@tangle-network/agent-eval 0.94.0 → 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 +5 -65
- package/dist/analyst/index.js.map +1 -1
- package/dist/{kind-factory-0BhLSI27.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-6dhErQbs.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-Ce17tDlG.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-Bbid3WoO.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/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/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 +1863 -1316
- 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-2K6UUZ7P.js → chunk-2T4EZACH.js} +1 -1
- package/dist/chunk-2T4EZACH.js.map +1 -0
- package/dist/{chunk-CTBHKLEU.js → chunk-77T4STFI.js} +59 -86
- package/dist/chunk-77T4STFI.js.map +1 -0
- package/dist/{chunk-EGPMSBEZ.js → chunk-7QTQKIDD.js} +178 -177
- package/dist/chunk-7QTQKIDD.js.map +1 -0
- package/dist/{chunk-MIFZUPEK.js → chunk-AQ5WQAIV.js} +21 -6
- package/dist/chunk-AQ5WQAIV.js.map +1 -0
- 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-TBDR6PAI.js → chunk-IZCEK2HR.js} +2 -2
- package/dist/{chunk-SD2YFWQQ.js → chunk-KKWJD5E6.js} +20 -20
- 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-E4GH6USR.js → chunk-NZEQVRH5.js} +2 -2
- package/dist/chunk-NZEQVRH5.js.map +1 -0
- package/dist/{chunk-MPQWFX6Y.js → chunk-PSWWQXHF.js} +13 -88
- package/dist/chunk-PSWWQXHF.js.map +1 -0
- package/dist/{chunk-Q5LIB7BC.js → chunk-S4SYLDFX.js} +2 -2
- package/dist/chunk-S4SYLDFX.js.map +1 -0
- package/dist/{chunk-KW53MSA5.js → chunk-X74V6ESX.js} +2 -2
- package/dist/{chunk-QMUEXQJS.js → chunk-YBIGNSCZ.js} +81 -4
- package/dist/chunk-YBIGNSCZ.js.map +1 -0
- package/dist/{chunk-2KNZHH3P.js → chunk-Z6L6YSU6.js} +2 -2
- 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/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-B6Ljo_dI.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 +127 -17
- package/dist/contract/index.js.map +1 -1
- package/dist/{insight-report-DWl3z9tl.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-Acf9CGhw.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-DlOz8PBx.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 +1 -1
- 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-B0C2_fVO.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-BxY0cKfs.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 -5665
- package/dist/index.d.ts.map +1 -0
- package/dist/index.js +66 -31
- 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-Bj7g0rqu.d.ts → llm-client.d.ts} +21 -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/{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/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 -104
- package/dist/prm/index.d.ts.map +1 -0
- 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-Cc6UHvUb.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-BWdoyleI.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/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 +612 -612
- package/dist/rl.js.map +1 -1
- package/dist/{run-campaign-7WNXMDSN.js → run-campaign-WXY7KI67.js} +2 -2
- 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-DeODGLra.d.ts → sandbox-harness.d.ts} +15 -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-CCJpTGOS.d.ts → statistics.d.ts} +46 -48
- 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-BDOFevaT.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-C2rqGH_l.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-D2t12mMw.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-BcFXE6LG.d.ts → trace/store.d.ts} +21 -12
- 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 -960
- package/dist/traces.d.ts.map +1 -0
- package/dist/traces.js +41 -11
- package/dist/{trajectory-2TkpSEVh.d.ts → trajectory.d.ts} +8 -9
- package/dist/trajectory.d.ts.map +1 -0
- package/dist/{types-C7DGg5ex.d.ts → types.d.ts} +31 -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/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 +1 -1
- 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-BPmzuVPk.d.ts +0 -101
- package/dist/chunk-2K6UUZ7P.js.map +0 -1
- package/dist/chunk-CTBHKLEU.js.map +0 -1
- package/dist/chunk-E4GH6USR.js.map +0 -1
- package/dist/chunk-EGPMSBEZ.js.map +0 -1
- package/dist/chunk-KWRRMR3J.js.map +0 -1
- package/dist/chunk-MIFZUPEK.js.map +0 -1
- package/dist/chunk-MPQWFX6Y.js.map +0 -1
- package/dist/chunk-Q5LIB7BC.js.map +0 -1
- package/dist/chunk-QMUEXQJS.js.map +0 -1
- package/dist/chunk-SD2YFWQQ.js.map +0 -1
- package/dist/control-D6qwHXIR.d.ts +0 -259
- package/dist/corpus-B8A4BDR3.d.ts +0 -560
- package/dist/failure-cluster-DH9Flgcf.d.ts +0 -76
- package/dist/harness-optimizer-mOl9XX_O.d.ts +0 -106
- package/dist/index-Bx3gZ8xl.d.ts +0 -159
- package/dist/pre-registration-mAnCugl9.d.ts +0 -523
- package/dist/provenance-P-bCL2Fo.d.ts +0 -441
- package/dist/query-B7GGjRox.d.ts +0 -32
- package/dist/release-report-BEbWmVYj.d.ts +0 -233
- package/dist/run-critic-CmMf05uV.d.ts +0 -56
- package/dist/run-improvement-loop-DBahB8Ax.d.ts +0 -427
- package/dist/runtime-trajectory-BDgfGZSr.d.ts +0 -49
- package/dist/semantic-concept-judge-B9MgmBnM.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-TBDR6PAI.js.map → chunk-IZCEK2HR.js.map} +0 -0
- /package/dist/{chunk-KW53MSA5.js.map → chunk-X74V6ESX.js.map} +0 -0
- /package/dist/{chunk-2KNZHH3P.js.map → chunk-Z6L6YSU6.js.map} +0 -0
- /package/dist/{run-campaign-7WNXMDSN.js.map → run-campaign-WXY7KI67.js.map} +0 -0
|
@@ -1,7 +1,3 @@
|
|
|
1
|
-
import { T as TraceEmitter } from './emitter-C2rqGH_l.js';
|
|
2
|
-
import { F as FailureClass } from './schema-m0gsnbt3.js';
|
|
3
|
-
import { T as TraceStore } from './store-BcFXE6LG.js';
|
|
4
|
-
|
|
5
1
|
/**
|
|
6
2
|
* Policy-based agent control runtime.
|
|
7
3
|
*
|
|
@@ -13,10 +9,12 @@ import { T as TraceStore } from './store-BcFXE6LG.js';
|
|
|
13
9
|
* critic/revise, driver intervention, specialist calls, and human escalation
|
|
14
10
|
* are all just actions chosen by the control policy.
|
|
15
11
|
*/
|
|
16
|
-
|
|
17
|
-
type
|
|
18
|
-
type
|
|
19
|
-
|
|
12
|
+
import { TraceEmitter } from './trace/emitter';
|
|
13
|
+
import type { FailureClass } from './trace/schema';
|
|
14
|
+
import type { TraceStore } from './trace/store';
|
|
15
|
+
export type ControlSeverity = 'info' | 'warning' | 'error' | 'critical';
|
|
16
|
+
export type ControlActionFailureMode = 'continue' | 'stop';
|
|
17
|
+
export interface ControlEvalResult {
|
|
20
18
|
/** Stable validator or judge id. */
|
|
21
19
|
id: string;
|
|
22
20
|
/** Whether this check passed. */
|
|
@@ -34,12 +32,12 @@ interface ControlEvalResult {
|
|
|
34
32
|
/** Structured details for downstream control policies and reports. */
|
|
35
33
|
metadata?: Record<string, unknown>;
|
|
36
34
|
}
|
|
37
|
-
interface ControlBudget {
|
|
35
|
+
export interface ControlBudget {
|
|
38
36
|
maxSteps: number;
|
|
39
37
|
maxWallMs?: number;
|
|
40
38
|
maxCostUsd?: number;
|
|
41
39
|
}
|
|
42
|
-
interface ControlStopPolicies<TState, TAction> {
|
|
40
|
+
export interface ControlStopPolicies<TState, TAction> {
|
|
43
41
|
/**
|
|
44
42
|
* Stop after N consecutive steps with no state fingerprint change and
|
|
45
43
|
* less than `minScoreDelta` score movement. Disabled when omitted.
