@tangle-network/agent-eval 0.95.0 → 0.96.0
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 +17 -0
- package/dist/adapters/http.d.ts +17 -10
- package/dist/adapters/langchain.d.ts +14 -7
- package/dist/adapters/otel.d.ts +25 -13
- package/dist/{rl/adversarial.d.ts → adversarial-DIVcDoI_.d.ts} +7 -6
- package/dist/analyst/index.d.ts +236 -28
- package/dist/analyst/index.js +1 -1
- package/dist/{trace-analyst/analyst.d.ts → analyst-C8HHvfJp.d.ts} +14 -12
- package/dist/{contract/analyze-runs.d.ts → analyze-runs-DtT6F_6T.d.ts} +10 -9
- package/dist/authenticity/index.d.ts +16 -15
- package/dist/{baseline.d.ts → baseline-Bbid3WoO.d.ts} +44 -8
- package/dist/belief-state/index.d.ts +605 -14
- package/dist/benchmarks/index.d.ts +5 -23
- package/dist/builder-eval/index.d.ts +250 -5
- package/dist/calibration-BPmzuVPk.d.ts +101 -0
- package/dist/campaign/index.d.ts +1452 -38
- package/dist/{chunk-AQ5WQAIV.js → chunk-3NHEO6ZC.js} +2 -2
- package/dist/chunk-3NHEO6ZC.js.map +1 -0
- package/dist/cli.d.ts +0 -2
- package/dist/{contract/intake/code-agent-session.d.ts → code-agent-session-CPHRCb4-.d.ts} +17 -15
- package/dist/contract/index.d.ts +824 -107
- package/dist/contract/index.js +145 -1
- package/dist/contract/index.js.map +1 -1
- package/dist/control-Doncu-B_.d.ts +259 -0
- package/dist/{control-runtime.d.ts → control-runtime-Acf9CGhw.d.ts} +26 -23
- package/dist/control.d.ts +10 -11
- package/dist/corpus-D4YW9UoJ.d.ts +560 -0
- package/dist/{cost-ledger.d.ts → cost-ledger-DuSqlw5B.d.ts} +11 -10
- package/dist/{counterfactual.d.ts → counterfactual-DlOz8PBx.d.ts} +13 -12
- package/dist/{dataset.d.ts → dataset-BbGkaN2I.d.ts} +14 -11
- package/dist/{analyst/registry.d.ts → default-registry-GyE8X5SP.d.ts} +37 -8
- package/dist/diagnose.d.ts +252 -1
- package/dist/{trace/emitter.d.ts → emitter-C2rqGH_l.d.ts} +12 -9
- package/dist/{errors.d.ts → errors-CzMUYo7b.d.ts} +11 -10
- package/dist/failure-cluster-DH9Flgcf.d.ts +76 -0
- package/dist/{feedback-trajectory.d.ts → feedback-trajectory-BxY0cKfs.d.ts} +38 -36
- package/dist/fuzz.d.ts +547 -1
- package/dist/gepa-Dprxvz8r.d.ts +414 -0
- package/dist/governance/index.d.ts +135 -5
- package/dist/harness-optimizer-mOl9XX_O.d.ts +106 -0
- package/dist/hosted/index.d.ts +239 -10
- package/dist/index-_Y4oNOOb.d.ts +159 -0
- package/dist/index.d.ts +5660 -277
- package/dist/index.js +1 -1
- package/dist/{contract/insight-report.d.ts → insight-report-BnRjTibG.d.ts} +19 -16
- package/dist/{trace/integrity.d.ts → integrity-D2t12mMw.d.ts} +14 -11
- package/dist/{judge-calibration.d.ts → judge-calibration-0p2QcWNE.d.ts} +17 -16
- package/dist/{analyst/kind-factory.d.ts → kind-factory-X3eDYbKn.d.ts} +61 -10
- package/dist/knowledge/index.d.ts +103 -3
- package/dist/{llm-client.d.ts → llm-client-Bj7g0rqu.d.ts} +23 -21
- package/dist/matrix/index.d.ts +30 -12
- package/dist/meta-eval/index.d.ts +182 -6
- package/dist/{multi-layer-verifier.d.ts → multi-layer-verifier-DUZXrPDA.d.ts} +15 -12
- package/dist/multishot/index.d.ts +290 -7
- package/dist/{rl/off-policy.d.ts → off-policy-DiwuKKg7.d.ts} +9 -8
- package/dist/openapi.json +1 -1
- package/dist/{meta-eval/outcome-store.d.ts → outcome-store-rnXLEqSn.d.ts} +8 -7
- package/dist/{pareto.d.ts → pareto-E-pembql.d.ts} +10 -9
- package/dist/perf/index.d.ts +119 -13
- package/dist/pipelines/index.d.ts +173 -8
- package/dist/pre-registration-nfUdc9EQ.d.ts +483 -0
- package/dist/prm/index.d.ts +104 -5
- package/dist/provenance-CIxfBnkl.d.ts +426 -0
- package/dist/query-B7GGjRox.d.ts +32 -0
- package/dist/{trace/raw-provider-sink.d.ts → raw-provider-sink-C46HDghv.d.ts} +14 -13
- package/dist/{red-team.d.ts → red-team-BWdoyleI.d.ts} +16 -14
- package/dist/{trace/redact.d.ts → redact-B40YG2M_.d.ts} +8 -7
- package/dist/release-report-pidWUMZ2.d.ts +233 -0
- package/dist/reporting.d.ts +16 -15
- package/dist/{eval-campaign.d.ts → researcher-Jr8ME1dZ.d.ts} +156 -34
- package/dist/rl.d.ts +1193 -1
- package/dist/{prm/rubric.d.ts → rubric-Cc6UHvUb.d.ts} +13 -10
- package/dist/{meta-eval/rubric-predictive-validity.d.ts → rubric-predictive-validity-C2hDKM8Z.d.ts} +11 -8
- package/dist/run-critic-CmMf05uV.d.ts +56 -0
- package/dist/{run-record.d.ts → run-record-CP2ObebC.d.ts} +117 -15
- package/dist/runtime-trajectory-BOUUjI0y.d.ts +49 -0
- package/dist/{trace/schema.d.ts → schema-m0gsnbt3.d.ts} +30 -29
- package/dist/semantic-concept-judge-DSBB2Cfp.d.ts +624 -0
- package/dist/{sequential.d.ts → sequential-5iSVfzl2.d.ts} +10 -9
- package/dist/{series-convergence.d.ts → series-convergence-D5OWMBg6.d.ts} +5 -4
- package/dist/sink-fetch-B1Yg4Til.d.ts +101 -0
- package/dist/{statistics.d.ts → statistics-CCJpTGOS.d.ts} +48 -46
- package/dist/{trace/store.d.ts → store-BcFXE6LG.d.ts} +12 -21
- package/dist/{trace-analyst/types.d.ts → store-C1YxJDEK.d.ts} +74 -18
- package/dist/storyboard/index.d.ts +81 -16
- package/dist/{summary-report.d.ts → summary-report-CInXwsza.d.ts} +160 -23
- package/dist/telemetry/{sink-file.d.ts → file.d.ts} +7 -5
- package/dist/telemetry/index.d.ts +35 -17
- package/dist/{sandbox-harness.d.ts → test-graded-scenario-DeODGLra.d.ts} +60 -15
- package/dist/{locked-jsonl-appender.d.ts → testing-C21CHsq2.d.ts} +4 -3
- package/dist/testing.d.ts +1 -5
- package/dist/traces.d.ts +976 -4
- package/dist/{trajectory.d.ts → trajectory-2TkpSEVh.d.ts} +9 -8
- package/dist/{analyst/types.d.ts → types-B5x54y6n.d.ts} +112 -19
- package/dist/{campaign/types.d.ts → types-BMahhhio.d.ts} +47 -44
- package/dist/{matrix/types.d.ts → types-BUxNaJ8c.d.ts} +11 -9
- package/dist/{types.d.ts → types-C7DGg5ex.d.ts} +33 -31
- package/dist/{verdict.d.ts → verdict-C9MlYujm.d.ts} +3 -2
- package/dist/wire/index.d.ts +570 -13
- package/dist/workflow/index.d.ts +496 -22
- package/package.json +2 -2
- package/dist/action-policy.d.ts +0 -24
- package/dist/action-policy.d.ts.map +0 -1
- package/dist/action-policy.test.d.ts +0 -2
- package/dist/action-policy.test.d.ts.map +0 -1
- package/dist/active-learning.d.ts +0 -41
- package/dist/active-learning.d.ts.map +0 -1
- package/dist/adapters/http.d.ts.map +0 -1
- package/dist/adapters/langchain.d.ts.map +0 -1
- package/dist/adapters/otel.d.ts.map +0 -1
- package/dist/agent-profile-cell.d.ts +0 -101
- package/dist/agent-profile-cell.d.ts.map +0 -1
- package/dist/agent-profile.d.ts +0 -27
- package/dist/agent-profile.d.ts.map +0 -1
- package/dist/agent-profile.test.d.ts +0 -2
- package/dist/agent-profile.test.d.ts.map +0 -1
- package/dist/analyst/adapters.d.ts +0 -62
- package/dist/analyst/adapters.d.ts.map +0 -1
- package/dist/analyst/analyst.test.d.ts +0 -2
- package/dist/analyst/analyst.test.d.ts.map +0 -1
- package/dist/analyst/ax-service.d.ts +0 -27
- package/dist/analyst/ax-service.d.ts.map +0 -1
- package/dist/analyst/behavioral-analyst.d.ts +0 -28
- package/dist/analyst/behavioral-analyst.d.ts.map +0 -1
- package/dist/analyst/chat-client.d.ts +0 -91
- package/dist/analyst/chat-client.d.ts.map +0 -1
- package/dist/analyst/default-registry.d.ts +0 -27
- package/dist/analyst/default-registry.d.ts.map +0 -1
- package/dist/analyst/default-registry.test.d.ts +0 -2
- package/dist/analyst/default-registry.test.d.ts.map +0 -1
- package/dist/analyst/finding-signature.d.ts +0 -48
- package/dist/analyst/finding-signature.d.ts.map +0 -1
- package/dist/analyst/finding-subject.d.ts +0 -146
- package/dist/analyst/finding-subject.d.ts.map +0 -1
- package/dist/analyst/finding-subject.test.d.ts +0 -2
- package/dist/analyst/finding-subject.test.d.ts.map +0 -1
- package/dist/analyst/findings-store.d.ts +0 -75
- package/dist/analyst/findings-store.d.ts.map +0 -1
- package/dist/analyst/index.d.ts.map +0 -1
- package/dist/analyst/kind-factory.d.ts.map +0 -1
- package/dist/analyst/kinds/failure-mode.d.ts +0 -19
- package/dist/analyst/kinds/failure-mode.d.ts.map +0 -1
- package/dist/analyst/kinds/improvement.d.ts +0 -23
- package/dist/analyst/kinds/improvement.d.ts.map +0 -1
- package/dist/analyst/kinds/index.d.ts +0 -22
- package/dist/analyst/kinds/index.d.ts.map +0 -1
- package/dist/analyst/kinds/kinds.test.d.ts +0 -2
- package/dist/analyst/kinds/kinds.test.d.ts.map +0 -1
- package/dist/analyst/kinds/knowledge-gap.d.ts +0 -28
- package/dist/analyst/kinds/knowledge-gap.d.ts.map +0 -1
- package/dist/analyst/kinds/knowledge-poisoning.d.ts +0 -22
- package/dist/analyst/kinds/knowledge-poisoning.d.ts.map +0 -1
- package/dist/analyst/kinds/skill-usage.d.ts +0 -84
- package/dist/analyst/kinds/skill-usage.d.ts.map +0 -1
- package/dist/analyst/kinds/skill-usage.test.d.ts +0 -2
- package/dist/analyst/kinds/skill-usage.test.d.ts.map +0 -1
- package/dist/analyst/parse-tolerant.d.ts +0 -26
- package/dist/analyst/parse-tolerant.d.ts.map +0 -1
- package/dist/analyst/parse-tolerant.test.d.ts +0 -2
- package/dist/analyst/parse-tolerant.test.d.ts.map +0 -1
- package/dist/analyst/registry.budget.test.d.ts +0 -2
- package/dist/analyst/registry.budget.test.d.ts.map +0 -1
- package/dist/analyst/registry.d.ts.map +0 -1
- package/dist/analyst/steer-firewall.d.ts +0 -35
- package/dist/analyst/steer-firewall.d.ts.map +0 -1
- package/dist/analyst/steer-firewall.test.d.ts +0 -2
- package/dist/analyst/steer-firewall.test.d.ts.map +0 -1
- package/dist/analyst/structure-findings.d.ts +0 -37
- package/dist/analyst/structure-findings.d.ts.map +0 -1
- package/dist/analyst/structure-findings.test.d.ts +0 -2
- package/dist/analyst/structure-findings.test.d.ts.map +0 -1
- package/dist/analyst/tool-groups.d.ts +0 -34
- package/dist/analyst/tool-groups.d.ts.map +0 -1
- package/dist/analyst/types.d.ts.map +0 -1
- package/dist/anti-slop.d.ts +0 -59
- package/dist/anti-slop.d.ts.map +0 -1
- package/dist/artifact-validator.d.ts +0 -74
- package/dist/artifact-validator.d.ts.map +0 -1
- package/dist/attestation.d.ts +0 -63
- package/dist/attestation.d.ts.map +0 -1
- package/dist/attestation.test.d.ts +0 -2
- package/dist/attestation.test.d.ts.map +0 -1
- package/dist/authenticity/index.d.ts.map +0 -1
- package/dist/authenticity/index.test.d.ts +0 -2
- package/dist/authenticity/index.test.d.ts.map +0 -1
- package/dist/auto-pr.d.ts +0 -120
- package/dist/auto-pr.d.ts.map +0 -1
- package/dist/baseline.d.ts.map +0 -1
- package/dist/behavior-dsl.d.ts +0 -73
- package/dist/behavior-dsl.d.ts.map +0 -1
- package/dist/belief-state/calibration.d.ts +0 -10
- package/dist/belief-state/calibration.d.ts.map +0 -1
- package/dist/belief-state/calibration.test.d.ts +0 -2
- package/dist/belief-state/calibration.test.d.ts.map +0 -1
- package/dist/belief-state/code-agent-corpus.d.ts +0 -66
- package/dist/belief-state/code-agent-corpus.d.ts.map +0 -1
- package/dist/belief-state/code-agent-corpus.test.d.ts +0 -2
- package/dist/belief-state/code-agent-corpus.test.d.ts.map +0 -1
- package/dist/belief-state/code-agent-evidence.d.ts +0 -22
- package/dist/belief-state/code-agent-evidence.d.ts.map +0 -1
- package/dist/belief-state/code-agent-evidence.test.d.ts +0 -2
- package/dist/belief-state/code-agent-evidence.test.d.ts.map +0 -1
- package/dist/belief-state/extract.d.ts +0 -7
- package/dist/belief-state/extract.d.ts.map +0 -1
- package/dist/belief-state/extract.test.d.ts +0 -2
- package/dist/belief-state/extract.test.d.ts.map +0 -1
- package/dist/belief-state/index.d.ts.map +0 -1
- package/dist/belief-state/ope.d.ts +0 -17
- package/dist/belief-state/ope.d.ts.map +0 -1
- package/dist/belief-state/ope.test.d.ts +0 -2
- package/dist/belief-state/ope.test.d.ts.map +0 -1
- package/dist/belief-state/phase0-measurement.d.ts +0 -55
- package/dist/belief-state/phase0-measurement.d.ts.map +0 -1
- package/dist/belief-state/report.d.ts +0 -17
- package/dist/belief-state/report.d.ts.map +0 -1
- package/dist/belief-state/report.test.d.ts +0 -2
- package/dist/belief-state/report.test.d.ts.map +0 -1
- package/dist/belief-state/research-evidence.d.ts +0 -23
- package/dist/belief-state/research-evidence.d.ts.map +0 -1
- package/dist/belief-state/research-evidence.test.d.ts +0 -2
- package/dist/belief-state/research-evidence.test.d.ts.map +0 -1
- package/dist/belief-state/runtime-benchmark-corpus.d.ts +0 -32
- package/dist/belief-state/runtime-benchmark-corpus.d.ts.map +0 -1
- package/dist/belief-state/runtime-hooks.d.ts +0 -87
- package/dist/belief-state/runtime-hooks.d.ts.map +0 -1
- package/dist/belief-state/runtime-hooks.test.d.ts +0 -2
