@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
|
@@ -0,0 +1,34 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Built-in rubrics shipped with agent-eval.
|
|
3
|
+
*
|
|
4
|
+
* A rubric is a set of scoring axes plus a system prompt that tells the
|
|
5
|
+
* judging LLM how to grade against those axes. Built-in rubrics are
|
|
6
|
+
* curated for use cases that recur across Tangle projects — call them
|
|
7
|
+
* by name from any client.
|
|
8
|
+
*
|
|
9
|
+
* Adding a rubric:
|
|
10
|
+
* 1. Define the Rubric object below with a clear `description` and
|
|
11
|
+
* named `dimensions`.
|
|
12
|
+
* 2. Register it in `BUILTIN_RUBRICS` at the bottom.
|
|
13
|
+
* 3. Add a test in `tests/wire/rubrics.test.ts`.
|
|
14
|
+
*
|
|
15
|
+
* Custom rubrics: callers pass `rubric` inline to /v1/judge instead of
|
|
16
|
+
* `rubricName` — see schemas.ts.
|
|
17
|
+
*/
|
|
18
|
+
import type { Rubric } from './schemas';
|
|
19
|
+
export declare const BUILTIN_RUBRICS: Record<string, Rubric>;
|
|
20
|
+
/** Get a built-in rubric by name, or undefined. */
|
|
21
|
+
export declare function getBuiltinRubric(name: string): Rubric | undefined;
|
|
22
|
+
/** List built-in rubrics with their stable versions. */
|
|
23
|
+
export declare function listBuiltinRubrics(): {
|
|
24
|
+
name: string;
|
|
25
|
+
description: string;
|
|
26
|
+
dimensions: {
|
|
27
|
+
id: string;
|
|
28
|
+
description: string;
|
|
29
|
+
weight: number;
|
|
30
|
+
}[];
|
|
31
|
+
failureModes: string[];
|
|
32
|
+
rubricVersion: string;
|
|
33
|
+
}[];
|
|
34
|
+
//# sourceMappingURL=rubrics.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"rubrics.d.ts","sourceRoot":"","sources":["../../src/wire/rubrics.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AACH,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,WAAW,CAAA;AA4EvC,eAAO,MAAM,eAAe,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAElD,CAAA;AAED,mDAAmD;AACnD,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAEjE;AAED,wDAAwD;AACxD,wBAAgB,kBAAkB;;;;;;;;;;IAYjC"}
|
|
@@ -0,0 +1,410 @@
|
|
|
1
|
+
import { z } from 'zod';
|
|
2
|
+
export declare const RubricDimensionSchema: z.ZodObject<{
|
|
3
|
+
id: z.ZodString;
|
|
4
|
+
description: z.ZodString;
|
|
5
|
+
weight: z.ZodDefault<z.ZodNumber>;
|
|
6
|
+
min: z.ZodDefault<z.ZodNumber>;
|
|
7
|
+
max: z.ZodDefault<z.ZodNumber>;
|
|
8
|
+
}, z.core.$strip>;
|
|
9
|
+
export declare const FailureModeSchema: z.ZodObject<{
|
|
10
|
+
id: z.ZodString;
|
|
11
|
+
description: z.ZodString;
|
|
12
|
+
}, z.core.$strip>;
|
|
13
|
+
export declare const RubricSchema: z.ZodObject<{
|
|
14
|
+
name: z.ZodString;
|
|
15
|
+
description: z.ZodString;
|
|
16
|
+
systemPrompt: z.ZodString;
|
|
17
|
+
dimensions: z.ZodArray<z.ZodObject<{
|
|
18
|
+
id: z.ZodString;
|
|
19
|
+
description: z.ZodString;
|
|
20
|
+
weight: z.ZodDefault<z.ZodNumber>;
|
|
21
|
+
min: z.ZodDefault<z.ZodNumber>;
|
|
22
|
+
max: z.ZodDefault<z.ZodNumber>;
|
|
23
|
+
}, z.core.$strip>>;
|
|
24
|
+
failureModes: z.ZodDefault<z.ZodArray<z.ZodObject<{
|
|
25
|
+
id: z.ZodString;
|
|
26
|
+
description: z.ZodString;
|
|
27
|
+
}, z.core.$strip>>>;
|
|
28
|
+
wins: z.ZodDefault<z.ZodArray<z.ZodObject<{
|
|
29
|
+
id: z.ZodString;
|
|
30
|
+
description: z.ZodString;
|
|
31
|
+
}, z.core.$strip>>>;
|
|
32
|
+
}, z.core.$strip>;
|
|
33
|
+
export declare const JudgeRequestSchema: z.ZodObject<{
|
|
34
|
+
rubricName: z.ZodOptional<z.ZodString>;
|
|
35
|
+
rubric: z.ZodOptional<z.ZodObject<{
|
|
36
|
+
name: z.ZodString;
|
|
37
|
+
description: z.ZodString;
|
|
38
|
+
systemPrompt: z.ZodString;
|
|
39
|
+
dimensions: z.ZodArray<z.ZodObject<{
|
|
40
|
+
id: z.ZodString;
|
|
41
|
+
description: z.ZodString;
|
|
42
|
+
weight: z.ZodDefault<z.ZodNumber>;
|
|
43
|
+
min: z.ZodDefault<z.ZodNumber>;
|
|
44
|
+
max: z.ZodDefault<z.ZodNumber>;
|
|
45
|
+
}, z.core.$strip>>;
|
|
46
|
+
failureModes: z.ZodDefault<z.ZodArray<z.ZodObject<{
|
|
47
|
+
id: z.ZodString;
|
|
48
|
+
description: z.ZodString;
|
|
49
|
+
}, z.core.$strip>>>;
|
|
50
|
+
wins: z.ZodDefault<z.ZodArray<z.ZodObject<{
|
|
51
|
+
id: z.ZodString;
|
|
52
|
+
description: z.ZodString;
|
|
53
|
+
}, z.core.$strip>>>;
|
|
54
|
+
}, z.core.$strip>>;
