@tangle-network/agent-eval 0.95.0 → 0.95.1
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/{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/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 +688 -106
- 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-C1NCIZ9o.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/{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-CncDq9qE.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/{matrix/types.d.ts → types-BUxNaJ8c.d.ts} +11 -9
- package/dist/{types.d.ts → types-C7DGg5ex.d.ts} +33 -31
- package/dist/{campaign/types.d.ts → types-DQRY8ZT-.d.ts} +47 -44
- 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/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
package/dist/judges.d.ts
DELETED
|
@@ -1,74 +0,0 @@
|
|
|
1
|
-
import { JudgeError } from './errors';
|
|
2
|
-
import type { JudgeFn } from './types';
|
|
3
|
-
/**
|
|
4
|
-
* A judge's LLM response could not be parsed into scored dimensions.
|
|
5
|
-
* Thrown instead of fabricating a `{ dimension: 'parse_error', score: 0 }`
|
|
6
|
-
* row — a synthetic zero is indistinguishable from a real low score
|
|
7
|
-
* downstream. Carries the raw response for forensics. Callers (executor,
|
|
8
|
-
* ensemble wrappers) catch this per-judge and record a failed judge.
|
|
9
|
-
*/
|
|
10
|
-
export declare class JudgeParseError extends JudgeError {
|
|
11
|
-
/** Name of the judge whose response failed to parse. */
|
|
12
|
-
readonly judgeName: string;
|
|
13
|
-
/** The raw (truncated) model response that failed to parse. */
|
|
14
|
-
readonly raw: string;
|
|
15
|
-
constructor(judgeName: string, raw: string, options?: {
|
|
16
|
-
cause?: unknown;
|
|
17
|
-
});
|
|
18
|
-
}
|
|
19
|
-
/**
|
|
20
|
-
* Create a domain expert judge with a configurable domain.
|
|
21
|
-
*
|
|
22
|
-
* The judge evaluates professional accuracy and depth.
|
|
23
|
-
*
|
|
24
|
-
* @deprecated Legacy `JudgeFn` factory tied to the fixed gpt-4o prompt shape.
|
|
25
|
-
* Build judges as campaign `JudgeConfig`s (src/campaign/types.ts) — or
|
|
26
|
-
* multi-model panels via `ensembleJudge` (src/judge-panel.ts) — which are
|
|
27
|
-
* pluggable, fail-loud, and drive the campaign/improvement-loop engines.
|
|
28
|
-
*/
|
|
29
|
-
export declare function createDomainExpertJudge(domain: string): JudgeFn;
|
|
30
|
-
/**
|
|
31
|
-
* Code execution judge — evaluates whether code blocks are valid and runnable.
|
|
32
|
-
*
|
|
33
|
-
* @deprecated Legacy `JudgeFn` factory tied to the fixed gpt-4o prompt shape.
|
|
34
|
-
* Build judges as campaign `JudgeConfig`s (src/campaign/types.ts) — or
|
|
35
|
-
* multi-model panels via `ensembleJudge` (src/judge-panel.ts).
|
|
36
|
-
*/
|
|
37
|
-
export declare const codeExecutionJudge: JudgeFn;
|
|
38
|
-
/**
|
|
39
|
-
* Coherence judge — evaluates multi-turn consistency and progression.
|
|
40
|
-
*
|
|
41
|
-
* @deprecated Legacy `JudgeFn` factory tied to the fixed gpt-4o prompt shape.
|
|
42
|
-
* Build judges as campaign `JudgeConfig`s (src/campaign/types.ts) — or
|
|
43
|
-
* multi-model panels via `ensembleJudge` (src/judge-panel.ts).
|
|
44
|
-
*/
|
|
45
|
-
export declare const coherenceJudge: JudgeFn;
|
|
46
|
-
/**
|
|
47
|
-
* Adversarial judge — red-teams agent responses.
|
|
48
|
-
*
|
|
49
|
-
* @deprecated Legacy `JudgeFn` factory tied to the fixed gpt-4o prompt shape.
|
|
50
|
-
* Build judges as campaign `JudgeConfig`s (src/campaign/types.ts) — or
|
|
51
|
-
* multi-model panels via `ensembleJudge` (src/judge-panel.ts).
|
|
52
|
-
*/
|
|
53
|
-
export declare const adversarialJudge: JudgeFn;
|
|
54
|
-
/**
|
|
55
|
-
* Create a custom judge with a fully custom prompt.
|
|
56
|
-
*
|
|
57
|
-
* @deprecated Legacy `JudgeFn` factory tied to the fixed gpt-4o prompt shape.