- package/dist/belief-state/runtime-hooks.test.d.ts.map +0 -1
- package/dist/belief-state/selective.d.ts +0 -15
- package/dist/belief-state/selective.d.ts.map +0 -1
- package/dist/belief-state/selective.test.d.ts +0 -2
- package/dist/belief-state/selective.test.d.ts.map +0 -1
- package/dist/belief-state/shadow-probe.d.ts +0 -80
- package/dist/belief-state/shadow-probe.d.ts.map +0 -1
- package/dist/belief-state/shadow-probe.test.d.ts +0 -2
- package/dist/belief-state/shadow-probe.test.d.ts.map +0 -1
- package/dist/belief-state/types.d.ts +0 -195
- package/dist/belief-state/types.d.ts.map +0 -1
- package/dist/belief-state/types.test.d.ts +0 -2
- package/dist/belief-state/types.test.d.ts.map +0 -1
- package/dist/benchmark.d.ts +0 -14
- package/dist/benchmark.d.ts.map +0 -1
- package/dist/benchmarks/index.d.ts.map +0 -1
- package/dist/benchmarks/routing/dataset.d.ts +0 -34
- package/dist/benchmarks/routing/dataset.d.ts.map +0 -1
- package/dist/benchmarks/routing/index.d.ts +0 -34
- package/dist/benchmarks/routing/index.d.ts.map +0 -1
- package/dist/benchmarks/types.d.ts +0 -49
- package/dist/benchmarks/types.d.ts.map +0 -1
- package/dist/bisector.d.ts +0 -81
- package/dist/bisector.d.ts.map +0 -1
- package/dist/budget-guard.d.ts +0 -31
- package/dist/budget-guard.d.ts.map +0 -1
- package/dist/builder-eval/builder-session.d.ts +0 -111
- package/dist/builder-eval/builder-session.d.ts.map +0 -1
- package/dist/builder-eval/correlation.d.ts +0 -32
- package/dist/builder-eval/correlation.d.ts.map +0 -1
- package/dist/builder-eval/index.d.ts.map +0 -1
- package/dist/builder-eval/project-registry.d.ts +0 -51
- package/dist/builder-eval/project-registry.d.ts.map +0 -1
- package/dist/builder-eval/three-layer-eval.d.ts +0 -55
- package/dist/builder-eval/three-layer-eval.d.ts.map +0 -1
- package/dist/campaign/analyst-surface.d.ts +0 -108
- package/dist/campaign/analyst-surface.d.ts.map +0 -1
- package/dist/campaign/analyst-surface.test.d.ts +0 -2
- package/dist/campaign/analyst-surface.test.d.ts.map +0 -1
- package/dist/campaign/auto-pr.d.ts +0 -46
- package/dist/campaign/auto-pr.d.ts.map +0 -1
- package/dist/campaign/distillation/agreement-judge.d.ts +0 -69
- package/dist/campaign/distillation/agreement-judge.d.ts.map +0 -1
- package/dist/campaign/distillation/cli.d.ts +0 -35
- package/dist/campaign/distillation/cli.d.ts.map +0 -1
- package/dist/campaign/distillation/distillation.test.d.ts +0 -2
- package/dist/campaign/distillation/distillation.test.d.ts.map +0 -1
- package/dist/campaign/distillation/gold-scenarios.d.ts +0 -54
- package/dist/campaign/distillation/gold-scenarios.d.ts.map +0 -1
- package/dist/campaign/distillation/run-distillation.d.ts +0 -119
- package/dist/campaign/distillation/run-distillation.d.ts.map +0 -1
- package/dist/campaign/gates/compose.d.ts +0 -12
- package/dist/campaign/gates/compose.d.ts.map +0 -1
- package/dist/campaign/gates/default-production-gate.d.ts +0 -58
- package/dist/campaign/gates/default-production-gate.d.ts.map +0 -1
- package/dist/campaign/gates/heldout-gate.d.ts +0 -12
- package/dist/campaign/gates/heldout-gate.d.ts.map +0 -1
- package/dist/campaign/gates/promotion-policy.d.ts +0 -125
- package/dist/campaign/gates/promotion-policy.d.ts.map +0 -1
- package/dist/campaign/gates/promotion-policy.test.d.ts +0 -2
- package/dist/campaign/gates/promotion-policy.test.d.ts.map +0 -1
- package/dist/campaign/gates/sequential.d.ts +0 -146
- package/dist/campaign/gates/sequential.d.ts.map +0 -1
- package/dist/campaign/gates/sequential.test.d.ts +0 -2
- package/dist/campaign/gates/sequential.test.d.ts.map +0 -1
- package/dist/campaign/gates/statistical-heldout.d.ts +0 -99
- package/dist/campaign/gates/statistical-heldout.d.ts.map +0 -1
- package/dist/campaign/gates/statistical-heldout.test.d.ts +0 -2
- package/dist/campaign/gates/statistical-heldout.test.d.ts.map +0 -1
- package/dist/campaign/index.d.ts.map +0 -1
- package/dist/campaign/labeled-store/fs-adapter.d.ts +0 -59
- package/dist/campaign/labeled-store/fs-adapter.d.ts.map +0 -1
- package/dist/campaign/presets/compare-proposers.d.ts +0 -146
- package/dist/campaign/presets/compare-proposers.d.ts.map +0 -1
- package/dist/campaign/presets/playback.d.ts +0 -120
- package/dist/campaign/presets/playback.d.ts.map +0 -1
- package/dist/campaign/presets/playback.test.d.ts +0 -2
- package/dist/campaign/presets/playback.test.d.ts.map +0 -1
- package/dist/campaign/presets/run-eval.d.ts +0 -14
- package/dist/campaign/presets/run-eval.d.ts.map +0 -1
- package/dist/campaign/presets/run-improvement-loop.d.ts +0 -63
- package/dist/campaign/presets/run-improvement-loop.d.ts.map +0 -1
- package/dist/campaign/presets/run-improvement-loop.test.d.ts +0 -2
- package/dist/campaign/presets/run-improvement-loop.test.d.ts.map +0 -1
- package/dist/campaign/presets/run-optimization.d.ts +0 -92
- package/dist/campaign/presets/run-optimization.d.ts.map +0 -1
- package/dist/campaign/presets/run-profile-matrix.d.ts +0 -151
- package/dist/campaign/presets/run-profile-matrix.d.ts.map +0 -1
- package/dist/campaign/presets/run-skill-opt.d.ts +0 -96
- package/dist/campaign/presets/run-skill-opt.d.ts.map +0 -1
- package/dist/campaign/proposers/_findings-text.d.ts +0 -22
- package/dist/campaign/proposers/_findings-text.d.ts.map +0 -1
- package/dist/campaign/proposers/ace.d.ts +0 -33
- package/dist/campaign/proposers/ace.d.ts.map +0 -1
- package/dist/campaign/proposers/ace.test.d.ts +0 -2
- package/dist/campaign/proposers/ace.test.d.ts.map +0 -1
- package/dist/campaign/proposers/analysis-edit.d.ts +0 -32
- package/dist/campaign/proposers/analysis-edit.d.ts.map +0 -1
- package/dist/campaign/proposers/evolutionary.d.ts +0 -20
- package/dist/campaign/proposers/evolutionary.d.ts.map +0 -1
- package/dist/campaign/proposers/fapo.d.ts +0 -120
- package/dist/campaign/proposers/fapo.d.ts.map +0 -1
- package/dist/campaign/proposers/gepa.d.ts +0 -86
- package/dist/campaign/proposers/gepa.d.ts.map +0 -1
- package/dist/campaign/proposers/halo.d.ts +0 -44
- package/dist/campaign/proposers/halo.d.ts.map +0 -1
- package/dist/campaign/proposers/halo.test.d.ts +0 -2
- package/dist/campaign/proposers/halo.test.d.ts.map +0 -1
- package/dist/campaign/proposers/memory.d.ts +0 -47
- package/dist/campaign/proposers/memory.d.ts.map +0 -1
- package/dist/campaign/proposers/memory.test.d.ts +0 -2
- package/dist/campaign/proposers/memory.test.d.ts.map +0 -1
- package/dist/campaign/proposers/skill-opt.d.ts +0 -88
- package/dist/campaign/proposers/skill-opt.d.ts.map +0 -1
- package/dist/campaign/proposers/trace-analyst.d.ts +0 -48
- package/dist/campaign/proposers/trace-analyst.d.ts.map +0 -1
- package/dist/campaign/proposers/trace-analyst.test.d.ts +0 -2
- package/dist/campaign/proposers/trace-analyst.test.d.ts.map +0 -1
- package/dist/campaign/provenance.d.ts +0 -185
- package/dist/campaign/provenance.d.ts.map +0 -1
- package/dist/campaign/run-campaign.d.ts +0 -90
- package/dist/campaign/run-campaign.d.ts.map +0 -1
- package/dist/campaign/score-utils.d.ts +0 -26
- package/dist/campaign/score-utils.d.ts.map +0 -1
- package/dist/campaign/skill-patch.d.ts +0 -62
- package/dist/campaign/skill-patch.d.ts.map +0 -1
- package/dist/campaign/storage.d.ts +0 -38
- package/dist/campaign/storage.d.ts.map +0 -1
- package/dist/campaign/types.d.ts.map +0 -1
- package/dist/campaign/worktree/index.d.ts +0 -53
- package/dist/campaign/worktree/index.d.ts.map +0 -1
- package/dist/canary.d.ts +0 -101
- package/dist/canary.d.ts.map +0 -1
- package/dist/causal-attribution.d.ts +0 -45
- package/dist/causal-attribution.d.ts.map +0 -1
- package/dist/chunk-AQ5WQAIV.js.map +0 -1
- package/dist/ci-gate.d.ts +0 -44
- package/dist/ci-gate.d.ts.map +0 -1
- package/dist/cli.d.ts.map +0 -1
- package/dist/client.d.ts +0 -77
- package/dist/client.d.ts.map +0 -1
- package/dist/client.test.d.ts +0 -2
- package/dist/client.test.d.ts.map +0 -1
- package/dist/command-runner.d.ts +0 -74
- package/dist/command-runner.d.ts.map +0 -1
- package/dist/command-runner.test.d.ts +0 -2
- package/dist/command-runner.test.d.ts.map +0 -1
- package/dist/completion-verifier.d.ts +0 -147
- package/dist/completion-verifier.d.ts.map +0 -1
- package/dist/completion-verifier.test.d.ts +0 -9
- package/dist/completion-verifier.test.d.ts.map +0 -1
- package/dist/concurrency.d.ts +0 -23
- package/dist/concurrency.d.ts.map +0 -1
- package/dist/contamination-guard.d.ts +0 -81
- package/dist/contamination-guard.d.ts.map +0 -1
- package/dist/contract/analyze-runs.d.ts.map +0 -1
- package/dist/contract/define-agent-eval.d.ts +0 -52
- package/dist/contract/define-agent-eval.d.ts.map +0 -1
- package/dist/contract/diff.d.ts +0 -114
- package/dist/contract/diff.d.ts.map +0 -1
- package/dist/contract/index.d.ts.map +0 -1
- package/dist/contract/insight-report.d.ts.map +0 -1
- package/dist/contract/insight-types-fwd.d.ts +0 -7
- package/dist/contract/insight-types-fwd.d.ts.map +0 -1
- package/dist/contract/intake/agent-trace.d.ts +0 -97
- package/dist/contract/intake/agent-trace.d.ts.map +0 -1
- package/dist/contract/intake/code-agent-session.d.ts.map +0 -1
- package/dist/contract/intake/feedback-table.d.ts +0 -87
- package/dist/contract/intake/feedback-table.d.ts.map +0 -1
- package/dist/contract/intake/index.d.ts +0 -22
- package/dist/contract/intake/index.d.ts.map +0 -1
- package/dist/contract/intake/otel-spans.d.ts +0 -33
- package/dist/contract/intake/otel-spans.d.ts.map +0 -1
- package/dist/contract/self-improve.d.ts +0 -284
- package/dist/contract/self-improve.d.ts.map +0 -1
- package/dist/control-runtime.d.ts.map +0 -1
- package/dist/control-runtime.test.d.ts +0 -2
- package/dist/control-runtime.test.d.ts.map +0 -1
- package/dist/control.d.ts.map +0 -1
- package/dist/convergence.d.ts +0 -29
- package/dist/convergence.d.ts.map +0 -1
- package/dist/cost-ledger.d.ts.map +0 -1
- package/dist/cost-ledger.test.d.ts +0 -2
- package/dist/cost-ledger.test.d.ts.map +0 -1
- package/dist/cost-report.d.ts +0 -43
- package/dist/cost-report.d.ts.map +0 -1
- package/dist/cost-report.test.d.ts +0 -2
- package/dist/cost-report.test.d.ts.map +0 -1
- package/dist/cost-tracker.d.ts +0 -76
- package/dist/cost-tracker.d.ts.map +0 -1
- package/dist/counterfactual.d.ts.map +0 -1
- package/dist/cross-trace-diff.d.ts +0 -56
- package/dist/cross-trace-diff.d.ts.map +0 -1
- package/dist/dataset.d.ts.map +0 -1
- package/dist/deploy-gate-layer.d.ts +0 -125
- package/dist/deploy-gate-layer.d.ts.map +0 -1
- package/dist/deploy-gate-layer.test.d.ts +0 -2
- package/dist/deploy-gate-layer.test.d.ts.map +0 -1
- package/dist/description-length-gate.d.ts +0 -119
- package/dist/description-length-gate.d.ts.map +0 -1
- package/dist/detectors/edge.test.d.ts +0 -2
- package/dist/detectors/edge.test.d.ts.map +0 -1
- package/dist/detectors/index.d.ts +0 -81
- package/dist/detectors/index.d.ts.map +0 -1
- package/dist/detectors/index.test.d.ts +0 -2
- package/dist/detectors/index.test.d.ts.map +0 -1
- package/dist/diagnose/causal-sweep.d.ts +0 -100
- package/dist/diagnose/causal-sweep.d.ts.map +0 -1
- package/dist/diagnose/index.d.ts +0 -36
- package/dist/diagnose/index.d.ts.map +0 -1
- package/dist/diagnose/remediation.d.ts +0 -68
- package/dist/diagnose/remediation.d.ts.map +0 -1
- package/dist/diagnose/repair.d.ts +0 -77
- package/dist/diagnose/repair.d.ts.map +0 -1
- package/dist/discover-personas.d.ts +0 -35
- package/dist/discover-personas.d.ts.map +0 -1
- package/dist/driver.d.ts +0 -95
- package/dist/driver.d.ts.map +0 -1
- package/dist/driver.test.d.ts +0 -8
- package/dist/driver.test.d.ts.map +0 -1
- package/dist/dual-agent-bench.d.ts +0 -81
- package/dist/dual-agent-bench.d.ts.map +0 -1
- package/dist/error-count-extractor.d.ts +0 -47
- package/dist/error-count-extractor.d.ts.map +0 -1
- package/dist/error-count-extractor.test.d.ts +0 -2
- package/dist/error-count-extractor.test.d.ts.map +0 -1
- package/dist/errors.d.ts.map +0 -1
- package/dist/eval-campaign.d.ts.map +0 -1
- package/dist/eval-campaign.test.d.ts +0 -2
- package/dist/eval-campaign.test.d.ts.map +0 -1
- package/dist/eval-tools.d.ts +0 -55
- package/dist/eval-tools.d.ts.map +0 -1
- package/dist/eval-trace-store.d.ts +0 -107
- package/dist/eval-trace-store.d.ts.map +0 -1
- package/dist/eval-trace-store.test.d.ts +0 -2