|
|
@@ -57,7 +55,7 @@ interface ControlStopPolicies<TState, TAction> {
|
|
|
57
55
|
/** Override the default JSON/string fingerprint for repeated-action checks. */
|
|
58
56
|
actionFingerprint?: (action: TAction) => string;
|
|
59
57
|
}
|
|
60
|
-
interface ControlContext<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult> {
|
|
58
|
+
export interface ControlContext<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult> {
|
|
61
59
|
intent: string;
|
|
62
60
|
state: TState;
|
|
63
61
|
evals: TEval[];
|
|
@@ -70,7 +68,7 @@ interface ControlContext<TState, TAction, TActionResult, TEval extends ControlEv
|
|
|
70
68
|
abortSignal: AbortSignal;
|
|
71
69
|
emitter?: TraceEmitter;
|
|
72
70
|
}
|
|
73
|
-
type ControlDecision<TAction> = {
|
|
71
|
+
export type ControlDecision<TAction> = {
|
|
74
72
|
type: 'continue';
|
|
75
73
|
action: TAction;
|
|
76
74
|
reason?: string;
|
|
@@ -80,26 +78,26 @@ type ControlDecision<TAction> = {
|
|
|
80
78
|
pass?: boolean;
|
|
81
79
|
score?: number;
|
|
82
80
|
};
|
|
83
|
-
interface StopDecision {
|
|
81
|
+
export interface StopDecision {
|
|
84
82
|
stop: boolean;
|
|
85
83
|
pass: boolean;
|
|
86
84
|
reason: string;
|
|
87
85
|
score?: number;
|
|
88
86
|
failureClass?: FailureClass;
|
|
89
87
|
}
|
|
90
|
-
interface ControlActionOutcome<TActionResult> {
|
|
88
|
+
export interface ControlActionOutcome<TActionResult> {
|
|
91
89
|
ok: boolean;
|
|
92
90
|
result?: TActionResult;
|
|
93
91
|
error?: string;
|
|
94
92
|
costUsd?: number;
|
|
95
93
|
durationMs: number;
|
|
96
94
|
}
|
|
97
|
-
interface ControlRuntimeError {
|
|
95
|
+
export interface ControlRuntimeError {
|
|
98
96
|
phase: 'observe' | 'validate' | 'decide' | 'act' | 'stop-policy' | 'on-step' | 'trace';
|
|
99
97
|
stepIndex: number;
|
|
100
98
|
message: string;
|
|
101
99
|
}
|
|
102
|
-
interface ControlStep<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult> {
|
|
100
|
+
export interface ControlStep<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult> {
|
|
103
101
|
index: number;
|
|
104
102
|
decision: ControlDecision<TAction>;
|
|
105
103
|
beforeState: TState;
|
|
@@ -110,7 +108,7 @@ interface ControlStep<TState, TAction, TActionResult, TEval extends ControlEvalR
|
|
|
110
108
|
startedAt: string;
|
|
111
109
|
endedAt: string;
|
|
112
110
|
}
|
|
113
|
-
interface ControlRunResult<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult> {
|
|
111
|
+
export interface ControlRunResult<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult> {
|
|
114
112
|
intent: string;
|
|
115
113
|
pass: boolean;
|
|
116
114
|
completed: boolean;
|
|
@@ -127,7 +125,7 @@ interface ControlRunResult<TState, TAction, TActionResult, TEval extends Control
|
|
|
127
125
|
runtimeErrors: ControlRuntimeError[];
|
|
128
126
|
stoppedBy: 'policy' | 'stop-policy' | 'budget' | 'abort' | 'runtime-error';
|
|
129
127
|
}
|
|
130
|
-
interface ControlRuntimeConfig<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult> {
|
|
128
|
+
export interface ControlRuntimeConfig<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult> {
|
|
131
129
|
intent: string;
|
|
132
130
|
budget?: Partial<ControlBudget>;
|
|
133
131
|
signal?: AbortSignal;
|
|
@@ -172,11 +170,10 @@ interface ControlRuntimeConfig<TState, TAction, TActionResult, TEval extends Con
|
|
|
172
170
|
projectId?: string;
|
|
173
171
|
variantId?: string;
|
|
174
172
|
}
|
|
175
|
-
declare function runAgentControlLoop<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult>(config: ControlRuntimeConfig<TState, TAction, TActionResult, TEval>): Promise<ControlRunResult<TState, TAction, TActionResult, TEval>>;
|
|
176
|
-
declare function stopOnNoProgress<TState, TAction>(maxNoProgressSteps: number, options?: Omit<ControlStopPolicies<TState, TAction>, 'maxNoProgressSteps'>): ControlStopPolicies<TState, TAction>;
|
|
177
|
-
declare function stopOnRepeatedAction<TState, TAction>(maxRepeatedActions: number, options?: Omit<ControlStopPolicies<TState, TAction>, 'maxRepeatedActions'>): ControlStopPolicies<TState, TAction>;
|
|
178
|
-
declare function objectiveEval(input: Omit<ControlEvalResult, 'objective'>): ControlEvalResult;
|
|
179
|
-
declare function subjectiveEval(input: Omit<ControlEvalResult, 'objective'>): ControlEvalResult;
|
|
180
|
-
declare function allCriticalPassed(evals: ControlEvalResult[]): boolean;
|
|
181
|
-
|
|
182
|
-
export { type ControlEvalResult as C, type StopDecision as S, type ControlRunResult as a, type ControlStep as b, type ControlActionFailureMode as c, type ControlActionOutcome as d, type ControlBudget as e, type ControlContext as f, type ControlDecision as g, type ControlRuntimeConfig as h, type ControlRuntimeError as i, type ControlSeverity as j, type ControlStopPolicies as k, allCriticalPassed as l, stopOnRepeatedAction as m, subjectiveEval as n, objectiveEval as o, runAgentControlLoop as r, stopOnNoProgress as s };
|
|
173
|
+
export declare function runAgentControlLoop<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult>(config: ControlRuntimeConfig<TState, TAction, TActionResult, TEval>): Promise<ControlRunResult<TState, TAction, TActionResult, TEval>>;
|
|
174
|
+
export declare function stopOnNoProgress<TState, TAction>(maxNoProgressSteps: number, options?: Omit<ControlStopPolicies<TState, TAction>, 'maxNoProgressSteps'>): ControlStopPolicies<TState, TAction>;
|
|
175
|
+
export declare function stopOnRepeatedAction<TState, TAction>(maxRepeatedActions: number, options?: Omit<ControlStopPolicies<TState, TAction>, 'maxRepeatedActions'>): ControlStopPolicies<TState, TAction>;
|
|
176
|
+
export declare function objectiveEval(input: Omit<ControlEvalResult, 'objective'>): ControlEvalResult;
|
|
177
|
+
export declare function subjectiveEval(input: Omit<ControlEvalResult, 'objective'>): ControlEvalResult;
|
|
178
|
+
export declare function allCriticalPassed(evals: ControlEvalResult[]): boolean;
|
|
179
|
+
