@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
|
@@ -1,68 +1,3 @@
|
|
|
1
|
-
import { S as Span } from '../schema-m0gsnbt3.js';
|
|
2
|
-
|
|
3
|
-
/**
|
|
4
|
-
* Code-edit extraction + the "watch the agent write code" animation.
|
|
5
|
-
*
|
|
6
|
-
* The base storyboard (./index) compresses a run into scenes but keeps scenes
|
|
7
|
-
* payload-free (title + narration + evidence span ids). To actually SHOW the
|
|
8
|
-
* code an agent wrote, this module pulls the concrete edit (path, diff,
|
|
9
|
-
* before/after, language) out of the raw edit-tool spans and renders a
|
|
10
|
-
* self-contained HTML replay that types each file out, character by character,
|
|
11
|
-
* like an editor recording.
|
|
12
|
-
*
|
|
13
|
-
* Additive on purpose: it reuses the canonical `Span` (trace/schema) and the
|
|
14
|
-
* `Storyboard` IR (./index) without modifying the reducer/compiler — so it
|
|
15
|
-
* composes with the existing pipeline and a Remotion consumer can read the same
|
|
16
|
-
* `CodeEdit[]` to render an MP4.
|
|
17
|
-
*/
|
|
18
|
-
|
|
19
|
-
/** A concrete code change lifted from a tool span. */
|
|
20
|
-
interface CodeEdit {
|
|
21
|
-
/** Span the edit came from — back-reference into the trace. */
|
|
22
|
-
spanId: string;
|
|
23
|
-
path: string;
|
|
24
|
-
language?: string;
|
|
25
|
-
/** Unified diff, when the tool carried one. */
|
|
26
|
-
diff?: string;
|
|
27
|
-
/** Full file body before / after, when present — enables true keystroke
|
|
28
|
-
* animation rather than only flashing a diff. */
|
|
29
|
-
before?: string;
|
|
30
|
-
after?: string;
|
|
31
|
-
additions: number;
|
|
32
|
-
deletions: number;
|
|
33
|
-
startedAt: number;
|
|
34
|
-
}
|
|
35
|
-
/** Extract a CodeEdit from a single span, or undefined if it is not an edit. */
|
|
36
|
-
declare function codeEditFromSpan(span: Span): CodeEdit | undefined;
|
|
37
|
-
/** All code edits in a run, in chronological order. */
|
|
38
|
-
declare function extractCodeEdits(spans: readonly Span[]): CodeEdit[];
|
|
39
|
-
/** Pair each code-edit (`diff`) scene with its concrete edit, matched via the
|
|
40
|
-
* scene's evidence span ids. Scenes without a recoverable edit are dropped. */
|
|
41
|
-
declare function codeEditsForStoryboard(storyboard: Storyboard, spans: readonly Span[]): Array<{
|
|
42
|
-
scene: Scene;
|
|
43
|
-
edit: CodeEdit;
|
|
44
|
-
}>;
|
|
45
|
-
/** The text the animation types for an edit: prefer the full new file body,
|
|
46
|
-
* then the added lines of a diff, then a minimal placeholder so the scene is
|
|
47
|
-
* never blank. */
|
|
48
|
-
declare function editAnimationText(edit: CodeEdit): string;
|
|
49
|
-
interface CodeAnimationOptions {
|
|
50
|
-
/** Page title. Default "Agent writing code". */
|
|
51
|
-
title?: string;
|
|
52
|
-
/** Characters typed per animation tick. Higher = faster. Default 4. */
|
|
53
|
-
charsPerTick?: number;
|
|
54
|
-
/** Pause between files, ms. Default 900. */
|
|
55
|
-
pauseBetweenMs?: number;
|
|
56
|
-
}
|
|
57
|
-
/**
|
|
58
|
-
* Render a self-contained HTML page that animates the agent writing each file,
|
|
59
|
-
* character by character, with a filename tab, line numbers, and a blinking
|
|
60
|
-
* cursor — the shareable "watch it build" clip. No build step, no assets, no
|
|
61
|
-
* deps. The same `CodeEdit[]` feeds a Remotion MP4 renderer in a consumer
|
|
62
|
-
* package; this is the dependency-free baseline.
|
|
63
|
-
*/
|
|
64
|
-
declare function renderCodeAnimationHtml(source: readonly Span[] | readonly CodeEdit[], opts?: CodeAnimationOptions): string;
|
|
65
|
-
|
|
66
1
|
/**
|
|
67
2
|
* Storyboard — compile a raw agent trace into a small set of meaningful scenes,
|
|
68
3
|
* then render them as a shareable, watchable replay.
|
|
@@ -82,12 +17,12 @@ declare function renderCodeAnimationHtml(source: readonly Span[] | readonly Code
|
|
|
82
17
|
* the structure + visuals do not depend on one. Renderers that need a browser
|
|
83
18
|
* (Remotion, React) live in consumers; this module only emits strings.
|
|
84
19
|
*/
|
|
85
|
-
|
|
20
|
+
import type { Span } from '../trace/schema';
|
|
86
21
|
/** What the agent meaningfully did — the compressed vocabulary. */
|
|
87
|
-
type SemanticKind = 'understood_task' | 'user_message' | 'agent_reply' | 'reasoned' | 'ran_command' | 'read_file' | 'edited_code' | 'searched' | 'used_browser' | 'used_computer' | 'called_api' | 'called_tool' | 'evaluated' | 'observed_failure' | 'completed';
|
|
22
|
+
export type SemanticKind = 'understood_task' | 'user_message' | 'agent_reply' | 'reasoned' | 'ran_command' | 'read_file' | 'edited_code' | 'searched' | 'used_browser' | 'used_computer' | 'called_api' | 'called_tool' | 'evaluated' | 'observed_failure' | 'completed';
|
|
88
23
|
/** The modality-typed payload a scene shows. This is what makes any agent
|
|
89
24
|
* action — screen, browser, shell, code, API — actually viewable. */
|
|
90
|
-
type SceneVisual = {
|
|
25
|
+
export type SceneVisual = {
|
|
91
26
|
type: 'screenshot';
|
|
92
27
|
src: string;
|
|
93
28
|
caption?: string;
|
|
@@ -121,7 +56,7 @@ type SceneVisual = {
|
|
|
121
56
|
} | {
|
|
122
57
|
type: 'none';
|
|
123
58
|
};
|
|
124
|
-
interface SemanticEvent {
|
|
59
|
+
export interface SemanticEvent {
|
|
125
60
|
kind: SemanticKind;
|
|
126
61
|
/** One-line headline (the ticket/scene title). */
|
|
127
62
|
title: string;
|
|
@@ -136,7 +71,7 @@ interface SemanticEvent {
|
|
|
136
71
|
startTs: number;
|
|
137
72
|
endTs: number;
|
|
138
73
|
}
|
|
139
|
-
interface ReduceOptions {
|
|
74
|
+
export interface ReduceOptions {
|
|
140
75
|
/** Collapse adjacent same-kind moments into one. Default true. */
|
|
141
76
|
collapseAdjacent?: boolean;
|
|
142
77
|
}
|
|
@@ -148,9 +83,9 @@ interface ReduceOptions {
|
|
|
148
83
|
* compression step), carrying the latest visual; failures and conversation
|
|
149
84
|
* turns never collapse.
