ReferenceReplaySplitComparison {
|
|
118
|
+
split: ReferenceReplaySplit;
|
|
119
|
+
baselineF1: number;
|
|
120
|
+
candidateF1: number;
|
|
121
|
+
f1Delta: number;
|
|
122
|
+
baselineRecall: number;
|
|
123
|
+
candidateRecall: number;
|
|
124
|
+
recallDelta: number;
|
|
125
|
+
}
|
|
126
|
+
export interface ReferenceReplayPromotionDecision {
|
|
127
|
+
promote: boolean;
|
|
128
|
+
reason: string;
|
|
129
|
+
aggregateDelta: number;
|
|
130
|
+
comparisons: ReferenceReplaySplitComparison[];
|
|
131
|
+
regressions: ReferenceReplaySplitComparison[];
|
|
132
|
+
}
|
|
133
|
+
export interface ReferenceReplayCaseRun<Input = unknown> {
|
|
134
|
+
caseId: string;
|
|
135
|
+
split: ReferenceReplaySplit;
|
|
136
|
+
input: Input;
|
|
137
|
+
metadata?: Record<string, unknown>;
|
|
138
|
+
references: ReferenceReplayItem[];
|
|
139
|
+
candidates: ReferenceReplayCandidate[];
|
|
140
|
+
score: ReferenceReplayScenarioScore;
|
|
141
|
+
durationMs: number;
|
|
142
|
+
error?: string;
|
|
143
|
+
}
|
|
144
|
+
export interface ReferenceReplayRun<Input = unknown> {
|
|
145
|
+
id: string;
|
|
146
|
+
variantId?: string;
|
|
147
|
+
startedAt: number;
|
|
148
|
+
completedAt: number;
|
|
149
|
+
durationMs: number;
|
|
150
|
+
cases: ReferenceReplayCaseRun<Input>[];
|
|
151
|
+
score: ReferenceReplayScore;
|
|
152
|
+
metadata?: Record<string, unknown>;
|
|
153
|
+
}
|
|
154
|
+
export interface ReferenceReplayRunOptions<Input = unknown> extends ReferenceReplayScoreOptions {
|
|
155
|
+
adapter: ReferenceReplayAdapterLike<Input>;
|
|
156
|
+
runId?: string;
|
|
157
|
+
variantId?: string;
|
|
158
|
+
metadata?: Record<string, unknown>;
|
|
159
|
+
store?: ReferenceReplayRunStore<Input>;
|
|
160
|
+
abortSignal?: AbortSignal;
|
|
161
|
+
continueOnError?: boolean;
|
|
162
|
+
now?: () => number;
|
|
163
|
+
}
|
|
164
|
+
export interface ReferenceReplayRunStore<Input = unknown> {
|
|
165
|
+
save(run: ReferenceReplayRun<Input>): Promise<void>;
|
|
166
|
+
list(): Promise<ReferenceReplayRun<Input>[]>;
|
|
167
|
+
}
|
|
168
|
+
export declare function runReferenceReplay<Input = unknown>(cases: ReferenceReplayCase<Input>[], options: ReferenceReplayRunOptions<Input>): Promise<ReferenceReplayRun<Input>>;
|
|
169
|
+
export declare function decideReferenceReplayRunPromotion(baseline: ReferenceReplayRun, candidate: ReferenceReplayRun, policy?: ReferenceReplayPromotionPolicy): ReferenceReplayPromotionDecision;
|
|
170
|
+
export declare function inMemoryReferenceReplayStore<Input = unknown>(initial?: ReferenceReplayRun<Input>[]): ReferenceReplayRunStore<Input>;
|
|
171
|
+
export declare function jsonlReferenceReplayStore<Input = unknown>(path: string): ReferenceReplayRunStore<Input>;
|
|
172
|
+
export declare function scoreReferenceReplay(scenarios: ReferenceReplayScenario[], options?: ReferenceReplayScoreOptions): ReferenceReplayScore;
|
|
173
|
+
export declare function compareReferenceReplay(baseline: ReferenceReplayScore, candidate: ReferenceReplayScore): ReferenceReplaySplitComparison[];
|
|
174
|
+
export declare function decideReferenceReplayPromotion(baseline: ReferenceReplayScore, candidate: ReferenceReplayScore, policy?: ReferenceReplayPromotionPolicy): ReferenceReplayPromotionDecision;
|
|
175
|
+
export declare function defaultReferenceReplayMatcher(reference: ReferenceReplayItem, candidate: ReferenceReplayCandidate): ReferenceMatchResult;
|
|
176
|
+
//# sourceMappingURL=reference-replay.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"reference-replay.d.ts","sourceRoot":"","sources":["../src/reference-replay.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAMH,MAAM,MAAM,oBAAoB,GAAG,OAAO,GAAG,KAAK,GAAG,MAAM,GAAG,SAAS,CAAA;AACvE,MAAM,MAAM,4BAA4B,GAAG,iBAAiB,GAAG,eAAe,CAAA;AAE9E,MAAM,WAAW,mBAAmB;IAClC,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,EAAE,MAAM,CAAA;IACb,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAA;IACf,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,wBAAwB;IACvC,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,EAAE,MAAM,CAAA;IACb,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAA;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,uBAAuB;IACtC,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,CAAC,EAAE,oBAAoB,CAAA;IAC5B,UAAU,EAAE,mBAAmB,EAAE,CAAA;IACjC,UAAU,EAAE,wBAAwB,EAAE,CAAA;IACtC,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,mBAAmB,CAAC,KAAK,GAAG,OAAO;IAClD,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,CAAC,EAAE,oBAAoB,CAAA;IAC5B,KAAK,EAAE,KAAK,CAAA;IACZ,UAAU,EAAE,mBAAmB,EAAE,CAAA;IACjC,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,gCAAgC,CAAC,KAAK,GAAG,OAAO;IAC/D,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,EAAE,oBAAoB,CAAA;IAC3B,KAAK,EAAE,KAAK,CAAA;IACZ,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,yBAAyB;IACxC,KAAK,EAAE,MAAM,CAAA;IACb,SAAS,EAAE,MAAM,CAAA;IACjB,WAAW,CAAC,EAAE,WAAW,CAAA;CAC1B;AAED,MAAM,WAAW,sBAAsB,CAAC,KAAK,GAAG,OAAO;IACrD,GAAG,CACD,QAAQ,EAAE,gCAAgC,CAAC,KAAK,CAAC,EACjD,OAAO,EAAE,yBAAyB,GACjC,OAAO,CAAC,wBAAwB,EAAE,CAAC,CAAA;CACvC;AAED,MAAM,MAAM,wBAAwB,CAAC,KAAK,GAAG,OAAO,IAAI,CACtD,QAAQ,EAAE,gCAAgC,CAAC,KAAK,CAAC,EACjD,OAAO,EAAE,yBAAyB,KAC/B,OAAO,CAAC,wBAAwB,EAAE,CAAC,CAAA;AAExC,MAAM,MAAM,0BAA0B,CAAC,KAAK,GAAG,OAAO,IAClD,sBAAsB,CAAC,KAAK,CAAC,GAC7B,wBAAwB,CAAC,KAAK,CAAC,CAAA;AAEnC,MAAM,WAAW,oBAAoB;IACnC,UAAU,EAAE,MAAM,CAAA;IAClB,WAAW,EAAE,MAAM,CAAA;IACnB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAA;IAC1B,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,EAAE,OAAO,CAAA;IAChB,MAAM,EAAE,MAAM,CAAA;IACd,MAAM,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,4BAA4B;IAC3C,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,EAAE,oBAAoB,CAAA;IAC3B,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,EAAE,MAAM,CAAA;IACb,cAAc,EAAE,MAAM,CAAA;IACtB,aAAa,EAAE,MAAM,CAAA;IACrB,WAAW,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,MAAM,EAAE,MAAM,CAAA;IACd,EAAE,EAAE,MAAM,CAAA;IACV,OAAO,EAAE,oBAAoB,EAAE,CAAA;CAChC;AAED,MAAM,WAAW,wBAAwB;IACvC,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,EAAE,MAAM,CAAA;IACb,cAAc,EAAE,MAAM,CAAA;IACtB,aAAa,EAAE,MAAM,CAAA;IACrB,WAAW,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,MAAM,EAAE,MAAM,CAAA;IACd,EAAE,EAAE,MAAM,CAAA;IACV,cAAc,EAAE,MAAM,CAAA;CACvB;AAED,MAAM,WAAW,oBAAoB;IACnC,SAAS,EAAE,4BAA4B,EAAE,CAAA;IACzC,SAAS,EAAE,wBAAwB,CAAA;IACnC,OAAO,EAAE,OAAO,CAAC,MAAM,CAAC,oBAAoB,EAAE,wBAAwB,CAAC,CAAC,CAAA;CACzE;AAED,MAAM,WAAW,oBAAoB;IACnC,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,MAAM,sBAAsB,GAAG,CACnC,SAAS,EAAE,mBAAmB,EAC9B,SAAS,EAAE,wBAAwB,EACnC,QAAQ,EAAE,uBAAuB,KAC9B,oBAAoB,CAAA;AAEzB,MAAM,WAAW,2BAA2B;IAC1C,OAAO,CAAC,EAAE,sBAAsB,CAAA;IAChC,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,aAAa,CAAC,EAAE,4BAA4B,CAAA;IAC5C,cAAc,CAAC,EAAE,OAAO,CAAA;IACxB,MAAM,CAAC,EAAE,oBAAoB,EAAE,CAAA;CAChC;AAED,MAAM,WAAW,8BAA8B;IAC7C,uEAAuE;IACvE,cAAc,CAAC,EAAE,oBAAoB,EAAE,CAAA;IACvC,wEAAwE;IACxE,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,gEAAgE;IAChE,aAAa,CAAC,EAAE,MAAM,CAAA;IACtB,2EAA2E;IAC3E,2BAA2B,CAAC,EAAE,OAAO,CAAA;CACtC;AAED,MAAM,WAAW,8BAA8B;IAC7C,KAAK,EAAE,oBAAoB,CAAA;IAC3B,UAAU,EAAE,MAAM,CAAA;IAClB,WAAW,EAAE,MAAM,CAAA;IACnB,OAAO,EAAE,MAAM,CAAA;IACf,cAAc,EAAE,MAAM,CAAA;IACtB,eAAe,EAAE,MAAM,CAAA;IACvB,WAAW,EAAE,MAAM,CAAA;CACpB;AAED,MAAM,WAAW,gCAAgC;IAC/C,OAAO,EAAE,OAAO,CAAA;IAChB,MAAM,EAAE,MAAM,CAAA;IACd,cAAc,EAAE,MAAM,CAAA;IACtB,WAAW,EAAE,8BAA8B,EAAE,CAAA;IAC7C,WAAW,EAAE,8BAA8B,EAAE,CAAA;CAC9C;AAED,MAAM,WAAW,sBAAsB,CAAC,KAAK,GAAG,OAAO;IACrD,MAAM,EAAE,MAAM,CAAA;IACd,KAAK,EAAE,oBAAoB,CAAA;IAC3B,KAAK,EAAE,KAAK,CAAA;IACZ,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAClC,UAAU,EAAE,mBAAmB,EAAE,CAAA;IACjC,UAAU,EAAE,wBAAwB,EAAE,CAAA;IACtC,KAAK,EAAE,4BAA4B,CAAA;IACnC,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,kBAAkB,CAAC,KAAK,GAAG,OAAO;IACjD,EAAE,EAAE,MAAM,CAAA;IACV,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,SAAS,EAAE,MAAM,CAAA;IACjB,WAAW,EAAE,MAAM,CAAA;IACnB,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,EAAE,sBAAsB,CAAC,KAAK,CAAC,EAAE,CAAA;IACtC,KAAK,EAAE,oBAAoB,CAAA;IAC3B,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,yBAAyB,CAAC,KAAK,GAAG,OAAO,CAAE,SAAQ,2BAA2B;IAC7F,OAAO,EAAE,0BAA0B,CAAC,KAAK,CAAC,CAAA;IAC1C,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAClC,KAAK,CAAC,EAAE,uBAAuB,CAAC,KAAK,CAAC,CAAA;IACtC,WAAW,CAAC,EAAE,WAAW,CAAA;IACzB,eAAe,CAAC,EAAE,OAAO,CAAA;IACzB,GAAG,CAAC,EAAE,MAAM,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,uBAAuB,CAAC,KAAK,GAAG,OAAO;IACtD,IAAI,CAAC,GAAG,EAAE,kBAAkB,CAAC,KAAK,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IACnD,IAAI,IAAI,OAAO,CAAC,kBAAkB,CAAC,KAAK,CAAC,EAAE,CAAC,CAAA;CAC7C;AAKD,wBAAsB,kBAAkB,CAAC,KAAK,GAAG,OAAO,EACtD,KAAK,EAAE,mBAAmB,CAAC,KAAK,CAAC,EAAE,EACnC,OAAO,EAAE,yBAAyB,CAAC,KAAK,CAAC,GACxC,OAAO,CAAC,kBAAkB,CAAC,KAAK,CAAC,CAAC,CA4FpC;AAED,wBAAgB,iCAAiC,CAC/C,QAAQ,EAAE,kBAAkB,EAC5B,SAAS,EAAE,kBAAkB,EAC7B,MAAM,GAAE,8BAAmC,GAC1C,gCAAgC,CAElC;AAED,wBAAgB,4BAA4B,CAAC,KAAK,GAAG,OAAO,EAC1D,OAAO,GAAE,kBAAkB,CAAC,KAAK,CAAC,EAAO,GACxC,uBAAuB,CAAC,KAAK,CAAC,CAUhC;AAkBD,wBAAgB,yBAAyB,CAAC,KAAK,GAAG,OAAO,EACvD,IAAI,EAAE,MAAM,GACX,uBAAuB,CAAC,KAAK,CAAC,CAgBhC;AAED,wBAAgB,oBAAoB,CAClC,SAAS,EAAE,uBAAuB,EAAE,EACpC,OAAO,GAAE,2BAAgC,GACxC,oBAAoB,CAkBtB;AAED,wBAAgB,sBAAsB,CACpC,QAAQ,EAAE,oBAAoB,EAC9B,SAAS,EAAE,oBAAoB,GAC9B,8BAA8B,EAAE,CAkBlC;AAED,wBAAgB,8BAA8B,CAC5C,QAAQ,EAAE,oBAAoB,EAC9B,SAAS,EAAE,oBAAoB,EAC/B,MAAM,GAAE,8BAAmC,GAC1C,gCAAgC,CAuElC;AAED,wBAAgB,6BAA6B,CAC3C,SAAS,EAAE,mBAAmB,EAC9B,SAAS,EAAE,wBAAwB,GAClC,oBAAoB,CAgBtB"}
|
|
@@ -0,0 +1,79 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Reflective mutation — primitives for trace-conditioned prompt rewriting.