- package/dist/eval-trace-store.test.d.ts.map +0 -1
- package/dist/executor.d.ts +0 -38
- package/dist/executor.d.ts.map +0 -1
- package/dist/executor.test.d.ts +0 -10
- package/dist/executor.test.d.ts.map +0 -1
- package/dist/experiment-tracker.d.ts +0 -178
- package/dist/experiment-tracker.d.ts.map +0 -1
- package/dist/experiment-tracker.test.d.ts +0 -2
- package/dist/experiment-tracker.test.d.ts.map +0 -1
- package/dist/failure-taxonomy.d.ts +0 -38
- package/dist/failure-taxonomy.d.ts.map +0 -1
- package/dist/feedback-trajectory.d.ts.map +0 -1
- package/dist/feedback-trajectory.test.d.ts +0 -2
- package/dist/feedback-trajectory.test.d.ts.map +0 -1
- package/dist/flow-layer.d.ts +0 -90
- package/dist/flow-layer.d.ts.map +0 -1
- package/dist/flow-layer.test.d.ts +0 -2
- package/dist/flow-layer.test.d.ts.map +0 -1
- package/dist/fuzz/capsule.d.ts +0 -46
- package/dist/fuzz/capsule.d.ts.map +0 -1
- package/dist/fuzz/cube.d.ts +0 -36
- package/dist/fuzz/cube.d.ts.map +0 -1
- package/dist/fuzz/explorer-cost.test.d.ts +0 -2
- package/dist/fuzz/explorer-cost.test.d.ts.map +0 -1
- package/dist/fuzz/explorer.d.ts +0 -64
- package/dist/fuzz/explorer.d.ts.map +0 -1
- package/dist/fuzz/fuzz-agent.d.ts +0 -16
- package/dist/fuzz/fuzz-agent.d.ts.map +0 -1
- package/dist/fuzz/fuzz-agent.test.d.ts +0 -2
- package/dist/fuzz/fuzz-agent.test.d.ts.map +0 -1
- package/dist/fuzz/gates.d.ts +0 -33
- package/dist/fuzz/gates.d.ts.map +0 -1
- package/dist/fuzz/index.d.ts +0 -26
- package/dist/fuzz/index.d.ts.map +0 -1
- package/dist/fuzz/policies.d.ts +0 -28
- package/dist/fuzz/policies.d.ts.map +0 -1
- package/dist/fuzz/tools.d.ts +0 -20
- package/dist/fuzz/tools.d.ts.map +0 -1
- package/dist/fuzz/types.d.ts +0 -307
- package/dist/fuzz/types.d.ts.map +0 -1
- package/dist/golden-matcher.d.ts +0 -71
- package/dist/golden-matcher.d.ts.map +0 -1
- package/dist/governance/eu-ai-act.d.ts +0 -37
- package/dist/governance/eu-ai-act.d.ts.map +0 -1
- package/dist/governance/index.d.ts.map +0 -1
- package/dist/governance/nist-ai-rmf.d.ts +0 -15
- package/dist/governance/nist-ai-rmf.d.ts.map +0 -1
- package/dist/governance/soc2.d.ts +0 -12
- package/dist/governance/soc2.d.ts.map +0 -1
- package/dist/governance/types.d.ts +0 -66
- package/dist/governance/types.d.ts.map +0 -1
- package/dist/harness-optimizer.d.ts +0 -82
- package/dist/harness-optimizer.d.ts.map +0 -1
- package/dist/held-out-gate.d.ts +0 -135
- package/dist/held-out-gate.d.ts.map +0 -1
- package/dist/hosted/client.d.ts +0 -73
- package/dist/hosted/client.d.ts.map +0 -1
- package/dist/hosted/from-env.test.d.ts +0 -8
- package/dist/hosted/from-env.test.d.ts.map +0 -1
- package/dist/hosted/index.d.ts.map +0 -1
- package/dist/hosted/types.d.ts +0 -159
- package/dist/hosted/types.d.ts.map +0 -1
- package/dist/index.d.ts.map +0 -1
- package/dist/integrity/backend-integrity.d.ts +0 -71
- package/dist/integrity/backend-integrity.d.ts.map +0 -1
- package/dist/integrity/preflight.d.ts +0 -72
- package/dist/integrity/preflight.d.ts.map +0 -1
- package/dist/integrity/preflight.test.d.ts +0 -2
- package/dist/integrity/preflight.test.d.ts.map +0 -1
- package/dist/integrity/single-backend.d.ts +0 -67
- package/dist/integrity/single-backend.d.ts.map +0 -1
- package/dist/intent-match-judge.d.ts +0 -69
- package/dist/intent-match-judge.d.ts.map +0 -1
- package/dist/intent-match-judge.test.d.ts +0 -2
- package/dist/intent-match-judge.test.d.ts.map +0 -1
- package/dist/judge-calibration.d.ts.map +0 -1
- package/dist/judge-ensemble.d.ts +0 -66
- package/dist/judge-ensemble.d.ts.map +0 -1
- package/dist/judge-ensemble.test.d.ts +0 -8
- package/dist/judge-ensemble.test.d.ts.map +0 -1
- package/dist/judge-families.d.ts +0 -38
- package/dist/judge-families.d.ts.map +0 -1
- package/dist/judge-panel.d.ts +0 -65
- package/dist/judge-panel.d.ts.map +0 -1
- package/dist/judge-retry.d.ts +0 -70
- package/dist/judge-retry.d.ts.map +0 -1
- package/dist/judge-runner.d.ts +0 -36
- package/dist/judge-runner.d.ts.map +0 -1
- package/dist/judge-runner.test.d.ts +0 -2
- package/dist/judge-runner.test.d.ts.map +0 -1
- package/dist/judges.d.ts +0 -74
- package/dist/judges.d.ts.map +0 -1
- package/dist/keyword-coverage-judge.d.ts +0 -89
- package/dist/keyword-coverage-judge.d.ts.map +0 -1
- package/dist/keyword-coverage-judge.test.d.ts +0 -2
- package/dist/keyword-coverage-judge.test.d.ts.map +0 -1
- package/dist/knowledge/index.d.ts.map +0 -1
- package/dist/knowledge/readiness.d.ts +0 -26
- package/dist/knowledge/readiness.d.ts.map +0 -1
- package/dist/knowledge/types.d.ts +0 -75
- package/dist/knowledge/types.d.ts.map +0 -1
- package/dist/live-proof.d.ts +0 -62
- package/dist/live-proof.d.ts.map +0 -1
- package/dist/llm-client.d.ts.map +0 -1
- package/dist/llm-client.test.d.ts +0 -2
- package/dist/llm-client.test.d.ts.map +0 -1
- package/dist/locked-jsonl-appender.d.ts.map +0 -1
- package/dist/matrix/aggregation.d.ts +0 -16
- package/dist/matrix/aggregation.d.ts.map +0 -1
- package/dist/matrix/index.d.ts.map +0 -1
- package/dist/matrix/runner.d.ts +0 -15
- package/dist/matrix/runner.d.ts.map +0 -1
- package/dist/matrix/types.d.ts.map +0 -1
- package/dist/meta-eval/calibration.d.ts +0 -47
- package/dist/meta-eval/calibration.d.ts.map +0 -1
- package/dist/meta-eval/correlation-study.d.ts +0 -53
- package/dist/meta-eval/correlation-study.d.ts.map +0 -1
- package/dist/meta-eval/index.d.ts.map +0 -1
- package/dist/meta-eval/outcome-store.d.ts.map +0 -1
- package/dist/meta-eval/rubric-predictive-validity.d.ts.map +0 -1
- package/dist/meta-eval/sentinel.d.ts +0 -169
- package/dist/meta-eval/sentinel.d.ts.map +0 -1
- package/dist/metrics.d.ts +0 -63
- package/dist/metrics.d.ts.map +0 -1
- package/dist/model-seats.d.ts +0 -71
- package/dist/model-seats.d.ts.map +0 -1
- package/dist/model-seats.test.d.ts +0 -2
- package/dist/model-seats.test.d.ts.map +0 -1
- package/dist/muffled-gate-scanner.d.ts +0 -102
- package/dist/muffled-gate-scanner.d.ts.map +0 -1
- package/dist/multi-layer-verifier.d.ts.map +0 -1
- package/dist/multi-layer-verifier.test.d.ts +0 -2
- package/dist/multi-layer-verifier.test.d.ts.map +0 -1
- package/dist/multi-toolchain-layer.d.ts +0 -80
- package/dist/multi-toolchain-layer.d.ts.map +0 -1
- package/dist/multi-toolchain-layer.test.d.ts +0 -2
- package/dist/multi-toolchain-layer.test.d.ts.map +0 -1
- package/dist/multishot/default-tools.d.ts +0 -34
- package/dist/multishot/default-tools.d.ts.map +0 -1
- package/dist/multishot/index.d.ts.map +0 -1
- package/dist/multishot/judges.d.ts +0 -32
- package/dist/multishot/judges.d.ts.map +0 -1
- package/dist/multishot/matrix.d.ts +0 -107
- package/dist/multishot/matrix.d.ts.map +0 -1
- package/dist/multishot/multishot.d.ts +0 -23
- package/dist/multishot/multishot.d.ts.map +0 -1
- package/dist/multishot/router.d.ts +0 -37
- package/dist/multishot/router.d.ts.map +0 -1
- package/dist/multishot/types.d.ts +0 -60
- package/dist/multishot/types.d.ts.map +0 -1
- package/dist/observability.d.ts +0 -71
- package/dist/observability.d.ts.map +0 -1
- package/dist/oracle.d.ts +0 -55
- package/dist/oracle.d.ts.map +0 -1
- package/dist/orthogonality.d.ts +0 -35
- package/dist/orthogonality.d.ts.map +0 -1
- package/dist/otel-pipeline.d.ts +0 -31
- package/dist/otel-pipeline.d.ts.map +0 -1
- package/dist/paraphrase.d.ts +0 -107
- package/dist/paraphrase.d.ts.map +0 -1
- package/dist/pareto.d.ts.map +0 -1
- package/dist/partition-held-out.d.ts +0 -70
- package/dist/partition-held-out.d.ts.map +0 -1
- package/dist/partition-held-out.test.d.ts +0 -2
- package/dist/partition-held-out.test.d.ts.map +0 -1
- package/dist/perf/index.d.ts.map +0 -1
- package/dist/perf/integrity.d.ts +0 -30
- package/dist/perf/integrity.d.ts.map +0 -1
- package/dist/perf/journey.d.ts +0 -45
- package/dist/perf/journey.d.ts.map +0 -1
- package/dist/perf/ratchet.d.ts +0 -47
- package/dist/perf/ratchet.d.ts.map +0 -1
- package/dist/pipelines/budget-breach.d.ts +0 -31
- package/dist/pipelines/budget-breach.d.ts.map +0 -1
- package/dist/pipelines/budget-breach.test.d.ts +0 -2
- package/dist/pipelines/budget-breach.test.d.ts.map +0 -1
- package/dist/pipelines/failure-cluster.d.ts +0 -38
- package/dist/pipelines/failure-cluster.d.ts.map +0 -1
- package/dist/pipelines/failure-cluster.test.d.ts +0 -2
- package/dist/pipelines/failure-cluster.test.d.ts.map +0 -1
- package/dist/pipelines/first-divergence.d.ts +0 -26
- package/dist/pipelines/first-divergence.d.ts.map +0 -1
- package/dist/pipelines/first-divergence.test.d.ts +0 -2
- package/dist/pipelines/first-divergence.test.d.ts.map +0 -1
- package/dist/pipelines/index.d.ts.map +0 -1
- package/dist/pipelines/judge-agreement.d.ts +0 -26
- package/dist/pipelines/judge-agreement.d.ts.map +0 -1
- package/dist/pipelines/judge-agreement.test.d.ts +0 -2
- package/dist/pipelines/judge-agreement.test.d.ts.map +0 -1
- package/dist/pipelines/regression.d.ts +0 -23
- package/dist/pipelines/regression.d.ts.map +0 -1
- package/dist/pipelines/regression.test.d.ts +0 -2
- package/dist/pipelines/regression.test.d.ts.map +0 -1
- package/dist/pipelines/stuck-loop.d.ts +0 -32
- package/dist/pipelines/stuck-loop.d.ts.map +0 -1
- package/dist/pipelines/stuck-loop.test.d.ts +0 -2
- package/dist/pipelines/stuck-loop.test.d.ts.map +0 -1
- package/dist/pipelines/tool-waste.d.ts +0 -34
- package/dist/pipelines/tool-waste.d.ts.map +0 -1
- package/dist/pipelines/tool-waste.test.d.ts +0 -2
- package/dist/pipelines/tool-waste.test.d.ts.map +0 -1
- package/dist/playbook.d.ts +0 -16
- package/dist/playbook.d.ts.map +0 -1
- package/dist/pr-review-benchmark.d.ts +0 -88
- package/dist/pr-review-benchmark.d.ts.map +0 -1
- package/dist/pr-review-benchmark.test.d.ts +0 -2
- package/dist/pr-review-benchmark.test.d.ts.map +0 -1
- package/dist/pre-registration.d.ts +0 -125
- package/dist/pre-registration.d.ts.map +0 -1
- package/dist/prm/builtin-rubrics.d.ts +0 -33
- package/dist/prm/builtin-rubrics.d.ts.map +0 -1
- package/dist/prm/index.d.ts.map +0 -1
- package/dist/prm/inference.d.ts +0 -29
- package/dist/prm/inference.d.ts.map +0 -1
- package/dist/prm/inference.test.d.ts +0 -2
- package/dist/prm/inference.test.d.ts.map +0 -1
- package/dist/prm/rubric.d.ts.map +0 -1
- package/dist/prm/training-export.d.ts +0 -38
- package/dist/prm/training-export.d.ts.map +0 -1
- package/dist/produced-state.d.ts +0 -63
- package/dist/produced-state.d.ts.map +0 -1
- package/dist/produced-state.test.d.ts +0 -8
- package/dist/produced-state.test.d.ts.map +0 -1
- package/dist/profile/baselines.d.ts +0 -37
- package/dist/profile/baselines.d.ts.map +0 -1
- package/dist/profile/index.d.ts +0 -105
- package/dist/profile/index.d.ts.map +0 -1
- package/dist/promotion-gate.d.ts +0 -93
- package/dist/promotion-gate.d.ts.map +0 -1
- package/dist/prompt-registry.d.ts +0 -41
- package/dist/prompt-registry.d.ts.map +0 -1
- package/dist/propose-review-control.d.ts +0 -49
- package/dist/propose-review-control.d.ts.map +0 -1
- package/dist/propose-review-control.test.d.ts +0 -2
- package/dist/propose-review-control.test.d.ts.map +0 -1
- package/dist/propose-review.d.ts +0 -155
- package/dist/propose-review.d.ts.map +0 -1
- package/dist/red-team.d.ts.map +0 -1
- package/dist/reference-replay-steering.d.ts +0 -11
- package/dist/reference-replay-steering.d.ts.map +0 -1
- package/dist/reference-replay.d.ts +0 -176
- package/dist/reference-replay.d.ts.map +0 -1
- package/dist/reflective-mutation.d.ts +0 -79
- package/dist/reflective-mutation.d.ts.map +0 -1
- package/dist/registry.d.ts +0 -31
- package/dist/registry.d.ts.map +0 -1
- package/dist/release-confidence.d.ts +0 -128
- package/dist/release-confidence.d.ts.map +0 -1
- package/dist/release-report.d.ts +0 -11
- package/dist/release-report.d.ts.map +0 -1
- package/dist/replay.d.ts +0 -120
- package/dist/replay.d.ts.map +0 -1
- package/dist/reporter.d.ts +0 -14
- package/dist/reporter.d.ts.map +0 -1
- package/dist/reporting.d.ts.map +0 -1
- package/dist/researcher.d.ts +0 -140
- package/dist/researcher.d.ts.map +0 -1
- package/dist/reviewer.d.ts +0 -118
- package/dist/reviewer.d.ts.map +0 -1
- package/dist/reviewer.test.d.ts +0 -2
- package/dist/reviewer.test.d.ts.map +0 -1
- package/dist/reward-model-export.d.ts +0 -60
- package/dist/reward-model-export.d.ts.map +0 -1
- package/dist/rl/active-curriculum.d.ts +0 -110