//# sourceMappingURL=control-runtime.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"control-runtime.d.ts","sourceRoot":"","sources":["../src/control-runtime.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAGH,OAAO,EAAmB,YAAY,EAAE,MAAM,iBAAiB,CAAA;AAC/D,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAA;AAClD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,eAAe,CAAA;AAE/C,MAAM,MAAM,eAAe,GAAG,MAAM,GAAG,SAAS,GAAG,OAAO,GAAG,UAAU,CAAA;AACvE,MAAM,MAAM,wBAAwB,GAAG,UAAU,GAAG,MAAM,CAAA;AAE1D,MAAM,WAAW,iBAAiB;IAChC,oCAAoC;IACpC,EAAE,EAAE,MAAM,CAAA;IACV,iCAAiC;IACjC,MAAM,EAAE,OAAO,CAAA;IACf,sDAAsD;IACtD,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,gFAAgF;IAChF,QAAQ,CAAC,EAAE,eAAe,CAAA;IAC1B,6BAA6B;IAC7B,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,8DAA8D;IAC9D,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,4EAA4E;IAC5E,SAAS,CAAC,EAAE,OAAO,CAAA;IACnB,sEAAsE;IACtE,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,aAAa;IAC5B,QAAQ,EAAE,MAAM,CAAA;IAChB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,UAAU,CAAC,EAAE,MAAM,CAAA;CACpB;AAED,MAAM,WAAW,mBAAmB,CAAC,MAAM,EAAE,OAAO;IAClD;;;OAGG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAA;IAC3B;;;OAGG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAA;IAC3B,qEAAqE;IACrE,aAAa,CAAC,EAAE,MAAM,CAAA;IACtB,0EAA0E;IAC1E,gBAAgB,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,MAAM,CAAA;IAC5C,+EAA+E;IAC/E,iBAAiB,CAAC,EAAE,CAAC,MAAM,EAAE,OAAO,KAAK,MAAM,CAAA;CAChD;AAED,MAAM,WAAW,cAAc,CAC7B,MAAM,EACN,OAAO,EACP,aAAa,EACb,KAAK,SAAS,iBAAiB,GAAG,iBAAiB;IAEnD,MAAM,EAAE,MAAM,CAAA;IACd,KAAK,EAAE,MAAM,CAAA;IACb,KAAK,EAAE,KAAK,EAAE,CAAA;IACd,OAAO,EAAE,WAAW,CAAC,MAAM,EAAE,OAAO,EAAE,aAAa,EAAE,KAAK,CAAC,EAAE,CAAA;IAC7D,MAAM,EAAE,aAAa,CAAA;IACrB,SAAS,EAAE,MAAM,CAAA;IACjB,MAAM,EAAE,MAAM,CAAA;IACd,YAAY,EAAE,MAAM,CAAA;IACpB,gBAAgB,CAAC,EAAE,MAAM,CAAA;IACzB,WAAW,EAAE,WAAW,CAAA;IACxB,OAAO,CAAC,EAAE,YAAY,CAAA;CACvB;AAED,MAAM,MAAM,eAAe,CAAC,OAAO,IAC/B;IACE,IAAI,EAAE,UAAU,CAAA;IAChB,MAAM,EAAE,OAAO,CAAA;IACf,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB,GACD;IACE,IAAI,EAAE,MAAM,CAAA;IACZ,MAAM,EAAE,MAAM,CAAA;IACd,IAAI,CAAC,EAAE,OAAO,CAAA;IACd,KAAK,CAAC,EAAE,MAAM,CAAA;CACf,CAAA;AAEL,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,OAAO,CAAA;IACb,IAAI,EAAE,OAAO,CAAA;IACb,MAAM,EAAE,MAAM,CAAA;IACd,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,YAAY,CAAC,EAAE,YAAY,CAAA;CAC5B;AAED,MAAM,WAAW,oBAAoB,CAAC,aAAa;IACjD,EAAE,EAAE,OAAO,CAAA;IACX,MAAM,CAAC,EAAE,aAAa,CAAA;IACtB,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,UAAU,EAAE,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,mBAAmB;IAClC,KAAK,EAAE,SAAS,GAAG,UAAU,GAAG,QAAQ,GAAG,KAAK,GAAG,aAAa,GAAG,SAAS,GAAG,OAAO,CAAA;IACtF,SAAS,EAAE,MAAM,CAAA;IACjB,OAAO,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,WAAW,CAC1B,MAAM,EACN,OAAO,EACP,aAAa,EACb,KAAK,SAAS,iBAAiB,GAAG,iBAAiB;IAEnD,KAAK,EAAE,MAAM,CAAA;IACb,QAAQ,EAAE,eAAe,CAAC,OAAO,CAAC,CAAA;IAClC,WAAW,EAAE,MAAM,CAAA;IACnB,UAAU,EAAE,MAAM,CAAA;IAClB,WAAW,EAAE,KAAK,EAAE,CAAA;IACpB,UAAU,EAAE,KAAK,EAAE,CAAA;IACnB,aAAa,CAAC,EAAE,oBAAoB,CAAC,aAAa,CAAC,CAAA;IACnD,SAAS,EAAE,MAAM,CAAA;IACjB,OAAO,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,gBAAgB,CAC/B,MAAM,EACN,OAAO,EACP,aAAa,EACb,KAAK,SAAS,iBAAiB,GAAG,iBAAiB;IAEnD,MAAM,EAAE,MAAM,CAAA;IACd,IAAI,EAAE,OAAO,CAAA;IACb,SAAS,EAAE,OAAO,CAAA;IAClB,MAAM,EAAE,MAAM,CAAA;IACd,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,KAAK,EAAE,WAAW,CAAC,MAAM,EAAE,OAAO,EAAE,aAAa,EAAE,KAAK,CAAC,EAAE,CAAA;IAC3D,UAAU,EAAE,MAAM,GAAG,SAAS,CAAA;IAC9B,UAAU,EAAE,KAAK,EAAE,CAAA;IACnB,MAAM,EAAE,MAAM,CAAA;IACd,YAAY,EAAE,MAAM,CAAA;IACpB,6FAA6F;IAC7F,KAAK,EAAE,MAAM,GAAG,IAAI,CAAA;IACpB,YAAY,CAAC,EAAE,YAAY,CAAA;IAC3B,aAAa,EAAE,mBAAmB,EAAE,CAAA;IACpC,SAAS,EAAE,QAAQ,GAAG,aAAa,GAAG,QAAQ,GAAG,OAAO,GAAG,eAAe,CAAA;CAC3E;AAED,MAAM,WAAW,oBAAoB,CACnC,MAAM,EACN,OAAO,EACP,aAAa,EACb,KAAK,SAAS,iBAAiB,GAAG,iBAAiB;IAEnD,MAAM,EAAE,MAAM,CAAA;IACd,MAAM,CAAC,EAAE,OAAO,CAAC,aAAa,CAAC,CAAA;IAC/B,MAAM,CAAC,EAAE,WAAW,CAAA;IACpB,iGAAiG;IACjG,aAAa,CAAC,EAAE,wBAAwB,CAAA;IACxC;;;OAGG;IACH,gBAAgB,CAAC,EAAE,CAAC,GAAG,EAAE;QACvB,MAAM,EAAE,OAAO,CAAA;QACf,MAAM,EAAE,aAAa,CAAA;QACrB,KAAK,EAAE,MAAM,CAAA;QACb,KAAK,EAAE,KAAK,EAAE,CAAA;QACd,OAAO,EAAE,WAAW,CAAC,MAAM,EAAE,OAAO,EAAE,aAAa,EAAE,KAAK,CAAC,EAAE,CAAA;KAC9D,KAAK,MAAM,GAAG,SAAS,CAAA;IAExB,2FAA2F;IAC3F,OAAO,EAAE,CAAC,GAAG,EAAE;QACb,OAAO,EAAE,WAAW,CAAC,MAAM,EAAE,OAAO,EAAE,aAAa,EAAE,KAAK,CAAC,EAAE,CAAA;QAC7D,WAAW,EAAE,WAAW,CAAA;KACzB,KAAK,OAAO,CAAC,MAAM,CAAC,GAAG,MAAM,CAAA;IAE9B,gGAAgG;IAChG,QAAQ,EAAE,CAAC,GAAG,EAAE;QACd,MAAM,EAAE,MAAM,CAAA;QACd,KAAK,EAAE,MAAM,CAAA;QACb,OAAO,EAAE,WAAW,CAAC,MAAM,EAAE,OAAO,EAAE,aAAa,EAAE,KAAK,CAAC,EAAE,CAAA;QAC7D,WAAW,EAAE,WAAW,CAAA;KACzB,KAAK,OAAO,CAAC,KAAK,EAAE,CAAC,GAAG,KAAK,EAAE,CAAA;IAEhC,uGAAuG;IACvG,MAAM,EAAE,CACN,GAAG,EAAE,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,aAAa,EAAE,KAAK,CAAC,KACvD,OAAO,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC,GAAG,eAAe,CAAC,OAAO,CAAC,CAAA;IAEjE,iDAAiD;IACjD,GAAG,EAAE,CACH,MAAM,EAAE,OAAO,EACf,GAAG,EAAE,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,aAAa,EAAE,KAAK,CAAC,KACvD,OAAO,CAAC,aAAa,CAAC,GAAG,aAAa,CAAA;IAE3C,yEAAyE;IACzE,UAAU,CAAC,EAAE,CACX,GAAG,EAAE,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,aAAa,EAAE,KAAK,CAAC,KACvD,OAAO,CAAC,YAAY,CAAC,GAAG,YAAY,CAAA;IAEzC,0DAA0D;IAC1D,MAAM,CAAC,EAAE,CAAC,IAAI,EAAE,WAAW,CAAC,MAAM,EAAE,OAAO,EAAE,aAAa,EAAE,KAAK,CAAC,KAAK,OAAO,CAAC,IAAI,CAAC,GAAG,IAAI,CAAA;IAE3F,kFAAkF;IAClF,YAAY,CAAC,EAAE,mBAAmB,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAEnD,yEAAyE;IACzE,KAAK,CAAC,EAAE,UAAU,CAAA;IAClB,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,SAAS,CAAC,EAAE,MAAM,CAAA;CACnB;AAOD,wBAAsB,mBAAmB,CACvC,MAAM,EACN,OAAO,EACP,aAAa,EACb,KAAK,SAAS,iBAAiB,GAAG,iBAAiB,EAEnD,MAAM,EAAE,oBAAoB,CAAC,MAAM,EAAE,OAAO,EAAE,aAAa,EAAE,KAAK,CAAC,GAClE,OAAO,CAAC,gBAAgB,CAAC,MAAM,EAAE,OAAO,EAAE,aAAa,EAAE,KAAK,CAAC,CAAC,CAgmBlE;AAED,wBAAgB,gBAAgB,CAAC,MAAM,EAAE,OAAO,EAC9C,kBAAkB,EAAE,MAAM,EAC1B,OAAO,GAAE,IAAI,CAAC,mBAAmB,CAAC,MAAM,EAAE,OAAO,CAAC,EAAE,oBAAoB,CAAM,GAC7E,mBAAmB,CAAC,MAAM,EAAE,OAAO,CAAC,CAEtC;AAED,wBAAgB,oBAAoB,CAAC,MAAM,EAAE,OAAO,EAClD,kBAAkB,EAAE,MAAM,EAC1B,OAAO,GAAE,IAAI,CAAC,mBAAmB,CAAC,MAAM,EAAE,OAAO,CAAC,EAAE,oBAAoB,CAAM,GAC7E,mBAAmB,CAAC,MAAM,EAAE,OAAO,CAAC,CAEtC;AAED,wBAAgB,aAAa,CAAC,KAAK,EAAE,IAAI,CAAC,iBAAiB,EAAE,WAAW,CAAC,GAAG,iBAAiB,CAE5F;AAED,wBAAgB,cAAc,CAAC,KAAK,EAAE,IAAI,CAAC,iBAAiB,EAAE,WAAW,CAAC,GAAG,iBAAiB,CAE7F;AAoDD,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,iBAAiB,EAAE,GAAG,OAAO,CAIrE"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"control-runtime.test.d.ts","sourceRoot":"","sources":["../src/control-runtime.test.ts"],"names":[],"mappings":""}