|
|
3
|
+
*
|
|
4
|
+
* Used by `prompt-evolution.ts` (and any consumer running iterative
|
|
5
|
+
* improvement). Given a parent prompt + concrete trace evidence (top trials,
|
|
6
|
+
* bottom trials, missed expectations), produce an LLM-ready prompt that
|
|
7
|
+
* proposes targeted mutations — not blind rephrasings.
|
|
8
|
+
*
|
|
9
|
+
* Why this lives outside `prompt-evolution.ts`: any consumer that wants to
|
|
10
|
+
* run reflective rewriting WITHOUT the population/Pareto machinery can
|
|
11
|
+
* import these primitives directly.
|
|
12
|
+
*
|
|
13
|
+
* Quality bar (vs. naive "mutate this prompt"):
|
|
14
|
+
* - Show parent ↔ children diff, not just one variant
|
|
15
|
+
* - Quote specific missed goldens with their match phrases
|
|
16
|
+
* - Surface the model's actual emitted output side-by-side with what was expected
|
|
17
|
+
* - Quote concrete mutation primitives so the model has a vocabulary
|
|
18
|
+
*/
|
|
19
|
+
export interface TrialTrace {
|
|
20
|
+
/** Stable id for the trial — surfaces in the prompt for grounding. */
|
|
21
|
+
id: string;
|
|
22
|
+
/** Score the trial received on its primary metric. */
|
|
23
|
+
score: number;
|
|
24
|
+
/** Candidate inputs the agent was given (e.g., the fixture or scenario). */
|
|
25
|
+
inputName?: string;
|
|
26
|
+
/**
|
|
27
|
+
* Goldens / expectations this trial was tested against, with whether each
|
|
28
|
+
* was matched. The reflection prompt quotes the missed ones specifically.
|
|
29
|
+
*/
|
|
30
|
+
expectations?: Array<{
|
|
31
|
+
id: string;
|
|
32
|
+
phrase: string;
|
|
33
|
+
matched: boolean;
|
|
34
|
+
}>;
|
|
35
|
+
/** Free-form text — what the agent actually emitted (e.g., findings, plan). */
|
|
36
|
+
emitted?: string;
|
|
37
|
+
/** The judge's free-form note on WHY this trial scored low — the failure
|
|
38
|
+
* evidence the reflection targets. Should be a generalizable pattern (which
|
|
39
|
+
* checks/lines/dimensions failed, and how), never case-specific ground
|
|
40
|
+
* truth (that would teach memorization, not capability). */
|
|
41
|
+
failureNote?: string;
|
|
42
|
+
/** Optional structured metrics (recall, precision, cost, latency). */
|
|
43
|
+
metrics?: Record<string, number>;
|
|
44
|
+
}
|
|
45
|
+
export interface ReflectionContext {
|
|
46
|
+
/** What is being mutated — appears in the system prompt for orientation. */
|
|
47
|
+
target: string;
|
|
48
|
+
/** Current variant's payload — JSON-serialised for the prompt. */
|
|
49
|
+
parentPayload: unknown;
|
|
50
|
+
/** Best-performing trials this generation. */
|
|
51
|
+
topTrials: TrialTrace[];
|
|
52
|
+
/** Worst-performing trials this generation — the missed-golden source. */
|
|
53
|
+
bottomTrials: TrialTrace[];
|
|
54
|
+
/** How many children the mutator should propose. */
|
|
55
|
+
childCount: number;
|
|
56
|
+
/** Optional: domain-specific mutation primitives the model can pick from. */
|
|
57
|
+
mutationPrimitives?: string[];
|
|
58
|
+
}
|
|
59
|
+
export declare const DEFAULT_MUTATION_PRIMITIVES: string[];
|
|
60
|
+
/**
|
|
61
|
+
* Build the LLM-ready reflection prompt. Output is plain text — pass it as
|
|
62
|
+
* the user message. The system message should be small and stable (e.g.
|
|
63
|
+
* "Output ONLY a JSON object matching the schema below.").
|
|
64
|
+
*/
|
|
65
|
+
export declare function buildReflectionPrompt(ctx: ReflectionContext): string;
|
|
66
|
+
/** Render analyst findings + an optional analysis report into a prompt evidence
|
|
67
|
+
* block. A findings producer diagnoses the generation's traces; this surfaces
|
|
68
|
+
* that diagnosis to the proposer so a mutation targets a NAMED root cause instead of blindly
|
|
69
|
+
* rephrasing. Duck-types the finding shape (claim / severity / area /
|
|
70
|
+
* recommended_action) so any findings producer's rows render. Returns null when
|
|
71
|
+
* there is nothing to add, so callers append nothing. */
|
|
72
|
+
export declare function renderAnalystEvidence(findings: unknown[], report?: unknown): string | null;
|
|
73
|
+
export interface ReflectionProposal {
|
|
74
|
+
label: string;
|
|
75
|
+
rationale: string;
|
|
76
|
+
payload: unknown;
|
|
77
|
+
}
|
|
78
|
+
export declare function parseReflectionResponse(raw: string, maxProposals?: number): ReflectionProposal[];
|
|
79
|
+
//# sourceMappingURL=reflective-mutation.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"reflective-mutation.d.ts","sourceRoot":"","sources":["../src/reflective-mutation.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAEH,MAAM,WAAW,UAAU;IACzB,sEAAsE;IACtE,EAAE,EAAE,MAAM,CAAA;IACV,sDAAsD;IACtD,KAAK,EAAE,MAAM,CAAA;IACb,4EAA4E;IAC5E,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB;;;OAGG;IACH,YAAY,CAAC,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,OAAO,CAAA;KAAE,CAAC,CAAA;IACtE,+EAA+E;IAC/E,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB;;;iEAG6D;IAC7D,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,sEAAsE;IACtE,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CACjC;AAED,MAAM,WAAW,iBAAiB;IAChC,4EAA4E;IAC5E,MAAM,EAAE,MAAM,CAAA;IACd,kEAAkE;IAClE,aAAa,EAAE,OAAO,CAAA;IACtB,8CAA8C;IAC9C,SAAS,EAAE,UAAU,EAAE,CAAA;IACvB,0EAA0E;IAC1E,YAAY,EAAE,UAAU,EAAE,CAAA;IAC1B,oDAAoD;IACpD,UAAU,EAAE,MAAM,CAAA;IAClB,6EAA6E;IAC7E,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAA;CAC9B;AAED,eAAO,MAAM,2BAA2B,EAAE,MAAM,EAO/C,CAAA;AAED;;;;GAIG;AACH,wBAAgB,qBAAqB,CAAC,GAAG,EAAE,iBAAiB,GAAG,MAAM,CAqFpE;AAED;;;;;0DAK0D;AAC1D,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,MAAM,CAAC,EAAE,OAAO,GAAG,MAAM,GAAG,IAAI,CAuC1F;AAWD,MAAM,WAAW,kBAAkB;IACjC,KAAK,EAAE,MAAM,CAAA;IACb,SAAS,EAAE,MAAM,CAAA;IACjB,OAAO,EAAE,OAAO,CAAA;CACjB;AA0DD,wBAAgB,uBAAuB,CAAC,GAAG,EAAE,MAAM,EAAE,YAAY,CAAC,EAAE,MAAM,GAAG,kBAAkB,EAAE,CA8EhG"}
|
|
@@ -0,0 +1,31 @@
|
|
|
1
|
+
import type { Scenario, ScenarioFile } from './types';
|
|
2
|
+
/**
|
|
3
|
+
* ScenarioRegistry — manages scenario discovery and filtering.