|
|
55
|
+
content: z.ZodString;
|
|
56
|
+
context: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
57
|
+
model: z.ZodOptional<z.ZodString>;
|
|
58
|
+
}, z.core.$strip>;
|
|
59
|
+
export declare const JudgeResultSchema: z.ZodObject<{
|
|
60
|
+
composite: z.ZodNumber;
|
|
61
|
+
dimensions: z.ZodRecord<z.ZodString, z.ZodNumber>;
|
|
62
|
+
failureModes: z.ZodDefault<z.ZodArray<z.ZodString>>;
|
|
63
|
+
wins: z.ZodDefault<z.ZodArray<z.ZodString>>;
|
|
64
|
+
rationale: z.ZodString;
|
|
65
|
+
rubricVersion: z.ZodString;
|
|
66
|
+
model: z.ZodString;
|
|
67
|
+
durationMs: z.ZodNumber;
|
|
68
|
+
}, z.core.$strip>;
|
|
69
|
+
export declare const RubricInfoSchema: z.ZodObject<{
|
|
70
|
+
name: z.ZodString;
|
|
71
|
+
description: z.ZodString;
|
|
72
|
+
dimensions: z.ZodArray<z.ZodObject<{
|
|
73
|
+
id: z.ZodString;
|
|
74
|
+
description: z.ZodString;
|
|
75
|
+
weight: z.ZodNumber;
|
|
76
|
+
}, z.core.$strip>>;
|
|
77
|
+
failureModes: z.ZodDefault<z.ZodArray<z.ZodString>>;
|
|
78
|
+
rubricVersion: z.ZodString;
|
|
79
|
+
}, z.core.$strip>;
|
|
80
|
+
export declare const ListRubricsResponseSchema: z.ZodObject<{
|
|
81
|
+
rubrics: z.ZodArray<z.ZodObject<{
|
|
82
|
+
name: z.ZodString;
|
|
83
|
+
description: z.ZodString;
|
|
84
|
+
dimensions: z.ZodArray<z.ZodObject<{
|
|
85
|
+
id: z.ZodString;
|
|
86
|
+
description: z.ZodString;
|
|
87
|
+
weight: z.ZodNumber;
|
|
88
|
+
}, z.core.$strip>>;
|
|
89
|
+
failureModes: z.ZodDefault<z.ZodArray<z.ZodString>>;
|
|
90
|
+
rubricVersion: z.ZodString;
|
|
91
|
+
}, z.core.$strip>>;
|
|
92
|
+
}, z.core.$strip>;
|
|
93
|
+
export declare const VersionResponseSchema: z.ZodObject<{
|
|
94
|
+
package: z.ZodString;
|
|
95
|
+
version: z.ZodString;
|
|
96
|
+
wireVersion: z.ZodString;
|
|
97
|
+
apiSurface: z.ZodArray<z.ZodString>;
|
|
98
|
+
}, z.core.$strip>;
|
|
99
|
+
export declare const HealthResponseSchema: z.ZodObject<{
|
|
100
|
+
status: z.ZodLiteral<"ok">;
|
|
101
|
+
uptimeSec: z.ZodNumber;
|
|
102
|
+
}, z.core.$strip>;
|
|
103
|
+
/**
|
|
104
|
+
* Minimal `TraceEvent` shape that the production runtime emits.
|
|
105
|
+
* Matches `trace/schema.ts` `TraceEvent` but is duplicated here as a
|
|
106
|
+
* wire schema so non-TypeScript clients can validate without depending
|
|
107
|
+
* on internal types.
|
|
108
|
+
*/
|
|
109
|
+
export declare const TraceEventSchema: z.ZodObject<{
|
|
110
|
+
eventId: z.ZodString;
|
|
111
|
+
runId: z.ZodString;
|
|
112
|
+
spanId: z.ZodOptional<z.ZodString>;
|
|
113
|
+
kind: z.ZodEnum<{
|
|
114
|
+
policy_violation: "policy_violation";
|
|
115
|
+
custom: "custom";
|
|
116
|
+
error: "error";
|
|
117
|
+
log: "log";
|
|
118
|
+
budget_decrement: "budget_decrement";
|
|
119
|
+
budget_breach: "budget_breach";
|
|
120
|
+
state_mutation: "state_mutation";
|
|
121
|
+
redaction_applied: "redaction_applied";
|
|
122
|
+
}>;
|
|
123
|
+
timestamp: z.ZodNumber;
|
|
124
|
+
payload: z.ZodRecord<z.ZodString, z.ZodUnknown>;
|
|
125
|
+
}, z.core.$strip>;
|
|
126
|
+
export declare const TracesIngestRequestSchema: z.ZodObject<{
|
|
127
|
+
events: z.ZodArray<z.ZodObject<{
|
|
128
|
+
eventId: z.ZodString;
|
|
129
|
+
runId: z.ZodString;
|
|
130
|
+
spanId: z.ZodOptional<z.ZodString>;
|
|
131
|
+
kind: z.ZodEnum<{
|
|
132
|
+
policy_violation: "policy_violation";
|
|
133
|
+
custom: "custom";
|
|
134
|
+
error: "error";
|
|
135
|
+
log: "log";
|
|
136
|
+
budget_decrement: "budget_decrement";
|
|
137
|
+
budget_breach: "budget_breach";
|
|
138
|
+
state_mutation: "state_mutation";
|
|
139
|
+
redaction_applied: "redaction_applied";
|
|
140
|
+
}>;
|
|
141
|
+
timestamp: z.ZodNumber;
|
|
142
|
+
payload: z.ZodRecord<z.ZodString, z.ZodUnknown>;
|
|
143
|
+
}, z.core.$strip>>;
|
|
144
|
+
}, z.core.$strip>;
|
|
145
|
+
export declare const TracesIngestResponseSchema: z.ZodObject<{
|
|
146
|
+
accepted: z.ZodNumber;
|
|
147
|
+
rejected: z.ZodNumber;
|
|
148
|
+
errors: z.ZodDefault<z.ZodArray<z.ZodObject<{
|
|
149
|
+
eventId: z.ZodString;
|
|
150
|
+
message: z.ZodString;
|
|
151
|
+
}, z.core.$strip>>>;
|
|
152
|
+
}, z.core.$strip>;
|
|
153
|
+
export declare const FeedbackLabelSchema: z.ZodObject<{
|
|
154
|
+
id: z.ZodOptional<z.ZodString>;