|
|
58
|
-
* Build judges as campaign `JudgeConfig`s (src/campaign/types.ts) — or
|
|
59
|
-
* multi-model panels via `ensembleJudge` (src/judge-panel.ts).
|
|
60
|
-
*/
|
|
61
|
-
export declare function createCustomJudge(name: string, systemPrompt: string, opts?: {
|
|
62
|
-
model?: string;
|
|
63
|
-
temperature?: number;
|
|
64
|
-
maxTokens?: number;
|
|
65
|
-
}): JudgeFn;
|
|
66
|
-
/**
|
|
67
|
-
* Default judge set (domain must be provided for domain expert)
|
|
68
|
-
*
|
|
69
|
-
* @deprecated Legacy `JudgeFn` factory tied to the fixed gpt-4o prompt shape.
|
|
70
|
-
* Build judges as campaign `JudgeConfig`s (src/campaign/types.ts) — or
|
|
71
|
-
* multi-model panels via `ensembleJudge` (src/judge-panel.ts).
|
|
72
|
-
*/
|
|
73
|
-
export declare function defaultJudges(domain: string): JudgeFn[];
|
|
74
|
-
//# sourceMappingURL=judges.d.ts.map
|
package/dist/judges.d.ts.map
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"judges.d.ts","sourceRoot":"","sources":["../src/judges.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,UAAU,EAAE,MAAM,UAAU,CAAA;AACrC,OAAO,KAAK,EAAE,OAAO,EAA0B,MAAM,SAAS,CAAA;AAE9D;;;;;;GAMG;AACH,qBAAa,eAAgB,SAAQ,UAAU;IAC7C,wDAAwD;IACxD,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;IAC1B,+DAA+D;IAC/D,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAA;gBAER,SAAS,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,OAAO,CAAA;KAAE;CAK1E;AAED;;;;;;;;;GASG;AACH,wBAAgB,uBAAuB,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAsC/D;AAED;;;;;;GAMG;AACH,eAAO,MAAM,kBAAkB,EAAE,OA4ChC,CAAA;AAED;;;;;;GAMG;AACH,eAAO,MAAM,cAAc,EAAE,OAuC5B,CAAA;AAED;;;;;;GAMG;AACH,eAAO,MAAM,gBAAgB,EAAE,OAgC9B,CAAA;AAED;;;;;;GAMG;AACH,wBAAgB,iBAAiB,CAC/B,IAAI,EAAE,MAAM,EACZ,YAAY,EAAE,MAAM,EACpB,IAAI,CAAC,EAAE;IAAE,KAAK,CAAC,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,MAAM,CAAA;CAAE,GAClE,OAAO,CA2BT;AAED;;;;;;GAMG;AACH,wBAAgB,aAAa,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,EAAE,CAEvD"}
|
|
@@ -1,89 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Keyword-coverage judge — baseline complement to the semantic concept
|
|
3
|
-
* judge.
|
|
4
|
-
*
|
|
5
|
-
* Where {@link runSemanticConceptJudge} uses an LLM to read source code
|
|
6
|
-
* and decide whether a concept is REALLY implemented (not just
|
|
7
|
-
* keyword-mentioned), this judge does the cheap, deterministic version:
|
|
8
|
-
* fetch the served preview, concatenate every linked CSS/JS asset, and
|
|
9
|
-
* substring-match each expected concept's keywords against the
|
|
10
|
-
* concatenated haystack. Optional `requiredElement` selector adds a
|
|
11
|
-
* structural gate so "supply counter" can require an actual `<input>` or
|
|
12
|
-
* `<table>`, not just a comment containing the word.
|
|
13
|
-
*
|
|
14
|
-
* Use both judges. Keyword coverage is a fast 0-cost gate — a stub page
|
|
15
|
-
* with the right keywords passes here, fails the semantic judge. Score
|
|
16
|
-
* divergence between the two is itself a signal: high keyword coverage
|
|
17
|
-
* + low semantic = "the agent slapped the right words on the right
|
|
18
|
-
* scaffold but didn't wire any of it up."
|
|
19
|
-
*
|
|
20
|
-
* Pure functions, soft-fail on fetch error, no LLM dependency.