- package/dist/rl/active-curriculum.d.ts.map +0 -1
- package/dist/rl/adaptation-eval.d.ts +0 -109
- package/dist/rl/adaptation-eval.d.ts.map +0 -1
- package/dist/rl/adversarial.d.ts.map +0 -1
- package/dist/rl/compute-curves.d.ts +0 -127
- package/dist/rl/compute-curves.d.ts.map +0 -1
- package/dist/rl/contamination.d.ts +0 -117
- package/dist/rl/contamination.d.ts.map +0 -1
- package/dist/rl/corpus.d.ts +0 -55
- package/dist/rl/corpus.d.ts.map +0 -1
- package/dist/rl/corpus.test.d.ts +0 -2
- package/dist/rl/corpus.test.d.ts.map +0 -1
- package/dist/rl/dataset.d.ts +0 -102
- package/dist/rl/dataset.d.ts.map +0 -1
- package/dist/rl/dataset.test.d.ts +0 -2
- package/dist/rl/dataset.test.d.ts.map +0 -1
- package/dist/rl/exporters.d.ts +0 -141
- package/dist/rl/exporters.d.ts.map +0 -1
- package/dist/rl/index.d.ts +0 -49
- package/dist/rl/index.d.ts.map +0 -1
- package/dist/rl/off-policy.d.ts.map +0 -1
- package/dist/rl/predictive-validity-researcher.d.ts +0 -69
- package/dist/rl/predictive-validity-researcher.d.ts.map +0 -1
- package/dist/rl/preferences.d.ts +0 -141
- package/dist/rl/preferences.d.ts.map +0 -1
- package/dist/rl/process-reward.d.ts +0 -122
- package/dist/rl/process-reward.d.ts.map +0 -1
- package/dist/rl/reward-hacking.d.ts +0 -104
- package/dist/rl/reward-hacking.d.ts.map +0 -1
- package/dist/rl/rl-campaign.d.ts +0 -85
- package/dist/rl/rl-campaign.d.ts.map +0 -1
- package/dist/rl/run-record-adapters.d.ts +0 -56
- package/dist/rl/run-record-adapters.d.ts.map +0 -1
- package/dist/rl/sim-fidelity.d.ts +0 -166
- package/dist/rl/sim-fidelity.d.ts.map +0 -1
- package/dist/rl/sim-fidelity.test.d.ts +0 -2
- package/dist/rl/sim-fidelity.test.d.ts.map +0 -1
- package/dist/rl/tournament.d.ts +0 -115
- package/dist/rl/tournament.d.ts.map +0 -1
- package/dist/rl/verifiable-reward.d.ts +0 -124
- package/dist/rl/verifiable-reward.d.ts.map +0 -1
- package/dist/run-critic.d.ts +0 -23
- package/dist/run-critic.d.ts.map +0 -1
- package/dist/run-evidence.d.ts +0 -32
- package/dist/run-evidence.d.ts.map +0 -1
- package/dist/run-record.d.ts.map +0 -1
- package/dist/run-record.test.d.ts +0 -2
- package/dist/run-record.test.d.ts.map +0 -1
- package/dist/run-score.d.ts +0 -31
- package/dist/run-score.d.ts.map +0 -1
- package/dist/runtime-trajectory.d.ts +0 -47
- package/dist/runtime-trajectory.d.ts.map +0 -1
- package/dist/sandbox-harness.d.ts.map +0 -1
- package/dist/sandbox-harness.test.d.ts +0 -2
- package/dist/sandbox-harness.test.d.ts.map +0 -1
- package/dist/sandbox-pool.d.ts +0 -74
- package/dist/sandbox-pool.d.ts.map +0 -1
- package/dist/sandbox-pool.test.d.ts +0 -2
- package/dist/sandbox-pool.test.d.ts.map +0 -1
- package/dist/scorecard.d.ts +0 -133
- package/dist/scorecard.d.ts.map +0 -1
- package/dist/scorecard.test.d.ts +0 -2
- package/dist/scorecard.test.d.ts.map +0 -1
- package/dist/self-play.d.ts +0 -69
- package/dist/self-play.d.ts.map +0 -1
- package/dist/semantic-concept-judge.d.ts +0 -135
- package/dist/semantic-concept-judge.d.ts.map +0 -1
- package/dist/semantic-concept-judge.test.d.ts +0 -2
- package/dist/semantic-concept-judge.test.d.ts.map +0 -1
- package/dist/sequential.d.ts.map +0 -1
- package/dist/series-convergence.d.ts.map +0 -1
- package/dist/slo.d.ts +0 -48
- package/dist/slo.d.ts.map +0 -1
- package/dist/state-continuity.d.ts +0 -47
- package/dist/state-continuity.d.ts.map +0 -1
- package/dist/statistics.d.ts.map +0 -1
- package/dist/statistics.test.d.ts +0 -2
- package/dist/statistics.test.d.ts.map +0 -1
- package/dist/steering-optimizer.d.ts +0 -58
- package/dist/steering-optimizer.d.ts.map +0 -1
- package/dist/steering.d.ts +0 -24
- package/dist/steering.d.ts.map +0 -1
- package/dist/storyboard/code-edit.d.ts +0 -64
- package/dist/storyboard/code-edit.d.ts.map +0 -1
- package/dist/storyboard/code-edit.test.d.ts +0 -2
- package/dist/storyboard/code-edit.test.d.ts.map +0 -1
- package/dist/storyboard/index.d.ts.map +0 -1
- package/dist/storyboard/index.test.d.ts +0 -2
- package/dist/storyboard/index.test.d.ts.map +0 -1
- package/dist/summary-report.d.ts.map +0 -1
- package/dist/telemetry/client.d.ts +0 -35
- package/dist/telemetry/client.d.ts.map +0 -1
- package/dist/telemetry/index.d.ts.map +0 -1
- package/dist/telemetry/schema.d.ts +0 -61
- package/dist/telemetry/schema.d.ts.map +0 -1
- package/dist/telemetry/sink-fetch.d.ts +0 -39
- package/dist/telemetry/sink-fetch.d.ts.map +0 -1
- package/dist/telemetry/sink-file.d.ts.map +0 -1
- package/dist/test-graded-scenario.d.ts +0 -42
- package/dist/test-graded-scenario.d.ts.map +0 -1
- package/dist/testing.d.ts.map +0 -1
- package/dist/tool-use-metrics.d.ts +0 -35
- package/dist/tool-use-metrics.d.ts.map +0 -1
- package/dist/trace/capture-fetch.d.ts +0 -48
- package/dist/trace/capture-fetch.d.ts.map +0 -1
- package/dist/trace/capture-fetch.test.d.ts +0 -2
- package/dist/trace/capture-fetch.test.d.ts.map +0 -1
- package/dist/trace/emitter.d.ts.map +0 -1
- package/dist/trace/extract-usage.d.ts +0 -46
- package/dist/trace/extract-usage.d.ts.map +0 -1
- package/dist/trace/extract-usage.test.d.ts +0 -2
- package/dist/trace/extract-usage.test.d.ts.map +0 -1
- package/dist/trace/index.d.ts +0 -15
- package/dist/trace/index.d.ts.map +0 -1
- package/dist/trace/integrity.d.ts.map +0 -1
- package/dist/trace/otel-bridge.d.ts +0 -29
- package/dist/trace/otel-bridge.d.ts.map +0 -1
- package/dist/trace/otel-export.d.ts +0 -52
- package/dist/trace/otel-export.d.ts.map +0 -1
- package/dist/trace/otel.d.ts +0 -57
- package/dist/trace/otel.d.ts.map +0 -1
- package/dist/trace/otlp-attributes.d.ts +0 -17
- package/dist/trace/otlp-attributes.d.ts.map +0 -1
- package/dist/trace/query.d.ts +0 -29
- package/dist/trace/query.d.ts.map +0 -1
- package/dist/trace/query.test.d.ts +0 -2
- package/dist/trace/query.test.d.ts.map +0 -1
- package/dist/trace/raw-provider-sink.d.ts.map +0 -1
- package/dist/trace/redact.d.ts.map +0 -1
- package/dist/trace/schema.d.ts.map +0 -1
- package/dist/trace/store-to-otlp.d.ts +0 -72
- package/dist/trace/store-to-otlp.d.ts.map +0 -1
- package/dist/trace/store-to-otlp.test.d.ts +0 -2
- package/dist/trace/store-to-otlp.test.d.ts.map +0 -1
- package/dist/trace/store.d.ts.map +0 -1
- package/dist/trace/store.test.d.ts +0 -2
- package/dist/trace/store.test.d.ts.map +0 -1
- package/dist/trace-analyst/analyst.d.ts.map +0 -1
- package/dist/trace-analyst/analyst.test.d.ts +0 -2
- package/dist/trace-analyst/analyst.test.d.ts.map +0 -1
- package/dist/trace-analyst/behavioral-metrics.d.ts +0 -40
- package/dist/trace-analyst/behavioral-metrics.d.ts.map +0 -1
- package/dist/trace-analyst/behavioral-metrics.test.d.ts +0 -2
- package/dist/trace-analyst/behavioral-metrics.test.d.ts.map +0 -1
- package/dist/trace-analyst/hook.d.ts +0 -55
- package/dist/trace-analyst/hook.d.ts.map +0 -1
- package/dist/trace-analyst/index.d.ts +0 -18
- package/dist/trace-analyst/index.d.ts.map +0 -1
- package/dist/trace-analyst/insights.d.ts +0 -71
- package/dist/trace-analyst/insights.d.ts.map +0 -1
- package/dist/trace-analyst/insights.test.d.ts +0 -2
- package/dist/trace-analyst/insights.test.d.ts.map +0 -1
- package/dist/trace-analyst/otlp-flatten.d.ts +0 -42
- package/dist/trace-analyst/otlp-flatten.d.ts.map +0 -1
- package/dist/trace-analyst/otlp-span.d.ts +0 -85
- package/dist/trace-analyst/otlp-span.d.ts.map +0 -1
- package/dist/trace-analyst/otlp-span.test.d.ts +0 -2
- package/dist/trace-analyst/otlp-span.test.d.ts.map +0 -1
- package/dist/trace-analyst/otlp-to-run-records.d.ts +0 -115
- package/dist/trace-analyst/otlp-to-run-records.d.ts.map +0 -1
- package/dist/trace-analyst/otlp-to-run-records.test.d.ts +0 -2
- package/dist/trace-analyst/otlp-to-run-records.test.d.ts.map +0 -1
- package/dist/trace-analyst/otlp-to-run-records.timestamps.test.d.ts +0 -2
- package/dist/trace-analyst/otlp-to-run-records.timestamps.test.d.ts.map +0 -1
- package/dist/trace-analyst/prompts.d.ts +0 -6
- package/dist/trace-analyst/prompts.d.ts.map +0 -1
- package/dist/trace-analyst/store-otlp.d.ts +0 -126
- package/dist/trace-analyst/store-otlp.d.ts.map +0 -1
- package/dist/trace-analyst/store-otlp.test.d.ts +0 -8
- package/dist/trace-analyst/store-otlp.test.d.ts.map +0 -1
- package/dist/trace-analyst/store-otlp.timestamps.test.d.ts +0 -2
- package/dist/trace-analyst/store-otlp.timestamps.test.d.ts.map +0 -1
- package/dist/trace-analyst/store.d.ts +0 -63
- package/dist/trace-analyst/store.d.ts.map +0 -1
- package/dist/trace-analyst/tools.d.ts +0 -44
- package/dist/trace-analyst/tools.d.ts.map +0 -1
- package/dist/trace-analyst/tools.test.d.ts +0 -10
- package/dist/trace-analyst/tools.test.d.ts.map +0 -1
- package/dist/trace-analyst/types.d.ts.map +0 -1
- package/dist/trace-contracts.d.ts +0 -180
- package/dist/trace-contracts.d.ts.map +0 -1
- package/dist/traced-analyst.d.ts +0 -26
- package/dist/traced-analyst.d.ts.map +0 -1
- package/dist/traced-judges.d.ts +0 -27
- package/dist/traced-judges.d.ts.map +0 -1
- package/dist/traces.d.ts.map +0 -1
- package/dist/trajectory.d.ts.map +0 -1
- package/dist/types.d.ts.map +0 -1
- package/dist/ui-finding.d.ts +0 -104
- package/dist/ui-finding.d.ts.map +0 -1
- package/dist/verdict-cache.d.ts +0 -78
- package/dist/verdict-cache.d.ts.map +0 -1
- package/dist/verdict-cache.test.d.ts +0 -2
- package/dist/verdict-cache.test.d.ts.map +0 -1
- package/dist/verdict.d.ts.map +0 -1
- package/dist/visual-diff.d.ts +0 -32
- package/dist/visual-diff.d.ts.map +0 -1
- package/dist/wire/handlers.d.ts +0 -54
- package/dist/wire/handlers.d.ts.map +0 -1
- package/dist/wire/index.d.ts.map +0 -1
- package/dist/wire/openapi.d.ts +0 -3
- package/dist/wire/openapi.d.ts.map +0 -1
- package/dist/wire/rpc.d.ts +0 -21
- package/dist/wire/rpc.d.ts.map +0 -1
- package/dist/wire/rubrics.d.ts +0 -34
- package/dist/wire/rubrics.d.ts.map +0 -1
- package/dist/wire/schemas.d.ts +0 -410
- package/dist/wire/schemas.d.ts.map +0 -1
- package/dist/wire/server.d.ts +0 -60
- package/dist/wire/server.d.ts.map +0 -1
- package/dist/workflow/event-schema.d.ts +0 -5
- package/dist/workflow/event-schema.d.ts.map +0 -1
- package/dist/workflow/feedback-pack.d.ts +0 -99
- package/dist/workflow/feedback-pack.d.ts.map +0 -1
- package/dist/workflow/index.d.ts.map +0 -1
- package/dist/workflow/intelligence-export.d.ts +0 -62
- package/dist/workflow/intelligence-export.d.ts.map +0 -1
- package/dist/workflow/partner-report.d.ts +0 -49
- package/dist/workflow/partner-report.d.ts.map +0 -1
- package/dist/workflow/phase-graph.d.ts +0 -43
- package/dist/workflow/phase-graph.d.ts.map +0 -1
- package/dist/workflow/promotion-gate.d.ts +0 -61
- package/dist/workflow/promotion-gate.d.ts.map +0 -1
- package/dist/workflow/run-record.d.ts +0 -12
- package/dist/workflow/run-record.d.ts.map +0 -1
- package/dist/workflow/runtime-adapter.d.ts +0 -20
- package/dist/workflow/runtime-adapter.d.ts.map +0 -1
- package/dist/workflow/sanitize.d.ts +0 -21
- package/dist/workflow/sanitize.d.ts.map +0 -1
- package/dist/workflow/schema.d.ts +0 -5
- package/dist/workflow/schema.d.ts.map +0 -1
- package/dist/workflow/summary.d.ts +0 -43
- package/dist/workflow/summary.d.ts.map +0 -1
- package/dist/workflow/trace-event-fields.d.ts +0 -6
- package/dist/workflow/trace-event-fields.d.ts.map +0 -1
- package/dist/workflow/trajectory.d.ts +0 -15
- package/dist/workflow/trajectory.d.ts.map +0 -1
- package/dist/workflow/types.d.ts +0 -68
- package/dist/workflow/types.d.ts.map +0 -1
- package/dist/workspace-inspector.d.ts +0 -67
- package/dist/workspace-inspector.d.ts.map +0 -1
- package/dist/wrangler-deploy-runner.test.d.ts +0 -2
- package/dist/wrangler-deploy-runner.test.d.ts.map +0 -1
|
@@ -1,74 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Artifact validators.
|
|
3
|
-
*
|
|
4
|
-
* Generic "score a produced artifact" primitive. Tax uses it for PDF form
|
|
5
|
-
* correctness, research for sourced briefs, browser for task assertions, coding
|
|
6
|
-
* for social posts. One interface, many validators; all plug into
|
|
7
|
-
* `BenchmarkRunner` the same way.
|
|
8
|
-
*
|
|
9
|
-
* A validator receives an `Artifact` (file on disk, JSON blob, text, binary)
|
|
10
|
-
* plus a `ValidationContext` (scenario id, the turns that produced it) and
|
|
11
|
-
* returns a `ValidationResult` with pass/fail + 0..1 score + structured
|
|
12
|
-
* issues.