|
package/dist/control.d.ts
CHANGED
|
@@ -1,9 +1,11 @@
|
|
|
1
|
-
export {
|
|
2
|
-
export {
|
|
3
|
-
|
|
4
|
-
|
|
5
|
-
|
|
6
|
-
|
|
7
|
-
|
|
8
|
-
|
|
9
|
-
|
|
1
|
+
export type { ActionExecutionPolicy, ActionPolicyDecision, } from './action-policy';
|
|
2
|
+
export { evaluateActionPolicy } from './action-policy';
|
|
3
|
+
export type { ControlActionFailureMode, ControlActionOutcome, ControlBudget, ControlContext, ControlDecision, ControlEvalResult, ControlRunResult, ControlRuntimeConfig, ControlRuntimeError, ControlSeverity, ControlStep, ControlStopPolicies, StopDecision, } from './control-runtime';
|
|
4
|
+
export { allCriticalPassed, objectiveEval, runAgentControlLoop, stopOnNoProgress, stopOnRepeatedAction, subjectiveEval, } from './control-runtime';
|
|
5
|
+
export type { ProposeReviewConfig, ProposeReviewReport, } from './propose-review';
|
|
6
|
+
export { runProposeReview } from './propose-review';
|
|
7
|
+
export type { ProposeReviewControlAction, ProposeReviewControlConfig, ProposeReviewControlResult, ProposeReviewControlState, } from './propose-review-control';
|
|
8
|
+
export { runProposeReviewAsControlLoop } from './propose-review-control';
|
|
9
|
+
export type { ControlRunToRunRecordOptions, RunEvidenceMetadata, } from './run-evidence';
|
|
10
|
+
export { controlRunToRunRecord, scoreFromEvals, } from './run-evidence';
|
|
11
|
+
//# sourceMappingURL=control.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"control.d.ts","sourceRoot":"","sources":["../src/control.ts"],"names":[],"mappings":"AAAA,YAAY,EACV,qBAAqB,EACrB,oBAAoB,GACrB,MAAM,iBAAiB,CAAA;AACxB,OAAO,EAAE,oBAAoB,EAAE,MAAM,iBAAiB,CAAA;AACtD,YAAY,EACV,wBAAwB,EACxB,oBAAoB,EACpB,aAAa,EACb,cAAc,EACd,eAAe,EACf,iBAAiB,EACjB,gBAAgB,EAChB,oBAAoB,EACpB,mBAAmB,EACnB,eAAe,EACf,WAAW,EACX,mBAAmB,EACnB,YAAY,GACb,MAAM,mBAAmB,CAAA;AAC1B,OAAO,EACL,iBAAiB,EACjB,aAAa,EACb,mBAAmB,EACnB,gBAAgB,EAChB,oBAAoB,EACpB,cAAc,GACf,MAAM,mBAAmB,CAAA;AAC1B,YAAY,EACV,mBAAmB,EACnB,mBAAmB,GACpB,MAAM,kBAAkB,CAAA;AACzB,OAAO,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAA;AACnD,YAAY,EACV,0BAA0B,EAC1B,0BAA0B,EAC1B,0BAA0B,EAC1B,yBAAyB,GAC1B,MAAM,0BAA0B,CAAA;AACjC,OAAO,EAAE,6BAA6B,EAAE,MAAM,0BAA0B,CAAA;AACxE,YAAY,EACV,4BAA4B,EAC5B,mBAAmB,GACpB,MAAM,gBAAgB,CAAA;AACvB,OAAO,EACL,qBAAqB,EACrB,cAAc,GACf,MAAM,gBAAgB,CAAA"}
|
package/dist/control.js
CHANGED
|
@@ -4,7 +4,7 @@ import {
|
|
|
4
4
|
runProposeReview,
|
|
5
5
|
runProposeReviewAsControlLoop,
|
|
6
6
|
scoreFromEvals
|
|
7
|
-
} from "./chunk-
|
|
7
|
+
} from "./chunk-NZEQVRH5.js";
|
|
8
8
|
import {
|
|
9
9
|
allCriticalPassed,
|
|
10
10
|
objectiveEval,
|
|
@@ -13,7 +13,7 @@ import {
|
|
|
13
13
|
stopOnRepeatedAction,
|
|
14
14
|
subjectiveEval
|
|
15
15
|
} from "./chunk-YEHAEDUD.js";
|
|
16
|
-
import "./chunk-
|
|
16
|
+
import "./chunk-LO6IOIJ2.js";
|
|
17
17
|
import "./chunk-TVVP3ZZQ.js";
|
|
18
18
|
import "./chunk-VSMTAMNK.js";
|
|
19
19
|
import "./chunk-3BFEG2F6.js";
|
|
@@ -0,0 +1,29 @@
|
|
|
1
|
+
import type { CompletionCriterion, DriverState } from './types';
|
|
2
|
+
/**
|
|
3
|
+
* ConvergenceTracker — tracks completion percentage over turns.
|
|
4
|
+
*
|
|
5
|
+
* Produces convergence curves showing how quickly the agent reaches
|
|
6
|
+
* completion criteria.
|
|
7
|
+
*/
|
|
8
|
+
export declare class ConvergenceTracker {
|
|
9
|
+
private criteria;
|
|
10
|
+
private history;
|
|
11
|
+
constructor(criteria: CompletionCriterion[]);
|
|
12
|
+
/** Evaluate criteria against current state, record result */
|
|
13
|
+
record(turn: number, state: DriverState): {
|
|
14
|
+
completionPercent: number;
|
|
15
|
+
complete: boolean;
|
|
16
|
+
criteriaStatus: Record<string, boolean | number>;
|
|
17
|
+
};
|
|
18
|
+
/** Get convergence curve */
|
|
19
|
+
getCurve(): number[];
|
|
20
|
+
/** Get full history with per-criterion status */
|
|
21
|
+
getHistory(): {
|
|
22
|
+
turn: number;
|
|
23
|
+
completionPercent: number;
|
|
24
|
+
criteriaStatus: Record<string, boolean | number>;
|
|
25
|
+
}[];
|
|
26
|
+
/** Find the turn where completion first reached 100% (or null) */
|
|
27
|
+
getTurnToCompletion(): number | null;
|
|
28
|
+
}
|
|
29
|
+
//# sourceMappingURL=convergence.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"convergence.d.ts","sourceRoot":"","sources":["../src/convergence.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,mBAAmB,EAAE,WAAW,EAAE,MAAM,SAAS,CAAA;AAE/D;;;;;GAKG;AACH,qBAAa,kBAAkB;IAC7B,OAAO,CAAC,QAAQ,CAAuB;IACvC,OAAO,CAAC,OAAO,CAIP;gBAEI,QAAQ,EAAE,mBAAmB,EAAE;IAI3C,6DAA6D;IAC7D,MAAM,CACJ,IAAI,EAAE,MAAM,EACZ,KAAK,EAAE,WAAW,GACjB;QACD,iBAAiB,EAAE,MAAM,CAAA;QACzB,QAAQ,EAAE,OAAO,CAAA;QACjB,cAAc,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,GAAG,MAAM,CAAC,CAAA;KACjD;IA4BD,4BAA4B;IAC5B,QAAQ,IAAI,MAAM,EAAE;IAIpB,iDAAiD;IACjD,UAAU;cAnDF,MAAM;2BACO,MAAM;wBACT,MAAM,CAAC,MAAM,EAAE,OAAO,GAAG,MAAM,CAAC;;IAqDlD,kEAAkE;IAClE,mBAAmB,IAAI,MAAM,GAAG,IAAI;CAIrC"}
|
|
@@ -17,8 +17,8 @@
|
|
|
17
17
|
*/
|
|
18
18
|
/** Logical role of an LLM call. Free-form union — consumers add their own
|
|
19
19
|
* channels; the rollup keys on whatever string is supplied. */
|
|
20
|
-
type CostChannel = 'agent' | 'judge' | 'verifier' | 'analyst' | 'driver' | (string & {});
|
|
21
|
-
interface CostUsage {
|
|
20
|
+
export type CostChannel = 'agent' | 'judge' | 'verifier' | 'analyst' | 'driver' | (string & {});
|
|
21
|
+
export interface CostUsage {
|
|
22
22
|
inputTokens: number;
|
|
23
23
|
outputTokens: number;
|
|
24
24
|
cachedTokens?: number;
|
|
@@ -30,8 +30,8 @@ interface CostUsage {
|
|
|
30
30
|
* means any cost computed is `costUnknown` and the 0 must not aggregate as a
|
|
31
31
|
* measured cost.