|
|
155
|
+
source: z.ZodEnum<{
|
|
156
|
+
judge: "judge";
|
|
157
|
+
system: "system";
|
|
158
|
+
user: "user";
|
|
159
|
+
policy: "policy";
|
|
160
|
+
environment: "environment";
|
|
161
|
+
metric: "metric";
|
|
162
|
+
}>;
|
|
163
|
+
kind: z.ZodEnum<{
|
|
164
|
+
approve: "approve";
|
|
165
|
+
reject: "reject";
|
|
166
|
+
select: "select";
|
|
167
|
+
edit: "edit";
|
|
168
|
+
rank: "rank";
|
|
169
|
+
rate: "rate";
|
|
170
|
+
comment: "comment";
|
|
171
|
+
metric_outcome: "metric_outcome";
|
|
172
|
+
policy_block: "policy_block";
|
|
173
|
+
revision_request: "revision_request";
|
|
174
|
+
}>;
|
|
175
|
+
value: z.ZodUnknown;
|
|
176
|
+
reason: z.ZodOptional<z.ZodString>;
|
|
177
|
+
severity: z.ZodOptional<z.ZodEnum<{
|
|
178
|
+
error: "error";
|
|
179
|
+
info: "info";
|
|
180
|
+
critical: "critical";
|
|
181
|
+
warning: "warning";
|
|
182
|
+
}>>;
|
|
183
|
+
createdAt: z.ZodString;
|
|
184
|
+
metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
185
|
+
}, z.core.$strip>;
|
|
186
|
+
export declare const FeedbackAttemptSchema: z.ZodObject<{
|
|
187
|
+
id: z.ZodString;
|
|
188
|
+
stepIndex: z.ZodNumber;
|
|
189
|
+
artifactType: z.ZodEnum<{
|
|
190
|
+
action: "action";
|
|
191
|
+
decision: "decision";
|
|
192
|
+
text: "text";
|
|
193
|
+
code: "code";
|
|
194
|
+
plan: "plan";
|
|
195
|
+
research: "research";
|
|
196
|
+
ui: "ui";
|
|
197
|
+
data: "data";
|
|
198
|
+
other: "other";
|
|
199
|
+
}>;
|
|
200
|
+
artifact: z.ZodUnknown;
|
|
201
|
+
options: z.ZodOptional<z.ZodArray<z.ZodUnknown>>;
|
|
202
|
+
proposedAction: z.ZodOptional<z.ZodObject<{
|
|
203
|
+
type: z.ZodString;
|
|
204
|
+
risk: z.ZodOptional<z.ZodEnum<{
|
|
205
|
+
medium: "medium";
|
|
206
|
+
low: "low";
|
|
207
|
+
high: "high";
|
|
208
|
+
}>>;
|
|
209
|
+
costUsd: z.ZodOptional<z.ZodNumber>;
|
|
210
|
+
externalSideEffect: z.ZodOptional<z.ZodBoolean>;
|
|
211
|
+
requiresApproval: z.ZodOptional<z.ZodBoolean>;
|
|
212
|
+
metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
213
|
+
}, z.core.$strip>>;
|
|
214
|
+
feedback: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
215
|
+
id: z.ZodOptional<z.ZodString>;
|
|
216
|
+
source: z.ZodEnum<{
|
|
217
|
+
judge: "judge";
|
|
218
|
+
system: "system";
|
|
219
|
+
user: "user";
|
|
220
|
+
policy: "policy";
|
|
221
|
+
environment: "environment";
|
|
222
|
+
metric: "metric";
|
|
223
|
+
}>;
|
|
224
|
+
kind: z.ZodEnum<{
|
|
225
|
+
approve: "approve";
|
|
226
|
+
reject: "reject";
|
|
227
|
+
select: "select";
|
|
228
|
+
edit: "edit";
|
|
229
|
+
rank: "rank";
|
|
230
|
+
rate: "rate";
|
|
231
|
+
comment: "comment";
|
|
232
|
+
metric_outcome: "metric_outcome";
|
|
233
|
+
policy_block: "policy_block";
|
|
234
|
+
revision_request: "revision_request";
|
|
235
|
+
}>;
|
|
236
|
+
value: z.ZodUnknown;
|
|
237
|
+
reason: z.ZodOptional<z.ZodString>;
|
|
238
|
+
severity: z.ZodOptional<z.ZodEnum<{
|
|
239
|
+
error: "error";
|
|
240
|
+
info: "info";
|
|
241
|
+
critical: "critical";
|
|
242
|
+
warning: "warning";
|
|
243
|
+
}>>;
|
|
244
|
+
createdAt: z.ZodString;
|
|
245
|
+
metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
246
|
+
}, z.core.$strip>>>;
|
|
247
|
+
createdAt: z.ZodString;
|
|
248
|
+
metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
249
|
+
}, z.core.$strip>;
|
|
250
|
+
export declare const FeedbackTrajectorySchema: z.ZodObject<{
|
|
251
|
+
id: z.ZodString;
|
|
252
|
+
projectId: z.ZodOptional<z.ZodString>;
|
|
253
|
+
scenarioId: z.ZodOptional<z.ZodString>;
|
|
254
|
+
task: z.ZodObject<{
|
|
255
|
+
intent: z.ZodString;
|
|
256
|
+
context: z.ZodOptional<z.ZodUnknown>;
|
|
257
|
+
}, z.core.$strip>;
|
|
258
|
+
attempts: z.ZodDefault<z.ZodArray<z.ZodObject<{
|
|
259
|
+
id: z.ZodString;
|
|
260
|
+
stepIndex: z.ZodNumber;
|
|
261
|
+
artifactType: z.ZodEnum<{
|
|
262
|
+
action: "action";
|
|
263
|
+
decision: "decision";
|
|
264
|
+
text: "text";
|
|
265
|
+
code: "code";
|
|
266
|
+
plan: "plan";
|
|
267
|
+
research: "research";
|
|
268
|
+
ui: "ui";
|
|
269
|
+
data: "data";
|
|
270
|
+
other: "other";
|
|
271
|
+
}>;
|
|
272
|
+
artifact: z.ZodUnknown;
|
|
273
|
+
options: z.ZodOptional<z.ZodArray<z.ZodUnknown>>;
|
|
274
|
+
