@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
|
@@ -1,6 +1,3 @@
|
|
|
1
|
-
import { A as AgentEvalError, C as CaptureIntegrityError } from './errors-CzMUYo7b.js';
|
|
2
|
-
import { R as RawProviderSink, P as ProviderRedactor } from './raw-provider-sink-C46HDghv.js';
|
|
3
|
-
|
|
4
1
|
/**
|
|
5
2
|
* LLM client with graceful degrade.
|
|
6
3
|
*
|
|
@@ -22,8 +19,9 @@ import { R as RawProviderSink, P as ProviderRedactor } from './raw-provider-sink
|
|
|
22
19
|
* output (semantic concept judge, reviewer directives, critic scores). Primitives
|
|
23
20
|
* that need free-form text use `callLlm` and parse output themselves.
|
|
24
21
|
*/
|
|
25
|
-
|
|
26
|
-
|
|
22
|
+
import { AgentEvalError, CaptureIntegrityError } from './errors';
|
|
23
|
+
import { type ProviderRedactor, type RawProviderSink } from './trace/raw-provider-sink';
|
|
24
|
+
export interface LlmMessage {
|
|
27
25
|
role: 'system' | 'user' | 'assistant';
|
|
28
26
|
/**
|
|
29
27
|
* Either a plain text content string OR a multimodal content array
|
|
@@ -40,7 +38,7 @@ interface LlmMessage {
|
|
|
40
38
|
};
|
|
41
39
|
}>;
|
|
42
40
|
}
|
|
43
|
-
interface LlmCallRequest {
|
|
41
|
+
export interface LlmCallRequest {
|
|
44
42
|
model: string;
|
|
45
43
|
messages: LlmMessage[];
|
|
46
44
|
/** Optional JSON-mode response format (response_format: json_object). */
|
|
@@ -55,14 +53,14 @@ interface LlmCallRequest {
|
|
|
55
53
|
/** Per-call timeout, default 300s. */
|
|
56
54
|
timeoutMs?: number;
|
|
57
55
|
}
|
|
58
|
-
interface LlmUsage {
|
|
56
|
+
export interface LlmUsage {
|
|
59
57
|
promptTokens: number;
|
|
60
58
|
completionTokens: number;
|
|
61
59
|
totalTokens: number;
|
|
62
60
|
/** Proxies populate this when prompt caching is on. */
|
|
63
61
|
cachedPromptTokens?: number;
|
|
64
62
|
}
|
|
65
|
-
interface LlmCallResult {
|
|
63
|
+
export interface LlmCallResult {
|
|
66
64
|
/** The text content of the first choice. Empty string if none. */
|
|
67
65
|
content: string;
|
|
68
66
|
usage: LlmUsage;
|
|
@@ -94,13 +92,13 @@ interface LlmCallResult {
|
|
|
94
92
|
/** Raw response body. */
|
|
95
93
|
raw: Record<string, unknown>;
|
|
96
94
|
}
|
|
97
|
-
declare class LlmCallError extends AgentEvalError {
|
|
95
|
+
export declare class LlmCallError extends AgentEvalError {
|
|
98
96
|
readonly status: number;
|
|
99
97
|
readonly body: string;
|
|
100
98
|
readonly model: string;
|
|
101
99
|
constructor(message: string, status: number, body: string, model: string);
|
|
102
100
|
}
|
|
103
|
-
interface LlmClientOptions {
|
|
101
|
+
export interface LlmClientOptions {
|
|
104
102
|
/** Base URL (without trailing slash). Must end at the `/v1` prefix. */
|
|
105
103
|
baseUrl?: string;
|
|
106
104
|
/** Bearer token — either `apiKey` or `bearer` populates `Authorization: Bearer ...`. */
|
|
@@ -165,38 +163,39 @@ interface LlmClientOptions {
|
|
|
165
163
|
* treated identically whether it surfaces in the HTTP client or a
|
|
166
164
|
* TCloud-backed judge.
|
|
167
165
|
*/
|
|
168
|
-
declare function isTransientLlmError(err: unknown): boolean;
|
|
166
|
+
export declare function isTransientLlmError(err: unknown): boolean;
|
|
169
167
|
/** Exponential backoff: 500ms, 1s, 2s, 4s, ... capped at 16s. Attempt is 0-indexed. */
|
|
170
|
-
declare function backoffMs(attempt: number): number;
|
|
168
|
+
export declare function backoffMs(attempt: number): number;
|
|
171
169
|
/**
|
|
172
170
|
* Strip a ```json / ``` code fence if the model emitted one.
|
|
173
171
|
* Idempotent for naked JSON. Some models (claude-code via router, certain
|
|
174
172
|
* deepseek models) wrap output even under json_object.
|
|
175
173
|
*/
|
|
176
|
-
declare function stripFencedJson(raw: string): string;
|
|
174
|
+
export declare function stripFencedJson(raw: string): string;
|
|
175
|
+
export declare function extractJsonPayload(raw: string): string;
|
|
177
176
|
/**
|
|
178
177
|
* Low-level call. Returns raw content + usage + cost. Retries on transient
|
|
179
178
|
* failures; does NOT degrade schema here — callers that want graceful
|
|
180
179
|
* degrade use `callLlmJson`.
|
|
181
180
|
*/
|
|
182
|
-
declare function callLlm(req: LlmCallRequest, opts?: LlmClientOptions): Promise<LlmCallResult>;
|
|
181
|
+
export declare function callLlm(req: LlmCallRequest, opts?: LlmClientOptions): Promise<LlmCallResult>;
|
|
183
182
|
/**
|
|
184
183
|
* Structured-output call. Returns parsed JSON plus the raw result envelope.
|
|
185
184
|
* Degrades `jsonSchema` → `jsonMode` on a 400 that names the schema param —
|
|
186
185
|
* critical for deepseek-v3/v4, kimi-k2.6, and other models that don't accept
|
|
187
186
|
* the `response_format.json_schema` shape but DO accept `json_object`.