|
|
32
32
|
*/
|
|
33
|
-
declare function modelPriceKey(model: string): string | null;
|
|
34
|
-
interface CostResult {
|
|
33
|
+
export declare function modelPriceKey(model: string): string | null;
|
|
34
|
+
export interface CostResult {
|
|
35
35
|
costUsd: number;
|
|
36
36
|
/** True when `model` has no pricing — the 0 is "not priced", NOT "free". */
|
|
37
37
|
costUnknown: boolean;
|
|
@@ -43,8 +43,8 @@ interface CostResult {
|
|
|
43
43
|
* model's input rate when present (no separate cache-discount table — callers
|
|
44
44
|
* that need provider-specific cache pricing supply `actualCostUsd` upstream).
|
|
45
45
|
*/
|
|
46
|
-
declare function costForUsage(model: string, usage: CostUsage): CostResult;
|
|
47
|
-
interface CostLedgerEntry extends CostUsage {
|
|
46
|
+
export declare function costForUsage(model: string, usage: CostUsage): CostResult;
|
|
47
|
+
export interface CostLedgerEntry extends CostUsage {
|
|
48
48
|
model: string;
|
|
49
49
|
channel: CostChannel;
|
|
50
50
|
costUsd: number;
|
|
@@ -55,7 +55,7 @@ interface CostLedgerEntry extends CostUsage {
|
|
|
55
55
|
tags?: Record<string, string>;
|
|
56
56
|
timestamp: number;
|
|
57
57
|
}
|
|
58
|
-
interface ChannelRollup {
|
|
58
|
+
export interface ChannelRollup {
|
|
59
59
|
channel: CostChannel;
|
|
60
60
|
calls: number;
|
|
61
61
|
inputTokens: number;
|
|
@@ -65,7 +65,7 @@ interface ChannelRollup {
|
|
|
65
65
|
/** Calls whose model was unpriced (their costUsd is 0-but-unknown). */
|
|
66
66
|
unpricedCalls: number;
|
|
67
67
|
}
|
|
68
|
-
interface CostLedgerSummary {
|
|
68
|
+
export interface CostLedgerSummary {
|
|
69
69
|
totalCalls: number;
|
|
70
70
|
inputTokens: number;
|
|
71
71
|
outputTokens: number;
|
|
@@ -86,7 +86,7 @@ interface CostLedgerSummary {
|
|
|
86
86
|
* answers "dollars per finished task", the metric every optimizer's
|
|
87
87
|
* quality-vs-cost tradeoff needs.
|
|
88
88
|
*/
|
|
89
|
-
declare class CostLedger {
|
|
89
|
+
export declare class CostLedger {
|
|
90
90
|
private readonly entries;
|
|
91
91
|
private completedTasks;
|
|
92
92
|
/**
|
|
@@ -109,5 +109,4 @@ declare class CostLedger {
|
|
|
109
109
|
/** Total spend divided by completed tasks; null when nothing completed. */
|
|
110
110
|
costPerCompletedTask(): number | null;
|
|
111
111
|
}
|
|
112
|
-
|
|
113
|
-
export { CostLedger as C, type CostChannel as a, type ChannelRollup as b, type CostLedgerEntry as c, type CostLedgerSummary as d, type CostResult as e, type CostUsage as f, costForUsage as g, modelPriceKey as m };
|
|
112
|
+
//# sourceMappingURL=cost-ledger.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"cost-ledger.d.ts","sourceRoot":"","sources":["../src/cost-ledger.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAKH;gEACgE;AAChE,MAAM,MAAM,WAAW,GAAG,OAAO,GAAG,OAAO,GAAG,UAAU,GAAG,SAAS,GAAG,QAAQ,GAAG,CAAC,MAAM,GAAG,EAAE,CAAC,CAAA;AAE/F,MAAM,WAAW,SAAS;IACxB,WAAW,EAAE,MAAM,CAAA;IACnB,YAAY,EAAE,MAAM,CAAA;IACpB,YAAY,CAAC,EAAE,MAAM,CAAA;CACtB;AAED;;;;;;GAMG;AACH,wBAAgB,aAAa,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAE1D;AAED,MAAM,WAAW,UAAU;IACzB,OAAO,EAAE,MAAM,CAAA;IACf,4EAA4E;IAC5E,WAAW,EAAE,OAAO,CAAA;CACrB;AAED;;;;;;GAMG;AACH,wBAAgB,YAAY,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,SAAS,GAAG,UAAU,CAQxE;AAED,MAAM,WAAW,eAAgB,SAAQ,SAAS;IAChD,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,EAAE,WAAW,CAAA;IACpB,OAAO,EAAE,MAAM,CAAA;IACf,WAAW,EAAE,OAAO,CAAA;IACpB,4DAA4D;IAC5D,aAAa,CAAC,EAAE,MAAM,CAAA;IACtB,0DAA0D;IAC1D,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC7B,SAAS,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,WAAW,aAAa;IAC5B,OAAO,EAAE,WAAW,CAAA;IACpB,KAAK,EAAE,MAAM,CAAA;IACb,WAAW,EAAE,MAAM,CAAA;IACnB,YAAY,EAAE,MAAM,CAAA;IACpB,YAAY,EAAE,MAAM,CAAA;IACpB,OAAO,EAAE,MAAM,CAAA;IACf,uEAAuE;IACvE,aAAa,EAAE,MAAM,CAAA;CACtB;AAED,MAAM,WAAW,iBAAiB;IAChC,UAAU,EAAE,MAAM,CAAA;IAClB,WAAW,EAAE,MAAM,CAAA;IACnB,YAAY,EAAE,MAAM,CAAA;IACpB,YAAY,EAAE,MAAM,CAAA;IACpB,YAAY,EAAE,MAAM,CAAA;IACpB,qDAAqD;IACrD,SAAS,EAAE,aAAa,EAAE,CAAA;IAC1B;2DACuD;IACvD,cAAc,EAAE,MAAM,EAAE,CAAA;IACxB,4EAA4E;IAC5E,WAAW,EAAE,OAAO,CAAA;CACrB;AAED;;;;;;GAMG;AACH,qBAAa,UAAU;IACrB,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAwB;IAChD,OAAO,CAAC,cAAc,CAAI;IAE1B;;;;OAIG;IACH,MAAM,CAAC,KAAK,EAAE;QACZ,KAAK,EAAE,MAAM,CAAA;QACb,OAAO,EAAE,WAAW,CAAA;QACpB,KAAK,EAAE,SAAS,CAAA;QAChB,aAAa,CAAC,EAAE,MAAM,CAAA;QACtB,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;QAC7B,SAAS,CAAC,EAAE,MAAM,CAAA;KACnB,GAAG,eAAe;IAqBnB,+EAA+E;IAC/E,aAAa,CAAC,KAAK,SAAI,GAAG,IAAI;IAS9B,IAAI,IAAI,eAAe,EAAE;IAIzB,OAAO,IAAI,iBAAiB;IA0C5B,2EAA2E;IAC3E,oBAAoB,IAAI,MAAM,GAAG,IAAI;CAItC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"cost-ledger.test.d.ts","sourceRoot":"","sources":["../src/cost-ledger.test.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,43 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Program cost report — a thin projection over `CostLedger.summary()` that
|
|
3
|
+
* adds the per-model rollup the summary lacks, plus `attachCostToReport`, the
|
|
4
|
+
* one way every artifact (capsules, campaign results, diagnose reports) gets
|
|
5
|
+
* its cost stamp.