proposedAction: z.ZodOptional<z.ZodObject<{
|
|
275
|
+
type: z.ZodString;
|
|
276
|
+
risk: z.ZodOptional<z.ZodEnum<{
|
|
277
|
+
medium: "medium";
|
|
278
|
+
low: "low";
|
|
279
|
+
high: "high";
|
|
280
|
+
}>>;
|
|
281
|
+
costUsd: z.ZodOptional<z.ZodNumber>;
|
|
282
|
+
externalSideEffect: z.ZodOptional<z.ZodBoolean>;
|
|
283
|
+
requiresApproval: z.ZodOptional<z.ZodBoolean>;
|
|
284
|
+
metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
285
|
+
}, z.core.$strip>>;
|
|
286
|
+
feedback: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
287
|
+
id: z.ZodOptional<z.ZodString>;
|
|
288
|
+
source: z.ZodEnum<{
|
|
289
|
+
judge: "judge";
|
|
290
|
+
system: "system";
|
|
291
|
+
user: "user";
|
|
292
|
+
policy: "policy";
|
|
293
|
+
environment: "environment";
|
|
294
|
+
metric: "metric";
|
|
295
|
+
}>;
|
|
296
|
+
kind: z.ZodEnum<{
|
|
297
|
+
approve: "approve";
|
|
298
|
+
reject: "reject";
|
|
299
|
+
select: "select";
|
|
300
|
+
edit: "edit";
|
|
301
|
+
rank: "rank";
|
|
302
|
+
rate: "rate";
|
|
303
|
+
comment: "comment";
|
|
304
|
+
metric_outcome: "metric_outcome";
|
|
305
|
+
policy_block: "policy_block";
|
|
306
|
+
revision_request: "revision_request";
|
|
307
|
+
}>;
|
|
308
|
+
value: z.ZodUnknown;
|
|
309
|
+
reason: z.ZodOptional<z.ZodString>;
|
|
310
|
+
severity: z.ZodOptional<z.ZodEnum<{
|
|
311
|
+
error: "error";
|
|
312
|
+
info: "info";
|
|
313
|
+
critical: "critical";
|
|
314
|
+
warning: "warning";
|
|
315
|
+
}>>;
|
|
316
|
+
createdAt: z.ZodString;
|
|
317
|
+
metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
318
|
+
}, z.core.$strip>>>;
|
|
319
|
+
createdAt: z.ZodString;
|
|
320
|
+
metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
321
|
+
}, z.core.$strip>>>;
|
|
322
|
+
labels: z.ZodDefault<z.ZodArray<z.ZodObject<{
|
|
323
|
+
id: z.ZodOptional<z.ZodString>;
|
|
324
|
+
source: z.ZodEnum<{
|
|
325
|
+
judge: "judge";
|
|
326
|
+
system: "system";
|
|
327
|
+
user: "user";
|
|
328
|
+
policy: "policy";
|
|
329
|
+
environment: "environment";
|
|
330
|
+
metric: "metric";
|
|
331
|
+
}>;
|
|
332
|
+
kind: z.ZodEnum<{
|
|
333
|
+
approve: "approve";
|
|
334
|
+
reject: "reject";
|
|
335
|
+
select: "select";
|
|
336
|
+
edit: "edit";
|
|
337
|
+
rank: "rank";
|
|
338
|
+
rate: "rate";
|
|
339
|
+
comment: "comment";
|
|
340
|
+
metric_outcome: "metric_outcome";
|
|
341
|
+
policy_block: "policy_block";
|
|
342
|
+
revision_request: "revision_request";
|
|
343
|
+
}>;
|
|
344
|
+
value: z.ZodUnknown;
|
|
345
|
+
reason: z.ZodOptional<z.ZodString>;
|
|
346
|
+
severity: z.ZodOptional<z.ZodEnum<{
|
|
347
|
+
error: "error";
|
|
348
|
+
info: "info";
|
|
349
|
+
critical: "critical";
|
|
350
|
+
warning: "warning";
|
|
351
|
+
}>>;
|
|
352
|
+
createdAt: z.ZodString;
|
|
353
|
+
metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
354
|
+
}, z.core.$strip>>>;
|
|
355
|
+
outcome: z.ZodOptional<z.ZodObject<{
|
|
356
|
+
success: z.ZodOptional<z.ZodBoolean>;
|
|
357
|
+
score: z.ZodOptional<z.ZodNumber>;
|
|
358
|
+
metrics: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodNumber>>;
|
|
359
|
+
costUsd: z.ZodOptional<z.ZodNumber>;
|
|
360
|
+
detail: z.ZodOptional<z.ZodString>;
|
|
361
|
+
observedAt: z.ZodOptional<z.ZodString>;
|
|
362
|
+
metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
363
|
+
}, z.core.$strip>>;
|
|
364
|
+
split: z.ZodOptional<z.ZodEnum<{
|
|
365
|
+
train: "train";
|
|
366
|
+
dev: "dev";
|
|
367
|
+
test: "test";
|
|
368
|
+
holdout: "holdout";
|
|
369
|
+
}>>;
|
|
370
|
+
tags: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodString>>;
|
|
371
|
+
createdAt: z.ZodString;
|
|
372
|
+
updatedAt: z.ZodOptional<z.ZodString>;
|
|
373
|
+
metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
374
|
+
}, z.core.$strip>;
|
|
375
|
+
export declare const FeedbackIngestResponseSchema: z.ZodObject<{
|
|
376
|
+
id: z.ZodString;
|
|
377
|
+
persisted: z.ZodBoolean;
|
|
378
|
+
}, z.core.$strip>;
|
|
379
|
+
export type TraceEvent = z.infer<typeof TraceEventSchema>;
|
|
380
|
+
export type TracesIngestRequest = z.infer<typeof TracesIngestRequestSchema>;
|
|
381
|
+
export type TracesIngestResponse = z.infer<typeof TracesIngestResponseSchema>;
|
|
382
|
+