|
|
150
85
|
*/
|
|
151
|
-
declare function reduceToSemanticEvents(spans: readonly Span[], opts?: ReduceOptions): SemanticEvent[];
|
|
152
|
-
type SceneType = 'title_card' | 'prompt' | 'reply' | 'reasoning' | 'terminal' | 'file' | 'diff' | 'search' | 'browser' | 'computer' | 'api' | 'tool' | 'eval' | 'error' | 'summary';
|
|
153
|
-
interface Scene {
|
|
86
|
+
export declare function reduceToSemanticEvents(spans: readonly Span[], opts?: ReduceOptions): SemanticEvent[];
|
|
87
|
+
export type SceneType = 'title_card' | 'prompt' | 'reply' | 'reasoning' | 'terminal' | 'file' | 'diff' | 'search' | 'browser' | 'computer' | 'api' | 'tool' | 'eval' | 'error' | 'summary';
|
|
88
|
+
export interface Scene {
|
|
154
89
|
sceneType: SceneType;
|
|
155
90
|
title: string;
|
|
156
91
|
narration: string;
|
|
@@ -159,12 +94,12 @@ interface Scene {
|
|
|
159
94
|
durationMs: number;
|
|
160
95
|
evidenceSpanIds: string[];
|
|
161
96
|
}
|
|
162
|
-
interface Storyboard {
|
|
97
|
+
export interface Storyboard {
|
|
163
98
|
title: string;
|
|
164
99
|
totalMs: number;
|
|
165
100
|
scenes: Scene[];
|
|
166
101
|
}
|
|
167
|
-
interface CompileOptions {
|
|
102
|
+
export interface CompileOptions {
|
|
168
103
|
/** Headline for the title card. Default "Agent run". */
|
|
169
104
|
title?: string;
|
|
170
105
|
/** Max action scenes between the title + summary cards. Default 16. */
|
|
@@ -175,10 +110,10 @@ interface CompileOptions {
|
|
|
175
110
|
* (importance 5) and edit (4) is always kept; the rest fill the budget by
|
|
176
111
|
* importance, then re-sort chronological. A title card opens, a summary closes.
|
|
177
112
|
*/
|
|
178
|
-
declare function compileStoryboard(events: readonly SemanticEvent[], opts?: CompileOptions): Storyboard;
|
|
113
|
+
export declare function compileStoryboard(events: readonly SemanticEvent[], opts?: CompileOptions): Storyboard;
|
|
179
114
|
/** Render the storyboard as a Markdown timeline — the human-readable shot list. */
|
|
180
|
-
declare function renderStoryboardMarkdown(storyboard: Storyboard): string;
|
|
181
|
-
interface HtmlRenderOptions {
|
|
115
|
+
export declare function renderStoryboardMarkdown(storyboard: Storyboard): string;
|
|
116
|
+
export interface HtmlRenderOptions {
|
|
182
117
|
/** Document title + on-page heading. Defaults to the storyboard title. */
|
|
183
118
|
title?: string;
|
|
184
119
|
}
|
|
@@ -189,6 +124,6 @@ interface HtmlRenderOptions {
|
|
|
189
124
|
* / browser / prose); controls let a viewer scrub. This is the "free clip" a
|
|
190
125
|
* run produces; an MP4 is the same storyboard fed to Remotion in a consumer.
|
|
191
126
|
*/
|
|
192
|
-
declare function renderStoryboardHtml(storyboard: Storyboard, opts?: HtmlRenderOptions): string;
|
|
193
|
-
|
|
194
|
-
|
|
127
|
+
export declare function renderStoryboardHtml(storyboard: Storyboard, opts?: HtmlRenderOptions): string;
|
|
128
|
+
export * from './code-edit';
|
|
129
|
+
//# sourceMappingURL=index.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/storyboard/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,iBAAiB,CAAA;AAE3C,mEAAmE;AACnE,MAAM,MAAM,YAAY,GACpB,iBAAiB,GACjB,cAAc,GACd,aAAa,GACb,UAAU,GACV,aAAa,GACb,WAAW,GACX,aAAa,GACb,UAAU,GACV,cAAc,GACd,eAAe,GACf,YAAY,GACZ,aAAa,GACb,WAAW,GACX,kBAAkB,GAClB,WAAW,CAAA;AAEf;sEACsE;AACtE,MAAM,MAAM,WAAW,GACnB;IAAE,IAAI,EAAE,YAAY,CAAC;IAAC,GAAG,EAAE,MAAM,CAAC;IAAC,OAAO,CAAC,EAAE,MAAM,CAAA;CAAE,GACrD;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,GAAG,CAAC,EAAE,MAAM,CAAC;IAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,UAAU,CAAC,EAAE,MAAM,CAAA;CAAE,GACvE;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,GAC7C;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GAC/C;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,MAAM,CAAC,EAAE,MAAM,CAAA;CAAE,GACtD;IACE,IAAI,EAAE,KAAK,CAAA;IACX,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,GAAG,EAAE,MAAM,CAAA;IACX,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,QAAQ,CAAC,EAAE,MAAM,CAAA;CAClB,GACD;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,IAAI,EAAE,MAAM,CAAA;CAAE,GAC/B;IAAE,IAAI,EAAE,MAAM,CAAA;CAAE,CAAA;AAEpB,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,YAAY,CAAA;IAClB,kDAAkD;IAClD,KAAK,EAAE,MAAM,CAAA;IACb,2BAA2B;IAC3B,OAAO,EAAE,MAAM,CAAA;IACf,oDAAoD;IACpD,MAAM,EAAE,WAAW,CAAA;IACnB,yDAAyD;IACzD,eAAe,EAAE,MAAM,EAAE,CAAA;IACzB,6EAA6E;IAC7E,UAAU,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAA;IAC7B,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,EAAE,MAAM,CAAA;CACd;AAED,MAAM,WAAW,aAAa;IAC5B,kEAAkE;IAClE,gBAAgB,CAAC,EAAE,OAAO,CAAA;CAC3B;AAwSD;;;;;;;GAOG;AACH,wBAAgB,sBAAsB,CACpC,KAAK,EAAE,SAAS,IAAI,EAAE,EACtB,IAAI,GAAE,aAAkB,GACvB,aAAa,EAAE,CA0CjB;AAID,MAAM,MAAM,SAAS,GACjB,YAAY,GACZ,QAAQ,GACR,OAAO,GACP,WAAW,GACX,UAAU,GACV,MAAM,GACN,MAAM,GACN,QAAQ,GACR,SAAS,GACT,UAAU,GACV,KAAK,GACL,MAAM,GACN,MAAM,GACN,OAAO,GACP,SAAS,CAAA;AAEb,MAAM,WAAW,KAAK;IACpB,SAAS,EAAE,SAAS,CAAA;IACpB,KAAK,EAAE,MAAM,CAAA;IACb,SAAS,EAAE,MAAM,CAAA;IACjB,+CAA+C;IAC/C,MAAM,EAAE,WAAW,CAAA;IACnB,UAAU,EAAE,MAAM,CAAA;IAClB,eAAe,EAAE,MAAM,EAAE,CAAA;CAC1B;AAED,MAAM,WAAW,UAAU;IACzB,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,EAAE,MAAM,CAAA;IACf,MAAM,EAAE,KAAK,EAAE,CAAA;CAChB;AAED,MAAM,WAAW,cAAc;IAC7B,wDAAwD;IACxD,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,uEAAuE;IACvE,SAAS,CAAC,EAAE,MAAM,CAAA;CACnB;AA4BD;;;;GAIG;AACH,wBAAgB,iBAAiB,CAC/B,MAAM,EAAE,SAAS,aAAa,EAAE,EAChC,IAAI,GAAE,cAAmB,GACxB,UAAU,CAoDZ;AAuGD,mFAAmF;AACnF,wBAAgB,wBAAwB,CAAC,UAAU,EAAE,UAAU,GAAG,MAAM,CAsBvE;AAED,MAAM,WAAW,iBAAiB;IAChC,0EAA0E;IAC1E,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED;;;;;;GAMG;AACH,wBAAgB,oBAAoB,CAAC,UAAU,EAAE,UAAU,EAAE,IAAI,GAAE,iBAAsB,GAAG,MAAM,CAsHjG;AAKD,cAAc,aAAa,CAAA"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index.test.d.ts","sourceRoot":"","sources":["../../src/storyboard/index.test.ts"],"names":[],"mappings":""}
|
|
@@ -1,141 +1,3 @@
|
|
|
1
|
-
import { R as RunRecord } from './run-record-e7vj1uZQ.js';
|
|
2
|
-
import { F as FailureClusterReport } from './failure-cluster-CL7IVgkJ.js';
|
|
3
|
-
|
|
4
|
-
/**
|
|
5
|
-
* HeldOutGate — first-class held-out paired-delta promotion gate.