|
|
6
|
+
*
|
|
7
|
+
* Honesty contract carried through from the ledger: `total.unknownEntries`
|
|
8
|
+
* and `perModel[].unpriced` surface the costUnknown axis — a $0 from an
|
|
9
|
+
* unpriced model is a lower bound, never a measured zero.
|
|
10
|
+
*/
|
|
11
|
+
import type { ChannelRollup, CostLedger } from './cost-ledger';
|
|
12
|
+
export interface ModelCostRollup {
|
|
13
|
+
model: string;
|
|
14
|
+
usd: number;
|
|
15
|
+
entries: number;
|
|
16
|
+
/** ≥1 entry for this model was costUnknown — `usd` is a lower bound. An
|
|
17
|
+
* `actualCostUsd` override clears the flag for that entry (the dollars are
|
|
18
|
+
* observed, even when the model has no pricing). */
|
|
19
|
+
unpriced: boolean;
|
|
20
|
+
}
|
|
21
|
+
export interface CostReport {
|
|
22
|
+
/** Per-channel breakdown — `CostLedgerSummary.byChannel` verbatim. */
|
|
23
|
+
perChannel: ChannelRollup[];
|
|
24
|
+
total: {
|
|
25
|
+
usd: number;
|
|
26
|
+
/** Entries whose cost was unknown — non-zero means `usd` is a lower bound. */
|
|
27
|
+
unknownEntries: number;
|
|
28
|
+
};
|
|
29
|
+
/** Per-model spend, sorted by model id. */
|
|
30
|
+
perModel: ModelCostRollup[];
|
|
31
|
+
}
|
|
32
|
+
/** Project a ledger into the program cost report. Pure — no I/O, no clock. */
|
|
33
|
+
export declare function costReport(ledger: CostLedger): CostReport;
|
|
34
|
+
/**
|
|
35
|
+
* Stamp a report-shaped object with its cost projection under the `cost` key.
|
|
36
|
+
* Generic so capsules, campaign results, and diagnose reports all stamp the
|
|
37
|
+
* same way. Throws when the report already carries a `cost` key — silently
|
|
38
|
+
* overwriting an existing stamp would corrupt the artifact's provenance.
|
|
39
|
+
*/
|
|
40
|
+
export declare function attachCostToReport<R extends object>(report: R, ledger: CostLedger): R & {
|
|
41
|
+
cost: CostReport;
|
|
42
|
+
};
|
|
43
|
+
//# sourceMappingURL=cost-report.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"cost-report.d.ts","sourceRoot":"","sources":["../src/cost-report.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,OAAO,KAAK,EAAE,aAAa,EAAE,UAAU,EAAE,MAAM,eAAe,CAAA;AAG9D,MAAM,WAAW,eAAe;IAC9B,KAAK,EAAE,MAAM,CAAA;IACb,GAAG,EAAE,MAAM,CAAA;IACX,OAAO,EAAE,MAAM,CAAA;IACf;;yDAEqD;IACrD,QAAQ,EAAE,OAAO,CAAA;CAClB;AAED,MAAM,WAAW,UAAU;IACzB,sEAAsE;IACtE,UAAU,EAAE,aAAa,EAAE,CAAA;IAC3B,KAAK,EAAE;QACL,GAAG,EAAE,MAAM,CAAA;QACX,8EAA8E;QAC9E,cAAc,EAAE,MAAM,CAAA;KACvB,CAAA;IACD,2CAA2C;IAC3C,QAAQ,EAAE,eAAe,EAAE,CAAA;CAC5B;AAED,8EAA8E;AAC9E,wBAAgB,UAAU,CAAC,MAAM,EAAE,UAAU,GAAG,UAAU,CAuBzD;AAED;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,CAAC,SAAS,MAAM,EACjD,MAAM,EAAE,CAAC,EACT,MAAM,EAAE,UAAU,GACjB,CAAC,GAAG;IAAE,IAAI,EAAE,UAAU,CAAA;CAAE,CAO1B"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"cost-report.test.d.ts","sourceRoot":"","sources":["../src/cost-report.test.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,76 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Cost tracker — token + USD accounting per scenario and per run.
|
|
3
|
+
*
|
|
4
|
+
* Adapted from generic usage-event accounting. Every
|
|
5
|
+
* optimizer needs to know "is the quality gain worth the cost delta?",
|
|
6
|
+
* and every dashboard needs dollars-per-completed-task. MODEL_PRICING
|
|
7
|
+
* from metrics.ts stays authoritative for estimate math; this module
|
|
8
|
+
* adds the aggregation + per-scenario roll-up that was duplicated
|
|
9
|
+
* across 4 verticals.
|
|
10
|
+
*/
|
|
11
|
+
export interface TokenSpec {
|
|
12
|
+
inputTokens: number;
|
|
13
|
+
outputTokens: number;
|
|
14
|
+
cachedTokens?: number;
|
|
15
|
+
reasoningTokens?: number;
|
|
16
|
+
}
|
|
17
|
+
export interface CostEntry extends TokenSpec {
|
|
18
|
+
scenarioId: string;
|
|
19
|
+
model: string;
|
|
20
|
+
/** Override estimate with an observed cost (e.g. from provider response). */
|
|
21
|
+
actualCostUsd?: number;
|
|
22
|
+
timestamp: number;
|
|
23
|
+
/** Free-form tags (variant id, round #, etc.). */
|
|
24
|
+
tags?: Record<string, string>;
|
|
25
|
+
}
|
|
26
|
+
export interface ScenarioCost {
|
|
27
|
+
scenarioId: string;
|
|
28
|
+
entries: CostEntry[];
|
|
29
|
+
totalInputTokens: number;
|
|
30
|
+
totalOutputTokens: number;
|
|
31
|
+
totalCachedTokens: number;
|
|
32
|
+
totalCostUsd: number;
|
|
33
|
+
/** Pass flag — set by consumer via markOutcome; used for cost-per-completed-task. */
|
|
34
|
+
completed?: boolean;
|
|
35
|
+
}
|
|
36
|
+
export declare class CostTracker {
|
|
37
|
+
private byScenario;
|
|
38
|
+
record(entry: Omit<CostEntry, 'timestamp'> & {
|
|
39
|
+
timestamp?: number;
|
|
40
|
+
}): CostEntry;
|
|
41
|
+
markOutcome(scenarioId: string, completed: boolean): void;
|
|
42
|
+
/**
|
|
43
|
+
* Convenience: record + markOutcome in one call from a
|
|
44
|
+
* `{ usage, verdict }`-shaped response (starter-foundry's
|
|
45
|
+
* `invokeMetaJudge` returns this shape; consumers that wrap any
|
|
46
|
+
* judge/critic can follow the same convention).
|
|
47
|
+
*
|
|
48
|
+
* `usage.model` must be present in `MODEL_PRICING` for cost math to
|
|
49
|
+
* populate; otherwise totalCostUsd stays at 0 for the entry but
|
|
50
|
+
* tokens still aggregate.