|
|
21
|
-
*/
|
|
22
|
-
export interface KeywordConceptSpec {
|
|
23
|
-
name: string;
|
|
24
|
-
keywords: string[];
|
|
25
|
-
/**
|
|
26
|
-
* Optional CSS selector that must match in the HTML for the concept
|
|
27
|
-
* to count as present. Tiny subset:
|
|
28
|
-
* - `tag` (e.g. `form`)
|
|
29
|
-
* - `tag[attr="value"]` (e.g. `input[type="number"]`)
|
|
30
|
-
* - `tag[attr]` (presence only)
|
|
31
|
-
* Anything more complex is rejected with `null` (treated as
|
|
32
|
-
* "unenforced", not "failed").
|
|
33
|
-
*/
|
|
34
|
-
requiredElement?: string;
|
|
35
|
-
}
|
|
36
|
-
export interface KeywordCoverageFinding {
|
|
37
|
-
concept: string;
|
|
38
|
-
found: boolean;
|
|
39
|
-
matchedKeywords: string[];
|
|
40
|
-
/** True iff the optional requiredElement selector matched; null when no selector. */
|
|
41
|
-
requiredElementPresent: boolean | null;
|
|
42
|
-
}
|
|
43
|
-
export interface KeywordCoverageResult {
|
|
44
|
-
/** 0..1 share of concepts satisfied. */
|
|
45
|
-
score: number;
|
|
46
|
-
presentCount: number;
|
|
47
|
-
totalCount: number;
|
|
48
|
-
findings: KeywordCoverageFinding[];
|
|
49
|
-
durationMs: number;
|
|
50
|
-
/** Total bytes assembled across html + linked assets. */
|
|
51
|
-
totalAssembledBytes: number;
|
|
52
|
-
/** Soft-failure reason if the audit couldn't run. */
|
|
53
|
-
error?: string;
|
|
54
|
-
}
|
|
55
|
-
export interface KeywordCoverageOptions {
|
|
56
|
-
/** Override fetch implementation — for tests. */
|
|
57
|
-
fetch?: typeof fetch;
|
|
58
|
-
/** Per-asset fetch timeout (default 3s). */
|
|
59
|
-
assetTimeoutMs?: number;
|
|
60
|
-
/** Initial-HTML fetch timeout (default 5s). */
|
|
61
|
-
htmlTimeoutMs?: number;
|
|
62
|
-
}
|
|
63
|
-
/**
|
|
64
|
-
* Element-presence check using a tiny CSS-selector subset. Returns
|
|
65
|
-
* null when the selector isn't supported — caller treats that as
|
|
66
|
-
* "unenforced" rather than "failed."
|
|
67
|
-
*/
|
|
68
|
-
export declare function htmlContainsElement(html: string, selector: string): boolean | null;
|
|
69
|
-
/**
|
|
70
|
-
* Pull every `<link rel=stylesheet href>` and `<script src>` from a
|
|
71
|
-
* raw HTML body. Returns absolute URLs resolved against `baseUrl`.
|
|
72
|
-
* Permissive regex — agent-authored markup doesn't always quote
|
|
73
|
-
* attributes the same way.
|
|
74
|
-
*/
|
|
75
|
-
export declare function extractAssetUrls(html: string, baseUrl: string): string[];
|
|
76
|
-
/**
|
|
77
|
-
* Score expected concepts against an already-fetched HTML payload + any
|
|
78
|
-
* pre-fetched CSS/JS assets. Use when the runner has the bytes in hand
|
|
79
|
-
* and doesn't want a fresh HTTP round-trip — e.g. sandbox runtime where
|
|
80
|
-
* the preview content was fetched via curl from inside the container.
|
|
81
|
-
*/
|
|
82
|
-
export declare function runKeywordCoverageJudge(html: string, expectedConcepts: ReadonlyArray<KeywordConceptSpec>, assets?: ReadonlyArray<string>): KeywordCoverageResult;
|
|
83
|
-
/**
|
|
84
|
-
* URL-fetch flavor — GET the preview, parallel-fetch every linked
|
|
85
|
-
* stylesheet + script (with bounded timeouts, soft-fail individually),
|
|
86
|
-
* then score via {@link runKeywordCoverageJudge}.