|
|
6
|
-
*
|
|
7
|
-
* Encodes the "honesty override" pattern that lived inline in
|
|
8
|
-
* `~/webb/redteam/scripts/agent-eval-autoresearch.ts:138–171`.
|
|
9
|
-
* The optimizer's best-guess is one thing; what we should actually
|
|
10
|
-
* ship is another. The gate is the line between them.
|
|
11
|
-
*
|
|
12
|
-
* A candidate is promoted iff ALL three pass:
|
|
13
|
-
*
|
|
14
|
-
* 1. **Productive runs**: the candidate has at least
|
|
15
|
-
* `minProductiveRuns` paired observations on items where BOTH
|
|
16
|
-
* candidate and baseline produced a real (non-silent) score.
|
|
17
|
-
* 2. **Paired delta**: the lower bound of the bootstrap CI on the
|
|
18
|
-
* median per-item delta (candidate − baseline) on the HOLDOUT
|
|
19
|
-
* split is strictly greater than `pairedDeltaThreshold`.
|
|
20
|
-
* 3. **Overfit gap**: the candidate's gap between search-split
|
|
21
|
-
* score and holdout-split score is no worse (more positive)
|
|
22
|
-
* than the baseline's gap by more than `overfitGapThreshold`.
|
|
23
|
-
* "Better on search, worse on holdout" is the canonical
|
|
24
|
-
* overfit pattern; this catches it.
|
|
25
|
-
*
|
|
26
|
-
* The decision carries a machine-readable `rejectionCode` plus an
|
|
27
|
-
* `evidence` block with every number the gate looked at, so the
|
|
28
|
-
* downstream researcher / paper / dashboard can re-derive the
|
|
29
|
-
* verdict without re-running.
|
|
30
|
-
*
|
|
31
|
-
* See also:
|
|
32
|
-
* - `src/statistics.ts` for `pairedBootstrap` + `wilcoxonSignedRank`
|
|
33
|
-
* - `src/run-record.ts` for the input row schema
|
|
34
|
-
* - `src/reference-replay.ts` for the older, reference-replay-
|
|
35
|
-
* specific promotion path (still useful for replay-style evals).
|
|
36
|
-
*/
|
|
37
|
-
|
|
38
|
-
type HeldOutGateRejectionCode = 'few_runs' | 'negative_delta' | 'overfit_gap' | 'cost_ceiling';
|
|
39
|
-
interface HeldOutGateConfig {
|
|
40
|
-
/** Minimum number of paired (candidate, baseline) holdout observations
|
|
41
|
-
* required before the gate will even consider promoting. Default 3. */
|
|
42
|
-
minProductiveRuns?: number;
|
|
43
|
-
/** The bootstrap-CI lower bound on the median paired holdout delta
|
|
44
|
-
* must exceed this to promote. Default 0. */
|
|
45
|
-
pairedDeltaThreshold?: number;
|
|
46
|
-
/** Maximum allowed worsening of (search − holdout) gap relative to
|
|
47
|
-
* baseline. Default 0.15 (i.e. candidate may overfit by up to 15
|
|
48
|
-
* absolute score points more than baseline before rejection). */
|
|
49
|
-
overfitGapThreshold?: number;
|
|
50
|
-
/** Stable label of the baseline candidate. Required — paper-grade
|
|
51
|
-
* evaluation never compares two unlabelled candidates. */
|
|
52
|
-
baselineKey: string;
|
|
53
|
-
/** Confidence level for the bootstrap CI. Default 0.95. */
|
|
54
|
-
confidence?: number;
|
|
55
|
-
/** Bootstrap resamples. Default 2000. */
|
|
56
|
-
bootstrapResamples?: number;
|
|
57
|
-
/** Optional deterministic seed for the bootstrap. Default undefined
|
|
58
|
-
* (Math.random). */
|
|
59
|
-
seed?: number;
|
|
60
|
-
/**
|
|
61
|
-
* Hard ceiling on the candidate's median per-task USD cost. When the
|
|
62
|
-
* candidate clears the quality gates (paired-delta + overfit-gap) but
|
|
63
|
-
* its median cost exceeds this number, the gate rejects with
|
|
64
|
-
* `cost_ceiling`. Default `undefined` = no cost ceiling, behaving
|
|
65
|
-
* exactly like the pre-cost gate.
|
|
66
|
-
*
|
|
67
|
-
* This exists because "we ship the better prompt" is only an honest
|
|
68
|
-
* pitch when the better prompt also fits a customer-stated budget.
|
|
69
|
-
* Cost is read from `RunRecord.costUsd` (already mandatory on every
|
|
70
|
-
* run) so no new schema is required.
|
|
71
|
-
*/
|
|
72
|
-
costPerTaskCeiling?: number;
|
|
73
|
-
}
|
|
74
|
-
interface GateEvidence {
|
|
75
|
-
/** Number of paired (candidate, baseline) holdout observations used. */
|
|
76
|
-
productiveRuns: number;
|
|
77
|
-
/** Median of (candidate − baseline) paired holdout deltas. */
|
|
78
|
-
medianPairedDelta: number;
|
|
79
|
-
/** Bootstrap CI on the median paired holdout delta. */
|
|
80
|
-
pairedCI: {
|
|
81
|
-
low: number;
|
|
82
|
-
high: number;
|
|
83
|
-
};
|
|
84
|
-
/** Wilcoxon signed-rank p-value on the paired holdout deltas. */
|
|
85
|
-
pairedPValue: number;
|
|
86
|
-
/** Mean candidate score on the search split (NaN if none). */
|
|
87
|
-
searchScore: number;
|
|
88
|
-
/** Mean candidate score on the holdout split (NaN if none). */
|
|
89
|
-
holdoutScore: number;
|
|
90
|
-
/** Candidate (search − holdout) gap. */
|
|
91
|
-
overfitGap: number;
|
|
92
|
-
/** Baseline (search − holdout) gap. */
|
|
93
|
-
baselineOverfitGap: number;
|
|
94
|
-
/** Median per-task USD cost across the candidate's runs. Recorded
|
|
95
|
-
* even when no `costPerTaskCeiling` is configured so downstream
|
|
96
|
-
* dashboards (intelligence.tangle.tools) can render \$/task per
|
|
97
|
-
* generation regardless of gating policy. */
|
|
98
|
-
medianCandidateCost: number;
|
|
99
|
-
/** Median per-task USD cost across the baseline runs, for
|
|
100
|
-
* symmetric reporting. */
|
|
101
|
-
medianBaselineCost: number;
|
|
102
|
-
}
|
|
103
|
-
interface GateDecision {
|
|
104
|
-
/** Final promote/no-promote verdict. */
|
|
105
|
-
promote: boolean;
|
|
106
|
-
/** The candidate that was evaluated. */
|
|
107
|
-
candidateId: string;
|
|
108
|
-
/** The baseline it was compared against. */
|
|
109
|
-
baselineId: string;
|
|
110
|
-
/** Every number the gate looked at, for audit + paper export. */
|
|
111
|
-
evidence: GateEvidence;
|
|
112
|
-
/** Human-readable reason. */
|
|
113
|
-
reason: string;
|
|
114
|
-
/** Machine-readable rejection code, or null on promote. */
|
|
115
|
-
rejectionCode: HeldOutGateRejectionCode | null;
|
|
116
|
-
}
|
|
117
|
-
/**
|
|
118
|
-
* Held-out paired-delta promotion gate. Construct once with config,
|
|
119
|
-
* call `evaluate(candidateRuns, baselineRuns)` per (candidate,
|
|
120
|
-
* baseline) pair. Stateless across calls.