|
|
188
187
|
*/
|
|
189
|
-
declare function callLlmJson<T = unknown>(req: LlmCallRequest, opts?: LlmClientOptions): Promise<{
|
|
188
|
+
export declare function callLlmJson<T = unknown>(req: LlmCallRequest, opts?: LlmClientOptions): Promise<{
|
|
190
189
|
value: T;
|
|
191
190
|
result: LlmCallResult;
|
|
192
191
|
}>;
|
|
193
|
-
type LlmRouteAssertionReason = 'no_explicit_base_url' | 'base_url_blocked' | 'base_url_not_allowed' | 'no_auth' | 'wrong_provider';
|
|
194
|
-
declare class LlmRouteAssertionError extends CaptureIntegrityError {
|
|
192
|
+
export type LlmRouteAssertionReason = 'no_explicit_base_url' | 'base_url_blocked' | 'base_url_not_allowed' | 'no_auth' | 'wrong_provider';
|
|
193
|
+
export declare class LlmRouteAssertionError extends CaptureIntegrityError {
|
|
195
194
|
readonly reason: LlmRouteAssertionReason;
|
|
196
195
|
readonly baseUrl: string;
|
|
197
196
|
constructor(message: string, reason: LlmRouteAssertionReason, baseUrl: string);
|
|
198
197
|
}
|
|
199
|
-
interface LlmRouteRequirements {
|
|
198
|
+
export interface LlmRouteRequirements {
|
|
200
199
|
/**
|
|
201
200
|
* Throw if `opts.baseUrl` is undefined, i.e. the call would fall back to
|
|
202
201
|
* `DEFAULT_BASE_URL`. Set this for evaluation runs where silently using
|
|
@@ -228,7 +227,7 @@ interface LlmRouteRequirements {
|
|
|
228
227
|
* Throws `LlmRouteAssertionError`. Pure — no I/O — so it's safe to call
|
|
229
228
|
* from constructors and CI gates.
|
|
230
229
|
*/
|
|
231
|
-
declare function assertLlmRoute(opts: LlmClientOptions, req?: LlmRouteRequirements): void;
|
|
230
|
+
export declare function assertLlmRoute(opts: LlmClientOptions, req?: LlmRouteRequirements): void;
|
|
232
231
|
/**
|
|
233
232
|
* Probe whether a model is reachable. Returns latency + null error on
|
|
234
233
|
* success; `ok=false` + error message on any failure (HTTP, timeout,
|
|
@@ -240,7 +239,7 @@ declare function assertLlmRoute(opts: LlmClientOptions, req?: LlmRouteRequiremen
|
|
|
240
239
|
* for short prompts, so don't tighten this further. We don't validate
|
|
241
240
|
* content; HTTP 200 means reachable.
|
|
242
241
|
*/
|
|
243
|
-
declare function probeLlm(model: string, opts?: LlmClientOptions & {
|
|
242
|
+
export declare function probeLlm(model: string, opts?: LlmClientOptions & {
|
|
244
243
|
timeoutMs?: number;
|
|
245
244
|
}): Promise<{
|
|
246
245
|
ok: boolean;
|
|
@@ -252,7 +251,7 @@ declare function probeLlm(model: string, opts?: LlmClientOptions & {
|
|
|
252
251
|
* Thin wrapper around the free functions; exists for callers that want
|
|
253
252
|
* to inject a single configured instance into multiple primitives.
|
|
254
253
|
*/
|
|
255
|
-
declare class LlmClient {
|
|
254
|
+
export declare class LlmClient {
|
|
256
255
|
private readonly opts;
|
|
257
256
|
constructor(opts?: LlmClientOptions);
|
|
258
257
|
call(req: LlmCallRequest, per?: LlmClientOptions): Promise<LlmCallResult>;
|
|
@@ -261,5 +260,4 @@ declare class LlmClient {
|
|
|
261
260
|
result: LlmCallResult;
|
|
262
261
|
}>;
|
|
263
262
|
}
|
|
264
|
-
|
|
265
|
-
export { type LlmClientOptions as L, type LlmRouteRequirements as a, type LlmCallRequest as b, type LlmCallResult as c, LlmCallError as d, LlmClient as e, type LlmMessage as f, LlmRouteAssertionError as g, type LlmUsage as h, assertLlmRoute as i, backoffMs as j, callLlm as k, callLlmJson as l, isTransientLlmError as m, probeLlm as p, stripFencedJson as s };
|
|
263
|
+
//# sourceMappingURL=llm-client.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"llm-client.d.ts","sourceRoot":"","sources":["../src/llm-client.