export type FeedbackTrajectory = z.infer<typeof FeedbackTrajectorySchema>;
|
|
383
|
+
export type FeedbackIngestResponse = z.infer<typeof FeedbackIngestResponseSchema>;
|
|
384
|
+
export declare const ErrorResponseSchema: z.ZodObject<{
|
|
385
|
+
error: z.ZodObject<{
|
|
386
|
+
code: z.ZodString;
|
|
387
|
+
message: z.ZodString;
|
|
388
|
+
details: z.ZodOptional<z.ZodUnknown>;
|
|
389
|
+
}, z.core.$strip>;
|
|
390
|
+
}, z.core.$strip>;
|
|
391
|
+
export type RubricDimension = z.infer<typeof RubricDimensionSchema>;
|
|
392
|
+
export type FailureMode = z.infer<typeof FailureModeSchema>;
|
|
393
|
+
export type Rubric = z.infer<typeof RubricSchema>;
|
|
394
|
+
export type JudgeRequest = z.infer<typeof JudgeRequestSchema>;
|
|
395
|
+
export type JudgeResult = z.infer<typeof JudgeResultSchema>;
|
|
396
|
+
export type RubricInfo = z.infer<typeof RubricInfoSchema>;
|
|
397
|
+
export type ListRubricsResponse = z.infer<typeof ListRubricsResponseSchema>;
|
|
398
|
+
export type VersionResponse = z.infer<typeof VersionResponseSchema>;
|
|
399
|
+
export type ErrorResponse = z.infer<typeof ErrorResponseSchema>;
|
|
400
|
+
/**
|
|
401
|
+
* Bump on any breaking change to a request/response schema.
|
|
402
|
+
* Non-breaking (additive) changes don't require a bump.
|
|
403
|
+
*/
|
|
404
|
+
export declare const WIRE_VERSION = "1.0.0";
|
|
405
|
+
/**
|
|
406
|
+
* Stable hash of a rubric. Used to make scores comparable across runs:
|
|
407
|
+
* if the rubricVersion matches, the rubric was identical.
|
|
408
|
+
*/
|
|
409
|
+
export declare function hashRubric(rubric: Rubric): string;
|
|
410
|
+
//# sourceMappingURL=schemas.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"schemas.d.ts","sourceRoot":"","sources":["../../src/wire/schemas.ts"],"names":[],"mappings":"AAaA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAA;AAMvB,eAAO,MAAM,qBAAqB;;;;;;iBAkBL,CAAA;AAE7B,eAAO,MAAM,iBAAiB;;;iBAKL,CAAA;AAIzB,eAAO,MAAM,YAAY;;;;;;;;;;;;;;;;;;;iBA6BL,CAAA;AAIpB,eAAO,MAAM,kBAAkB;;;;;;;;;;;;;;;;;;;;;;;;;iBA2BL,CAAA;AAE1B,eAAO,MAAM,iBAAiB;;;;;;;;;iBA6BL,CAAA;AAIzB,eAAO,MAAM,gBAAgB;;;;;;;;;;iBAUL,CAAA;AAExB,eAAO,MAAM,yBAAyB;;;;;;;;;;;;iBAIL,CAAA;AAIjC,eAAO,MAAM,qBAAqB;;;;;iBAWL,CAAA;AAE7B,eAAO,MAAM,oBAAoB;;;iBAKL,CAAA;AAI5B;;;;;GAKG;AACH,eAAO,MAAM,gBAAgB;;;;;;;;;;;;;;;;iBA0BL,CAAA;AAExB,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;iBAQL,CAAA;AAEjC,eAAO,MAAM,0BAA0B;;;;;;;iBAiBL,CAAA;AAElC,eAAO,MAAM,mBAAmB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAsBL,CAAA;AAE3B,eAAO,MAAM,qBAAqB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA+BL,CAAA;AAE7B,eAAO,MAAM,wBAAwB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA4BL,CAAA;AAEhC,eAAO,MAAM,4BAA4B;;;iBAKL,CAAA;AAEpC,MAAM,MAAM,UAAU,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,gBAAgB,CAAC,CAAA;AACzD,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAA;AAC3E,MAAM,MAAM,oBAAoB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,0BAA0B,CAAC,CAAA;AAC7E,MAAM,MAAM,kBAAkB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,wBAAwB,CAAC,CAAA;AACzE,MAAM,MAAM,sBAAsB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,4BAA4B,CAAC,CAAA;AAIjF,eAAO,MAAM,mBAAmB;;;;;;iBAcL,CAAA;AAI3B,MAAM,MAAM,eAAe,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,qBAAqB,CAAC,CAAA;AACnE,MAAM,MAAM,WAAW,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,iBAAiB,CAAC,CAAA;AAC3D,MAAM,MAAM,MAAM,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,YAAY,CAAC,CAAA;AACjD,MAAM,MAAM,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAA;AAC7D,MAAM,MAAM,WAAW,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,iBAAiB,CAAC,CAAA;AAC3D,MAAM,MAAM,UAAU,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,gBAAgB,CAAC,CAAA;AACzD,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAA;AAC3E,MAAM,MAAM,eAAe,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,qBAAqB,CAAC,CAAA;AACnE,MAAM,MAAM,aAAa,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,mBAAmB,CAAC,CAAA;AAI/D;;;GAGG;AACH,eAAO,MAAM,YAAY,UAAU,CAAA;AAEnC;;;GAGG;AACH,wBAAgB,UAAU,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,CAQjD"}
|
|
@@ -0,0 +1,60 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* HTTP transport for the wire protocol.