|
|
4
|
+
*
|
|
5
|
+
* Each agent registers its scenarios. The registry handles conversion
|
|
6
|
+
* from ScenarioFile format to the framework's Scenario type.
|
|
7
|
+
*/
|
|
8
|
+
export declare class ScenarioRegistry {
|
|
9
|
+
private scenarios;
|
|
10
|
+
private scenarioFiles;
|
|
11
|
+
/** Register scenarios from ScenarioFile format */
|
|
12
|
+
registerFiles(files: ScenarioFile[]): void;
|
|
13
|
+
/** Register pre-built Scenario objects directly */
|
|
14
|
+
register(scenarios: Scenario[]): void;
|
|
15
|
+
/** Get all scenarios */
|
|
16
|
+
all(): Scenario[];
|
|
17
|
+
/** Get scenarios filtered by category */
|
|
18
|
+
byCategory(category: string): Scenario[];
|
|
19
|
+
/** List all categories with counts */
|
|
20
|
+
listCategories(): {
|
|
21
|
+
category: string;
|
|
22
|
+
count: number;
|
|
23
|
+
}[];
|
|
24
|
+
/** Get scenarios filtered by persona */
|
|
25
|
+
byPersona(persona: string): Scenario[];
|
|
26
|
+
/** Get a single scenario by ID */
|
|
27
|
+
byId(id: string): Scenario | undefined;
|
|
28
|
+
/** Count total scenarios */
|
|
29
|
+
get count(): number;
|
|
30
|
+
}
|
|
31
|
+
//# sourceMappingURL=registry.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"registry.d.ts","sourceRoot":"","sources":["../src/registry.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,YAAY,EAAE,MAAM,SAAS,CAAA;AAErD;;;;;GAKG;AACH,qBAAa,gBAAgB;IAC3B,OAAO,CAAC,SAAS,CAAiB;IAClC,OAAO,CAAC,aAAa,CAAqB;IAE1C,kDAAkD;IAClD,aAAa,CAAC,KAAK,EAAE,YAAY,EAAE,GAAG,IAAI;IAK1C,mDAAmD;IACnD,QAAQ,CAAC,SAAS,EAAE,QAAQ,EAAE,GAAG,IAAI;IAIrC,wBAAwB;IACxB,GAAG,IAAI,QAAQ,EAAE;IAIjB,yCAAyC;IACzC,UAAU,CAAC,QAAQ,EAAE,MAAM,GAAG,QAAQ,EAAE;IAKxC,sCAAsC;IACtC,cAAc,IAAI;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,EAAE;IAQvD,wCAAwC;IACxC,SAAS,CAAC,OAAO,EAAE,MAAM,GAAG,QAAQ,EAAE;IAItC,kCAAkC;IAClC,IAAI,CAAC,EAAE,EAAE,MAAM,GAAG,QAAQ,GAAG,SAAS;IAItC,4BAA4B;IAC5B,IAAI,KAAK,IAAI,MAAM,CAElB;CACF"}
|
|
@@ -0,0 +1,128 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Release confidence gate.
|
|
3
|
+
*
|
|
4
|
+
* This is the production-facing composition layer over the lower-level
|
|
5
|
+
* primitives:
|
|
6
|
+
* - Dataset manifests prove corpus/version coverage.
|
|
7
|
+
* - RunRecord rows prove reproducible search/holdout outcomes.
|
|
8
|
+
* - Multi-shot trace evidence carries turn counts and ASI diagnostics.
|
|
9
|
+
* - HeldOutGate decisions remain the paired promotion authority.
|
|
10
|
+
*
|
|
11
|
+
* The gate is intentionally pure and conservative. Missing declared evidence
|
|
12
|
+
* fails closed instead of being treated as a neutral zero.
|
|
13
|
+
*/
|
|
14
|
+
import type { DatasetManifest, DatasetScenario, DatasetSplit } from './dataset';
|
|
15
|
+
import type { GateDecision } from './held-out-gate';
|
|
16
|
+
import type { RunRecord, RunSplitTag } from './run-record';
|
|
17
|
+
/** Severity of an actionable finding attached to a run/trace. */
|
|
18
|
+
export type AsiSeverity = 'info' | 'warning' | 'error' | 'critical';
|
|
19
|
+
/** Actionable side-info — a diagnosed finding the loop can act on. */
|
|
20
|
+
export interface ActionableSideInfo {
|
|
21
|
+
/** Stable expectation/check id when available. */
|
|
22
|
+
expectationId?: string;
|
|
23
|
+
/** Human-readable diagnosis of what happened. */
|
|
24
|
+
message: string;
|
|
25
|
+
severity?: AsiSeverity;
|
|
26
|
+
/** Concrete trace excerpt, file path, tool call, screenshot id, etc. */
|
|
27
|
+
evidence?: string;
|
|
28
|
+
/** Prompt/tool/context surface likely responsible. */
|
|
29
|
+
responsibleSurface?: string;
|
|
30
|
+
/** Suggested fix in natural language. */
|
|
31
|
+
suggestion?: string;
|
|
32
|
+
/** Whether this expectation was satisfied. Defaults to false for ASI rows. */
|
|
33
|
+
matched?: boolean;
|
|
34
|
+
metadata?: Record<string, unknown>;
|
|
35
|
+
}
|
|
36
|
+
export type ReleaseConfidenceStatus = 'pass' | 'warn' | 'fail';
|
|
37
|
+
export type ReleaseConfidenceAxisName = 'corpus' | 'quality' | 'generalization' | 'diagnostics' | 'efficiency';
|
|
38
|
+
export interface ReleaseTraceEvidence {
|
|
39
|
+
scenarioId: string;
|
|
40
|
+
candidateId?: string;
|
|
41
|
+
split?: RunSplitTag;
|
|
42
|
+
score?: number;
|
|
43
|
+
ok?: boolean;
|
|
44
|
+
turnCount?: number;
|
|
45
|
+
costUsd?: number;
|
|
46
|
+
durationMs?: number;
|
|
47
|
+
failureMode?: string;
|
|
48
|
+
asi?: ActionableSideInfo[];
|
|
49
|
+
metadata?: Record<string, unknown>;
|
|
50
|
+
}
|
|
51
|
+
export interface ReleaseConfidenceThresholds {
|
|
52
|
+
/** Require a Dataset manifest or explicit scenarios. Default true. */
|
|
53
|
+
requireCorpus?: boolean;
|
|
54
|
+
minScenarioCount?: number;
|
|
55
|
+
minSearchRuns?: number;
|
|
56
|
+
minHoldoutRuns?: number;
|
|
57
|
+
/** Require at least one holdout scenario/run. Default true. */
|
|
58
|
+
requireHoldout?: boolean;
|
|
59
|
+
minPassRate?: number;
|
|
60
|
+
minMeanScore?: number;
|
|
61
|
+
/** Search mean may exceed holdout mean by at most this much. */
|
|
62
|
+
maxOverfitGap?: number;
|
|
63
|
+
maxMeanCostUsd?: number;
|
|
64
|
+
maxP95WallMs?: number;
|
|
65
|
+
/** Low-score/failed rows must carry ASI. Default true. */
|
|
66
|
+
requireAsiForFailures?: boolean;
|
|
67
|
+