|
|
13
|
-
*/
|
|
14
|
-
export interface Artifact {
|
|
15
|
-
/** Logical kind — validators type-guard on this */
|
|
16
|
-
kind: 'file' | 'json' | 'text' | 'binary' | string;
|
|
17
|
-
/** Filesystem-style path, optional */
|
|
18
|
-
path?: string;
|
|
19
|
-
/** String content for text/json/file kinds */
|
|
20
|
-
content?: string;
|
|
21
|
-
/** Binary content (if kind === 'binary') */
|
|
22
|
-
bytes?: Uint8Array;
|
|
23
|
-
/** Caller-supplied metadata (mimeType, sha256, size, etc.) */
|
|
24
|
-
metadata?: Record<string, unknown>;
|
|
25
|
-
}
|
|
26
|
-
export interface ValidationContext {
|
|
27
|
-
scenarioId: string;
|
|
28
|
-
turnIndex?: number;
|
|
29
|
-
/** Prior artifacts for multi-artifact scenarios */
|
|
30
|
-
priorArtifacts?: Artifact[];
|
|
31
|
-
/** Free-form hints the validator uses for domain-specific checks */
|
|
32
|
-
hints?: Record<string, unknown>;
|
|
33
|
-
}
|
|
34
|
-
export interface ValidationIssue {
|
|
35
|
-
severity: 'error' | 'warning' | 'info';
|
|
36
|
-
message: string;
|
|
37
|
-
/** Optional path into the artifact (e.g. JSON path or byte offset) */
|
|
38
|
-
locus?: string;
|
|
39
|
-
}
|
|
40
|
-
export interface ValidationResult {
|
|
41
|
-
pass: boolean;
|
|
42
|
-
/** 0–1 normalized score. Validators should be monotonic in pass-ness. */
|
|
43
|
-
score: number;
|
|
44
|
-
issues: ValidationIssue[];
|
|
45
|
-
/** Diagnostic payload for reporters */
|
|
46
|
-
evidence?: Record<string, unknown>;
|
|
47
|
-
}
|
|
48
|
-
export interface ArtifactValidator {
|
|
49
|
-
/** Stable identifier for the validator; appears in reports. */
|
|
50
|
-
name: string;
|
|
51
|
-
/** Optional description for human-facing reports. */
|
|
52
|
-
description?: string;
|
|
53
|
-
/** Called once per artifact; validators are expected to be pure + idempotent. */
|
|
54
|
-
validate(artifact: Artifact, context: ValidationContext): Promise<ValidationResult>;
|
|
55
|
-
}
|
|
56
|
-
/**
|
|
57
|
-
* Run every validator on the same artifact; aggregate pass as AND, score as
|
|
58
|
-
* (weighted) mean, issues concatenated. Weights default to 1 each.
|
|
59
|
-
*/
|
|
60
|
-
export declare function composeValidators(validators: ArtifactValidator[], options?: {
|
|
61
|
-
name?: string;
|
|
62
|
-
weights?: number[];
|
|
63
|
-
}): ArtifactValidator;
|
|
64
|
-
/** Pass if the artifact body matches a provided regex. */
|
|
65
|
-
export declare function regexMatch(name: string, pattern: RegExp): ArtifactValidator;
|
|
66
|
-
/** Pass if JSON parses and every required key is present. */
|
|
67
|
-
export declare function jsonHasKeys(name: string, requiredPaths: string[]): ArtifactValidator;
|
|
68
|
-
/** Pass if min ≤ byte length ≤ max. */
|
|
69
|
-
export declare function byteLengthRange(name: string, min: number, max: number): ArtifactValidator;
|
|
70
|
-
/** Pass if the artifact contains every required substring (case-insensitive by default). */
|
|
71
|
-
export declare function containsAll(name: string, required: string[], options?: {
|
|
72
|
-
caseSensitive?: boolean;
|
|
73
|
-
}): ArtifactValidator;
|
|
74
|
-
//# sourceMappingURL=artifact-validator.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"artifact-validator.d.ts","sourceRoot":"","sources":["../src/artifact-validator.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,MAAM,WAAW,QAAQ;IACvB,mDAAmD;IACnD,IAAI,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,GAAG,QAAQ,GAAG,MAAM,CAAA;IAClD,sCAAsC;IACtC,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,8CAA8C;IAC9C,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,4CAA4C;IAC5C,KAAK,CAAC,EAAE,UAAU,CAAA;IAClB,8DAA8D;IAC9D,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,iBAAiB;IAChC,UAAU,EAAE,MAAM,CAAA;IAClB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,mDAAmD;IACnD,cAAc,CAAC,EAAE,QAAQ,EAAE,CAAA;IAC3B,oEAAoE;IACpE,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAChC;AAED,MAAM,WAAW,eAAe;IAC9B,QAAQ,EAAE,OAAO,GAAG,SAAS,GAAG,MAAM,CAAA;IACtC,OAAO,EAAE,MAAM,CAAA;IACf,sEAAsE;IACtE,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,gBAAgB;IAC/B,IAAI,EAAE,OAAO,CAAA;IACb,yEAAyE;IACzE,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,eAAe,EAAE,CAAA;IACzB,uCAAuC;IACvC,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,iBAAiB;IAChC,+DAA+D;IAC/D,IAAI,EAAE,MAAM,CAAA;IACZ,qDAAqD;IACrD,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,iFAAiF;IACjF,QAAQ,CAAC,QAAQ,EAAE,QAAQ,EAAE,OAAO,EAAE,iBAAiB,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAAA;CACpF;AAMD;;;GAGG;AACH,wBAAgB,iBAAiB,CAC/B,UAAU,EAAE,iBAAiB,EAAE,EAC/B,OAAO,CAAC,EAAE;IAAE,IAAI,CAAC,EAAE,MAAM,CAAC;IAAC,OAAO,CAAC,EAAE,MAAM,EAAE,CAAA;CAAE,GAC9C,iBAAiB,CAyBnB;AAMD,0DAA0D;AAC1D,wBAAgB,UAAU,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,iBAAiB,CAe3E;AAED,6DAA6D;AAC7D,wBAAgB,WAAW,CAAC,IAAI,EAAE,MAAM,EAAE,aAAa,EAAE,MAAM,EAAE,GAAG,iBAAiB,CAoCpF;AAED,uCAAuC;AACvC,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,GAAG,iBAAiB,CAiBzF;AAED,4FAA4F;AAC5F,wBAAgB,WAAW,CACzB,IAAI,EAAE,MAAM,EACZ,QAAQ,EAAE,MAAM,EAAE,EAClB,OAAO,CAAC,EAAE;IAAE,aAAa,CAAC,EAAE,OAAO,CAAA;CAAE,GACpC,iBAAiB,CAsBnB"}
|
package/dist/attestation.d.ts
DELETED
|
@@ -1,63 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Reproducibility attestation for any serializable report object.
|
|
3
|
-
*
|
|
4
|
-
* `attest()` binds a report to its content address (sha-256 over canonical
|
|
5
|
-
* JSON) plus the provenance needed to reproduce it: model versions, seeds,
|
|
6
|
-
* price-table hash, code SHA, inputs hash. `verifyAttestation()` recomputes
|
|
7
|
-
* the address and answers "is this the exact report that provenance
|
|
8
|
-
* describes?" — any single-field tamper changes the hash.
|
|
9
|
-
*
|
|
10
|
-
* Layering: content-addressing is the substrate's job; cryptographic SIGNING
|
|
11
|
-
* (who vouches for the attestation, key management, transparency logs) is the
|
|
12
|
-
* consumer's layer on top. An `AttestedReport` is a stable byte-identical
|
|
13
|
-
* payload a consumer can sign — the substrate never holds keys.
|
|
14
|
-
*
|
|
15
|
-
* Generic by design: the report parameter is ANY value `canonicalJson`
|
|
16
|
-
* accepts (campaign results, fuzz capsules, scorecards, cost ledgers). Do not
|
|
17
|
-
* couple this module to a specific report schema.
|
|
18
|
-
*/
|
|
19
|
-
/** Hash scheme identifier carried by every attestation. A verifier rejects
|
|
20
|
-
* unknown algorithms instead of guessing. */
|
|
21
|
-
export declare const ATTESTATION_ALGORITHM: "sha256/canonical-json";
|
|
22
|
-
export interface AttestationProvenance {
|
|
23
|
-
/** Every model involved in producing the report, name → version/id. */
|
|
24
|
-
modelVersions: Record<string, string>;
|
|
25
|
-
/** RNG seeds the run was driven by, when seeded. */
|
|
26
|
-
seeds?: number[];
|
|
27
|
-
/** Content hash of the price table used for cost figures — cost numbers
|
|
28
|
-
* are only reproducible against the same prices. */
|
|
29
|
-
priceTableHash?: string;
|
|
30
|
-
/** Git SHA of the code that produced the report. */
|
|
31
|
-
codeSha: string;
|
|
32
|
-
/** Content hash of the input set (scenarios, dataset manifest, ...). */
|
|
33
|
-
inputsHash?: string;
|
|
34
|
-
/** ISO-8601 timestamp, caller-supplied — the substrate stays clock-free
|
|
35
|
-
* so attestation is deterministic and testable. */
|
|
36
|
-
createdAt: string;
|
|
37
|
-
}
|
|
38
|
-
export interface AttestedReport {
|
|
39
|
-
/** Hex sha-256 over the canonical JSON of the report. */
|
|
40
|
-
reportHash: string;
|
|
41
|
-
provenance: AttestationProvenance;
|
|
42
|
-
algorithm: typeof ATTESTATION_ALGORITHM;
|
|
43
|
-
}
|
|
44
|
-
export interface AttestationVerification {
|
|
45
|
-
valid: boolean;
|
|
46
|
-
/** Populated iff `valid` is false — names the exact mismatch. */
|
|
47
|
-
reason?: string;
|
|
48
|
-
}
|
|
49
|
-
/**
|
|
50
|
-
* Content-address a report and bind it to its provenance. Throws (via
|
|
51
|
-
* `canonicalJson`) if the report contains undefined / function / symbol /
|
|
52
|
-
* non-finite numbers — a report that cannot be unambiguously serialized
|
|
53
|
-
* cannot be attested.
|
|
54
|
-
*/
|
|
55
|
-
export declare function attest(report: unknown, provenance: AttestationProvenance): AttestedReport;
|
|
56
|
-
/**
|
|
57
|
-
* Verify a report against its attestation. Returns a typed outcome rather
|
|
58
|
-
* than throwing: an unverifiable report (e.g. one that no longer
|
|
59
|
-
* canonicalizes) is a verification failure with the cause in `reason`, not a
|
|
60
|
-
* crash — verifiers run in pipelines that must record WHY, not die.
|
|
61
|
-
*/
|
|
62
|
-
export declare function verifyAttestation(report: unknown, attested: AttestedReport): AttestationVerification;
|
|
63
|
-
//# sourceMappingURL=attestation.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"attestation.d.ts","sourceRoot":"","sources":["../src/attestation.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAIH;8CAC8C;AAC9C,eAAO,MAAM,qBAAqB,EAAG,uBAAgC,CAAA;AAErE,MAAM,WAAW,qBAAqB;IACpC,uEAAuE;IACvE,aAAa,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IACrC,oDAAoD;IACpD,KAAK,CAAC,EAAE,MAAM,EAAE,CAAA;IAChB;yDACqD;IACrD,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,oDAAoD;IACpD,OAAO,EAAE,MAAM,CAAA;IACf,wEAAwE;IACxE,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB;wDACoD;IACpD,SAAS,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,WAAW,cAAc;IAC7B,yDAAyD;IACzD,UAAU,EAAE,MAAM,CAAA;IAClB,UAAU,EAAE,qBAAqB,CAAA;IACjC,SAAS,EAAE,OAAO,qBAAqB,CAAA;CACxC;AAED,MAAM,WAAW,uBAAuB;IACtC,KAAK,EAAE,OAAO,CAAA;IACd,iEAAiE;IACjE,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAED;;;;;GAKG;AACH,wBAAgB,MAAM,CAAC,MAAM,EAAE,OAAO,EAAE,UAAU,EAAE,qBAAqB,GAAG,cAAc,CAMzF;AAED;;;;;GAKG;AACH,wBAAgB,iBAAiB,CAC/B,MAAM,EAAE,OAAO,EACf,QAAQ,EAAE,cAAc,GACvB,uBAAuB,CAuBzB"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"attestation.test.d.ts","sourceRoot":"","sources":["../src/attestation.test.ts"],"names":[],"mappings":""}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/authenticity/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AAEH,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,CAAA;IACZ,OAAO,CAAC,EAAE,MAAM,CAAA;CACjB;AAED,MAAM,WAAW,mBAAmB;IAClC,sDAAsD;IACtD,KAAK,EAAE,MAAM,CAAA;IACb,qEAAqE;IACrE,gBAAgB,CAAC,EAAE,MAAM,CAAA;IACzB,4EAA4E;IAC5E,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB;;0EAEsE;IACtE,QAAQ,EAAE,MAAM,CAAA;IAChB,4EAA4E;IAC5E,SAAS,EAAE,MAAM,CAAA;IACjB,2EAA2E;IAC3E,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,kFAAkF;IAClF,QAAQ,EAAE,MAAM,CAAA;IAChB,yDAAyD;IACzD,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,2CAA2C;IAC3C,OAAO,CAAC,EAAE;QAAE,QAAQ,CAAC,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,MAAM,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,CAAA;CAChF;AAED,MAAM,WAAW,kBAAkB;IACjC,iEAAiE;IACjE,QAAQ,EAAE,MAAM,CAAA;IAChB,uBAAuB,EAAE,OAAO,CAAA;IAChC,qBAAqB,EAAE,MAAM,CAAA;IAC7B,YAAY,EAAE,OAAO,CAAA;IACrB,SAAS,EAAE,OAAO,CAAA;IAClB,KAAK,EAAE,OAAO,CAAA;IACd;;sEAEkE;IAClE,kBAAkB,EAAE,OAAO,CAAA;IAC3B;0DACsD;IACtD,aAAa,EAAE,OAAO,CAAA;IACtB,QAAQ,EAAE,OAAO,CAAA;IACjB,qDAAqD;IACrD,WAAW,EAAE,MAAM,CAAA;IACnB,yDAAyD;IACzD,KAAK,EAAE,MAAM,EAAE,CAAA;CAChB;AAgDD;qCACqC;AACrC,wBAAgB,iBAAiB,CAC/B,KAAK,EAAE,SAAS,YAAY,EAAE,EAC9B,OAAO,EAAE,mBAAmB,GAC3B,kBAAkB,CA4FpB;AAED,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,OAAO,CAAA;IACd,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAED;8DAC8D;AAC9D,wBAAgB,YAAY,CAC1B,CAAC,EAAE,kBAAkB,EACrB,IAAI,GAAE;IAAE,KAAK,CAAC,EAAE,MAAM,CAAC;IAAC,eAAe,CAAC,EAAE,OAAO,CAAC;IAAC,oBAAoB,CAAC,EAAE,OAAO,CAAA;CAAO,GACvF,YAAY,CAsBd;AAID,MAAM,WAAW,kBAAkB;IACjC,kDAAkD;IAClD,SAAS,EAAE,MAAM,CAAA;IACjB,0DAA0D;IAC1D,OAAO,EAAE,MAAM,CAAA;IACf,oEAAoE;IACpE,SAAS,EAAE,MAAM,CAAA;IACjB,OAAO,EAAE,MAAM,CAAA;CAChB;AAED;mDACmD;AACnD,MAAM,MAAM,UAAU,GAAG,CAAC,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC,MAAM,CAAC,CAAA;AA0B1E;;;;;GAKG;AACH,wBAAsB,uBAAuB,CAC3C,KAAK,EAAE,SAAS,YAAY,EAAE,EAC9B,QAAQ,EAAE,UAAU,EACpB,IAAI,GAAE;IAAE,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,UAAU,CAAC,EAAE,MAAM,CAAA;CAAO,GAClD,OAAO,CAAC,kBAAkB,CAAC,CA+B7B;AAID,MAAM,WAAW,gBAAgB;IAC/B,8EAA8E;IAC9E,MAAM,EAAE,MAAM,CAAA;IACd,SAAS,EAAE,MAAM,CAAA;CAClB;AAED;;;;;;;GAOG;AACH,wBAAsB,gBAAgB,CACpC,KAAK,EAAE,SAAS,YAAY,EAAE,EAC9B,QAAQ,EAAE,UAAU,EACpB,IAAI,GAAE;IAAE,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,UAAU,CAAC,EAAE,MAAM,CAAA;CAAO,GACnE,OAAO,CAAC,gBAAgB,CAAC,CA4B3B;AAID,MAAM,MAAM,YAAY,GAAG,YAAY,GAAG,YAAY,GAAG,MAAM,CAAA;AAE/D,MAAM,WAAW,eAAgB,SAAQ,kBAAkB;IACzD,uEAAuE;IACvE,eAAe,EAAE,MAAM,CAAA;IACvB,IAAI,EAAE,YAAY,CAAA;IAClB,sEAAsE;IACtE,YAAY,EAAE,OAAO,CAAA;IACrB,yCAAyC;IACzC,QAAQ,CAAC,EAAE,gBAAgB,CAAA;CAC5B;AAED;;;;;;;;;;GAUG;AACH,wBAAsB,oBAAoB,CACxC,KAAK,EAAE,SAAS,YAAY,EAAE,EAC9B,OAAO,EAAE,mBAAmB,EAC5B,QAAQ,EAAE,UAAU,EACpB,IAAI,GAAE;IACJ,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,QAAQ,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC3B,iBAAiB,CAAC,EAAE,MAAM,CAAA;CACtB,GACL,OAAO,CAAC,eAAe,CAAC,CAsC1B"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.test.d.ts","sourceRoot":"","sources":["../../src/authenticity/index.test.ts"],"names":[],"mappings":""}
|
package/dist/auto-pr.d.ts
DELETED
|
@@ -1,120 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Automated pull request opener for the improvement loop.