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAEH,OAAO,EAAE,cAAc,EAAE,qBAAqB,EAAE,MAAM,UAAU,CAAA;AAChE,OAAO,EAEL,KAAK,gBAAgB,EAGrB,KAAK,eAAe,EACrB,MAAM,2BAA2B,CAAA;AAIlC,MAAM,WAAW,UAAU;IACzB,IAAI,EAAE,QAAQ,GAAG,MAAM,GAAG,WAAW,CAAA;IACrC;;;OAGG;IACH,OAAO,EACH,MAAM,GACN,KAAK,CACD;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,GAC9B;QAAE,IAAI,EAAE,WAAW,CAAC;QAAC,SAAS,EAAE;YAAE,GAAG,EAAE,MAAM,CAAC;YAAC,MAAM,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,CAAA;SAAE,CAAA;KAAE,CACtF,CAAA;CACN;AAED,MAAM,WAAW,cAAc;IAC7B,KAAK,EAAE,MAAM,CAAA;IACb,QAAQ,EAAE,UAAU,EAAE,CAAA;IACtB,yEAAyE;IACzE,QAAQ,CAAC,EAAE,OAAO,CAAA;IAClB,oFAAoF;IACpF,UAAU,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;KAAE,CAAA;IAC9D,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,sCAAsC;IACtC,SAAS,CAAC,EAAE,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,QAAQ;IACvB,YAAY,EAAE,MAAM,CAAA;IACpB,gBAAgB,EAAE,MAAM,CAAA;IACxB,WAAW,EAAE,MAAM,CAAA;IACnB,uDAAuD;IACvD,kBAAkB,CAAC,EAAE,MAAM,CAAA;CAC5B;AAED,MAAM,WAAW,aAAa;IAC5B,kEAAkE;IAClE,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,EAAE,QAAQ,CAAA;IACf;;;OAGG;IACH,OAAO,EAAE,MAAM,GAAG,IAAI,CAAA;IACtB,uDAAuD;IACvD,KAAK,EAAE,MAAM,CAAA;IACb,uEAAuE;IACvE,UAAU,EAAE,MAAM,CAAA;IAClB;;;;;;;OAOG;IACH,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAA;IAC5B;;;;;OAKG;IACH,YAAY,CAAC,EAAE,OAAO,CAAA;IACtB,yBAAyB;IACzB,GAAG,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAC7B;AAED,qBAAa,YAAa,SAAQ,cAAc;aAG5B,MAAM,EAAE,MAAM;aACd,IAAI,EAAE,MAAM;aACZ,KAAK,EAAE,MAAM;gBAH7B,OAAO,EAAE,MAAM,EACC,MAAM,EAAE,MAAM,EACd,IAAI,EAAE,MAAM,EACZ,KAAK,EAAE,MAAM;CAIhC;AAED,MAAM,WAAW,gBAAgB;IAC/B,uEAAuE;IACvE,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,wFAAwF;IACxF,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,yGAAyG;IACzG,UAAU,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAA;IAC5C,oDAAoD;IACpD,gBAAgB,CAAC,EAAE,MAAM,CAAA;IACzB;;;;;OAKG;IACH,MAAM,CAAC,EAAE,WAAW,CAAA;IACpB;;;;;;OAMG;IACH,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,iFAAiF;IACjF,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,qGAAqG;IACrG,KAAK,CAAC,EAAE,OAAO,KAAK,CAAA;IACpB;;;;;;OAMG;IACH,OAAO,CAAC,EAAE,eAAe,CAAA;IACzB;;;OAGG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,gFAAgF;IAChF,YAAY,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,CAAA;IAClD,4FAA4F;IAC5F,QAAQ,CAAC,EAAE,gBAAgB,CAAA;CAC5B;AA2CD;;;;;;;;;;;GAWG;AACH,wBAAgB,mBAAmB,CAAC,GAAG,EAAE,OAAO,GAAG,OAAO,CAEzD;AA6BD,uFAAuF;AACvF,wBAAgB,SAAS,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAEjD;AAoFD;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAInD;AAED,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAwBtD;AAsCD;;;;GAIG;AACH,wBAAsB,OAAO,CAC3B,GAAG,EAAE,cAAc,EACnB,IAAI,GAAE,gBAAqB,GAC1B,OAAO,CAAC,aAAa,CAAC,CA8OxB;AA6BD;;;;;GAKG;AACH,wBAAsB,WAAW,CAAC,CAAC,GAAG,OAAO,EAC3C,GAAG,EAAE,cAAc,EACnB,IAAI,GAAE,gBAAqB,GAC1B,OAAO,CAAC;IAAE,KAAK,EAAE,CAAC,CAAC;IAAC,MAAM,EAAE,aAAa,CAAA;CAAE,CAAC,CAe9C;AAiBD,MAAM,MAAM,uBAAuB,GAC/B,sBAAsB,GACtB,kBAAkB,GAClB,sBAAsB,GACtB,SAAS,GACT,gBAAgB,CAAA;AAEpB,qBAAa,sBAAuB,SAAQ,qBAAqB;aAG7C,MAAM,EAAE,uBAAuB;aAC/B,OAAO,EAAE,MAAM;gBAF/B,OAAO,EAAE,MAAM,EACC,MAAM,EAAE,uBAAuB,EAC/B,OAAO,EAAE,MAAM;CAIlC;AAED,MAAM,WAAW,oBAAoB;IACnC;;;;;OAKG;IACH,sBAAsB,CAAC,EAAE,OAAO,CAAA;IAChC;;;OAGG;IACH,eAAe,CAAC,EAAE,KAAK,CAAC,MAAM,GAAG,MAAM,CAAC,CAAA;IACxC,8DAA8D;IAC9D,eAAe,CAAC,EAAE,KAAK,CAAC,MAAM,GAAG,MAAM,CAAC,CAAA;IACxC,uDAAuD;IACvD,WAAW,CAAC,EAAE,OAAO,CAAA;IACrB;;;OAGG;IACH,gBAAgB,CAAC,EAAE,MAAM,CAAA;CAC1B;AAED;;;;;;;;GAQG;AACH,wBAAgB,cAAc,CAAC,IAAI,EAAE,gBAAgB,EAAE,GAAG,GAAE,oBAAyB,GAAG,IAAI,CAiD3F;AAWD;;;;;;;;;;GAUG;AACH,wBAAsB,QAAQ,CAC5B,KAAK,EAAE,MAAM,EACb,IAAI,GAAE,gBAAgB,GAAG;IAAE,SAAS,CAAC,EAAE,MAAM,CAAA;CAAO,GACnD,OAAO,CAAC;IAAE,EAAE,EAAE,OAAO,CAAC;IAAC,SAAS,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAoBnE;AAED;;;;GAIG;AACH,qBAAa,SAAS;IACR,OAAO,CAAC,QAAQ,CAAC,IAAI;gBAAJ,IAAI,GAAE,gBAAqB;IAExD,IAAI,CAAC,GAAG,EAAE,cAAc,EAAE,GAAG,CAAC,EAAE,gBAAgB,GAAG,OAAO,CAAC,aAAa,CAAC;IAIzE,QAAQ,CAAC,CAAC,GAAG,OAAO,EAClB,GAAG,EAAE,cAAc,EACnB,GAAG,CAAC,EAAE,gBAAgB,GACrB,OAAO,CAAC;QAAE,KAAK,EAAE,CAAC,CAAC;QAAC,MAAM,EAAE,aAAa,CAAA;KAAE,CAAC;CAGhD"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"llm-client.test.d.ts","sourceRoot":"","sources":["../src/llm-client.test.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,19 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* LockedJsonlAppender — mutex-serialized JSONL append helper for arbitrary
|
|
3
|
+
* payloads. The reference-replay store does the same thing for typed
|
|
4
|
+
* `ReferenceReplayRun` rows; this is the generic version used by
|
|
5
|
+
* `MutationTelemetry`, `TrialTelemetry`, and any other consumer that wants
|
|
6
|
+
* append-only durable telemetry without rolling its own lock.