|
|
121
|
-
*/
|
|
122
|
-
declare class HeldOutGate {
|
|
123
|
-
private readonly minProductiveRuns;
|
|
124
|
-
private readonly pairedDeltaThreshold;
|
|
125
|
-
private readonly overfitGapThreshold;
|
|
126
|
-
private readonly baselineKey;
|
|
127
|
-
private readonly confidence;
|
|
128
|
-
private readonly resamples;
|
|
129
|
-
private readonly seed?;
|
|
130
|
-
private readonly costPerTaskCeiling?;
|
|
131
|
-
constructor(config: HeldOutGateConfig);
|
|
132
|
-
/** Decide whether `candidate` should replace `baseline`. Pairing
|
|
133
|
-
* is by (experimentId, seed) — identical experiment + seed pairs
|
|
134
|
-
* the candidate run with the matching baseline run. Pairs without
|
|
135
|
-
* a holdout score on both sides are dropped. */
|
|
136
|
-
evaluate(candidate: RunRecord[], baseline: RunRecord[]): GateDecision;
|
|
137
|
-
}
|
|
138
|
-
|
|
139
1
|
/**
|
|
140
2
|
* Reporting helpers — production summaries and paper-quality figures — sit alongside `reporter.ts` rather
|
|
141
3
|
* than replacing it.
|
|
@@ -160,8 +22,10 @@ declare class HeldOutGate {
|
|
|
160
22
|
* what you'd hand to vega-lite, plotly, matplotlib, or your own
|
|
161
23
|
* Canvas renderer to draw the actual figure.
|
|
162
24
|
*/
|
|
163
|
-
|
|
164
|
-
|
|
25
|
+
import type { GateDecision } from './held-out-gate';
|
|
26
|
+
import type { FailureClusterReport } from './pipelines/failure-cluster';
|
|
27
|
+
import type { RunRecord } from './run-record';
|
|
28
|
+
export interface SummaryTableOptions {
|
|
165
29
|
/** Comparator candidate id. Wilcoxon + Cohen's d are computed
|
|
166
30
|
* versus this candidate. Required for paired stats columns. */
|
|
167
31
|
comparator?: string;
|
|
@@ -172,7 +36,7 @@ interface SummaryTableOptions {
|
|
|
172
36
|
/** FDR for BH adjustment of the comparison p-values. Default 0.05. */
|
|
173
37
|
fdr?: number;
|
|
174
38
|
}
|
|
175
|
-
interface SummaryTableRow {
|
|
39
|
+
export interface SummaryTableRow {
|
|
176
40
|
candidateId: string;
|
|
177
41
|
n: number;
|
|
178
42
|
mean: number;
|
|
@@ -183,7 +47,7 @@ interface SummaryTableRow {
|
|
|
183
47
|
/** Cohen's d vs comparator. NaN if no comparator. */
|
|
184
48
|
cohensD: number;
|
|
185
49
|
}
|
|
186
|
-
interface SummaryTable {
|
|
50
|
+
export interface SummaryTable {
|
|
187
51
|
rows: SummaryTableRow[];
|
|
188
52
|
comparator: string | null;
|
|
189
53
|
split: 'search' | 'holdout';
|
|
@@ -195,8 +59,8 @@ interface SummaryTable {
|
|
|
195
59
|
* bootstrap CI on the chosen split, and (when a comparator is given)
|
|
196
60
|
* BH-adjusted Wilcoxon p + Cohen's d versus that comparator.
|
|
197
61
|
*/
|
|
198
|
-
declare function summaryTable(runs: RunRecord[], opts?: SummaryTableOptions): SummaryTable;
|
|
199
|
-
interface ParetoPoint {
|
|
62
|
+
export declare function summaryTable(runs: RunRecord[], opts?: SummaryTableOptions): SummaryTable;
|
|
63
|
+
export interface ParetoPoint {
|
|
200
64
|
candidateId: string;
|
|
201
65
|
/** Mean USD cost per run on the chosen split. */
|
|
202
66
|
cost: number;
|
|
@@ -211,7 +75,7 @@ interface ParetoPoint {
|
|
|
211
75
|
* for it was passed in. */
|
|
212
76
|
gate?: 'promote' | 'reject_few_runs' | 'reject_negative_delta' | 'reject_overfit_gap' | null;
|
|
213
77
|
}
|
|
214
|
-
interface ParetoFigureSpec {
|
|
78
|
+
export interface ParetoFigureSpec {
|
|
215
79
|
kind: 'pareto-cost-quality';
|
|
216
80
|
split: 'search' | 'holdout';
|
|
217
81
|
points: ParetoPoint[];
|
|
@@ -225,11 +89,11 @@ interface ParetoFigureSpec {
|
|
|
225
89
|
* candidate id; if present, every point picks up the gate verdict
|
|
226
90
|
* for overlay.
|
|
227
91
|
*/
|
|
228
|
-
declare function paretoChart(runs: RunRecord[], opts?: {
|
|
92
|
+
export declare function paretoChart(runs: RunRecord[], opts?: {
|
|
229
93
|
split?: 'search' | 'holdout';
|
|
230
94
|
gateDecisions?: Record<string, GateDecision>;
|
|
231
95
|
}): ParetoFigureSpec;
|
|
232
|
-
interface GainDistributionBin {
|
|
96
|
+
export interface GainDistributionBin {
|
|
233
97
|
/** Inclusive lower edge. */
|
|
234
98
|
lo: number;
|
|
235
99
|
/** Exclusive upper edge (or inclusive if it's the last bin). */
|
|
@@ -237,7 +101,7 @@ interface GainDistributionBin {
|
|
|
237
101
|
/** Number of pairs whose delta lands in this bin. */
|
|
238
102
|
count: number;
|
|
239
103
|
}
|
|
240
|
-
interface GainDistributionFigureSpec {
|
|
104
|
+
export interface GainDistributionFigureSpec {
|
|
241
105
|
kind: 'gain-distribution';
|
|
242
106
|
candidateId: string;
|
|
243
107
|
comparator: string;
|
|
@@ -251,7 +115,7 @@ interface GainDistributionFigureSpec {
|
|
|
251
115
|
high: number;
|
|
252
116
|
};
|
|
253
117
|
}
|
|
254
|
-
interface GainDistributionOptions {
|
|
118
|
+
export interface GainDistributionOptions {
|
|
255
119
|
/** Number of histogram bins. Default 11 (so the centre is exact at 0). */
|
|
256
120
|
bins?: number;
|
|
257
121
|
/** Which split to use. Default 'holdout'. */
|
|
@@ -268,15 +132,15 @@ interface GainDistributionOptions {
|
|
|
268
132
|
* comparator), histogrammed. Includes the bootstrap CI on the median
|
|
269
133
|
* delta — same primitive the promotion gate uses.