|
|
3
|
-
*
|
|
4
|
-
* When `runImprovementLoop` ships a winner (`autoOnPromote: 'pr'`) it produces
|
|
5
|
-
* a promoted surface diff. To close the eval → prod → eval cycle the framework
|
|
6
|
-
* lands that change as a reviewable code change. This module does exactly that:
|
|
7
|
-
*
|
|
8
|
-
* 1. Stage a branch off `baseBranch`.
|
|
9
|
-
* 2. Write each `fileChange` into the worktree.
|
|
10
|
-
* 3. Commit + push.
|
|
11
|
-
* 4. Open a PR via the GitHub API.
|
|
12
|
-
*
|
|
13
|
-
* Two transports ship in core:
|
|
14
|
-
*
|
|
15
|
-
* - `ghCliClient(opts)` — shells out to the `gh` CLI. No extra deps,
|
|
16
|
-
* re-uses the developer machine's `gh auth` state, works with both
|
|
17
|
-
* github.com and GitHub Enterprise. This is the recommended default.
|
|
18
|
-
* - `httpGithubClient(opts)` — direct `fetch` against `api.github.com`
|
|
19
|
-
* with a bearer token. Useful in CI where `gh` may not be installed.
|
|
20
|
-
*
|
|
21
|
-
* Both implement the small `AutoPrClient` interface, so tests substitute
|
|
22
|
-
* a fake without spinning a process or network.
|
|
23
|
-
*/
|
|
24
|
-
export interface FileChange {
|
|
25
|
-
/** Repo-relative path. Forward slashes; no `..`. */
|
|
26
|
-
path: string;
|
|
27
|
-
/** New file contents. UTF-8. */
|
|
28
|
-
contents: string;
|
|
29
|
-
/** Optional explanatory comment shown in the commit body. */
|
|
30
|
-
rationale?: string;
|
|
31
|
-
}
|
|
32
|
-
export interface RepoRef {
|
|
33
|
-
owner: string;
|
|
34
|
-
name: string;
|
|
35
|
-
}
|
|
36
|
-
export interface ProposeAutomatedPullRequestInput {
|
|
37
|
-
repo: RepoRef;
|
|
38
|
-
/** Branch to base the PR on. Default `'main'`. */
|
|
39
|
-
baseBranch?: string;
|
|
40
|
-
/** New branch name. Use a prefix + a short stable id; no spaces. */
|
|
41
|
-
branchName: string;
|
|
42
|
-
fileChanges: FileChange[];
|
|
43
|
-
title: string;
|
|
44
|
-
body: string;
|
|
45
|
-
/** Optional GitHub usernames to request review from. */
|
|
46
|
-
reviewers?: string[];
|
|
47
|
-
/** Optional labels to apply. */
|
|
48
|
-
labels?: string[];
|
|
49
|
-
/** Commit author name. Default: derived from the GitHub client. */
|
|
50
|
-
authorName?: string;
|
|
51
|
-
/** Commit author email. Default: derived from the GitHub client. */
|
|
52
|
-
authorEmail?: string;
|
|
53
|
-
/** Dry-run — do not push or open a PR; just return the would-be plan. */
|
|
54
|
-
dryRun?: boolean;
|
|
55
|
-
}
|
|
56
|
-
export interface ProposeAutomatedPullRequestResult {
|
|
57
|
-
prUrl: string;
|
|
58
|
-
branchName: string;
|
|
59
|
-
headSha: string;
|
|
60
|
-
dryRun: boolean;
|
|
61
|
-
}
|
|
62
|
-
/** Pluggable transport for the auto-PR pipeline. */
|
|
63
|
-
export interface AutoPrClient {
|
|
64
|
-
/**
|
|
65
|
-
* Create a branch from `baseBranch`, write file changes, commit, push,
|
|
66
|
-
* and open a PR. Returns the PR's HTML url and head SHA.
|
|
67
|
-
*
|
|
68
|
-
* Implementations must be idempotent on `branchName`: if the branch
|
|
69
|
-
* already exists with the same head SHA as the would-be commit, return
|
|
70
|
-
* the existing PR rather than failing. This makes the production loop
|
|
71
|
-
* safe to retry on transient errors.
|
|
72
|
-
*/
|
|
73
|
-
proposeChange(input: ProposeAutomatedPullRequestInput): Promise<ProposeAutomatedPullRequestResult>;
|
|
74
|
-
}
|
|
75
|
-
export declare function proposeAutomatedPullRequest(client: AutoPrClient, input: ProposeAutomatedPullRequestInput): Promise<ProposeAutomatedPullRequestResult>;
|
|
76
|
-
export interface HttpGithubClientOptions {
|
|
77
|
-
/** Personal access token, GitHub App token, or `GITHUB_TOKEN` from Actions. */
|
|
78
|
-
token: string;
|
|
79
|
-
/** Override for GitHub Enterprise. Default `'https://api.github.com'`. */
|
|
80
|
-
apiBase?: string;
|
|
81
|
-
/** Test seam — defaults to global `fetch`. */
|
|
82
|
-
fetchImpl?: typeof fetch;
|
|
83
|
-
/** Test seam — clock for commit timestamps. */
|
|
84
|
-
now?: () => Date;
|
|
85
|
-
}
|
|
86
|
-
/**
|
|
87
|
-
* Direct REST-API GitHub client. No external deps.
|
|
88
|
-
*
|
|
89
|
-
* Idempotency strategy: before creating refs/commits/PRs, check whether
|
|
90
|
-
* the branch already exists at the desired tree. If so, return the
|
|
91
|
-
* existing PR (or open one if missing). Errors from concurrent runs
|
|
92
|
-
* (`Reference already exists`) are caught and treated as success.
|
|
93
|
-
*/
|
|
94
|
-
export declare function httpGithubClient(opts: HttpGithubClientOptions): AutoPrClient;
|
|
95
|
-
export interface GhCliClientOptions {
|
|
96
|
-
/** Override the CLI binary (`gh`). For testing. */
|
|
97
|
-
bin?: string;
|
|
98
|
-
/** Working directory containing a clone of `repo`. Default: process cwd. */
|
|
99
|
-
cwd?: string;
|
|
100
|
-
/** Test seam: process spawner. Default: node:child_process spawn. */
|
|
101
|
-
exec?: (bin: string, args: string[], opts: {
|
|
102
|
-
cwd: string;
|
|
103
|
-
stdin?: string;
|
|
104
|
-
}) => Promise<{
|
|
105
|
-
stdout: string;
|
|
106
|
-
stderr: string;
|
|
107
|
-
exitCode: number;
|
|
108
|
-
}>;
|
|
109
|
-
}
|
|
110
|
-
/**
|
|
111
|
-
* `gh` CLI transport. Requires:
|
|
112
|
-
* - `gh` installed and authenticated (`gh auth status`).
|
|
113
|
-
* - A local clone of the repo with a clean working tree.
|
|
114
|
-
* - `git` on PATH.
|
|
115
|
-
*
|
|
116
|
-
* Uses `gh api` for repo metadata and `gh pr create` for the PR. The
|
|
117
|
-
* actual commit lands via `git`, which keeps `gh`'s footprint minimal.
|
|
118
|
-
*/
|
|
119
|
-
export declare function ghCliClient(opts?: GhCliClientOptions): AutoPrClient;
|
|
120
|
-
//# sourceMappingURL=auto-pr.d.ts.map
|
package/dist/auto-pr.d.ts.map
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"auto-pr.d.ts","sourceRoot":"","sources":["../src/auto-pr.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;GAsBG;AAIH,MAAM,WAAW,UAAU;IACzB,oDAAoD;IACpD,IAAI,EAAE,MAAM,CAAA;IACZ,gCAAgC;IAChC,QAAQ,EAAE,MAAM,CAAA;IAChB,6DAA6D;IAC7D,SAAS,CAAC,EAAE,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,OAAO;IACtB,KAAK,EAAE,MAAM,CAAA;IACb,IAAI,EAAE,MAAM,CAAA;CACb;AAED,MAAM,WAAW,gCAAgC;IAC/C,IAAI,EAAE,OAAO,CAAA;IACb,kDAAkD;IAClD,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,oEAAoE;IACpE,UAAU,EAAE,MAAM,CAAA;IAClB,WAAW,EAAE,UAAU,EAAE,CAAA;IACzB,KAAK,EAAE,MAAM,CAAA;IACb,IAAI,EAAE,MAAM,CAAA;IACZ,wDAAwD;IACxD,SAAS,CAAC,EAAE,MAAM,EAAE,CAAA;IACpB,gCAAgC;IAChC,MAAM,CAAC,EAAE,MAAM,EAAE,CAAA;IACjB,mEAAmE;IACnE,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,oEAAoE;IACpE,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,yEAAyE;IACzE,MAAM,CAAC,EAAE,OAAO,CAAA;CACjB;AAED,MAAM,WAAW,iCAAiC;IAChD,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,EAAE,MAAM,CAAA;IAClB,OAAO,EAAE,MAAM,CAAA;IACf,MAAM,EAAE,OAAO,CAAA;CAChB;AAED,oDAAoD;AACpD,MAAM,WAAW,YAAY;IAC3B;;;;;;;;OAQG;IACH,aAAa,CAAC,KAAK,EAAE,gCAAgC,GAAG,OAAO,CAAC,iCAAiC,CAAC,CAAA;CACnG;AAED,wBAAsB,2BAA2B,CAC/C,MAAM,EAAE,YAAY,EACpB,KAAK,EAAE,gCAAgC,GACtC,OAAO,CAAC,iCAAiC,CAAC,CAG5C;AAoCD,MAAM,WAAW,uBAAuB;IACtC,+EAA+E;IAC/E,KAAK,EAAE,MAAM,CAAA;IACb,0EAA0E;IAC1E,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,8CAA8C;IAC9C,SAAS,CAAC,EAAE,OAAO,KAAK,CAAA;IACxB,+CAA+C;IAC/C,GAAG,CAAC,EAAE,MAAM,IAAI,CAAA;CACjB;AAyBD;;;;;;;GAOG;AACH,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,uBAAuB,GAAG,YAAY,CAoK5E;AAID,MAAM,WAAW,kBAAkB;IACjC,mDAAmD;IACnD,GAAG,CAAC,EAAE,MAAM,CAAA;IACZ,4EAA4E;IAC5E,GAAG,CAAC,EAAE,MAAM,CAAA;IACZ,qEAAqE;IACrE,IAAI,CAAC,EAAE,CACL,GAAG,EAAE,MAAM,EACX,IAAI,EAAE,MAAM,EAAE,EACd,IAAI,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAA;KAAE,KAClC,OAAO,CAAC;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;CACnE;AAED;;;;;;;;GAQG;AACH,wBAAgB,WAAW,CAAC,IAAI,GAAE,kBAAuB,GAAG,YAAY,CAgHvE"}
|
package/dist/baseline.d.ts.map
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"baseline.d.ts","sourceRoot":"","sources":["../src/baseline.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAIH,MAAM,WAAW,aAAa;IAC5B,+DAA+D;IAC/D,MAAM,EAAE,MAAM,CAAA;IACd,wCAAwC;IACxC,cAAc,EAAE,OAAO,CAAA;IACvB,QAAQ,EAAE,MAAM,EAAE,CAAA;IAClB,SAAS,EAAE,MAAM,EAAE,CAAA;CACpB;AAED,MAAM,WAAW,aAAa;IAC5B,MAAM,EAAE,MAAM,CAAA;IACd,YAAY,EAAE,MAAM,CAAA;IACpB,aAAa,EAAE,MAAM,CAAA;IACrB,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,EAAE,MAAM,CAAA;IACf,MAAM,EAAE,MAAM,CAAA;IACd,OAAO,EAAE,MAAM,CAAA;IACf,MAAM,EAAE,MAAM,CAAA;IACd,MAAM,EAAE,OAAO,CAAA;IACf,yEAAyE;IACzE,GAAG,EAAE,MAAM,CAAA;IACX,OAAO,EAAE,UAAU,GAAG,WAAW,GAAG,QAAQ,GAAG,UAAU,CAAA;CAC1D;AAED,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAE,aAAa,EAAE,CAAA;IACxB,6CAA6C;IAC7C,aAAa,EAAE,OAAO,CAAA;IACtB,2DAA2D;IAC3D,WAAW,EAAE,OAAO,CAAA;CACrB;AAED,MAAM,WAAW,eAAe;IAC9B,gFAAgF;IAChF,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,qEAAqE;IACrE,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,8EAA8E;IAC9E,mBAAmB,CAAC,EAAE,MAAM,CAAA;CAC7B;AAED;;;;;;GAMG;AACH,wBAAgB,iBAAiB,CAC/B,OAAO,EAAE,aAAa,EAAE,EACxB,OAAO,GAAE,eAAoB,GAC5B,cAAc,CAuDhB;AAMD,6DAA6D;AAC7D,wBAAgB,GAAG,CAAC,EAAE,EAAE,MAAM,EAAE,GAAG,MAAM,CAUxC;AAED;;;;GAIG;AACH,wBAAgB,WAAW,CAAC,CAAC,EAAE,MAAM,EAAE,EAAE,CAAC,EAAE,MAAM,EAAE,GAAG;IAAE,CAAC,EAAE,MAAM,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,CAAC,EAAE,MAAM,CAAA;CAAE,CAc1F"}
|
package/dist/behavior-dsl.d.ts
DELETED
|
@@ -1,73 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Behavior DSL — pytest-style assertions over a run's trajectory.
|
|
3
|
-
*
|
|
4
|
-
* Shape:
|
|
5
|
-
* expect(store, runId).toCall('search').withArgs({ q: /.+/ })
|
|
6
|
-
* expect(store, runId).toRefuse()
|
|
7
|
-
* expect(store, runId).toOutputMatch(/confirmed/i)
|
|
8
|
-
* expect(store, runId).toRespectBudget('tokens')
|
|
9
|
-
* expect(store, runId).toCompleteWithin({ wallMs: 30_000 })
|
|
10
|
-
*
|
|
11
|
-
* Each matcher returns an `Expectation` with `.check() → MatcherResult`
|
|
12
|
-
* so the DSL is composable with suite runners — you can collect all
|
|
13
|
-
* expectations into a report instead of throwing on first failure.