|
|
3
|
+
*
|
|
4
|
+
* Hono + @hono/node-server. Every endpoint:
|
|
5
|
+
* 1. Validates the request against its Zod schema.
|
|
6
|
+
* 2. Calls the matching handler in `handlers.ts`.
|
|
7
|
+
* 3. Renders 4xx for `WireError` with structured body, 500 for unexpected.
|
|
8
|
+
*
|
|
9
|
+
* The server holds optional `IngestionStores` (passed to `createApp`)
|
|
10
|
+
* to receive production traces and user feedback. With no stores wired,
|
|
11
|
+
* the ingestion endpoints return 503 — read endpoints (`/v1/judge`,
|
|
12
|
+
* `/v1/rubrics`, `/v1/version`) remain fully functional.
|
|
13
|
+
*
|
|
14
|
+
* Run via `agent-eval serve --port 5005`.
|
|
15
|
+
*/
|
|
16
|
+
import { type ServerType } from '@hono/node-server';
|
|
17
|
+
import { Hono } from 'hono';
|
|
18
|
+
import { type IngestionStores } from './handlers';
|
|
19
|
+
export interface CreateAppOptions {
|
|
20
|
+
/** Stores wired to the ingestion endpoints. */
|
|
21
|
+
stores?: IngestionStores;
|
|
22
|
+
/**
|
|
23
|
+
* Bearer-token auth. When provided, every endpoint EXCEPT `/healthz`
|
|
24
|
+
* and `/v1/version` requires `Authorization: Bearer <token>`. The
|
|
25
|
+
* token may be a static string OR a function for time-bounded /
|
|
26
|
+
* rotating tokens.
|
|
27
|
+
*
|
|
28
|
+
* Recommended for any server that accepts ingestion writes from the
|
|
29
|
+
* public internet. Read-only deployments may omit it.
|
|
30
|
+
*/
|
|
31
|
+
auth?: {
|
|
32
|
+
bearer: string | ((token: string) => boolean | Promise<boolean>);
|
|
33
|
+
};
|
|
34
|
+
}
|
|
35
|
+
export declare function createApp(opts?: CreateAppOptions): Hono<import("hono/types").BlankEnv, import("hono/types").BlankSchema, "/">;
|
|
36
|
+
export interface ServeOptions extends CreateAppOptions {
|
|
37
|
+
/** Default 5005. */
|
|
38
|
+
port?: number;
|
|
39
|
+
/** Default '127.0.0.1'. Set to '0.0.0.0' to listen on all interfaces. */
|
|
40
|
+
host?: string;
|
|
41
|
+
}
|
|
42
|
+
export declare function startServer(opts?: ServeOptions): ServerType;
|
|
43
|
+
export interface StartedServer {
|
|
44
|
+
server: ServerType;
|
|
45
|
+
/** The OS-assigned port. When opts.port was 0, this is the actual port the
|
|
46
|
+
* kernel bound — callers that need to dial back (smoke tests, sidecars
|
|
47
|
+
* registering with a parent) read this rather than guessing a free port. */
|
|
48
|
+
port: number;
|
|
49
|
+
/** Resolved host the server bound to (defaults to 127.0.0.1). */
|
|
50
|
+
host: string;
|
|
51
|
+
/** Close the server. Resolves once active connections have drained. */
|
|
52
|
+
close(): Promise<void>;
|
|
53
|
+
}
|
|
54
|
+
/**
|
|
55
|
+
* Promise-returning variant of `startServer` that resolves once the server is
|
|
56
|
+
* listening and surfaces the resolved bound port. Use this from smoke tests
|
|
57
|
+
* (`startServerAsync({ port: 0 })`) and any caller that needs to dial back.
|
|
58
|
+
*/
|
|
59
|
+
export declare function startServerAsync(opts?: ServeOptions): Promise<StartedServer>;
|
|
60
|
+
//# sourceMappingURL=server.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/wire/server.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AACH,OAAO,EAAE,KAAK,UAAU,EAAS,MAAM,mBAAmB,CAAA;AAC1D,OAAO,EAAE,IAAI,EAAE,MAAM,MAAM,CAAA;AAG3B,OAAO,EAML,KAAK,eAAe,EAErB,MAAM,YAAY,CAAA;AAMnB,MAAM,WAAW,gBAAgB;IAC/B,+CAA+C;IAC/C,MAAM,CAAC,EAAE,eAAe,CAAA;IACxB;;;;;;;;OAQG;IACH,IAAI,CAAC,EAAE;QACL,MAAM,EAAE,MAAM,GAAG,CAAC,CAAC,KAAK,EAAE,MAAM,KAAK,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC,CAAA;KACjE,CAAA;CACF;AAID,wBAAgB,SAAS,CAAC,IAAI,GAAE,gBAAqB,8EAqIpD;AAED,MAAM,WAAW,YAAa,SAAQ,gBAAgB;IACpD,oBAAoB;IACpB,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,yEAAyE;IACzE,IAAI,CAAC,EAAE,MAAM,CAAA;CACd;AAED,wBAAgB,WAAW,CAAC,IAAI,GAAE,YAAiB,GAAG,UAAU,CAQ/D;AAED,MAAM,WAAW,aAAa;IAC5B,MAAM,EAAE,UAAU,CAAA;IAClB;;iFAE6E;IAC7E,IAAI,EAAE,MAAM,CAAA;IACZ,iEAAiE;IACjE,IAAI,EAAE,MAAM,CAAA;IACZ,uEAAuE;IACvE,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAAA;CACvB;AAED;;;;GAIG;AACH,wBAAgB,gBAAgB,CAAC,IAAI,GAAE,YAAiB,GAAG,OAAO,CAAC,aAAa,CAAC,CA4BhF"}
|
|
@@ -0,0 +1,5 @@
|
|
|
1
|
+
import type { WorkflowTraceEventKind } from './types';