|
|
7
|
+
*
|
|
8
|
+
* Locks are per absolute file path (process-local). Cross-process
|
|
9
|
+
* concurrency is NOT addressed — that's an fcntl/flock problem.
|
|
10
|
+
*/
|
|
11
|
+
export declare class LockedJsonlAppender {
|
|
12
|
+
readonly path: string;
|
|
13
|
+
private readonly mutex;
|
|
14
|
+
constructor(path: string);
|
|
15
|
+
append(entry: unknown): Promise<void>;
|
|
16
|
+
}
|
|
17
|
+
/** Reset all internal mutex state — tests only. */
|
|
18
|
+
export declare function resetLockedAppendersForTesting(): void;
|
|
19
|
+
//# sourceMappingURL=locked-jsonl-appender.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"locked-jsonl-appender.d.ts","sourceRoot":"","sources":["../src/locked-jsonl-appender.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAiBH,qBAAa,mBAAmB;aAEF,IAAI,EAAE,MAAM;IADxC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAO;gBACD,IAAI,EAAE,MAAM;IAOlC,MAAM,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,CAAC,IAAI,CAAC;CAM5C;AAED,mDAAmD;AACnD,wBAAgB,8BAA8B,IAAI,IAAI,CAErD"}
|
|
@@ -0,0 +1,16 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Per-axis aggregation of cell runs into `AxisSummary` rows.
|
|
3
|
+
*
|
|
4
|
+
* Pure: consumes the final `cells: [{cell, runs}]` array and returns the
|
|
5
|
+
* `byAxis` table. Error runs contribute 0 to passRate and meanScore. Cost
|
|
6
|
+
* and duration always count — the budget was spent regardless.
|
|
7
|
+
*/
|
|
8
|
+
import type { AxisSummary, CellResult, MatrixAxis, MatrixCell, MatrixResult } from './types';
|
|
9
|
+
interface Row<Output> {
|
|
10
|
+
cell: MatrixCell;
|
|
11
|
+
result: CellResult<Output>;
|
|
12
|
+
}
|
|
13
|
+
export declare function summariseRows<Output>(rows: Row<Output>[], axisName: string, axisValue: string): AxisSummary;
|
|
14
|
+
export declare function buildByAxis<Output>(cells: MatrixResult<Output>['cells'], axes: MatrixAxis<unknown>[], aggregateBy: string[]): Record<string, Record<string, AxisSummary>>;
|
|
15
|
+
export {};
|
|
16
|
+
//# sourceMappingURL=aggregation.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"aggregation.d.ts","sourceRoot":"","sources":["../../src/matrix/aggregation.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,UAAU,EAAE,UAAU,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,SAAS,CAAA;AAE5F,UAAU,GAAG,CAAC,MAAM;IAClB,IAAI,EAAE,UAAU,CAAA;IAChB,MAAM,EAAE,UAAU,CAAC,MAAM,CAAC,CAAA;CAC3B;AAqBD,wBAAgB,aAAa,CAAC,MAAM,EAClC,IAAI,EAAE,GAAG,CAAC,MAAM,CAAC,EAAE,EACnB,QAAQ,EAAE,MAAM,EAChB,SAAS,EAAE,MAAM,GAChB,WAAW,CAyCb;AA2BD,wBAAgB,WAAW,CAAC,MAAM,EAChC,KAAK,EAAE,YAAY,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC,EACpC,IAAI,EAAE,UAAU,CAAC,OAAO,CAAC,EAAE,EAC3B,WAAW,EAAE,MAAM,EAAE,GACpB,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC,CAe7C"}
|
package/dist/matrix/index.d.ts
CHANGED
|
@@ -1,34 +1,16 @@
|
|
|
1
|
-
import { M as MatrixResult, a as MatrixAxis, A as AxisSummary, b as MatrixCell, C as CellResult, R as RunAgentMatrixOptions } from '../types-mn5Aqk7x.js';
|
|
2
|
-
export { D as DefaultVerdict } from '../verdict-C9MlYujm.js';
|
|
3
|
-
|
|
4
1
|
/**
|
|
5
|
-
*
|
|
2
|
+
* @tangle-network/agent-eval/matrix
|
|
6
3
|
*
|
|
7
|
-
*
|
|
8
|
-
*
|
|
9
|
-
*
|
|
10
|
-
|
|
11
|
-
|
|
12
|
-
interface Row<Output> {
|
|
13
|
-
cell: MatrixCell;
|
|
14
|
-
result: CellResult<Output>;
|
|
15
|
-
}
|
|
16
|
-
declare function summariseRows<Output>(rows: Row<Output>[], axisName: string, axisValue: string): AxisSummary;
|
|
17
|
-
declare function buildByAxis<Output>(cells: MatrixResult<Output>['cells'], axes: MatrixAxis<unknown>[], aggregateBy: string[]): Record<string, Record<string, AxisSummary>>;
|
|
18
|
-
|
|
19
|
-
/**
|
|
20
|
-
* N-axis cartesian runner.
|
|
21
|
-
*
|
|
22
|
-
* Expansion order: cartesian over `axes` in declared order, then `reps` as the
|
|
23
|
-
* inner-most dim → `ordinal = (cartIdx * reps) + rep`. The returned
|
|
24
|
-
* `cells[]` is sorted by `ordinal` so concurrent execution does not reorder
|
|
25
|
-
* the output.