|
|
14
|
-
*/
|
|
15
|
-
import type { BudgetLedgerEntry, Span, ToolSpan } from './trace/schema';
|
|
16
|
-
import type { TraceStore } from './trace/store';
|
|
17
|
-
import { type TraceContract } from './trace-contracts';
|
|
18
|
-
export interface MatcherResult {
|
|
19
|
-
ok: boolean;
|
|
20
|
-
detail: string;
|
|
21
|
-
evidence?: string;
|
|
22
|
-
}
|
|
23
|
-
export interface Expectation {
|
|
24
|
-
/** Human-facing label; used in reports. */
|
|
25
|
-
label: string;
|
|
26
|
-
check(): Promise<MatcherResult>;
|
|
27
|
-
}
|
|
28
|
-
export declare class BehaviorAssertion {
|
|
29
|
-
private store;
|
|
30
|
-
private runId;
|
|
31
|
-
constructor(store: TraceStore, runId: string);
|
|
32
|
-
toCall(toolName: string): CallExpectation;
|
|
33
|
-
toRefuse(markers?: RegExp[]): Expectation;
|
|
34
|
-
toOutputMatch(pattern: RegExp): Expectation;
|
|
35
|
-
toRespectBudget(dimension: keyof BudgetLedgerEntry['dimension'] | 'tokens' | 'wallMs' | 'calls' | 'usd'): Expectation;
|
|
36
|
-
toCompleteWithin(limits: {
|
|
37
|
-
wallMs?: number;
|
|
38
|
-
toolCalls?: number;
|
|
39
|
-
llmTurns?: number;
|
|
40
|
-
}): Expectation;
|
|
41
|
-
/** Evaluate a finite-trace temporal contract (`traceContract(...)`) over
|
|
42
|
-
* this run's span sequence. See `trace-contracts.ts` for the operators. */
|
|
43
|
-
toSatisfyContract(contract: TraceContract): Expectation;
|
|
44
|
-
toNeverCall(toolName: string): Expectation;
|
|
45
|
-
}
|
|
46
|
-
export declare class CallExpectation implements Expectation {
|
|
47
|
-
private store;
|
|
48
|
-
private runId;
|
|
49
|
-
private toolName;
|
|
50
|
-
private argMatchers;
|
|
51
|
-
private minCount;
|
|
52
|
-
private maxCount;
|
|
53
|
-
constructor(store: TraceStore, runId: string, toolName: string);
|
|
54
|
-
get label(): string;
|
|
55
|
-
withArgs(shape: Record<string, unknown | RegExp>): this;
|
|
56
|
-
times(n: number): this;
|
|
57
|
-
atLeast(n: number): this;
|
|
58
|
-
atMost(n: number): this;
|
|
59
|
-
check(): Promise<MatcherResult>;
|
|
60
|
-
}
|
|
61
|
-
export declare function expectAgent(store: TraceStore, runId: string): BehaviorAssertion;
|
|
62
|
-
/** Runs every expectation, collects results. Never throws. */
|
|
63
|
-
export declare function runExpectations(expectations: Expectation[]): Promise<{
|
|
64
|
-
results: Array<{
|
|
65
|
-
label: string;
|
|
66
|
-
result: MatcherResult;
|
|
67
|
-
}>;
|
|
68
|
-
pass: boolean;
|
|
69
|
-
passCount: number;
|
|
70
|
-
failCount: number;
|
|
71
|
-
}>;
|
|
72
|
-
export type { Span, ToolSpan };
|
|
73
|
-
//# sourceMappingURL=behavior-dsl.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"behavior-dsl.d.ts","sourceRoot":"","sources":["../src/behavior-dsl.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAGH,OAAO,KAAK,EAAE,iBAAiB,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,gBAAgB,CAAA;AACvE,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,eAAe,CAAA;AAC/C,OAAO,EAAyB,KAAK,aAAa,EAAE,MAAM,mBAAmB,CAAA;AAE7E,MAAM,WAAW,aAAa;IAC5B,EAAE,EAAE,OAAO,CAAA;IACX,MAAM,EAAE,MAAM,CAAA;IACd,QAAQ,CAAC,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,WAAW,WAAW;IAC1B,2CAA2C;IAC3C,KAAK,EAAE,MAAM,CAAA;IACb,KAAK,IAAI,OAAO,CAAC,aAAa,CAAC,CAAA;CAChC;AAED,qBAAa,iBAAiB;IAE1B,OAAO,CAAC,KAAK;IACb,OAAO,CAAC,KAAK;gBADL,KAAK,EAAE,UAAU,EACjB,KAAK,EAAE,MAAM;IAGvB,MAAM,CAAC,QAAQ,EAAE,MAAM,GAAG,eAAe;IAIzC,QAAQ,CAAC,OAAO,GAAE,MAAM,EAA4B,GAAG,WAAW;IAkBlE,aAAa,CAAC,OAAO,EAAE,MAAM,GAAG,WAAW;IAc3C,eAAe,CACb,SAAS,EAAE,MAAM,iBAAiB,CAAC,WAAW,CAAC,GAAG,QAAQ,GAAG,QAAQ,GAAG,OAAO,GAAG,KAAK,GACtF,WAAW;IAad,gBAAgB,CAAC,MAAM,EAAE;QACvB,MAAM,CAAC,EAAE,MAAM,CAAA;QACf,SAAS,CAAC,EAAE,MAAM,CAAA;QAClB,QAAQ,CAAC,EAAE,MAAM,CAAA;KAClB,GAAG,WAAW;IAuBf;gFAC4E;IAC5E,iBAAiB,CAAC,QAAQ,EAAE,aAAa,GAAG,WAAW;IAkBvD,WAAW,CAAC,QAAQ,EAAE,MAAM,GAAG,WAAW;CAe3C;AAED,qBAAa,eAAgB,YAAW,WAAW;IAM/C,OAAO,CAAC,KAAK;IACb,OAAO,CAAC,KAAK;IACb,OAAO,CAAC,QAAQ;IAPlB,OAAO,CAAC,WAAW,CAAwC;IAC3D,OAAO,CAAC,QAAQ,CAAI;IACpB,OAAO,CAAC,QAAQ,CAAW;gBAGjB,KAAK,EAAE,UAAU,EACjB,KAAK,EAAE,MAAM,EACb,QAAQ,EAAE,MAAM;IAG1B,IAAI,KAAK,IAAI,MAAM,CAElB;IAED,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,GAAG,MAAM,CAAC,GAAG,IAAI;IAKvD,KAAK,CAAC,CAAC,EAAE,MAAM,GAAG,IAAI;IAMtB,OAAO,CAAC,CAAC,EAAE,MAAM,GAAG,IAAI;IAKxB,MAAM,CAAC,CAAC,EAAE,MAAM,GAAG,IAAI;IAKjB,KAAK,IAAI,OAAO,CAAC,aAAa,CAAC;CAgBtC;AAED,wBAAgB,WAAW,CAAC,KAAK,EAAE,UAAU,EAAE,KAAK,EAAE,MAAM,GAAG,iBAAiB,CAE/E;AAED,8DAA8D;AAC9D,wBAAsB,eAAe,CAAC,YAAY,EAAE,WAAW,EAAE,GAAG,OAAO,CAAC;IAC1E,OAAO,EAAE,KAAK,CAAC;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,aAAa,CAAA;KAAE,CAAC,CAAA;IACxD,IAAI,EAAE,OAAO,CAAA;IACb,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,EAAE,MAAM,CAAA;CAClB,CAAC,CAWD;AAwBD,YAAY,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAA"}
|
|
@@ -1,10 +0,0 @@
|
|
|
1
|
-
import type { BeliefDecisionPoint, BeliefSelectivePolicy } from './types';
|
|
2
|
-
export type BeliefCalibrationRegion = 'all' | 'accepted' | 'rejected';
|
|
3
|
-
export interface BeliefCalibrationOptions {
|
|
4
|
-
bins?: number;
|
|
5
|
-
minPairs?: number;
|
|
6
|
-
policy?: BeliefSelectivePolicy;
|
|
7
|
-
region?: BeliefCalibrationRegion;
|
|
8
|
-
}
|
|
9
|
-
export declare function calibrateBeliefDecisions(points: BeliefDecisionPoint[], options?: BeliefCalibrationOptions): import("../meta-eval").CalibrationReport | null;
|
|
10
|
-
//# sourceMappingURL=calibration.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"calibration.d.ts","sourceRoot":"","sources":["../../src/belief-state/calibration.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,mBAAmB,EAAE,qBAAqB,EAAE,MAAM,SAAS,CAAA;AAEzE,MAAM,MAAM,uBAAuB,GAAG,KAAK,GAAG,UAAU,GAAG,UAAU,CAAA;AAErE,MAAM,WAAW,wBAAwB;IACvC,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,MAAM,CAAC,EAAE,qBAAqB,CAAA;IAC9B,MAAM,CAAC,EAAE,uBAAuB,CAAA;CACjC;AAED,wBAAgB,wBAAwB,CACtC,MAAM,EAAE,mBAAmB,EAAE,EAC7B,OAAO,GAAE,wBAA6B,mDAgBvC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"calibration.test.d.ts","sourceRoot":"","sources":["../../src/belief-state/calibration.test.ts"],"names":[],"mappings":""}
|
|
@@ -1,66 +0,0 @@
|
|
|
1
|
-
import type { CodeAgentSessionSource } from '../contract/intake/code-agent-session';
|
|
2
|
-
import type { RunRecord } from '../run-record';
|
|
3
|
-
import { type AnalyzeBeliefPolicyOptions } from './report';
|
|
4
|
-
import type { BeliefDecisionExtractionReport, BeliefDecisionKind, BeliefDecisionPoint, BeliefPolicyEvaluationReport, BeliefSelectivePolicy } from './types';
|
|
5
|
-
export type CodeAgentBeliefDecisionTargetId = 'failure-recovery' | 'tool-selection' | 'graph-completion';
|
|
6
|
-
export interface ExtractCodeAgentBeliefDecisionPointsOptions {
|
|
7
|
-
source: CodeAgentSessionSource;
|
|
8
|
-
entries: unknown[];
|
|
9
|
-
run: Pick<RunRecord, 'runId' | 'scenarioId' | 'outcome' | 'costUsd'>;
|
|
10
|
-
sourcePath?: string;
|
|
11
|
-
}
|
|
12
|
-
export interface BeliefDecisionInventoryBucket {
|
|
13
|
-
id: string;
|
|
14
|
-
kind?: BeliefDecisionKind;
|
|
15
|
-
targetId?: CodeAgentBeliefDecisionTargetId;
|
|
16
|
-
n: number;
|
|
17
|
-
withOutcome: number;
|
|
18
|
-
withConfidence: number;
|
|
19
|
-
withCandidateActions: number;
|
|
20
|
-
withBehaviorProb: number;
|
|
21
|
-
withTargetProb: number;
|
|
22
|
-
successRate: number | null;
|
|
23
|
-
meanScore: number | null;
|
|
24
|
-
meanConfidence: number | null;
|
|
25
|
-
}
|
|
26
|
-
export interface BeliefDecisionInventoryReport {
|
|
27
|
-
n: number;
|
|
28
|
-
byKind: BeliefDecisionInventoryBucket[];
|
|
29
|
-
byTarget: BeliefDecisionInventoryBucket[];
|
|
30
|
-
diagnostics: string[];
|
|
31
|
-
}
|
|
32
|
-
export interface BeliefDecisionTargetSelection {
|
|
33
|
-
id: CodeAgentBeliefDecisionTargetId;
|
|
34
|
-
label: string;
|
|
35
|
-
points: BeliefDecisionPoint[];
|
|
36
|
-
support: BeliefDecisionInventoryBucket;
|
|
37
|
-
reasons: string[];
|
|
38
|
-
}
|
|
39
|
-
export interface SelectBeliefDecisionTargetOptions {
|
|
40
|
-
minN?: number;
|
|
41
|
-
minOutcomeCoverage?: number;
|
|
42
|
-
preferredTargets?: CodeAgentBeliefDecisionTargetId[];
|
|
43
|
-
}
|
|
44
|
-
export interface AnalyzeBeliefDecisionCorpusOptions {
|
|
45
|
-
points: BeliefDecisionPoint[];
|
|
46
|
-
targetId?: CodeAgentBeliefDecisionTargetId;
|
|
47
|
-
minN?: number;
|
|
48
|
-
minOutcomeCoverage?: number;
|
|
49
|
-
minAccepted?: number;
|
|
50
|
-
confidenceThreshold?: number;
|
|
51
|
-
policy?: BeliefSelectivePolicy;
|
|
52
|
-
requireOpe?: boolean;
|
|
53
|
-
policyOptions?: Partial<AnalyzeBeliefPolicyOptions>;
|
|
54
|
-
}
|
|
55
|
-
export interface BeliefDecisionCorpusEvaluation {
|
|
56
|
-
inventory: BeliefDecisionInventoryReport;
|
|
57
|
-
target?: BeliefDecisionTargetSelection;
|
|
58
|
-
policy?: BeliefSelectivePolicy;
|
|
59
|
-
evaluation?: BeliefPolicyEvaluationReport;
|
|
60
|
-
diagnostics: string[];
|
|
61
|
-
}
|
|
62
|
-
export declare function extractCodeAgentBeliefDecisionPoints(options: ExtractCodeAgentBeliefDecisionPointsOptions): BeliefDecisionExtractionReport;
|
|
63
|
-
export declare function inventoryBeliefDecisionPoints(points: BeliefDecisionPoint[]): BeliefDecisionInventoryReport;
|
|
64
|
-
export declare function selectBeliefDecisionTarget(points: BeliefDecisionPoint[], options?: SelectBeliefDecisionTargetOptions): BeliefDecisionTargetSelection | null;
|
|
65
|
-
export declare function analyzeBeliefDecisionCorpus(options: AnalyzeBeliefDecisionCorpusOptions): BeliefDecisionCorpusEvaluation;
|
|
66
|
-