|
|
51
|
+
*/
|
|
52
|
+
recordVerdict(verdict: {
|
|
53
|
+
usage?: {
|
|
54
|
+
inputTokens: number;
|
|
55
|
+
outputTokens: number;
|
|
56
|
+
model: string;
|
|
57
|
+
cachedTokens?: number;
|
|
58
|
+
reasoningTokens?: number;
|
|
59
|
+
};
|
|
60
|
+
verdict?: 'pass' | 'fail' | 'borderline' | string;
|
|
61
|
+
}, scenarioId: string, tags?: Record<string, string>): CostEntry | null;
|
|
62
|
+
get(scenarioId: string): ScenarioCost | undefined;
|
|
63
|
+
list(): ScenarioCost[];
|
|
64
|
+
summary(): CostSummary;
|
|
65
|
+
}
|
|
66
|
+
export interface CostSummary {
|
|
67
|
+
scenarioCount: number;
|
|
68
|
+
completedCount: number;
|
|
69
|
+
totalInputTokens: number;
|
|
70
|
+
totalOutputTokens: number;
|
|
71
|
+
totalCostUsd: number;
|
|
72
|
+
avgCostPerScenarioUsd: number;
|
|
73
|
+
/** Total USD / completed scenarios — null when nothing completed. */
|
|
74
|
+
costPerCompletedTaskUsd: number | null;
|
|
75
|
+
}
|
|
76
|
+
//# sourceMappingURL=cost-tracker.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"cost-tracker.d.ts","sourceRoot":"","sources":["../src/cost-tracker.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAIH,MAAM,WAAW,SAAS;IACxB,WAAW,EAAE,MAAM,CAAA;IACnB,YAAY,EAAE,MAAM,CAAA;IACpB,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,eAAe,CAAC,EAAE,MAAM,CAAA;CACzB;AAED,MAAM,WAAW,SAAU,SAAQ,SAAS;IAC1C,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,EAAE,MAAM,CAAA;IACb,6EAA6E;IAC7E,aAAa,CAAC,EAAE,MAAM,CAAA;IACtB,SAAS,EAAE,MAAM,CAAA;IACjB,kDAAkD;IAClD,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CAC9B;AAED,MAAM,WAAW,YAAY;IAC3B,UAAU,EAAE,MAAM,CAAA;IAClB,OAAO,EAAE,SAAS,EAAE,CAAA;IACpB,gBAAgB,EAAE,MAAM,CAAA;IACxB,iBAAiB,EAAE,MAAM,CAAA;IACzB,iBAAiB,EAAE,MAAM,CAAA;IACzB,YAAY,EAAE,MAAM,CAAA;IACpB,qFAAqF;IACrF,SAAS,CAAC,EAAE,OAAO,CAAA;CACpB;AAED,qBAAa,WAAW;IACtB,OAAO,CAAC,UAAU,CAAkC;IAEpD,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,SAAS,EAAE,WAAW,CAAC,GAAG;QAAE,SAAS,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,SAAS;IAwB/E,WAAW,CAAC,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,OAAO,GAAG,IAAI;IAMzD;;;;;;;;;OASG;IACH,aAAa,CACX,OAAO,EAAE;QACP,KAAK,CAAC,EAAE;YACN,WAAW,EAAE,MAAM,CAAA;YACnB,YAAY,EAAE,MAAM,CAAA;YACpB,KAAK,EAAE,MAAM,CAAA;YACb,YAAY,CAAC,EAAE,MAAM,CAAA;YACrB,eAAe,CAAC,EAAE,MAAM,CAAA;SACzB,CAAA;QACD,OAAO,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,YAAY,GAAG,MAAM,CAAA;KAClD,EACD,UAAU,EAAE,MAAM,EAClB,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GAC5B,SAAS,GAAG,IAAI;IAenB,GAAG,CAAC,UAAU,EAAE,MAAM,GAAG,YAAY,GAAG,SAAS;IAIjD,IAAI,IAAI,YAAY,EAAE;IAItB,OAAO,IAAI,WAAW;CAiBvB;AAED,MAAM,WAAW,WAAW;IAC1B,aAAa,EAAE,MAAM,CAAA;IACrB,cAAc,EAAE,MAAM,CAAA;IACtB,gBAAgB,EAAE,MAAM,CAAA;IACxB,iBAAiB,EAAE,MAAM,CAAA;IACzB,YAAY,EAAE,MAAM,CAAA;IACpB,qBAAqB,EAAE,MAAM,CAAA;IAC7B,qEAAqE;IACrE,uBAAuB,EAAE,MAAM,GAAG,IAAI,CAAA;CACvC"}
|
|
@@ -1,7 +1,3 @@
|
|
|
1
|
-
import { T as TraceEmitter } from './emitter-C2rqGH_l.js';
|
|
2
|
-
import { T as TraceStore } from './store-BcFXE6LG.js';
|
|
3
|
-
import { T as Trajectory, a as TrajectoryStep } from './trajectory-2TkpSEVh.js';
|
|
4
|
-
|
|
5
1
|
/**
|
|
6
2
|
* Counterfactual replay — "what would have happened if we'd changed
|
|
7
3
|
* exactly one thing at turn N?"
|
|
@@ -15,8 +11,11 @@ import { T as Trajectory, a as TrajectoryStep } from './trajectory-2TkpSEVh.js';
|
|
|
15
11
|
* `parentRunId = originalRunId`, so downstream diff + correlation
|
|
16
12
|
* pipelines see them natively.
|
|
17
13
|
*/
|
|
18
|
-
|
|
19
|
-
type
|
|
14
|
+
import { TraceEmitter } from './trace/emitter';
|
|
15
|
+
import type { Span } from './trace/schema';
|
|
16
|
+
import type { TraceStore } from './trace/store';
|
|
17
|
+
import { type Trajectory, type TrajectoryStep } from './trajectory';
|
|
18
|
+
export type CounterfactualMutation = {
|
|
20
19
|
kind: 'swap-model';
|
|
21
20
|
at: number;
|
|
22
21
|
newModel: string;
|
|
@@ -37,7 +36,7 @@ type CounterfactualMutation = {
|
|
|
37
36
|
describe: string;
|
|
38
37
|
apply: (step: TrajectoryStep) => TrajectoryStep;
|
|
39
38
|
};
|
|
40
|
-
interface CounterfactualContext {
|
|
39
|
+
export interface CounterfactualContext {
|
|
41
40
|
originalRunId: string;
|
|
42
41
|
originalTrajectory: Trajectory;
|
|
43
42
|
/** Steps up to (but not including) the mutation point — the prefix the
|
|
@@ -49,7 +48,7 @@ interface CounterfactualContext {
|
|
|
49
48
|
* re-emit it or continue from there). */
|
|
50
49
|
mutatedStep: TrajectoryStep;
|
|
51
50
|
}
|
|
52
|
-
interface CounterfactualResult {
|
|
51
|
+
export interface CounterfactualResult {
|
|
53
52
|
counterfactualRunId: string;
|
|
54
53
|
originalRunId: string;
|
|
55
54
|
mutation: CounterfactualMutation;
|
|
@@ -60,7 +59,7 @@ interface CounterfactualResult {
|
|
|
60
59
|
deltaScore: number | null;
|
|
61
60
|
};
|
|
62
61
|
}
|
|
63
|
-
interface CounterfactualRunner {
|
|
62
|
+
export interface CounterfactualRunner {
|
|
64
63
|
/**
|
|
65
64
|
* Execute the agent from `ctx.prefix` with the mutation applied.
|
|
66
65
|
* MUST emit spans into the provided emitter so they become part of
|
|
@@ -68,18 +67,18 @@ interface CounterfactualRunner {
|
|
|
68
67
|
*/
|
|
69
68
|
executeFrom: (ctx: CounterfactualContext, emitter: TraceEmitter) => Promise<void>;
|
|
70
69
|
}
|
|
71
|
-
declare function runCounterfactual(store: TraceStore, originalRunId: string, mutation: CounterfactualMutation, runner: CounterfactualRunner): Promise<CounterfactualResult>;
|
|
70
|
+
export declare function runCounterfactual(store: TraceStore, originalRunId: string, mutation: CounterfactualMutation, runner: CounterfactualRunner): Promise<CounterfactualResult>;
|
|
72
71
|
/**
|
|
73
72
|
* Aggregate a batch of counterfactuals into a simple attribution table:
|
|
74
73
|
* which mutation kinds move outcomes most? (Useful when you run a grid
|
|
75
74
|
* over the same trajectory — swap-model at every llm span, swap-tool
|
|
76
75
|
* at every tool span — and want a ranked summary.)