|
|
270
134
|
*/
|
|
271
|
-
declare function gainHistogram(runs: RunRecord[], candidateId: string, comparator: string, opts?: GainDistributionOptions): GainDistributionFigureSpec;
|
|
272
|
-
type ResearchReportDecision = 'promote' | 'hold' | 'reject' | 'equivalent' | 'needs_more_data';
|
|
135
|
+
export declare function gainHistogram(runs: RunRecord[], candidateId: string, comparator: string, opts?: GainDistributionOptions): GainDistributionFigureSpec;
|
|
136
|
+
export type ResearchReportDecision = 'promote' | 'hold' | 'reject' | 'equivalent' | 'needs_more_data';
|
|
273
137
|
/**
|
|
274
138
|
* Hard floor below which a paired comparison is treated as uninformative
|
|
275
139
|
* regardless of `minPairs`. Mirrors the lower limit on Wilcoxon signed-rank
|
|
276
140
|
* exact tables; below this the test has no power to separate effect sizes.
|
|
277
141
|
*/
|
|
278
|
-
declare const RESEARCH_REPORT_HARD_PAIR_FLOOR = 6;
|
|
279
|
-
interface ResearchReportOptions {
|
|
142
|
+
export declare const RESEARCH_REPORT_HARD_PAIR_FLOOR = 6;
|
|
143
|
+
export interface ResearchReportOptions {
|
|
280
144
|
/** Human-readable report title. */
|
|
281
145
|
title?: string;
|
|
282
146
|
/** Comparator candidate id. Required for statistical decision guidance. */
|
|
@@ -333,14 +197,14 @@ interface ResearchReportOptions {
|
|
|
333
197
|
*/
|
|
334
198
|
preregistrationHash?: string;
|
|
335
199
|
}
|
|
336
|
-
interface ResearchReportRecommendation {
|
|
200
|
+
export interface ResearchReportRecommendation {
|
|
337
201
|
decision: ResearchReportDecision;
|
|
338
202
|
candidateId: string | null;
|
|
339
203
|
rationale: string[];
|
|
340
204
|
risks: string[];
|
|
341
205
|
nextActions: string[];
|
|
342
206
|
}
|
|
343
|
-
interface ResearchReportCandidate {
|
|
207
|
+
export interface ResearchReportCandidate {
|
|
344
208
|
candidateId: string;
|
|
345
209
|
n: number;
|
|
346
210
|
mean: number;
|
|
@@ -376,7 +240,7 @@ interface ResearchReportCandidate {
|
|
|
376
240
|
decision: ResearchReportDecision;
|
|
377
241
|
decisionReason: string;
|
|
378
242
|
}
|
|
379
|
-
interface ResearchReportMethodology {
|
|
243
|
+
export interface ResearchReportMethodology {
|
|
380
244
|
/**
|
|
381
245
|
* Plain-language assumptions the report depends on. Read these first when
|
|
382
246
|
* deciding whether the verdict is load-bearing for a launch decision.
|
|
@@ -391,7 +255,7 @@ interface ResearchReportMethodology {
|
|
|
391
255
|
/** Citations for the methodological choices above. */
|
|
392
256
|
citations: string[];
|
|
393
257
|
}
|
|
394
|
-
interface ResearchReport {
|
|
258
|
+
export interface ResearchReport {
|
|
395
259
|
kind: 'agent-eval-research-report';
|
|
396
260
|
title: string;
|
|
397
261
|
generatedAt: string;
|
|
@@ -440,6 +304,5 @@ interface ResearchReport {
|
|
|
440
304
|
* Async because the fingerprint uses Web Crypto via `hashJson`; deterministic
|
|
441
305
|
* for any fixed `runs`, `seed`, and ROPE.
|
|
442
306
|
*/
|
|
443
|
-
declare function researchReport(runs: RunRecord[], opts?: ResearchReportOptions): Promise<ResearchReport>;
|
|
444
|
-
|
|
445
|
-
export { type GainDistributionBin as G, HeldOutGate as H, type ParetoFigureSpec as P, RESEARCH_REPORT_HARD_PAIR_FLOOR as R, type SummaryTable as S, type GainDistributionFigureSpec as a, type GainDistributionOptions as b, type ParetoPoint as c, type ResearchReport as d, type ResearchReportCandidate as e, type ResearchReportDecision as f, type ResearchReportMethodology as g, type ResearchReportOptions as h, type ResearchReportRecommendation as i, type SummaryTableOptions as j, type SummaryTableRow as k, gainHistogram as l, type GateDecision as m, type GateEvidence as n, type HeldOutGateConfig as o, paretoChart as p, type HeldOutGateRejectionCode as q, researchReport as r, summaryTable as s };
|
|
307
|
+
export declare function researchReport(runs: RunRecord[], opts?: ResearchReportOptions): Promise<ResearchReport>;
|
|
308
|
+
//# sourceMappingURL=summary-report.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"summary-report.d.ts","sourceRoot":"","sources":["../src/summary-report.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AAEH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAA;AACnD,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,6BAA6B,CAAA;AAEvE,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,cAAc,CAAA;AAY7C,MAAM,WAAW,mBAAmB;IAClC;oEACgE;IAChE,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,0DAA0D;IAC1D,KAAK,CAAC,EAAE,QAAQ,GAAG,SAAS,CAAA;IAC5B,uEAAuE;IACvE,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,sEAAsE;IACtE,GAAG,CAAC,EAAE,MAAM,CAAA;CACb;AAED,MAAM,WAAW,eAAe;IAC9B,WAAW,EAAE,MAAM,CAAA;IACnB,CAAC,EAAE,MAAM,CAAA;IACT,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,MAAM,CAAA;IACd,+DAA+D;IAC/D,MAAM,EAAE,MAAM,CAAA;IACd,qDAAqD;IACrD,OAAO,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,eAAe,EAAE,CAAA;IACvB,UAAU,EAAE,MAAM,GAAG,IAAI,CAAA;IACzB,KAAK,EAAE,QAAQ,GAAG,SAAS,CAAA;IAC3B,uDAAuD;IACvD,QAAQ,EAAE,MAAM,CAAA;CACjB;AAED;;;;GAIG;