|
|
4
|
+
* N-axis cartesian runner over substrate types. The matrix is a runner +
|
|
5
|
+
* aggregator: cells carry caller-supplied values (AgentProfile from
|
|
6
|
+
* agent-interface, Driver / Validator from agent-runtime, rubric records, anything), and the
|
|
7
|
+
* runner returns per-axis pass / score / cost / duration summaries.
|
|
26
8
|
*
|
|
27
|
-
*
|
|
28
|
-
*
|
|
29
|
-
*
|
|
9
|
+
* No "Spec" / "Wrapper" / "Config" types around substrate. The only
|
|
10
|
+
* matrix-owned types are orchestration types: `MatrixAxis<V>`, `MatrixCell`,
|
|
11
|
+
* `CellResult`, `AxisSummary`, `MatrixResult`.
|
|
30
12
|
*/
|
|
31
|
-
|
|
32
|
-
|
|
33
|
-
|
|
34
|
-
|
|
13
|
+
export { buildByAxis, summariseRows } from './aggregation';
|
|
14
|
+
export { runAgentMatrix } from './runner';
|
|
15
|
+
export type { AxisSummary, CellResult, DefaultVerdict, MatrixAxis, MatrixCell, MatrixResult, RunAgentMatrixOptions, } from './types';
|
|
16
|
+
//# sourceMappingURL=index.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/matrix/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAE,WAAW,EAAE,aAAa,EAAE,MAAM,eAAe,CAAA;AAC1D,OAAO,EAAE,cAAc,EAAE,MAAM,UAAU,CAAA;AACzC,YAAY,EACV,WAAW,EACX,UAAU,EACV,cAAc,EACd,UAAU,EACV,UAAU,EACV,YAAY,EACZ,qBAAqB,GACtB,MAAM,SAAS,CAAA"}
|
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* N-axis cartesian runner.
|
|
3
|
+
*
|
|
4
|
+
* Expansion order: cartesian over `axes` in declared order, then `reps` as the
|
|
5
|
+
* inner-most dim → `ordinal = (cartIdx * reps) + rep`. The returned
|
|
6
|
+
* `cells[]` is sorted by `ordinal` so concurrent execution does not reorder
|
|
7
|
+
* the output.
|
|
8
|
+
*
|
|
9
|
+
* Scheduling is a sliding window of in-flight promises capped at
|
|
10
|
+
* `maxConcurrency`. The window stops admitting new cells when the cost
|
|
11
|
+
* ceiling trips or the abort signal fires; in-flight cells finish.
|
|
12
|
+
*/
|
|
13
|
+
import type { MatrixResult, RunAgentMatrixOptions } from './types';
|
|
14
|
+
export declare function runAgentMatrix<Output>(opts: RunAgentMatrixOptions<Output>): Promise<MatrixResult<Output>>;
|
|
15
|
+
//# sourceMappingURL=runner.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/matrix/runner.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAGH,OAAO,KAAK,EAIV,YAAY,EACZ,qBAAqB,EACtB,MAAM,SAAS,CAAA;AA8DhB,wBAAsB,cAAc,CAAC,MAAM,EACzC,IAAI,EAAE,qBAAqB,CAAC,MAAM,CAAC,GAClC,OAAO,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC,CAwJ/B"}
|
|
@@ -1,20 +1,17 @@
|
|
|
1
|
-
import { D as DefaultVerdict } from './verdict-C9MlYujm.js';
|
|
2
|
-
|
|
3
1
|
/**
|
|
4
|
-
* @experimental
|
|
5
|
-
*
|
|
6
2
|
* N-axis cartesian matrix over substrate types — types module.
|
|
7
3
|
*
|
|
8
4
|
* The matrix is a runner + aggregator. It iterates the cartesian product of
|
|
9
|
-
* caller-provided axes (any value type — `AgentProfile` from
|
|
10
|
-
* / `Validator` from agent-runtime, rubric records, thinking levels, anything)
|
|
5
|
+
* caller-provided axes (any value type — `AgentProfile` from agent-interface,
|
|
6
|
+
* `Driver` / `Validator` from agent-runtime, rubric records, thinking levels, anything)
|
|
11
7
|
* and aggregates per-axis pass/score/cost summaries. Substrate types are
|
|
12
8
|
* imported at the boundary by JSDoc only; the matrix never wraps them.