/** Score below this is considered a failure for ASI coverage. Default 0.5. */
|
|
68
|
+
failureScoreThreshold?: number;
|
|
69
|
+
}
|
|
70
|
+
export interface ReleaseConfidenceInput {
|
|
71
|
+
target: string;
|
|
72
|
+
candidateId?: string;
|
|
73
|
+
baselineId?: string;
|
|
74
|
+
dataset?: DatasetManifest;
|
|
75
|
+
scenarios?: readonly DatasetScenario[];
|
|
76
|
+
runs?: readonly RunRecord[];
|
|
77
|
+
traces?: readonly ReleaseTraceEvidence[];
|
|
78
|
+
gateDecision?: GateDecision | null;
|
|
79
|
+
thresholds?: ReleaseConfidenceThresholds;
|
|
80
|
+
}
|
|
81
|
+
export interface ReleaseConfidenceAxis {
|
|
82
|
+
name: ReleaseConfidenceAxisName;
|
|
83
|
+
status: ReleaseConfidenceStatus;
|
|
84
|
+
score: number;
|
|
85
|
+
detail: string;
|
|
86
|
+
}
|
|
87
|
+
export interface ReleaseConfidenceIssue {
|
|
88
|
+
axis: ReleaseConfidenceAxisName;
|
|
89
|
+
severity: 'critical' | 'warning';
|
|
90
|
+
code: string;
|
|
91
|
+
detail: string;
|
|
92
|
+
}
|
|
93
|
+
export interface ReleaseConfidenceMetrics {
|
|
94
|
+
scenarioCount: number;
|
|
95
|
+
searchRuns: number;
|
|
96
|
+
holdoutRuns: number;
|
|
97
|
+
passRate: number;
|
|
98
|
+
meanScore: number;
|
|
99
|
+
searchMeanScore: number;
|
|
100
|
+
holdoutMeanScore: number;
|
|
101
|
+
overfitGap: number;
|
|
102
|
+
meanCostUsd: number;
|
|
103
|
+
p95WallMs: number;
|
|
104
|
+
failedRows: number;
|
|
105
|
+
failuresWithAsi: number;
|
|
106
|
+
singleShotTraces: number;
|
|
107
|
+
multiShotTraces: number;
|
|
108
|
+
splitCounts: Record<DatasetSplit, number>;
|
|
109
|
+
domainCounts: Record<string, number>;
|
|
110
|
+
failureModeCounts: Record<string, number>;
|
|
111
|
+
responsibleSurfaceCounts: Record<string, number>;
|
|
112
|
+
}
|
|
113
|
+
export interface ReleaseConfidenceScorecard {
|
|
114
|
+
target: string;
|
|
115
|
+
candidateId: string | null;
|
|
116
|
+
baselineId: string | null;
|
|
117
|
+
status: ReleaseConfidenceStatus;
|
|
118
|
+
promote: boolean;
|
|
119
|
+
axes: ReleaseConfidenceAxis[];
|
|
120
|
+
issues: ReleaseConfidenceIssue[];
|
|
121
|
+
metrics: ReleaseConfidenceMetrics;
|
|
122
|
+
dataset: DatasetManifest | null;
|
|
123
|
+
gateDecision: GateDecision | null;
|
|
124
|
+
summary: string;
|
|
125
|
+
}
|
|
126
|
+
export declare function evaluateReleaseConfidence(input: ReleaseConfidenceInput): ReleaseConfidenceScorecard;
|
|
127
|
+
export declare function assertReleaseConfidence(input: ReleaseConfidenceInput): ReleaseConfidenceScorecard;
|
|
128
|
+
//# sourceMappingURL=release-confidence.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"release-confidence.d.ts","sourceRoot":"","sources":["../src/release-confidence.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,eAAe,EAAE,YAAY,EAAE,MAAM,WAAW,CAAA;AAE/E,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAA;AACnD,OAAO,KAAK,EAAE,SAAS,EAAE,WAAW,EAAE,MAAM,cAAc,CAAA;AAE1D,iEAAiE;AACjE,MAAM,MAAM,WAAW,GAAG,MAAM,GAAG,SAAS,GAAG,OAAO,GAAG,UAAU,CAAA;AAEnE,sEAAsE;AACtE,MAAM,WAAW,kBAAkB;IACjC,kDAAkD;IAClD,aAAa,CAAC,EAAE,MAAM,CAAA;IACtB,iDAAiD;IACjD,OAAO,EAAE,MAAM,CAAA;IACf,QAAQ,CAAC,EAAE,WAAW,CAAA;IACtB,wEAAwE;IACxE,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,sDAAsD;IACtD,kBAAkB,CAAC,EAAE,MAAM,CAAA;IAC3B,yCAAyC;IACzC,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,8EAA8E;IAC9E,OAAO,CAAC,EAAE,OAAO,CAAA;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,MAAM,uBAAuB,GAAG,MAAM,GAAG,MAAM,GAAG,MAAM,CAAA;AAC9D,MAAM,MAAM,yBAAyB,GACjC,QAAQ,GACR,SAAS,GACT,gBAAgB,GAChB,aAAa,GACb,YAAY,CAAA;AAEhB,MAAM,WAAW,oBAAoB;IACnC,UAAU,EAAE,MAAM,CAAA;IAClB,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,KAAK,CAAC,EAAE,WAAW,CAAA;IACnB,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,EAAE,CAAC,EAAE,OAAO,CAAA;IACZ,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,GAAG,CAAC,EAAE,kBAAkB,EAAE,CAAA;IAC1B,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,2BAA2B;IAC1C,sEAAsE;IACtE,aAAa,CAAC,EAAE,OAAO,CAAA;IACvB,gBAAgB,CAAC,EAAE,MAAM,CAAA;IACzB,aAAa,CAAC,EAAE,MAAM,CAAA;IACtB,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,+DAA+D;IAC/D,cAAc,CAAC,EAAE,OAAO,CAAA;IACxB,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,gEAAgE;IAChE,aAAa,CAAC,EAAE,MAAM,CAAA;IACtB,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,0DAA0D;IAC1D,qBAAqB,CAAC,EAAE,OAAO,CAAA;IAC/B,8EAA8E;IAC9E,qBAAqB,CAAC,EAAE,MAAM,CAAA;CAC/B;AAED,MAAM,WAAW,sBAAsB;IACrC,MAAM,EAAE,MAAM,CAAA;IACd,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,OAAO,CAAC,EAAE,eAAe,CAAA;IACzB,SAAS,CAAC,EAAE,SAAS,eAAe,EAAE,CAAA;IACtC,IAAI,CAAC,EAAE,SAAS,SAAS,EAAE,CAAA;IAC3B,MAAM,CAAC,EAAE,SAAS,oBAAoB,EAAE,CAAA;IACxC,YAAY,CAAC,EAAE,YAAY,GAAG,IAAI,CAAA;IAClC,UAAU,CAAC,EAAE,2BAA2B,CAAA;CACzC;AAED,MAAM,WAAW,qBAAqB;IACpC,IAAI,EAAE,yBAAyB,CAAA;IAC/B,MAAM,EAAE,uBAAuB,CAAA;IAC/B,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,sBAAsB;IACrC,IAAI,EAAE,yBAAyB,CAAA;IAC/B,QAAQ,EAAE,UAAU,GAAG,SAAS,CAAA;IAChC,IAAI,EAAE,MAAM,CAAA;IACZ,MAAM,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,wBAAwB;IACvC,aAAa,EAAE,MAAM,CAAA;IACrB,UAAU,EAAE,MAAM,CAAA;IAClB,WAAW,EAAE,MAAM,CAAA;IACnB,QAAQ,EAAE,MAAM,CAAA;IAChB,SAAS,EAAE,MAAM,CAAA;IACjB,eAAe,EAAE,MAAM,CAAA;IACvB,gBAAgB,EAAE,MAAM,CAAA;IACxB,UAAU,EAAE,MAAM,CAAA;IAClB,WAAW,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