|
|
77
76
|
*/
|
|
78
|
-
declare function attributeCounterfactuals(results: CounterfactualResult[]): Array<{
|
|
77
|
+
export declare function attributeCounterfactuals(results: CounterfactualResult[]): Array<{
|
|
79
78
|
mutationKind: CounterfactualMutation['kind'];
|
|
80
79
|
n: number;
|
|
81
80
|
meanAbsDelta: number;
|
|
82
81
|
meanSignedDelta: number;
|
|
83
82
|
}>;
|
|
84
|
-
|
|
85
|
-
|
|
83
|
+
export type { Span };
|
|
84
|
+
//# sourceMappingURL=counterfactual.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"counterfactual.d.ts","sourceRoot":"","sources":["../src/counterfactual.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAGH,OAAO,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAA;AAC9C,OAAO,KAAK,EAAW,IAAI,EAAY,MAAM,gBAAgB,CAAA;AAC7D,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,eAAe,CAAA;AAC/C,OAAO,EAAmB,KAAK,UAAU,EAAE,KAAK,cAAc,EAAE,MAAM,cAAc,CAAA;AAEpF,MAAM,MAAM,sBAAsB,GAC9B;IAAE,IAAI,EAAE,YAAY,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAA;CAAE,GACpD;IAAE,IAAI,EAAE,kBAAkB,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,OAAO,CAAA;CAAE,GAC5D;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,EAAE,EAAE,MAAM,CAAA;CAAE,GACtC;IAAE,IAAI,EAAE,uBAAuB,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GAC9D;IACE,IAAI,EAAE,QAAQ,CAAA;IACd,EAAE,EAAE,MAAM,CAAA;IACV,QAAQ,EAAE,MAAM,CAAA;IAChB,KAAK,EAAE,CAAC,IAAI,EAAE,cAAc,KAAK,cAAc,CAAA;CAChD,CAAA;AAEL,MAAM,WAAW,qBAAqB;IACpC,aAAa,EAAE,MAAM,CAAA;IACrB,kBAAkB,EAAE,UAAU,CAAA;IAC9B;0EACsE;IACtE,MAAM,EAAE,cAAc,EAAE,CAAA;IACxB,QAAQ,EAAE,sBAAsB,CAAA;IAChC;;8CAE0C;IAC1C,WAAW,EAAE,cAAc,CAAA;CAC5B;AAED,MAAM,WAAW,oBAAoB;IACnC,mBAAmB,EAAE,MAAM,CAAA;IAC3B,aAAa,EAAE,MAAM,CAAA;IACrB,QAAQ,EAAE,sBAAsB,CAAA;IAChC,gEAAgE;IAChE,KAAK,EAAE;QACL,oBAAoB,EAAE,MAAM,GAAG,IAAI,CAAA;QACnC,0BAA0B,EAAE,MAAM,GAAG,IAAI,CAAA;QACzC,UAAU,EAAE,MAAM,GAAG,IAAI,CAAA;KAC1B,CAAA;CACF;AAED,MAAM,WAAW,oBAAoB;IACnC;;;;OAIG;IACH,WAAW,EAAE,CAAC,GAAG,EAAE,qBAAqB,EAAE,OAAO,EAAE,YAAY,KAAK,OAAO,CAAC,IAAI,CAAC,CAAA;CAClF;AAED,wBAAsB,iBAAiB,CACrC,KAAK,EAAE,UAAU,EACjB,aAAa,EAAE,MAAM,EACrB,QAAQ,EAAE,sBAAsB,EAChC,MAAM,EAAE,oBAAoB,GAC3B,OAAO,CAAC,oBAAoB,CAAC,CA6C/B;AA0BD;;;;;GAKG;AACH,wBAAgB,wBAAwB,CAAC,OAAO,EAAE,oBAAoB,EAAE,GAAG,KAAK,CAAC;IAC/E,YAAY,EAAE,sBAAsB,CAAC,MAAM,CAAC,CAAA;IAC5C,CAAC,EAAE,MAAM,CAAA;IACT,YAAY,EAAE,MAAM,CAAA;IACpB,eAAe,EAAE,MAAM,CAAA;CACxB,CAAC,CA4BD;AAGD,YAAY,EAAE,IAAI,EAAE,CAAA"}
|
|
@@ -0,0 +1,56 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Full cross-trace diff — align two trajectories step-by-step, report
|
|
3
|
+
* per-step score deltas, attribute a variant's total outcome lead to
|
|
4
|
+
* specific turns.
|
|
5
|
+
*
|
|
6
|
+
* 0.5 shipped `firstDivergenceView` (finds the first differing step).
|
|
7
|
+
* This does the heavier work: full alignment via LCS, per-step
|
|
8
|
+
* contribution to score delta using PRM verdicts when available,
|
|
9
|
+
* fallback to structural heuristics (latency, token count, tool
|
|
10
|
+
* outcome) otherwise.
|
|
11
|
+
*/
|
|
12
|
+
import type { TraceStore } from './trace/store';
|
|
13
|
+
import { type TrajectoryStep } from './trajectory';
|
|
14
|
+
export type AlignmentOp = {
|
|
15
|
+
op: 'match';
|
|
16
|
+
a: TrajectoryStep;
|
|
17
|
+
b: TrajectoryStep;
|
|
18
|
+
} | {
|
|
19
|
+
op: 'insert';
|
|
20
|
+
b: TrajectoryStep;
|
|
21
|
+
} | {
|
|
22
|
+
op: 'delete';
|
|
23
|
+
a: TrajectoryStep;
|
|
24
|
+
} | {
|
|
25
|
+
op: 'replace';
|
|
26
|
+
a: TrajectoryStep;
|
|
27
|
+
b: TrajectoryStep;
|
|
28
|
+
};
|
|
29
|
+
export interface StepAttribution {
|
|
30
|
+
op: AlignmentOp;
|
|
31
|
+
/** Difference in PRM score (or null when not scored by a matching judge). */
|
|
32
|
+
prmDelta: number | null;
|
|
33
|
+
/** Difference in latency (endedAt - startedAt). */
|
|
34
|
+
latencyDeltaMs: number | null;
|
|
35
|
+
/** Difference in token count (LLM spans). */
|
|
36
|
+
tokenDelta: number | null;
|
|
37
|
+
/** Reason this step is / isn't considered a contributor to the outcome delta. */
|
|
38
|
+
note: string;
|
|
39
|
+
}
|
|
40
|
+
export interface CrossTraceDiff {
|
|
41
|
+
runA: string;
|
|
42
|
+
runB: string;
|
|
43
|
+
alignment: AlignmentOp[];
|
|
44
|
+
attributions: StepAttribution[];
|
|
45
|
+
/** Total score delta (B - A). */
|
|
46
|
+
totalScoreDelta: number | null;
|
|
47
|
+
/** Sum of PRM deltas across matched/replaced steps. Close to
|
|
48
|
+
* `totalScoreDelta` when PRM covers the trajectory; gap indicates
|
|
49
|
+
* unmodeled variance. */
|
|
50
|
+
prmDeltaSum: number;
|
|
51
|
+
}
|
|
52
|
+
export interface CrossTraceDiffOptions {
|
|
53
|
+
stepEquals?: (a: TrajectoryStep, b: TrajectoryStep) => boolean;
|
|
54
|
+
}
|
|
55
|
+
export declare function crossTraceDiff(store: TraceStore, runA: string, runB: string, options?: CrossTraceDiffOptions): Promise<CrossTraceDiff>;
|
|
56
|
+
//# sourceMappingURL=cross-trace-diff.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"cross-trace-diff.d.ts","sourceRoot":"","sources":["../src/cross-trace-diff.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAIH,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,eAAe,CAAA;AAC/C,OAAO,EAAmB,KAAK,cAAc,EAAE,MAAM,cAAc,CAAA;AAEnE,MAAM,MAAM,WAAW,GACnB;IAAE,EAAE,EAAE,OAAO,CAAC;IAAC,CAAC,EAAE,cAAc,CAAC;IAAC,CAAC,EAAE,cAAc,CAAA;CAAE,GACrD;IAAE,EAAE,EAAE,QAAQ,CAAC;IAAC,CAAC,EAAE,cAAc,CAAA;CAAE,GACnC;IAAE,EAAE,EAAE,QAAQ,CAAC;IAAC,CAAC,EAAE,cAAc,CAAA;CAAE,GACnC;IAAE,EAAE,EAAE,SAAS,CAAC;IAAC,CAAC,EAAE,cAAc,CAAC;IAAC,CAAC,EAAE,cAAc,CAAA;CAAE,CAAA;AAE3D,MAAM,WAAW,eAAe;IAC9B,EAAE,EAAE,WAAW,CAAA;IACf,6EAA6E;IAC7E,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;IACvB,mDAAmD;IACnD,cAAc,EAAE,MAAM,GAAG,IAAI,CAAA;IAC7B,6CAA6C;IAC7C,UAAU,EAAE,MAAM,GAAG,IAAI,CAAA;IACzB,iFAAiF;IACjF,IAAI,EAAE,MAAM,CAAA;CACb;AAED,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,MAAM,CAAA;IACZ,IAAI,EAAE,MAAM,CAAA;IACZ,SAAS,EAAE,WAAW,EAAE,CAAA;IACxB,YAAY,EAAE,eAAe,EAAE,CAAA;IAC/B,iCAAiC;IACjC,eAAe,EAAE,MAAM,GAAG,IAAI,CAAA;IAC9B;;8BAE0B;IAC1B,WAAW,EAAE,MAAM,CAAA;CACpB;AAED,MAAM,WAAW,qBAAqB;IACpC,UAAU,CAAC,EAAE,CAAC,CAAC,EAAE,cAAc,EAAE,CAAC,EAAE,cAAc,KAAK,OAAO,CAAA;CAC/D;AAED,wBAAsB,cAAc,CAClC,KAAK,EAAE,UAAU,EACjB,IAAI,EAAE,MAAM,EACZ,IAAI,EAAE,MAAM,EACZ,OAAO,GAAE,qBAA0B,GAClC,OAAO,CAAC,cAAc,CAAC,CAwBzB"}
|