|
|
13
9
|
*/
|
|
14
|
-
|
|
10
|
+
import type { DefaultVerdict } from '../verdict';
|
|
11
|
+
export type { DefaultVerdict } from '../verdict';
|
|
15
12
|
/** One axis = one dimension to iterate. `V` is the value type — pass any
|
|
16
13
|
* substrate type (AgentProfile, Driver, Validator, rubric record). */
|
|
17
|
-
interface MatrixAxis<V> {
|
|
14
|
+
export interface MatrixAxis<V> {
|
|
18
15
|
/** Axis name. Becomes the key in `MatrixResult.byAxis`. */
|
|
19
16
|
name: string;
|
|
20
17
|
/** Stable id per value. Used as the bucket key in aggregation. */
|
|
@@ -27,7 +24,7 @@ interface MatrixAxis<V> {
|
|
|
27
24
|
label?: (value: V, id: string) => string;
|
|
28
25
|
}
|
|
29
26
|
/** A cell carries one picked value from each axis, keyed by axis name. */
|
|
30
|
-
interface MatrixCell {
|
|
27
|
+
export interface MatrixCell {
|
|
31
28
|
axes: Record<string, {
|
|
32
29
|
id: string;
|
|
33
30
|
value: unknown;
|
|
@@ -37,7 +34,7 @@ interface MatrixCell {
|
|
|
37
34
|
/** Stable sort key — preserves cartesian order across concurrent execution. */
|
|
38
35
|
ordinal: number;
|
|
39
36
|
}
|
|
40
|
-
interface CellResult<Output> {
|
|
37
|
+
export interface CellResult<Output> {
|
|
41
38
|
output: Output;
|
|
42
39
|
verdict: DefaultVerdict;
|
|
43
40
|
costUsd: number;
|
|
@@ -50,7 +47,7 @@ interface CellResult<Output> {
|
|
|
50
47
|
kind: string;
|
|
51
48
|
};
|
|
52
49
|
}
|
|
53
|
-
interface AxisSummary {
|
|
50
|
+
export interface AxisSummary {
|
|
54
51
|
axisName: string;
|
|
55
52
|
axisValue: string;
|
|
56
53
|
cells: number;
|
|
@@ -61,7 +58,7 @@ interface AxisSummary {
|
|
|
61
58
|
totalCostUsd: number;
|
|
62
59
|
meanDurationMs: number;
|
|
63
60
|
}
|
|
64
|
-
interface MatrixResult<Output> {
|
|
61
|
+
export interface MatrixResult<Output> {
|
|
65
62
|
cells: Array<{
|
|
66
63
|
cell: MatrixCell;
|
|
67
64
|
runs: CellResult<Output>[];
|
|
@@ -83,7 +80,7 @@ interface MatrixResult<Output> {
|
|
|
83
80
|
/** Stable id-like string generated at the end of the run. */
|
|
84
81
|
matrixId: string;
|
|
85
82
|
}
|
|
86
|
-
interface RunAgentMatrixOptions<Output> {
|
|
83
|
+
export interface RunAgentMatrixOptions<Output> {
|
|
87
84
|
axes: MatrixAxis<unknown>[];
|
|
88
85
|
/** User-supplied cell executor. May throw; the matrix captures throws as
|
|
89
86
|
* `CellResult.error` and continues. */
|
|
@@ -106,5 +103,4 @@ interface RunAgentMatrixOptions<Output> {
|
|
|
106
103
|
* and suppresses scheduling of new ones. */
|
|
107
104
|
signal?: AbortSignal;
|
|
108
105
|
}
|
|
109
|
-
|
|
110
|
-
export type { AxisSummary as A, CellResult as C, MatrixResult as M, RunAgentMatrixOptions as R, MatrixAxis as a, MatrixCell as b };
|
|
106
|
+
//# sourceMappingURL=types.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/matrix/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,YAAY,CAAA;AAEhD,YAAY,EAAE,cAAc,EAAE,MAAM,YAAY,CAAA;AAEhD;uEACuE;AACvE,MAAM,WAAW,UAAU,CAAC,CAAC;IAC3B,2DAA2D;IAC3D,IAAI,EAAE,MAAM,CAAA;IACZ,kEAAkE;IAClE,MAAM,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,CAAC,CAAA;KAAE,CAAC,CAAA;IACvC;4DACwD;IACxD,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,CAAC,EAAE,EAAE,EAAE,MAAM,KAAK,MAAM,CAAA;CACzC;AAED,0EAA0E;AAC1E,MAAM,WAAW,UAAU;IACzB,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,OAAO,CAAA;KAAE,CAAC,CAAA;IACpD,gEAAgE;IAChE,GAAG,EAAE,MAAM,CAAA;IACX,+EAA+E;IAC/E,OAAO,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,UAAU,CAAC,MAAM;IAChC,MAAM,EAAE,MAAM,CAAA;IACd,OAAO,EAAE,cAAc,CAAA;IACvB,OAAO,EAAE,MAAM,CAAA;IACf,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,CAAC,EAAE,MAAM,CAAA;IACd;6CACyC;IACzC,KAAK,CAAC,EAAE;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAA;CAC1C;AAED,MAAM,WAAW,WAAW;IAC1B,QAAQ,EAAE,MAAM,CAAA;IAChB,SAAS,EAAE,MAAM,CAAA;IACjB,KAAK,EAAE,MAAM,CAAA;IACb,QAAQ,EAAE,MAAM,CAAA;IAChB,SAAS,EAAE,MAAM,CAAA;IACjB,QAAQ,EAAE,MAAM,CAAA;IAChB,QAAQ,EAAE,MAAM,CAAA;IAChB,YAAY,EAAE,MAAM,CAAA;IACpB,cAAc,EAAE,MAAM,CAAA;CACvB;AAED,MAAM,WAAW,YAAY,CAAC,MAAM;IAClC,KAAK,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,UAAU,CAAC;QAAC,IAAI,EAAE,UAAU,CAAC,MAAM,CAAC,EAAE,CAAA;KAAE,CAAC,CAAA;IAC9D;mEAC+D;IAC/D,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC,CAAA;IACnD,OAAO,EAAE;QACP,UAAU,EAAE,MAAM,CAAA;QAClB,YAAY,EAAE,MAAM,CAAA;QACpB;+CACuC;QACvC,YAAY,EAAE,MAAM,CAAA;QACpB,eAAe,EAAE,MAAM,CAAA;QACvB,gBAAgB,EAAE,MAAM,CAAA;QACxB,YAAY,EAAE,MAAM,CAAA;QACpB,UAAU,EAAE,MAAM,CAAA;KACnB,CAAA;IACD,6DAA6D;IAC7D,QAAQ,EAAE,MAAM,CAAA;CACjB;AAED,MAAM,WAAW,qBAAqB,CAAC,MAAM;IAC3C,IAAI,EAAE,UAAU,CAAC,OAAO,CAAC,EAAE,CAAA;IAC3B;4CACwC;IACxC,OAAO,EAAE,CAAC,IAAI,EAAE,UAAU,KAAK,OAAO,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAA;IAC1D,sCAAsC;IACtC,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,2DAA2D;IAC3D,MAAM,CAAC,EAAE,CAAC,IAAI,EAAE,IAAI,CAAC,UAAU,EAAE,KAAK,GAAG,SAAS,CAAC,KAAK,OAAO,CAAA;IAC/D,sEAAsE;IACtE,WAAW,CAAC,EAAE,MAAM,EAAE,CAAA;IACtB,iEAAiE;IACjE,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB;;sDAEkD;IAClD,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,+DAA+D;IAC/D,cAAc,CAAC,EAAE,CAAC,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,CAAC,MAAM,CAAC,KAAK,IAAI,CAAA;IACvE;iDAC6C;IAC7C,MAAM,CAAC,EAAE,WAAW,CAAA;CACrB"}
|
|
@@ -0,0 +1,47 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Calibration curve — binned "if eval says X, what does reality show?"