AACH,wBAAgB,YAAY,CAAC,IAAI,EAAE,SAAS,EAAE,EAAE,IAAI,GAAE,mBAAwB,GAAG,YAAY,CAuE5F;AAkDD,MAAM,WAAW,WAAW;IAC1B,WAAW,EAAE,MAAM,CAAA;IACnB,iDAAiD;IACjD,IAAI,EAAE,MAAM,CAAA;IACZ,sCAAsC;IACtC,OAAO,EAAE,MAAM,CAAA;IACf,+CAA+C;IAC/C,CAAC,EAAE,MAAM,CAAA;IACT;2CACuC;IACvC,UAAU,EAAE,OAAO,CAAA;IACnB;gCAC4B;IAC5B,IAAI,CAAC,EAAE,SAAS,GAAG,iBAAiB,GAAG,uBAAuB,GAAG,oBAAoB,GAAG,IAAI,CAAA;CAC7F;AAED,MAAM,WAAW,gBAAgB;IAC/B,IAAI,EAAE,qBAAqB,CAAA;IAC3B,KAAK,EAAE,QAAQ,GAAG,SAAS,CAAA;IAC3B,MAAM,EAAE,WAAW,EAAE,CAAA;IACrB,IAAI,EAAE;QAAE,CAAC,EAAE,SAAS,CAAC;QAAC,CAAC,EAAE,OAAO,CAAA;KAAE,CAAA;CACnC;AAED;;;;GAIG;AACH,wBAAgB,WAAW,CACzB,IAAI,EAAE,SAAS,EAAE,EACjB,IAAI,GAAE;IACJ,KAAK,CAAC,EAAE,QAAQ,GAAG,SAAS,CAAA;IAC5B,aAAa,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,CAAA;CACxC,GACL,gBAAgB,CA0ClB;AAgBD,MAAM,WAAW,mBAAmB;IAClC,4BAA4B;IAC5B,EAAE,EAAE,MAAM,CAAA;IACV,gEAAgE;IAChE,EAAE,EAAE,MAAM,CAAA;IACV,qDAAqD;IACrD,KAAK,EAAE,MAAM,CAAA;CACd;AAED,MAAM,WAAW,0BAA0B;IACzC,IAAI,EAAE,mBAAmB,CAAA;IACzB,WAAW,EAAE,MAAM,CAAA;IACnB,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,EAAE,QAAQ,GAAG,SAAS,CAAA;IAC3B,4BAA4B;IAC5B,CAAC,EAAE,MAAM,CAAA;IACT,IAAI,EAAE,mBAAmB,EAAE,CAAA;IAC3B,MAAM,EAAE,MAAM,CAAA;IACd,EAAE,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAA;CAClC;AAED,MAAM,WAAW,uBAAuB;IACtC,0EAA0E;IAC1E,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,6CAA6C;IAC7C,KAAK,CAAC,EAAE,QAAQ,GAAG,SAAS,CAAA;IAC5B,iDAAiD;IACjD,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,yCAAyC;IACzC,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,0BAA0B;IAC1B,IAAI,CAAC,EAAE,MAAM,CAAA;CACd;AAED;;;;GAIG;AACH,wBAAgB,aAAa,CAC3B,IAAI,EAAE,SAAS,EAAE,EACjB,WAAW,EAAE,MAAM,EACnB,UAAU,EAAE,MAAM,EAClB,IAAI,GAAE,uBAA4B,GACjC,0BAA0B,CAmE5B;AAID,MAAM,MAAM,sBAAsB,GAC9B,SAAS,GACT,MAAM,GACN,QAAQ,GACR,YAAY,GACZ,iBAAiB,CAAA;AAErB;;;;GAIG;AACH,eAAO,MAAM,+BAA+B,IAAI,CAAA;AAEhD,MAAM,WAAW,qBAAqB;IACpC,mCAAmC;IACnC,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,2EAA2E;IAC3E,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,sEAAsE;IACtE,KAAK,CAAC,EAAE,QAAQ,GAAG,SAAS,CAAA;IAC5B,yEAAyE;IACzE,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,gDAAgD;IAChD,GAAG,CAAC,EAAE,MAAM,CAAA;IACZ;;;;;;;OAOG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB;;;;;OAKG;IACH,IAAI,CAAC,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAA;IACpC;;;OAGG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB;;;OAGG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,8DAA8D;IAC9D,aAAa,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,CAAA;IAC5C,yDAAyD;IACzD,eAAe,CAAC,EAAE,oBAAoB,CAAA;IACtC,6FAA6F;IAC7F,YAAY,CAAC,EAAE,MAAM,EAAE,CAAA;IACvB,qFAAqF;IACrF,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,kDAAkD;IAClD,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB;;;;OAIG;IACH,mBAAmB,CAAC,EAAE,MAAM,CAAA;CAC7B;AAED,MAAM,WAAW,4BAA4B;IAC3C,QAAQ,EAAE,sBAAsB,CAAA;IAChC,WAAW,EAAE,MAAM,GAAG,IAAI,CAAA;IAC1B,SAAS,EAAE,MAAM,EAAE,CAAA;IACnB,KAAK,EAAE,MAAM,EAAE,CAAA;IACf,WAAW,EAAE,MAAM,EAAE,CAAA;CACtB;AAED,MAAM,WAAW,uBAAuB;IACtC,WAAW,EAAE,MAAM,CAAA;IACnB,CAAC,EAAE,MAAM,CAAA;IACT,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,MAAM,CAAA;IACd,MAAM,EAAE,MAAM,CAAA;IACd,OAAO,EAAE,MAAM,CAAA;IACf,qBAAqB,EAAE,MAAM,GAAG,IAAI,CAAA;IACpC,OAAO,EAAE,MAAM,CAAA;IACf,UAAU,EAAE,MAAM,GAAG,IAAI,CAAA;IACzB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;IACvB,MAAM,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAA;IAC5C;;;;;OAKG;IACH,iBAAiB,EAAE,MAAM,GAAG,IAAI,CAAA;IAChC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;IACvB;;;;;OAKG;IACH,GAAG,EAAE,MAAM,GAAG,IAAI,CAAA;IAClB,gBAAgB,EAAE,OAAO,CAAA;IACzB,IAAI,CAAC,EAAE,WAAW,CAAC,MAAM,CAAC,CAAA;IAC1B,QAAQ,EAAE,sBAAsB,CAAA;IAChC,cAAc,EAAE,MAAM,CAAA;CACvB;AAED,MAAM,WAAW,yBAAyB;IACxC;;;OAGG;IACH,WAAW,EAAE,MAAM,EAAE,CAAA;IACrB,0DAA0D;IAC1D,OAAO,EAAE,MAAM,EAAE,CAAA;IACjB,+EAA+E;IAC/E,YAAY,EAAE,MAAM,EAAE,CAAA;IACtB,oEAAoE;IACpE,cAAc,EAAE,MAAM,EAAE,CAAA;IACxB,sDAAsD;IACtD,SAAS,EAAE,MAAM,EAAE,CAAA;CACpB;AAED,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,4BAA4B,CAAA;IAClC,KAAK,EAAE,MAAM,CAAA;IACb,WAAW,EAAE,MAAM,CAAA;IACnB,KAAK,EAAE,QAAQ,GAAG,SAAS,CAAA;IAC3B,UAAU,EAAE,MAAM,GAAG,IAAI,CAAA;IACzB;;;;OAIG;IACH,cAAc,EAAE,MAAM,CAAA;IACtB,mBAAmB,EAAE,MAAM,GAAG,IAAI,CAAA;IAClC,IAAI,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAA;IAC1C,gBAAgB,EAAE,MAAM,EAAE,CAAA;IAC1B,cAAc,EAAE,4BAA4B,CAAA;IAC5C,UAAU,EAAE,uBAAuB,EAAE,CAAA;IACrC,OAAO,EAAE,YAAY,CAAA;IACrB,MAAM,EAAE;QACN,MAAM,EAAE,gBAAgB,CAAA;QACxB,KAAK,EAAE,0BAA0B,EAAE,CAAA;KACpC,CAAA;IACD,WAAW,EAAE,yBAAyB,CAAA;IACtC,eAAe,CAAC,EAAE,oBAAoB,CAAA;IACtC,QAAQ,EAAE,MAAM,CAAA;IAChB,IAAI,EAAE,MAAM,CAAA;CACb;AAoHD;;;;;;;;;;;;;;;;;;GAkBG;AACH,wBAAsB,cAAc,CAClC,IAAI,EAAE,SAAS,EAAE,EACjB,IAAI,GAAE,qBAA0B,GAC/B,OAAO,CAAC,cAAc,CAAC,CA2KzB"}
|
|
@@ -0,0 +1,35 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Telemetry client — thin wrapper that builds envelopes from `EmitArgs` and
|
|
3
|
+
* delegates to a `TelemetrySink`. Pure logic; no I/O. Use this from any
|
|
4
|
+
* runtime — Workers, Node, browser — and choose the sink accordingly.