|
|
2
|
+
export declare const WORKFLOW_TRACE_EVENT_KINDS: readonly WorkflowTraceEventKind[];
|
|
3
|
+
export declare function validateWorkflowTraceEventKind(kind: string): WorkflowTraceEventKind;
|
|
4
|
+
export declare function validateWorkflowTraceEventPayload(kind: WorkflowTraceEventKind, payload: Record<string, unknown>): Record<string, unknown>;
|
|
5
|
+
//# sourceMappingURL=event-schema.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"event-schema.d.ts","sourceRoot":"","sources":["../../src/workflow/event-schema.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,SAAS,CAAA;AAgCrD,eAAO,MAAM,0BAA0B,EAAE,SAAS,sBAAsB,EAAyB,CAAA;AAKjG,wBAAgB,8BAA8B,CAAC,IAAI,EAAE,MAAM,GAAG,sBAAsB,CAKnF;AAED,wBAAgB,iCAAiC,CAC/C,IAAI,EAAE,sBAAsB,EAC5B,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAC/B,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CA2FzB"}
|
|
@@ -0,0 +1,99 @@
|
|
|
1
|
+
import type { AnalystFinding, AnalystSeverity, EvidenceRef } from '../analyst/types';
|
|
2
|
+
import type { FailureClusterInsight } from '../contract/insight-report';
|
|
3
|
+
import type { LayerResult, VerificationReport } from '../multi-layer-verifier';
|
|
4
|
+
import type { FailureClusterReport } from '../pipelines/failure-cluster';
|
|
5
|
+
import type { WorkflowTraceEnvelope, WorkflowTraceSummary } from './types';
|
|
6
|
+
export type WorkflowFeedbackPackVersion = 'workflow-feedback-pack-v1';
|
|
7
|
+
export type WorkflowFeedbackSeverity = AnalystSeverity;
|
|
8
|
+
export interface WorkflowToolUsageSummary {
|
|
9
|
+
totalCalls: number;
|
|
10
|
+
erroredCalls: number;
|
|
11
|
+
byTool: Record<string, {
|
|
12
|
+
calls: number;
|
|
13
|
+
errors: number;
|
|
14
|
+
}>;
|
|
15
|
+
}
|
|
16
|
+
export interface WorkflowVerifierFindingSummary {
|
|
17
|
+
severity: WorkflowFeedbackSeverity;
|
|
18
|
+
message: string;
|
|
19
|
+
evidence?: string;
|
|
20
|
+
detail?: Record<string, unknown>;
|
|
21
|
+
}
|
|
22
|
+
export interface WorkflowVerifierLayerSummary {
|
|
23
|
+
layer: string;
|
|
24
|
+
status: LayerResult['status'];
|
|
25
|
+
score?: number;
|
|
26
|
+
durationMs: number;
|
|
27
|
+
reason?: string;
|
|
28
|
+
findings: WorkflowVerifierFindingSummary[];
|
|
29
|
+
diagnostics?: Record<string, number | null>;
|
|
30
|
+
}
|
|
31
|
+
export interface WorkflowVerifierSummary {
|
|
32
|
+
allPass: boolean;
|
|
33
|
+
blendedScore: number;
|
|
34
|
+
durationMs: number;
|
|
35
|
+
failedLayers: string[];
|
|
36
|
+
layers: WorkflowVerifierLayerSummary[];
|
|
37
|
+
}
|
|
38
|
+
export interface WorkflowFailureClusterInput {
|
|
39
|
+
id: string;
|
|
40
|
+
name: string;
|
|
41
|
+
share?: number;
|
|
42
|
+
runCount?: number;
|
|
43
|
+
exemplars?: readonly string[];
|
|
44
|
+
suggestedFix?: string;
|
|
45
|
+
metadata?: Record<string, unknown>;
|
|
46
|
+
}
|
|
47
|
+
export interface WorkflowFailureClusterSummary {
|
|
48
|
+
id: string;
|
|
49
|
+
name: string;
|
|
50
|
+
share: number;
|
|
51
|
+
runCount?: number;
|
|
52
|
+
exemplars: string[];
|
|
53
|
+
suggestedFix?: string;
|
|
54
|
+
source: 'failure-cluster-view' | 'insight-report' | 'custom';
|
|
55
|
+
metadata?: Record<string, unknown>;
|
|
56
|
+
}
|
|
57
|
+
export interface WorkflowAnalystFindingSummary {
|
|
58
|
+
findingId: string;
|
|
59
|
+
analystId: string;
|
|
60
|
+
severity: WorkflowFeedbackSeverity;
|
|
61
|
+
area: string;
|
|
62
|
+
claim: string;
|
|
63
|
+
confidence: number;
|
|
64
|
+
subject?: string;
|
|
65
|
+
recommendedAction?: string;
|
|
66
|
+
evidenceRefs: EvidenceRef[];
|
|
67
|
+
}
|
|
68
|
+
export interface WorkflowAnalystFeedbackPack {
|
|
69
|
+
schemaVersion: WorkflowFeedbackPackVersion;
|
|
70
|
+
runId: string;
|
|
71
|
+
generatedAt: string;
|
|
72
|
+
summary: WorkflowTraceSummary;
|
|
73
|
+
verifier?: WorkflowVerifierSummary;
|
|
74
|
+
toolUsage: WorkflowToolUsageSummary;
|
|
75
|
+
failureClusters: WorkflowFailureClusterSummary[];
|
|
76
|
+
findings: WorkflowAnalystFindingSummary[];
|
|
77
|
+
recommendations: string[];
|
|
78
|
+
driverContextLines: string[];
|
|
79
|
+
}
|
|
80
|
+
export interface WorkflowFeedbackPackLimits {
|
|
81
|
+
findings?: number;
|
|
82
|
+