,UAAU,EAAE,MAAM,CAAA;IAClB,eAAe,EAAE,MAAM,CAAA;IACvB,gBAAgB,EAAE,MAAM,CAAA;IACxB,eAAe,EAAE,MAAM,CAAA;IACvB,WAAW,EAAE,MAAM,CAAC,YAAY,EAAE,MAAM,CAAC,CAAA;IACzC,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IACpC,iBAAiB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IACzC,wBAAwB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CACjD;AAED,MAAM,WAAW,0BAA0B;IACzC,MAAM,EAAE,MAAM,CAAA;IACd,WAAW,EAAE,MAAM,GAAG,IAAI,CAAA;IAC1B,UAAU,EAAE,MAAM,GAAG,IAAI,CAAA;IACzB,MAAM,EAAE,uBAAuB,CAAA;IAC/B,OAAO,EAAE,OAAO,CAAA;IAChB,IAAI,EAAE,qBAAqB,EAAE,CAAA;IAC7B,MAAM,EAAE,sBAAsB,EAAE,CAAA;IAChC,OAAO,EAAE,wBAAwB,CAAA;IACjC,OAAO,EAAE,eAAe,GAAG,IAAI,CAAA;IAC/B,YAAY,EAAE,YAAY,GAAG,IAAI,CAAA;IACjC,OAAO,EAAE,MAAM,CAAA;CAChB;AAiBD,wBAAgB,yBAAyB,CACvC,KAAK,EAAE,sBAAsB,GAC5B,0BAA0B,CAyE5B;AAED,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,sBAAsB,GAAG,0BAA0B,CAMjG"}
|
|
@@ -0,0 +1,11 @@
|
|
|
1
|
+
import type { ReleaseConfidenceScorecard } from './release-confidence';
|
|
2
|
+
import type { RunRecord } from './run-record';
|
|
3
|
+
export interface RenderReleaseReportOptions {
|
|
4
|
+
title?: string;
|
|
5
|
+
runs?: readonly RunRecord[];
|
|
6
|
+
comparator?: string;
|
|
7
|
+
traceAnalystFindings?: readonly string[];
|
|
8
|
+
nextActions?: readonly string[];
|
|
9
|
+
}
|
|
10
|
+
export declare function renderReleaseReport(scorecard: ReleaseConfidenceScorecard, options?: RenderReleaseReportOptions): string;
|
|
11
|
+
//# sourceMappingURL=release-report.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"release-report.d.ts","sourceRoot":"","sources":["../src/release-report.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,0BAA0B,EAAE,MAAM,sBAAsB,CAAA;AACtE,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,cAAc,CAAA;AAG7C,MAAM,WAAW,0BAA0B;IACzC,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,IAAI,CAAC,EAAE,SAAS,SAAS,EAAE,CAAA;IAC3B,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,oBAAoB,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;IACxC,WAAW,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;CAChC;AAED,wBAAgB,mBAAmB,CACjC,SAAS,EAAE,0BAA0B,EACrC,OAAO,GAAE,0BAA+B,GACvC,MAAM,CAkFR"}
|
package/dist/replay.d.ts
ADDED
|
@@ -0,0 +1,120 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Replay-from-raw-events — turn every captured campaign run into a
|
|
3
|
+
* re-runnable artifact.
|
|
4
|
+
*
|
|
5
|
+
* `RawProviderSink` captures every provider HTTP envelope; `runEvalCampaign`
|
|
6
|
+
* makes that capture the default. Together they make every past run a
|
|
7
|
+
* complete fingerprint of what happened on the wire — enough to replay
|
|
8
|
+
* the run without burning new LLM cost.
|
|
9
|
+
*
|
|
10
|
+
* Three use cases this primitive enables:
|
|
11
|
+
*
|
|
12
|
+
* 1. **Post-hoc judging** — apply a new judge / rubric / scoring callback
|
|
13
|
+
* to last week's runs without re-calling any LLM. The cost of trying
|
|
14
|
+
* a new rubric drops from "another full sweep" to a CPU-bound replay.
|
|
15
|
+
* 2. **Determinism audits** — replay the same campaign and verify the
|
|
16
|
+
* raw responses match byte-for-byte. Any drift is a non-determinism
|
|
17
|
+
* bug (in the harness, the prompt builder, the sandbox, …).
|
|
18
|
+
* 3. **Free judge calibration** — run two judges on identical responses
|
|
19
|
+
* and measure inter-judge agreement without doubling LLM spend.
|
|
20
|
+
*
|
|
21
|
+
* The interface is deliberately fetch-shaped. Inject `createReplayFetch`
|
|
22
|
+
* into `LlmClientOptions.fetch` and every `callLlm` transparently reads
|
|
23
|
+
* from the cache instead of calling the network. No new code path through
|
|
24
|
+
* the LLM client is needed; the cache hit is invisible to the runner.
|
|
25
|
+
*/
|
|
26
|
+
import { ReplayError } from './errors';
|
|
27
|
+
import type { RawProviderEvent, RawProviderSink } from './trace/raw-provider-sink';
|
|
28
|
+
export declare class ReplayCacheMissError extends ReplayError {
|
|
29
|
+
readonly url: string;
|
|
30
|
+
readonly requestKey: string;
|
|
31
|
+
constructor(url: string, requestKey: string, message?: string);
|
|
32
|
+
}
|
|
33
|
+
export interface ReplayCacheEntry {
|
|
34
|
+
request: RawProviderEvent;
|
|
35
|
+
response: RawProviderEvent;
|
|
36
|
+
}
|
|
37
|
+
export interface ReplayCacheStats {
|
|
38
|
+
total: number;
|
|
39
|
+
byProvider: Record<string, number>;
|
|
40
|
+
byModel: Record<string, number>;
|
|
41
|
+
/** Spans for which we have a request but no response (run aborted mid-call). */
|
|
42
|
+
orphanRequests: number;
|
|
43
|
+
}
|
|
44
|
+
/**
|
|
45
|
+
* In-memory deterministic cache of (request → response) keyed on a stable
|
|
46
|
+
* hash of the request body. Built from a `RawProviderSink` containing
|
|
47
|
+
* paired `request` and `response` events from a previous run.
|
|
48
|
+
*
|
|
49
|
+
* The cache is the source of truth for replay; `createReplayFetch` is a
|
|
50
|
+
* thin wrapper that reads from it.
|
|
51
|
+
*/
|
|
52
|
+
export declare class ReplayCache {
|
|
53
|
+
private byKey;
|
|
54
|
+
private orphans;
|
|
55
|
+
private byProvider;
|
|
56
|
+
private byModel;
|
|
57
|
+
/**
|
|
58
|
+
* Build a cache from a sink's events. The sink must implement `list()`.