|
|
5
|
+
*
|
|
6
|
+
* For an opinionated singleton with env-var-driven sink wiring (the bad CLI
|
|
7
|
+
* pattern), see `./node-client.ts`.
|
|
8
|
+
*/
|
|
9
|
+
import type { TelemetryKind, TelemetryModel, TelemetrySource } from './schema';
|
|
10
|
+
import type { TelemetrySink } from './sink-fetch';
|
|
11
|
+
export interface EmitArgs {
|
|
12
|
+
kind: TelemetryKind;
|
|
13
|
+
runId: string;
|
|
14
|
+
parentRunId?: string;
|
|
15
|
+
ok: boolean;
|
|
16
|
+
durationMs: number;
|
|
17
|
+
data?: Record<string, unknown>;
|
|
18
|
+
metrics?: Record<string, number>;
|
|
19
|
+
tags?: Record<string, string>;
|
|
20
|
+
model?: TelemetryModel;
|
|
21
|
+
error?: string;
|
|
22
|
+
/** Override the source for this envelope. Falls back to `defaultSource`. */
|
|
23
|
+
source?: TelemetrySource;
|
|
24
|
+
}
|
|
25
|
+
export declare class TelemetryClient {
|
|
26
|
+
private readonly sink;
|
|
27
|
+
private readonly defaultSource;
|
|
28
|
+
constructor(sink: TelemetrySink, defaultSource: TelemetrySource);
|
|
29
|
+
emit(args: EmitArgs): void;
|
|
30
|
+
close(): Promise<void>;
|
|
31
|
+
}
|
|
32
|
+
export declare const SECRET_FLAGS: Set<string>;
|
|
33
|
+
/** Strip likely-secret values from argv, preserving structure. */
|
|
34
|
+
export declare function sanitiseArgv(argv: string[]): string[];
|
|
35
|
+
//# sourceMappingURL=client.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"client.d.ts","sourceRoot":"","sources":["../../src/telemetry/client.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,KAAK,EAAqB,aAAa,EAAE,cAAc,EAAE,eAAe,EAAE,MAAM,UAAU,CAAA;AAEjG,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,cAAc,CAAA;AAEjD,MAAM,WAAW,QAAQ;IACvB,IAAI,EAAE,aAAa,CAAA;IACnB,KAAK,EAAE,MAAM,CAAA;IACb,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,EAAE,EAAE,OAAO,CAAA;IACX,UAAU,EAAE,MAAM,CAAA;IAClB,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAC9B,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAChC,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC7B,KAAK,CAAC,EAAE,cAAc,CAAA;IACtB,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,4EAA4E;IAC5E,MAAM,CAAC,EAAE,eAAe,CAAA;CACzB;AAED,qBAAa,eAAe;IAExB,OAAO,CAAC,QAAQ,CAAC,IAAI;IACrB,OAAO,CAAC,QAAQ,CAAC,aAAa;gBADb,IAAI,EAAE,aAAa,EACnB,aAAa,EAAE,eAAe;IAGjD,IAAI,CAAC,IAAI,EAAE,QAAQ,GAAG,IAAI;IAwBpB,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC;CAG7B;AAWD,eAAO,MAAM,YAAY,aAA8D,CAAA;AAEvF,kEAAkE;AAClE,wBAAgB,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,MAAM,EAAE,CAgBrD"}
|
|
@@ -1,38 +1,20 @@
|
|
|
1
|
-
import { T as TelemetryKind, a as TelemetryModel, b as TelemetrySource, c as TelemetrySink } from '../sink-fetch-B1Yg4Til.js';
|
|
2
|
-
export { F as FanoutTelemetrySink, H as HttpTelemetrySink, I as InMemoryTelemetrySink, N as NullTelemetrySink, d as TELEMETRY_SCHEMA_VERSION, e as TelemetryEnvelope } from '../sink-fetch-B1Yg4Til.js';
|
|
3
|
-
|
|
4
1
|
/**
|
|
5
|
-
*
|
|
6
|
-
* delegates to a `TelemetrySink`. Pure logic; no I/O. Use this from any
|
|
7
|
-
* runtime — Workers, Node, browser — and choose the sink accordingly.
|
|
2
|
+
* Public entry for the telemetry sub-module.
|
|
8
3
|
*
|
|
9
|
-
*
|
|
10
|
-
*
|
|
4
|
+
* Workers-safe by default — only `fetch` + pure JS. The Node file sink is
|
|
5
|
+
* exported separately via './telemetry/file' so consumers that import this entry
|
|
6
|
+
* cannot accidentally pull `node:fs` into a Worker bundle.
|
|
7
|
+
*
|
|
8
|
+
* Consume:
|
|
9
|
+
* import { TelemetryClient, HttpTelemetrySink, FanoutTelemetrySink }
|
|
10
|
+
* from '@tangle-network/agent-eval/telemetry'
|
|
11
|
+
*
|
|
12
|
+
* For Node:
|
|
13
|
+
* import { FileTelemetrySink, defaultTelemetryDir }
|
|
14
|
+
* from '@tangle-network/agent-eval/telemetry/file'
|
|
11
15
|
*/
|
|
12
|
-
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
|
|
16
|
-
|
|
17
|
-
ok: boolean;
|
|
18
|
-
durationMs: number;
|
|
19
|
-
data?: Record<string, unknown>;
|
|
20
|
-
metrics?: Record<string, number>;
|
|
21
|
-
tags?: Record<string, string>;
|
|
22
|
-
model?: TelemetryModel;
|
|
23
|
-
error?: string;
|
|
24
|
-
/** Override the source for this envelope. Falls back to `defaultSource`. */
|
|
25
|
-
source?: TelemetrySource;
|
|
26
|
-
}
|
|
27
|
-
declare class TelemetryClient {
|
|
28
|
-
private readonly sink;
|
|
29
|
-
private readonly defaultSource;
|
|
30
|
-
constructor(sink: TelemetrySink, defaultSource: TelemetrySource);
|
|
31
|
-
emit(args: EmitArgs): void;
|
|
32
|
-
close(): Promise<void>;
|
|
33
|
-
}
|
|
34
|
-
declare const SECRET_FLAGS: Set<string>;
|
|
35
|
-
/** Strip likely-secret values from argv, preserving structure. */
|
|
36
|
-
declare function sanitiseArgv(argv: string[]): string[];
|
|
37
|
-
|
|
38
|
-
export { type EmitArgs, SECRET_FLAGS, TelemetryClient, TelemetryKind, TelemetryModel, TelemetrySink, TelemetrySource, sanitiseArgv };
|
|
16
|
+
export { type EmitArgs, SECRET_FLAGS, sanitiseArgv, TelemetryClient, } from './client';
|
|
17
|
+
export type { TelemetryEnvelope, TelemetryKind, TelemetryModel, TelemetrySource, } from './schema';
|
|
18
|
+
export { TELEMETRY_SCHEMA_VERSION } from './schema';
|
|
19
|
+
export { FanoutTelemetrySink, HttpTelemetrySink, InMemoryTelemetrySink, NullTelemetrySink, type TelemetrySink, } from './sink-fetch';
|
|
20
|
+
//# sourceMappingURL=index.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/telemetry/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,EACL,KAAK,QAAQ,EACb,YAAY,EACZ,YAAY,EACZ,eAAe,GAChB,MAAM,UAAU,CAAA;AACjB,YAAY,EACV,iBAAiB,EACjB,aAAa,EACb,cAAc,EACd,eAAe,GAChB,MAAM,UAAU,CAAA;AACjB,OAAO,EAAE,wBAAwB,EAAE,MAAM,UAAU,CAAA;AACnD,OAAO,EACL,mBAAmB,EACnB,iBAAiB,EACjB,qBAAqB,EACrB,iBAAiB,EACjB,KAAK,aAAa,GACnB,MAAM,cAAc,CAAA"}
|
|
@@ -0,0 +1,61 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Fleet telemetry envelope — agent-eval's portable observability shape.