//# sourceMappingURL=code-agent-corpus.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"code-agent-corpus.d.ts","sourceRoot":"","sources":["../../src/belief-state/code-agent-corpus.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,uCAAuC,CAAA;AACnF,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,eAAe,CAAA;AAE9C,OAAO,EAAE,KAAK,0BAA0B,EAAuB,MAAM,UAAU,CAAA;AAE/E,OAAO,KAAK,EAEV,8BAA8B,EAC9B,kBAAkB,EAElB,mBAAmB,EAEnB,4BAA4B,EAC5B,qBAAqB,EACtB,MAAM,SAAS,CAAA;AAEhB,MAAM,MAAM,+BAA+B,GACvC,kBAAkB,GAClB,gBAAgB,GAChB,kBAAkB,CAAA;AAEtB,MAAM,WAAW,2CAA2C;IAC1D,MAAM,EAAE,sBAAsB,CAAA;IAC9B,OAAO,EAAE,OAAO,EAAE,CAAA;IAClB,GAAG,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,GAAG,YAAY,GAAG,SAAS,GAAG,SAAS,CAAC,CAAA;IACpE,UAAU,CAAC,EAAE,MAAM,CAAA;CACpB;AAED,MAAM,WAAW,6BAA6B;IAC5C,EAAE,EAAE,MAAM,CAAA;IACV,IAAI,CAAC,EAAE,kBAAkB,CAAA;IACzB,QAAQ,CAAC,EAAE,+BAA+B,CAAA;IAC1C,CAAC,EAAE,MAAM,CAAA;IACT,WAAW,EAAE,MAAM,CAAA;IACnB,cAAc,EAAE,MAAM,CAAA;IACtB,oBAAoB,EAAE,MAAM,CAAA;IAC5B,gBAAgB,EAAE,MAAM,CAAA;IACxB,cAAc,EAAE,MAAM,CAAA;IACtB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAA;IAC1B,SAAS,EAAE,MAAM,GAAG,IAAI,CAAA;IACxB,cAAc,EAAE,MAAM,GAAG,IAAI,CAAA;CAC9B;AAED,MAAM,WAAW,6BAA6B;IAC5C,CAAC,EAAE,MAAM,CAAA;IACT,MAAM,EAAE,6BAA6B,EAAE,CAAA;IACvC,QAAQ,EAAE,6BAA6B,EAAE,CAAA;IACzC,WAAW,EAAE,MAAM,EAAE,CAAA;CACtB;AAED,MAAM,WAAW,6BAA6B;IAC5C,EAAE,EAAE,+BAA+B,CAAA;IACnC,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,mBAAmB,EAAE,CAAA;IAC7B,OAAO,EAAE,6BAA6B,CAAA;IACtC,OAAO,EAAE,MAAM,EAAE,CAAA;CAClB;AAED,MAAM,WAAW,iCAAiC;IAChD,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,kBAAkB,CAAC,EAAE,MAAM,CAAA;IAC3B,gBAAgB,CAAC,EAAE,+BAA+B,EAAE,CAAA;CACrD;AAED,MAAM,WAAW,kCAAkC;IACjD,MAAM,EAAE,mBAAmB,EAAE,CAAA;IAC7B,QAAQ,CAAC,EAAE,+BAA+B,CAAA;IAC1C,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,kBAAkB,CAAC,EAAE,MAAM,CAAA;IAC3B,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,mBAAmB,CAAC,EAAE,MAAM,CAAA;IAC5B,MAAM,CAAC,EAAE,qBAAqB,CAAA;IAC9B,UAAU,CAAC,EAAE,OAAO,CAAA;IACpB,aAAa,CAAC,EAAE,OAAO,CAAC,0BAA0B,CAAC,CAAA;CACpD;AAED,MAAM,WAAW,8BAA8B;IAC7C,SAAS,EAAE,6BAA6B,CAAA;IACxC,MAAM,CAAC,EAAE,6BAA6B,CAAA;IACtC,MAAM,CAAC,EAAE,qBAAqB,CAAA;IAC9B,UAAU,CAAC,EAAE,4BAA4B,CAAA;IACzC,WAAW,EAAE,MAAM,EAAE,CAAA;CACtB;AAyBD,wBAAgB,oCAAoC,CAClD,OAAO,EAAE,2CAA2C,GACnD,8BAA8B,CA0ChC;AAED,wBAAgB,6BAA6B,CAC3C,MAAM,EAAE,mBAAmB,EAAE,GAC5B,6BAA6B,CAyB/B;AAED,wBAAgB,0BAA0B,CACxC,MAAM,EAAE,mBAAmB,EAAE,EAC7B,OAAO,GAAE,iCAAsC,GAC9C,6BAA6B,GAAG,IAAI,CAiCtC;AAED,wBAAgB,2BAA2B,CACzC,OAAO,EAAE,kCAAkC,GAC1C,8BAA8B,CA6ChC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"code-agent-corpus.test.d.ts","sourceRoot":"","sources":["../../src/belief-state/code-agent-corpus.test.ts"],"names":[],"mappings":""}
|
|
@@ -1,22 +0,0 @@
|
|
|
1
|
-
import { type CodeAgentSessionDiagnostic, type CodeAgentSessionIntakeOptions, type CodeAgentSessionMetrics, type CodeAgentSessionSource } from '../contract/intake/code-agent-session';
|
|
2
|
-
import type { RunRecord } from '../run-record';
|
|
3
|
-
import { type BeliefDecisionInventoryReport } from './code-agent-corpus';
|
|
4
|
-
import { type BeliefDecisionResearchEvidencePacket, type BuildBeliefDecisionResearchEvidencePacketOptions } from './research-evidence';
|
|
5
|
-
import type { BeliefDecisionExtractionDiagnostic, BeliefDecisionPoint } from './types';
|
|
6
|
-
export interface CodeAgentBeliefSession extends CodeAgentSessionIntakeOptions {
|
|
7
|
-
source: CodeAgentSessionSource;
|
|
8
|
-
}
|
|
9
|
-
export interface BuildCodeAgentBeliefEvidenceCorpusOptions extends Omit<BuildBeliefDecisionResearchEvidencePacketOptions, 'points'> {
|
|
10
|
-
sessions: CodeAgentBeliefSession[];
|
|
11
|
-
}
|
|
12
|
-
export interface CodeAgentBeliefEvidenceCorpus {
|
|
13
|
-
runs: RunRecord[];
|
|
14
|
-
metrics: CodeAgentSessionMetrics[];
|
|
15
|
-
intakeDiagnostics: CodeAgentSessionDiagnostic[];
|
|
16
|
-
extractionDiagnostics: BeliefDecisionExtractionDiagnostic[];
|
|
17
|
-
decisions: BeliefDecisionPoint[];
|
|
18
|
-
inventory: BeliefDecisionInventoryReport;
|
|
19
|
-
evidence: BeliefDecisionResearchEvidencePacket;
|
|
20
|
-
}
|
|
21
|
-
export declare function buildCodeAgentBeliefEvidenceCorpus(options: BuildCodeAgentBeliefEvidenceCorpusOptions): CodeAgentBeliefEvidenceCorpus;
|
|
22
|
-
//# sourceMappingURL=code-agent-evidence.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"code-agent-evidence.d.ts","sourceRoot":"","sources":["../../src/belief-state/code-agent-evidence.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,KAAK,0BAA0B,EAC/B,KAAK,6BAA6B,EAElC,KAAK,uBAAuB,EAC5B,KAAK,sBAAsB,EAM5B,MAAM,uCAAuC,CAAA;AAC9C,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,eAAe,CAAA;AAC9C,OAAO,EACL,KAAK,6BAA6B,EAGnC,MAAM,qBAAqB,CAAA;AAC5B,OAAO,EACL,KAAK,oCAAoC,EACzC,KAAK,gDAAgD,EAEtD,MAAM,qBAAqB,CAAA;AAC5B,OAAO,KAAK,EAAE,kCAAkC,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAA;AAEtF,MAAM,WAAW,sBAAuB,SAAQ,6BAA6B;IAC3E,MAAM,EAAE,sBAAsB,CAAA;CAC/B;AAED,MAAM,WAAW,yCACf,SAAQ,IAAI,CAAC,gDAAgD,EAAE,QAAQ,CAAC;IACxE,QAAQ,EAAE,sBAAsB,EAAE,CAAA;CACnC;AAED,MAAM,WAAW,6BAA6B;IAC5C,IAAI,EAAE,SAAS,EAAE,CAAA;IACjB,OAAO,EAAE,uBAAuB,EAAE,CAAA;IAClC,iBAAiB,EAAE,0BAA0B,EAAE,CAAA;IAC/C,qBAAqB,EAAE,kCAAkC,EAAE,CAAA;IAC3D,SAAS,EAAE,mBAAmB,EAAE,CAAA;IAChC,SAAS,EAAE,6BAA6B,CAAA;IACxC,QAAQ,EAAE,oCAAoC,CAAA;CAC/C;AAED,wBAAgB,kCAAkC,CAChD,OAAO,EAAE,yCAAyC,GACjD,6BAA6B,CAwC/B"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"code-agent-evidence.test.d.ts","sourceRoot":"","sources":["../../src/belief-state/code-agent-evidence.test.ts"],"names":[],"mappings":""}
|
|
@@ -1,7 +0,0 @@
|
|
|
1
|
-
import type { TraceStore } from '../trace/store';
|
|
2
|
-
import type { BeliefDecisionExtractionReport } from './types';
|
|
3
|
-
export interface ExtractBeliefDecisionPointsOptions {
|
|
4
|
-
runIds?: string[];
|
|
5
|
-
}
|
|
6
|
-
export declare function extractBeliefDecisionPoints(store: TraceStore, options?: ExtractBeliefDecisionPointsOptions): Promise<BeliefDecisionExtractionReport>;
|
|
7
|
-
//# sourceMappingURL=extract.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"extract.d.ts","sourceRoot":"","sources":["../../src/belief-state/extract.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,gBAAgB,CAAA;AAChD,OAAO,KAAK,EAEV,8BAA8B,EAI/B,MAAM,SAAS,CAAA;AAGhB,MAAM,WAAW,kCAAkC;IACjD,MAAM,CAAC,EAAE,MAAM,EAAE,CAAA;CAClB;AAID,wBAAsB,2BAA2B,CAC/C,KAAK,EAAE,UAAU,EACjB,OAAO,GAAE,kCAAuC,GAC/C,OAAO,CAAC,8BAA8B,CAAC,CA8BzC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"extract.test.d.ts","sourceRoot":"","sources":["../../src/belief-state/extract.test.ts"],"names":[],"mappings":""}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/belief-state/index.ts"],"names":[],"mappings":"AAAA,cAAc,eAAe,CAAA;AAC7B,cAAc,qBAAqB,CAAA;AACnC,cAAc,uBAAuB,CAAA;AACrC,cAAc,WAAW,CAAA;AACzB,cAAc,OAAO,CAAA;AACrB,cAAc,sBAAsB,CAAA;AACpC,cAAc,UAAU,CAAA;AACxB,cAAc,qBAAqB,CAAA;AACnC,cAAc,4BAA4B,CAAA;AAC1C,cAAc,iBAAiB,CAAA;AAC/B,cAAc,aAAa,CAAA;AAC3B,cAAc,gBAAgB,CAAA;AAC9B,cAAc,SAAS,CAAA"}
|
|
@@ -1,17 +0,0 @@
|
|
|
1
|
-
import { type OffPolicyOptions, type OffPolicyTrajectory } from '../rl/off-policy';
|
|
2
|
-
import type { BeliefDecisionPoint, BeliefOpeReport, BeliefOpeTargetPolicy } from './types';
|
|
3
|
-
export interface BeliefOpeOptions extends OffPolicyOptions {
|
|
4
|
-
minEffectiveSampleSize?: number;
|
|
5
|
-
minEffectiveSampleRatio?: number;
|
|
6
|
-
maxDiagnostics?: number;
|
|
7
|
-
}
|
|
8
|
-
export interface BeliefOffPolicyTrajectoryReport {
|
|
9
|
-
targetPolicyId: string;
|
|
10
|
-
trajectories: OffPolicyTrajectory[];
|
|
11
|
-
dropped: number;
|
|
12
|
-
diagnostics: string[];
|
|
13
|
-
}
|
|
14
|
-
export declare function embeddedBeliefOpeTargetPolicy(id?: string): BeliefOpeTargetPolicy;
|
|
15
|
-
export declare function beliefDecisionsToOffPolicyTrajectories(points: BeliefDecisionPoint[], targetPolicy: BeliefOpeTargetPolicy, options?: Pick<BeliefOpeOptions, 'maxDiagnostics'>): BeliefOffPolicyTrajectoryReport;
|
|
16
|
-
export declare function evaluateBeliefOffPolicy(points: BeliefDecisionPoint[], targetPolicy: BeliefOpeTargetPolicy, options?: BeliefOpeOptions): BeliefOpeReport;
|
|
17
|
-
//# sourceMappingURL=ope.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"ope.d.ts","sourceRoot":"","sources":["../../src/belief-state/ope.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,KAAK,gBAAgB,EACrB,KAAK,mBAAmB,EAEzB,MAAM,kBAAkB,CAAA;AACzB,OAAO,KAAK,EACV,mBAAmB,EACnB,eAAe,EAEf,qBAAqB,EACtB,MAAM,SAAS,CAAA;AAEhB,MAAM,WAAW,gBAAiB,SAAQ,gBAAgB;IACxD,sBAAsB,CAAC,EAAE,MAAM,CAAA;IAC/B,uBAAuB,CAAC,EAAE,MAAM,CAAA;IAChC,cAAc,CAAC,EAAE,MAAM,CAAA;CACxB;AAED,MAAM,WAAW,+BAA+B;IAC9C,cAAc,EAAE,MAAM,CAAA;IACtB,YAAY,EAAE,mBAAmB,EAAE,CAAA;IACnC,OAAO,EAAE,MAAM,CAAA;IACf,WAAW,EAAE,MAAM,EAAE,CAAA;CACtB;AAED,wBAAgB,6BAA6B,CAAC,EAAE,SAAyB,GAAG,qBAAqB,CAUhG;AAED,wBAAgB,sCAAsC,CACpD,MAAM,EAAE,mBAAmB,EAAE,EAC7B,YAAY,EAAE,qBAAqB,EACnC,OAAO,GAAE,IAAI,CAAC,gBAAgB,EAAE,gBAAgB,CAAM,GACrD,+BAA+B,CA+CjC;AAED,wBAAgB,uBAAuB,CACrC,MAAM,EAAE,mBAAmB,EAAE,EAC7B,YAAY,EAAE,qBAAqB,EACnC,OAAO,GAAE,gBAAqB,GAC7B,eAAe,CAWjB"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"ope.test.d.ts","sourceRoot":"","sources":["../../src/belief-state/ope.test.ts"],"names":[],"mappings":""}
|
|
@@ -1,55 +0,0 @@
|
|
|
1
|
-
import type { RunSplitTag } from '../run-record';
|
|
2
|
-
import { type BeliefDecisionResearchEvidencePacket, type BuildBeliefDecisionResearchEvidencePacketOptions } from './research-evidence';
|
|
3
|
-
import { type RuntimeBeliefDecisionPoint, type RuntimeBeliefHookEvent } from './runtime-hooks';
|
|
4
|
-
import type { BeliefDecisionOutcome, BeliefDecisionPoint } from './types';
|
|
5
|
-
export interface RuntimeBeliefPhase0RunRecord {
|
|
6
|
-
runId: string;
|
|
7
|
-
scenarioId?: string;
|
|
8
|
-
splitTag: RunSplitTag;
|
|
9
|
-
}
|
|
10
|
-
export interface RuntimeBeliefDecisionLabel {
|
|
11
|
-
decisionId: string;
|
|
12
|
-
chosenAction: string;
|
|
13
|
-
outcome: BeliefDecisionOutcome;
|
|
14
|
-
confidence?: number;
|
|
15
|
-
behaviorProb?: number;
|
|
16
|
-
targetProb?: number;
|
|
17
|
-
qHat?: number | null;
|
|
18
|
-
costUsd?: number;
|
|
19
|
-
splitTag?: RunSplitTag;
|
|
20
|
-
metadata?: Record<string, unknown>;
|
|
21
|
-
}
|
|
22
|
-
export interface BuildRuntimeBeliefPhase0MeasurementOptions extends Omit<BuildBeliefDecisionResearchEvidencePacketOptions, 'points'> {
|
|
23
|
-
runs: RuntimeBeliefPhase0RunRecord[];
|
|
24
|
-
decisions: RuntimeBeliefDecisionPoint[];
|
|
25
|
-
events?: RuntimeBeliefHookEvent[];
|
|
26
|
-
labels: RuntimeBeliefDecisionLabel[];
|
|
27
|
-
baselinePolicyId?: string;
|
|
28
|
-
}
|
|
29
|
-
export interface RuntimeBeliefPhase0MeasurementSummary {
|
|
30
|
-
runCount: number;
|
|
31
|
-
producerDecisionCount: number;
|
|
32
|
-
lifecycleEventCount: number;
|
|
33
|
-
labelCount: number;
|
|
34
|
-
completedPointCount: number;
|
|
35
|
-
runJoinRate: number;
|
|
36
|
-
labelJoinRate: number;
|
|
37
|
-
missingRunRecordCount: number;
|
|
38
|
-
missingLabelCount: number;
|
|
39
|
-
withEvidence: number;
|
|
40
|
-
withOutcome: number;
|
|
41
|
-
withSplit: number;
|
|
42
|
-
withBehaviorProb: number;
|
|
43
|
-
withTargetProb: number;
|
|
44
|
-
baselinePolicyId: string;
|
|
45
|
-
packetStatus: BeliefDecisionResearchEvidencePacket['status'];
|
|
46
|
-
claimScope: BeliefDecisionResearchEvidencePacket['claimScope'];
|
|
47
|
-
}
|
|
48
|
-
export interface RuntimeBeliefPhase0Measurement {
|
|
49
|
-
points: BeliefDecisionPoint[];
|
|
50
|
-
packet: BeliefDecisionResearchEvidencePacket;
|
|
51
|
-
summary: RuntimeBeliefPhase0MeasurementSummary;
|
|
52
|
-
diagnostics: string[];
|
|
53
|
-
}
|
|
54
|
-
export declare function buildRuntimeBeliefPhase0Measurement(options: BuildRuntimeBeliefPhase0MeasurementOptions): RuntimeBeliefPhase0Measurement;
|
|
55
|
-
//# sourceMappingURL=phase0-measurement.d.ts.map
|