clusters?: number;
|
|
83
|
+
layerFindings?: number;
|
|
84
|
+
recommendations?: number;
|
|
85
|
+
contextLines?: number;
|
|
86
|
+
}
|
|
87
|
+
export interface BuildWorkflowAnalystFeedbackPackOptions {
|
|
88
|
+
envelope: WorkflowTraceEnvelope | unknown;
|
|
89
|
+
verifier?: VerificationReport;
|
|
90
|
+
analystFindings?: readonly AnalystFinding[];
|
|
91
|
+
failureClusters?: FailureClusterReport | FailureClusterInsight | readonly WorkflowFailureClusterInput[];
|
|
92
|
+
generatedAt?: string;
|
|
93
|
+
limits?: WorkflowFeedbackPackLimits;
|
|
94
|
+
}
|
|
95
|
+
export declare function buildWorkflowAnalystFeedbackPack(options: BuildWorkflowAnalystFeedbackPackOptions): WorkflowAnalystFeedbackPack;
|
|
96
|
+
export declare function renderWorkflowFeedbackPack(pack: WorkflowAnalystFeedbackPack, options?: {
|
|
97
|
+
maxChars?: number;
|
|
98
|
+
}): string;
|
|
99
|
+
//# sourceMappingURL=feedback-pack.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"feedback-pack.d.ts","sourceRoot":"","sources":["../../src/workflow/feedback-pack.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,eAAe,EAAE,WAAW,EAAE,MAAM,kBAAkB,CAAA;AACpF,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,4BAA4B,CAAA;AACvE,OAAO,KAAK,EAAE,WAAW,EAAE,kBAAkB,EAAE,MAAM,yBAAyB,CAAA;AAC9E,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,8BAA8B,CAAA;AAExE,OAAO,KAAK,EAAE,qBAAqB,EAAE,oBAAoB,EAAE,MAAM,SAAS,CAAA;AAE1E,MAAM,MAAM,2BAA2B,GAAG,2BAA2B,CAAA;AAErE,MAAM,MAAM,wBAAwB,GAAG,eAAe,CAAA;AAEtD,MAAM,WAAW,wBAAwB;IACvC,UAAU,EAAE,MAAM,CAAA;IAClB,YAAY,EAAE,MAAM,CAAA;IACpB,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;CAC1D;AAED,MAAM,WAAW,8BAA8B;IAC7C,QAAQ,EAAE,wBAAwB,CAAA;IAClC,OAAO,EAAE,MAAM,CAAA;IACf,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACjC;AAED,MAAM,WAAW,4BAA4B;IAC3C,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,WAAW,CAAC,QAAQ,CAAC,CAAA;IAC7B,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,UAAU,EAAE,MAAM,CAAA;IAClB,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,QAAQ,EAAE,8BAA8B,EAAE,CAAA;IAC1C,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC,CAAA;CAC5C;AAED,MAAM,WAAW,uBAAuB;IACtC,OAAO,EAAE,OAAO,CAAA;IAChB,YAAY,EAAE,MAAM,CAAA;IACpB,UAAU,EAAE,MAAM,CAAA;IAClB,YAAY,EAAE,MAAM,EAAE,CAAA;IACtB,MAAM,EAAE,4BAA4B,EAAE,CAAA;CACvC;AAED,MAAM,WAAW,2BAA2B;IAC1C,EAAE,EAAE,MAAM,CAAA;IACV,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,SAAS,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;IAC7B,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,6BAA6B;IAC5C,EAAE,EAAE,MAAM,CAAA;IACV,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,MAAM,CAAA;IACb,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,SAAS,EAAE,MAAM,EAAE,CAAA;IACnB,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,MAAM,EAAE,sBAAsB,GAAG,gBAAgB,GAAG,QAAQ,CAAA;IAC5D,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,6BAA6B;IAC5C,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,EAAE,MAAM,CAAA;IACjB,QAAQ,EAAE,wBAAwB,CAAA;IAClC,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,EAAE,MAAM,CAAA;IAClB,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,iBAAiB,CAAC,EAAE,MAAM,CAAA;IAC1B,YAAY,EAAE,WAAW,EAAE,CAAA;CAC5B;AAED,MAAM,WAAW,2BAA2B;IAC1C,aAAa,EAAE,2BAA2B,CAAA;IAC1C,KAAK,EAAE,MAAM,CAAA;IACb,WAAW,EAAE,MAAM,CAAA;IACnB,OAAO,EAAE,oBAAoB,CAAA;IAC7B,QAAQ,CAAC,EAAE,uBAAuB,CAAA;IAClC,SAAS,EAAE,wBAAwB,CAAA;IACnC,eAAe,EAAE,6BAA6B,EAAE,CAAA;IAChD,QAAQ,EAAE,6BAA6B,EAAE,CAAA;IACzC,eAAe,EAAE,MAAM,EAAE,CAAA;IACzB,kBAAkB,EAAE,MAAM,EAAE,CAAA;CAC7B;AAED,MAAM,WAAW,0BAA0B;IACzC,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,aAAa,CAAC,EAAE,MAAM,CAAA;IACtB,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,YAAY,CAAC,EAAE,MAAM,CAAA;CACtB;AAED,MAAM,WAAW,uCAAuC;IACtD,QAAQ,EAAE,qBAAqB,GAAG,OAAO,CAAA;IACzC,QAAQ,CAAC,EAAE,kBAAkB,CAAA;IAC7B,eAAe,CAAC,EAAE,SAAS,cAAc,EAAE,CAAA;IAC3C,eAAe,CAAC,EACZ,oBAAoB,GACpB,qBAAqB,GACrB,SAAS,2BAA2B,EAAE,CAAA;IAC1C,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,MAAM,CAAC,EAAE,0BAA0B,CAAA;CACpC;AAYD,wBAAgB,gCAAgC,CAC9C,OAAO,EAAE,uCAAuC,GAC/C,2BAA2B,CAuC7B;AAED,wBAAgB,0BAA0B,CACxC,IAAI,EAAE,2BAA2B,EACjC,OAAO,GAAE;IAAE,QAAQ,CAAC,EAAE,MAAM,CAAA;CAAO,GAClC,MAAM,CA6BR"}
|