|
|
3
|
+
*
|
|
4
|
+
* Companion to correlationStudy. Raw correlation is a single number;
|
|
5
|
+
* the calibration curve shows *where* the eval is well-calibrated vs
|
|
6
|
+
* overconfident / underconfident. Buckets the eval metric, computes
|
|
7
|
+
* mean outcome per bucket, reports expected-calibration-error (ECE).
|
|
8
|
+
*/
|
|
9
|
+
import type { TraceStore } from '../trace/store';
|
|
10
|
+
import type { EvalMetricSpec } from './correlation-study';
|
|
11
|
+
import type { OutcomeStore } from './outcome-store';
|
|
12
|
+
export interface CalibrationBin {
|
|
13
|
+
lower: number;
|
|
14
|
+
upper: number;
|
|
15
|
+
n: number;
|
|
16
|
+
evalMean: number;
|
|
17
|
+
outcomeMean: number;
|
|
18
|
+
/** |outcomeMean − evalMean|; contributes to ECE weighted by n/total. */
|
|
19
|
+
gap: number;
|
|
20
|
+
}
|
|
21
|
+
export interface CalibrationReport {
|
|
22
|
+
evalMetric: string;
|
|
23
|
+
outcomeMetric: string;
|
|
24
|
+
n: number;
|
|
25
|
+
bins: CalibrationBin[];
|
|
26
|
+
/** Expected Calibration Error — Σ (n_i/N) × |outcomeMean_i − evalMean_i|. */
|
|
27
|
+
ece: number;
|
|
28
|
+
/** Max bin gap — upper bound on miscalibration. */
|
|
29
|
+
maxGap: number;
|
|
30
|
+
}
|
|
31
|
+
export interface CalibrationOptions {
|
|
32
|
+
bins?: number;
|
|
33
|
+
/** Equal-width (fixed bin edges) or equal-frequency (quantile bins). */
|
|
34
|
+
binning?: 'equal-width' | 'equal-frequency';
|
|
35
|
+
/** Clip eval values to [lo, hi] before binning. */
|
|
36
|
+
range?: {
|
|
37
|
+
lo: number;
|
|
38
|
+
hi: number;
|
|
39
|
+
};
|
|
40
|
+
}
|
|
41
|
+
export interface CalibrationPair {
|
|
42
|
+
evalScore: number;
|
|
43
|
+
outcome: number;
|
|
44
|
+
}
|
|
45
|
+
export declare function calibrationCurve(traceStore: TraceStore, outcomeStore: OutcomeStore, evalMetric: EvalMetricSpec, outcomeMetric: string, options?: CalibrationOptions): Promise<CalibrationReport | null>;
|
|
46
|
+
export declare function calibrationFromPairs(inputPairs: CalibrationPair[], evalMetric: string, outcomeMetric: string, options?: CalibrationOptions): CalibrationReport | null;
|
|
47
|
+
//# sourceMappingURL=calibration.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"calibration.d.ts","sourceRoot":"","sources":["../../src/meta-eval/calibration.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAGH,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,gBAAgB,CAAA;AAChD,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAA;AACzD,OAAO,KAAK,EAAqB,YAAY,EAAE,MAAM,iBAAiB,CAAA;AAEtE,MAAM,WAAW,cAAc;IAC7B,KAAK,EAAE,MAAM,CAAA;IACb,KAAK,EAAE,MAAM,CAAA;IACb,CAAC,EAAE,MAAM,CAAA;IACT,QAAQ,EAAE,MAAM,CAAA;IAChB,WAAW,EAAE,MAAM,CAAA;IACnB,wEAAwE;IACxE,GAAG,EAAE,MAAM,CAAA;CACZ;AAED,MAAM,WAAW,iBAAiB;IAChC,UAAU,EAAE,MAAM,CAAA;IAClB,aAAa,EAAE,MAAM,CAAA;IACrB,CAAC,EAAE,MAAM,CAAA;IACT,IAAI,EAAE,cAAc,EAAE,CAAA;IACtB,6EAA6E;IAC7E,GAAG,EAAE,MAAM,CAAA;IACX,mDAAmD;IACnD,MAAM,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,kBAAkB;IACjC,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,wEAAwE;IACxE,OAAO,CAAC,EAAE,aAAa,GAAG,iBAAiB,CAAA;IAC3C,mDAAmD;IACnD,KAAK,CAAC,EAAE;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,EAAE,EAAE,MAAM,CAAA;KAAE,CAAA;CACnC;AAED,MAAM,WAAW,eAAe;IAC9B,SAAS,EAAE,MAAM,CAAA;IACjB,OAAO,EAAE,MAAM,CAAA;CAChB;AAED,wBAAsB,gBAAgB,CACpC,UAAU,EAAE,UAAU,EACtB,YAAY,EAAE,YAAY,EAC1B,UAAU,EAAE,cAAc,EAC1B,aAAa,EAAE,MAAM,EACrB,OAAO,GAAE,kBAAuB,GAC/B,OAAO,CAAC,iBAAiB,GAAG,IAAI,CAAC,CA8BnC;AAED,wBAAgB,oBAAoB,CAClC,UAAU,EAAE,eAAe,EAAE,EAC7B,UAAU,EAAE,MAAM,EAClB,aAAa,EAAE,MAAM,EACrB,OAAO,GAAE,kBAAuB,GAC/B,iBAAiB,GAAG,IAAI,CAsC1B"}
|
|
@@ -0,0 +1,53 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Correlation study — "does our eval score predict real-world outcomes?"