|
|
3
|
+
*
|
|
4
|
+
* Designed so any consumer (Node CLI, Cloudflare Worker, Lambda, browser
|
|
5
|
+
* extension) can emit structured rows describing one unit of work — a page
|
|
6
|
+
* audit, a tool call, an evolve round, a full agent run — to a central sink.
|
|
7
|
+
*
|
|
8
|
+
* The schema is intentionally a strict superset of agent-eval's `Run` shape
|
|
9
|
+
* so a future TraceStore adapter can promote envelopes into traces without
|
|
10
|
+
* translation.
|
|
11
|
+
*/
|
|
12
|
+
export declare const TELEMETRY_SCHEMA_VERSION = 1;
|
|
13
|
+
/** Discriminator for the unit of work this envelope describes. */
|
|
14
|
+
export type TelemetryKind = 'agent-run' | 'design-audit-page' | 'design-audit-run' | 'design-evolve-round' | 'design-evolve-run' | 'gepa-trial' | 'gepa-generation' | 'tool-call' | 'judge-verdict' | 'custom';
|
|
15
|
+
export interface TelemetryEnvelope {
|
|
16
|
+
schemaVersion: typeof TELEMETRY_SCHEMA_VERSION;
|
|
17
|
+
envelopeId: string;
|
|
18
|
+
runId: string;
|
|
19
|
+
timestamp: string;
|
|
20
|
+
parentRunId?: string;
|
|
21
|
+
source: TelemetrySource;
|
|
22
|
+
model?: TelemetryModel;
|
|
23
|
+
kind: TelemetryKind;
|
|
24
|
+
ok: boolean;
|
|
25
|
+
durationMs: number;
|
|
26
|
+
data: Record<string, unknown>;
|
|
27
|
+
metrics: Record<string, number>;
|
|
28
|
+
tags?: Record<string, string>;
|
|
29
|
+
error?: string;
|
|
30
|
+
}
|
|
31
|
+
export interface TelemetrySource {
|
|
32
|
+
/** Repo identity — basename of cwd plus git remote if discoverable. */
|
|
33
|
+
repo: string;
|
|
34
|
+
cwd: string;
|
|
35
|
+
gitSha?: string;
|
|
36
|
+
gitBranch?: string;
|
|
37
|
+
cliVersion: string;
|
|
38
|
+
/** What was invoked, e.g. `design-audit`, `bad run`, `gepa --target`. */
|
|
39
|
+
invocation: string;
|
|
40
|
+
/** Sanitised argv minus secrets. */
|
|
41
|
+
argv?: string[];
|
|
42
|
+
/**
|
|
43
|
+
* Multi-tenant identity. Set when the consumer runs inside a hosted
|
|
44
|
+
* product so a fleet rollup can group by tenant without leaking customer
|
|
45
|
+
* URLs or PII.
|
|
46
|
+
*/
|
|
47
|
+
tenantId?: string;
|
|
48
|
+
/** Optional sub-tenant identity (project, suite, walkthrough, customer). */
|
|
49
|
+
customerId?: string;
|
|
50
|
+
/** SHA-256 (12 hex) of the API key used to authenticate this run, when applicable. */
|
|
51
|
+
apiKeyHash?: string;
|
|
52
|
+
}
|
|
53
|
+
export interface TelemetryModel {
|
|
54
|
+
provider: string;
|
|
55
|
+
name: string;
|
|
56
|
+
/** SHA-256 (12 hex chars) of the prompt(s) used. */
|
|
57
|
+
promptHash?: string;
|
|
58
|
+
/** SHA-256 (12 hex chars) of the composed rubric body, if applicable. */
|
|
59
|
+
rubricHash?: string;
|
|
60
|
+
}
|
|
61
|
+
//# sourceMappingURL=schema.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"schema.d.ts","sourceRoot":"","sources":["../../src/telemetry/schema.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,eAAO,MAAM,wBAAwB,IAAI,CAAA;AAEzC,kEAAkE;AAClE,MAAM,MAAM,aAAa,GACrB,WAAW,GACX,mBAAmB,GACnB,kBAAkB,GAClB,qBAAqB,GACrB,mBAAmB,GACnB,YAAY,GACZ,iBAAiB,GACjB,WAAW,GACX,eAAe,GACf,QAAQ,CAAA;AAEZ,MAAM,WAAW,iBAAiB;IAChC,aAAa,EAAE,OAAO,wBAAwB,CAAA;IAC9C,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,EAAE,MAAM,CAAA;IACb,SAAS,EAAE,MAAM,CAAA;IACjB,WAAW,CAAC,EAAE,MAAM,CAAA;IAEpB,MAAM,EAAE,eAAe,CAAA;IACvB,KAAK,CAAC,EAAE,cAAc,CAAA;IACtB,IAAI,EAAE,aAAa,CAAA;IACnB,EAAE,EAAE,OAAO,CAAA;IACX,UAAU,EAAE,MAAM,CAAA;IAElB,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAC7B,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC/B,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAE7B,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,eAAe;IAC9B,uEAAuE;IACvE,IAAI,EAAE,MAAM,CAAA;IACZ,GAAG,EAAE,MAAM,CAAA;IACX,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,UAAU,EAAE,MAAM,CAAA;IAClB,yEAAyE;IACzE,UAAU,EAAE,MAAM,CAAA;IAClB,oCAAoC;IACpC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAA;IACf;;;;OAIG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,4EAA4E;IAC5E,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,sFAAsF;IACtF,UAAU,CAAC,EAAE,MAAM,CAAA;CACpB;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,EAAE,MAAM,CAAA;IAChB,IAAI,EAAE,MAAM,CAAA;IACZ,oDAAoD;IACpD,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,yEAAyE;IACzE,UAAU,CAAC,EAAE,MAAM,CAAA;CACpB"}
|