|
|
59
|
+
* Filter by `runId` / `spanId` to scope to a specific replay.
|
|
60
|
+
*/
|
|
61
|
+
static fromSink(sink: RawProviderSink, filter?: {
|
|
62
|
+
runId?: string;
|
|
63
|
+
spanId?: string;
|
|
64
|
+
}): Promise<ReplayCache>;
|
|
65
|
+
/** Build a cache from an in-memory event list. */
|
|
66
|
+
static fromEvents(events: RawProviderEvent[]): Promise<ReplayCache>;
|
|
67
|
+
/** Number of cacheable (request, response) pairs in the cache. */
|
|
68
|
+
size(): number;
|
|
69
|
+
stats(): ReplayCacheStats;
|
|
70
|
+
/** Iterate every cached `(request, response)` pair in insertion order. */
|
|
71
|
+
entries(): IterableIterator<ReplayCacheEntry>;
|
|
72
|
+
/**
|
|
73
|
+
* Look up a cached response by hashing the (model, messages, temperature,
|
|
74
|
+
* maxTokens, response_format) shape. Returns `undefined` on miss; the
|
|
75
|
+
* caller decides whether to throw, fall back to the network, or skip.
|
|
76
|
+
*/
|
|
77
|
+
lookup(requestBody: unknown): Promise<ReplayCacheEntry | undefined>;
|
|
78
|
+
}
|
|
79
|
+
export interface ReplayFetchOptions {
|
|
80
|
+
/**
|
|
81
|
+
* Behaviour on cache miss. Default `'throw'`. `'fallback'` calls the
|
|
82
|
+
* `fallbackFetch` (typically `globalThis.fetch`) so a partial replay can
|
|
83
|
+
* still complete; `'fail-closed'` returns a synthetic 599 response so the
|
|
84
|
+
* call site sees a non-retriable failure.
|
|
85
|
+
*/
|
|
86
|
+
onMiss?: 'throw' | 'fallback' | 'fail-closed';
|
|
87
|
+
fallbackFetch?: typeof fetch;
|
|
88
|
+
/** Optional callback fired once per replayed call (for telemetry / counters). */
|
|
89
|
+
onHit?: (info: {
|
|
90
|
+
url: string;
|
|
91
|
+
provider: string;
|
|
92
|
+
model: string;
|
|
93
|
+
}) => void;
|
|
94
|
+
/** Optional callback fired on cache miss before the `onMiss` policy applies. */
|
|
95
|
+
onMissNotify?: (info: {
|
|
96
|
+
url: string;
|
|
97
|
+
requestBody: unknown;
|
|
98
|
+
}) => void;
|
|
99
|
+
}
|
|
100
|
+
/**
|
|
101
|
+
* Build a `fetch`-shaped function that serves cached responses out of a
|
|
102
|
+
* `ReplayCache` for any URL ending in `/chat/completions`. Pass through
|
|
103
|
+
* `LlmClientOptions.fetch` and `callLlm` becomes free.
|
|
104
|
+
*
|
|
105
|
+
* Non-`/chat/completions` URLs are passed straight to the fallback fetch
|
|
106
|
+
* (default: `globalThis.fetch`). This matters because non-LLM HTTP work
|
|
107
|
+
* (judge HTTP servers, sandbox callbacks) sometimes flows through the same
|
|
108
|
+
* `fetch` and shouldn't be intercepted.
|
|
109
|
+
*/
|
|
110
|
+
export declare function createReplayFetch(cache: ReplayCache, opts?: ReplayFetchOptions): typeof fetch;
|
|
111
|
+
/**
|
|
112
|
+
* Convenience iterator over `(request, response)` pairs in a sink — for
|
|
113
|
+
* post-hoc scoring that doesn't need a `fetch` shim. The judge or scorer
|
|
114
|
+
* runs purely in-process over cached LLM outputs.
|
|
115
|
+
*/
|
|
116
|
+
export declare function iterateRawCalls(sink: RawProviderSink, filter?: {
|
|
117
|
+
runId?: string;
|
|
118
|
+
spanId?: string;
|
|
119
|
+
}): AsyncGenerator<ReplayCacheEntry>;
|
|
120
|
+
//# sourceMappingURL=replay.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"replay.d.ts","sourceRoot":"","sources":["../src/replay.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AAEH,OAAO,EAAE,WAAW,EAAE,MAAM,UAAU,CAAA;AAEtC,OAAO,KAAK,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,2BAA2B,CAAA;AAElF,qBAAa,oBAAqB,SAAQ,WAAW;aAEjC,GAAG,EAAE,MAAM;aACX,UAAU,EAAE,MAAM;gBADlB,GAAG,EAAE,MAAM,EACX,UAAU,EAAE,MAAM,EAClC,OAAO,CAAC,EAAE,MAAM;CAInB;AAED,MAAM,WAAW,gBAAgB;IAC/B,OAAO,EAAE,gBAAgB,CAAA;IACzB,QAAQ,EAAE,gBAAgB,CAAA;CAC3B;AAED,MAAM,WAAW,gBAAgB;IAC/B,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAClC,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC/B,gFAAgF;IAChF,cAAc,EAAE,MAAM,CAAA;CACvB;AAED;;;;;;;GAOG;AACH,qBAAa,WAAW;IACtB,OAAO,CAAC,KAAK,CAAsC;IACnD,OAAO,CAAC,OAAO,CAAI;IACnB,OAAO,CAAC,UAAU,CAA6B;IAC/C,OAAO,CAAC,OAAO,CAA6B;IAE5C;;;OAGG;WACU,QAAQ,CACnB,IAAI,EAAE,eAAe,EACrB,MAAM,GAAE;QAAE,KAAK,CAAC,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAO,GAC/C,OAAO,CAAC,WAAW,CAAC;IAQvB,kDAAkD;WACrC,UAAU,CAAC,MAAM,EAAE,gBAAgB,EAAE,GAAG,OAAO,CAAC,WAAW,CAAC;IA2BzE,kEAAkE;IAClE,IAAI,IAAI,MAAM;IAId,KAAK,IAAI,gBAAgB;IASzB,0EAA0E;IACzE,OAAO,IAAI,gBAAgB,CAAC,gBAAgB,CAAC;IAI9C;;;;OAIG;IACG,MAAM,CAAC,WAAW,EAAE,OAAO,GAAG,OAAO,CAAC,gBAAgB,GAAG,SAAS,CAAC;CAI1E;AAED,MAAM,WAAW,kBAAkB;IACjC;;;;;OAKG;IACH,MAAM,CAAC,EAAE,OAAO,GAAG,UAAU,GAAG,aAAa,CAAA;IAC7C,aAAa,CAAC,EAAE,OAAO,KAAK,CAAA;IAC5B,iFAAiF;IACjF,KAAK,CAAC,EAAE,CAAC,IAAI,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,CAAA;IACxE,gFAAgF;IAChF,YAAY,CAAC,EAAE,CAAC,IAAI,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,OAAO,CAAA;KAAE,KAAK,IAAI,CAAA;CACrE;AAED;;;;;;;;;GASG;AACH,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,WAAW,EAAE,IAAI,GAAE,kBAAuB,GAAG,OAAO,KAAK,CA+CjG;AAED;;;;GAIG;AACH,wBAAuB,eAAe,CACpC,IAAI,EAAE,eAAe,EACrB,MAAM,GAAE;IAAE,KAAK,CAAC,EAAE,MAAM,CAAC;IAAC,MAAM,CAAC,EAAE,MAAM,CAAA;CAAO,GAC/C,cAAc,CAAC,gBAAgB,CAAC,CAOlC"}
|
|
@@ -0,0 +1,14 @@
|
|
|
1
|
+
import type { BenchmarkReport, DriverResult } from './types';
|
|
2
|
+
/**
|
|
3
|
+
* Report generation utilities.