|
|
3
|
+
*
|
|
4
|
+
* This is the load-bearing signal. Takes a TraceStore + OutcomeStore,
|
|
5
|
+
* joins on runId, computes Pearson + Spearman + bootstrap CI for every
|
|
6
|
+
* (evalMetric, outcomeMetric) pair the caller declares.
|
|
7
|
+
*
|
|
8
|
+
* Without this number the framework is ornamental. With it and r > 0.6
|
|
9
|
+
* the framework is a moat — no other agent-eval tool publishes one.
|
|
10
|
+
*/
|
|
11
|
+
import type { Run } from '../trace/schema';
|
|
12
|
+
import type { TraceStore } from '../trace/store';
|
|
13
|
+
import type { OutcomeFilter, OutcomeStore } from './outcome-store';
|
|
14
|
+
export interface EvalMetricSpec {
|
|
15
|
+
id: string;
|
|
16
|
+
/** Extract a scalar from a run (defaults cover score/pass/durationMs/costUsd/tokens). */
|
|
17
|
+
extract?: (run: Run, store: TraceStore) => Promise<number | null>;
|
|
18
|
+
}
|
|
19
|
+
export interface OutcomePair {
|
|
20
|
+
evalMetric: string;
|
|
21
|
+
outcomeMetric: string;
|
|
22
|
+
}
|
|
23
|
+
export interface CorrelationResult {
|
|
24
|
+
evalMetric: string;
|
|
25
|
+
outcomeMetric: string;
|
|
26
|
+
n: number;
|
|
27
|
+
pearson: number;
|
|
28
|
+
spearman: number;
|
|
29
|
+
/** 95% bootstrap CI for Pearson. */
|
|
30
|
+
pearsonCi95: {
|
|
31
|
+
lower: number;
|
|
32
|
+
upper: number;
|
|
33
|
+
};
|
|
34
|
+
/** Rough verdict: 'strong' ≥ 0.7, 'moderate' ≥ 0.4, else 'weak'. */
|
|
35
|
+
verdict: 'strong' | 'moderate' | 'weak';
|
|
36
|
+
}
|
|
37
|
+
export interface CorrelationStudyResult {
|
|
38
|
+
pairs: CorrelationResult[];
|
|
39
|
+
joinedSamples: number;
|
|
40
|
+
skippedRuns: number;
|
|
41
|
+
}
|
|
42
|
+
export interface CorrelationStudyOptions {
|
|
43
|
+
/** Only join outcomes captured within this window after run.startedAt. */
|
|
44
|
+
maxCaptureLagMs?: number;
|
|
45
|
+
/** Restrict to a subset of outcomes (cohort, region, source). */
|
|
46
|
+
outcomeFilter?: OutcomeFilter;
|
|
47
|
+
/** Which outcome per run to use when multiple exist. Default 'latest'. */
|
|
48
|
+
reduction?: 'latest' | 'mean' | 'max';
|
|
49
|
+
/** Bootstrap iterations for the CI. Default 500. */
|
|
50
|
+
bootstrapIterations?: number;
|
|
51
|
+
}
|
|
52
|
+
export declare function correlationStudy(traceStore: TraceStore, outcomeStore: OutcomeStore, evalMetrics: EvalMetricSpec[], outcomeMetricNames: string[], options?: CorrelationStudyOptions): Promise<CorrelationStudyResult>;
|
|
53
|
+
//# sourceMappingURL=correlation-study.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"correlation-study.d.ts","sourceRoot":"","sources":["../../src/meta-eval/correlation-study.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAGH,OAAO,KAAK,EAAE,GAAG,EAAE,MAAM,iBAAiB,CAAA;AAC1C,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,gBAAgB,CAAA;AAChD,OAAO,KAAK,EAAqB,aAAa,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAA;AAErF,MAAM,WAAW,cAAc;IAC7B,EAAE,EAAE,MAAM,CAAA;IACV,yFAAyF;IACzF,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,EAAE,KAAK,EAAE,UAAU,KAAK,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,CAAA;CAClE;AAED,MAAM,WAAW,WAAW;IAC1B,UAAU,EAAE,MAAM,CAAA;IAClB,aAAa,EAAE,MAAM,CAAA;CACtB;AAED,MAAM,WAAW,iBAAiB;IAChC,UAAU,EAAE,MAAM,CAAA;IAClB,aAAa,EAAE,MAAM,CAAA;IACrB,CAAC,EAAE,MAAM,CAAA;IACT,OAAO,EAAE,MAAM,CAAA;IACf,QAAQ,EAAE,MAAM,CAAA;IAChB,oCAAoC;IACpC,WAAW,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAA;IAC7C,oEAAoE;IACpE,OAAO,EAAE,QAAQ,GAAG,UAAU,GAAG,MAAM,CAAA;CACxC;AAED,MAAM,WAAW,sBAAsB;IACrC,KAAK,EAAE,iBAAiB,EAAE,CAAA;IAC1B,aAAa,EAAE,MAAM,CAAA;IACrB,WAAW,EAAE,MAAM,CAAA;CACpB;AAED,MAAM,WAAW,uBAAuB;IACtC,0EAA0E;IAC1E,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,iEAAiE;IACjE,aAAa,CAAC,EAAE,aAAa,CAAA;IAC7B,0EAA0E;IAC1E,SAAS,CAAC,EAAE,QAAQ,GAAG,MAAM,GAAG,KAAK,CAAA;IACrC,oDAAoD;IACpD,mBAAmB,CAAC,EAAE,MAAM,CAAA;CAC7B;AAED,wBAAsB,gBAAgB,CACpC,UAAU,EAAE,UAAU,EACtB,YAAY,EAAE,YAAY,EAC1B,WAAW,EAAE,cAAc,EAAE,EAC7B,kBAAkB,EAAE,MAAM,EAAE,EAC5B,OAAO,GAAE,uBAA4B,GACpC,OAAO,CAAC,sBAAsB,CAAC,CA0EjC"}
|