|
|
4
|
+
*
|
|
5
|
+
* Outputs convergence curves, cost curves, quality curves,
|
|
6
|
+
* and per-persona summaries in markdown format.
|
|
7
|
+
*/
|
|
8
|
+
/** Generate a markdown report from benchmark results */
|
|
9
|
+
export declare function formatBenchmarkReport(report: BenchmarkReport): string;
|
|
10
|
+
/** Generate a markdown report from agent driver results */
|
|
11
|
+
export declare function formatDriverReport(results: DriverResult[]): string;
|
|
12
|
+
/** Print a compact summary to console */
|
|
13
|
+
export declare function printDriverSummary(results: DriverResult[]): void;
|
|
14
|
+
//# sourceMappingURL=reporter.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"reporter.d.ts","sourceRoot":"","sources":["../src/reporter.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,eAAe,EAAE,YAAY,EAAE,MAAM,SAAS,CAAA;AAE5D;;;;;GAKG;AAEH,wDAAwD;AACxD,wBAAgB,qBAAqB,CAAC,MAAM,EAAE,eAAe,GAAG,MAAM,CAwDrE;AAED,2DAA2D;AAC3D,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,YAAY,EAAE,GAAG,MAAM,CA6ClE;AAED,yCAAyC;AACzC,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,YAAY,EAAE,GAAG,IAAI,CAgBhE"}
|
package/dist/reporting.d.ts
CHANGED
|
@@ -1,15 +1,15 @@
|
|
|
1
|
-
export {
|
|
2
|
-
export {
|
|
3
|
-
export {
|
|
4
|
-
export {
|
|
5
|
-
export {
|
|
6
|
-
|
|
7
|
-
|
|
8
|
-
|
|
9
|
-
|
|
10
|
-
|
|
11
|
-
|
|
12
|
-
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
|
|
1
|
+
export type { RubricOutcomePair, RubricPredictiveValidityInput, RubricPredictiveValidityReport, RubricRanking, } from './meta-eval/rubric-predictive-validity';
|
|
2
|
+
export { rubricPredictiveValidity } from './meta-eval/rubric-predictive-validity';
|
|
3
|
+
export type { BootstrapOptions, BootstrapResult, JudgeReplayGateArgs, Verdict, } from './promotion-gate';
|
|
4
|
+
export { bootstrapCi, judgeReplayGate, } from './promotion-gate';
|
|
5
|
+
export type { ReleaseConfidenceAxis, ReleaseConfidenceAxisName, ReleaseConfidenceInput, ReleaseConfidenceIssue, ReleaseConfidenceMetrics, ReleaseConfidenceScorecard, ReleaseConfidenceStatus, ReleaseConfidenceThresholds, ReleaseTraceEvidence, } from './release-confidence';
|
|
6
|
+
export { assertReleaseConfidence, evaluateReleaseConfidence } from './release-confidence';
|
|
7
|
+
export type { RenderReleaseReportOptions } from './release-report';
|
|
8
|
+
export { renderReleaseReport } from './release-report';
|
|
9
|
+
export type { InterimReleaseConfidence, InterimReleaseConfidenceInput, PairedEvalueOptions, PairedEvalueSequence, PairedEvalueStep, SequentialDecision, } from './sequential';
|
|
10
|
+
export { evaluateInterimReleaseConfidence, pairedEvalueSequence, } from './sequential';
|
|
11
|
+
export type { PairedBootstrapOptions, PairedBootstrapResult, } from './statistics';
|
|
12
|
+
export { benjaminiHochberg, pairedBootstrap, wilcoxonSignedRank, } from './statistics';
|
|
13
|
+
export type { GainDistributionBin, GainDistributionFigureSpec, GainDistributionOptions, ParetoFigureSpec, ParetoPoint, ResearchReport, ResearchReportCandidate, ResearchReportDecision, ResearchReportMethodology, ResearchReportOptions, ResearchReportRecommendation, SummaryTable, SummaryTableOptions, SummaryTableRow, } from './summary-report';
|
|
14
|
+
export { gainHistogram, paretoChart, RESEARCH_REPORT_HARD_PAIR_FLOOR, researchReport, summaryTable, } from './summary-report';
|
|
15
|
+
//# sourceMappingURL=reporting.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"reporting.d.ts","sourceRoot":"","sources":["../src/reporting.ts"],"names":[],"mappings":"AAAA,YAAY,EACV,iBAAiB,EACjB,6BAA6B,EAC7B,8BAA8B,EAC9B,aAAa,GACd,MAAM,wCAAwC,CAAA;AAC/C,OAAO,EAAE,wBAAwB,EAAE,MAAM,wCAAwC,CAAA;AACjF,YAAY,EACV,gBAAgB,EAChB,eAAe,EACf,mBAAmB,EACnB,OAAO,GACR,MAAM,kBAAkB,CAAA;AACzB,OAAO,EACL,WAAW,EACX,eAAe,GAChB,MAAM,kBAAkB,CAAA;AACzB,YAAY,EACV,qBAAqB,EACrB,yBAAyB,EACzB,sBAAsB,EACtB,sBAAsB,EACtB,wBAAwB,EACxB,0BAA0B,EAC1B,uBAAuB,EACvB,2BAA2B,EAC3B,oBAAoB,GACrB,MAAM,sBAAsB,CAAA;AAC7B,OAAO,EAAE,uBAAuB,EAAE,yBAAyB,EAAE,MAAM,sBAAsB,CAAA;AACzF,YAAY,EAAE,0BAA0B,EAAE,MAAM,kBAAkB,CAAA;AAClE,OAAO,EAAE,mBAAmB,EAAE,MAAM,kBAAkB,CAAA;AACtD,YAAY,EACV,wBAAwB,EACxB,6BAA6B,EAC7B,mBAAmB,EACnB,oBAAoB,EACpB,gBAAgB,EAChB,kBAAkB,GACnB,MAAM,cAAc,CAAA;AACrB,OAAO,EACL,gCAAgC,EAChC,oBAAoB,GACrB,MAAM,cAAc,CAAA;AACrB,YAAY,EACV,sBAAsB,EACtB,qBAAqB,GACtB,MAAM,cAAc,CAAA;AACrB,OAAO,EACL,iBAAiB,EACjB,eAAe,EACf,kBAAkB,GACnB,MAAM,cAAc,CAAA;AACrB,YAAY,EACV,mBAAmB,EACnB,0BAA0B,EAC1B,uBAAuB,EACvB,gBAAgB,EAChB,WAAW,EACX,cAAc,EACd,uBAAuB,EACvB,sBAAsB,EACtB,yBAAyB,EACzB,qBAAqB,EACrB,4BAA4B,EAC5B,YAAY,EACZ,mBAAmB,EACnB,eAAe,GAChB,MAAM,kBAAkB,CAAA;AACzB,OAAO,EACL,aAAa,EACb,WAAW,EACX,+BAA+B,EAC/B,cAAc,EACd,YAAY,GACb,MAAM,kBAAkB,CAAA"}
|