|
|
87
|
-
*/
|
|
88
|
-
export declare function runKeywordCoverageJudgeUrl(previewUrl: string, expectedConcepts: ReadonlyArray<KeywordConceptSpec>, options?: KeywordCoverageOptions): Promise<KeywordCoverageResult>;
|
|
89
|
-
//# sourceMappingURL=keyword-coverage-judge.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"keyword-coverage-judge.d.ts","sourceRoot":"","sources":["../src/keyword-coverage-judge.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAIH,MAAM,WAAW,kBAAkB;IACjC,IAAI,EAAE,MAAM,CAAA;IACZ,QAAQ,EAAE,MAAM,EAAE,CAAA;IAClB;;;;;;;;OAQG;IACH,eAAe,CAAC,EAAE,MAAM,CAAA;CACzB;AAED,MAAM,WAAW,sBAAsB;IACrC,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,EAAE,OAAO,CAAA;IACd,eAAe,EAAE,MAAM,EAAE,CAAA;IACzB,qFAAqF;IACrF,sBAAsB,EAAE,OAAO,GAAG,IAAI,CAAA;CACvC;AAED,MAAM,WAAW,qBAAqB;IACpC,wCAAwC;IACxC,KAAK,EAAE,MAAM,CAAA;IACb,YAAY,EAAE,MAAM,CAAA;IACpB,UAAU,EAAE,MAAM,CAAA;IAClB,QAAQ,EAAE,sBAAsB,EAAE,CAAA;IAClC,UAAU,EAAE,MAAM,CAAA;IAClB,yDAAyD;IACzD,mBAAmB,EAAE,MAAM,CAAA;IAC3B,qDAAqD;IACrD,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,sBAAsB;IACrC,iDAAiD;IACjD,KAAK,CAAC,EAAE,OAAO,KAAK,CAAA;IACpB,4CAA4C;IAC5C,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,+CAA+C;IAC/C,aAAa,CAAC,EAAE,MAAM,CAAA;CACvB;AAID;;;;GAIG;AACH,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,GAAG,OAAO,GAAG,IAAI,CAsBlF;AAID;;;;;GAKG;AACH,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,MAAM,EAAE,CAiBxE;AAID;;;;;GAKG;AACH,wBAAgB,uBAAuB,CACrC,IAAI,EAAE,MAAM,EACZ,gBAAgB,EAAE,aAAa,CAAC,kBAAkB,CAAC,EACnD,MAAM,GAAE,aAAa,CAAC,MAAM,CAAM,GACjC,qBAAqB,CAkCvB;AAED;;;;GAIG;AACH,wBAAsB,0BAA0B,CAC9C,UAAU,EAAE,MAAM,EAClB,gBAAgB,EAAE,aAAa,CAAC,kBAAkB,CAAC,EACnD,OAAO,GAAE,sBAA2B,GACnC,OAAO,CAAC,qBAAqB,CAAC,CAgDhC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"keyword-coverage-judge.test.d.ts","sourceRoot":"","sources":["../src/keyword-coverage-judge.test.ts"],"names":[],"mappings":""}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/knowledge/index.ts"],"names":[],"mappings":"AAAA,cAAc,aAAa,CAAA;AAC3B,cAAc,SAAS,CAAA"}
|
|
@@ -1,26 +0,0 @@
|
|
|
1
|
-
import { type ControlEvalResult } from '../control-runtime';
|
|
2
|
-
import type { TraceEmitter } from '../trace/emitter';
|
|
3
|
-
import type { DataAcquisitionPlan, KnowledgeReadinessReport, KnowledgeRequirement, UserQuestion } from './types';
|
|
4
|
-
export interface ScoreKnowledgeReadinessOptions {
|
|
5
|
-
taskId: string;
|
|
6
|
-
requirements: KnowledgeRequirement[];
|
|
7
|
-
evidenceIds?: string[];
|
|
8
|
-
claimIds?: string[];
|
|
9
|
-
wikiPageIds?: string[];
|
|
10
|
-
userAnswers?: Record<string, string>;
|
|
11
|
-
metadata?: Record<string, unknown>;
|
|
12
|
-
now?: Date;
|
|
13
|
-
}
|
|
14
|
-
export declare function scoreKnowledgeReadiness(options: ScoreKnowledgeReadinessOptions): KnowledgeReadinessReport;
|
|
15
|
-
export declare function blockingKnowledgeEval(report: KnowledgeReadinessReport, options?: {
|
|
16
|
-
id?: string;
|
|
17
|
-
minimumScore?: number;
|
|
18
|
-
emitter?: TraceEmitter;
|
|
19
|
-
}): ControlEvalResult;
|
|
20
|
-
export declare function knowledgeReadinessTracePayload(report: KnowledgeReadinessReport, options?: {
|
|
21
|
-
passed?: boolean;
|
|
22
|
-
minimumScore?: number;
|
|
23
|
-
}): Record<string, unknown>;
|
|
24
|
-
export declare function userQuestionsForKnowledgeGaps(gaps: KnowledgeRequirement[]): UserQuestion[];
|
|
25
|
-
export declare function acquisitionPlansForKnowledgeGaps(gaps: KnowledgeRequirement[]): DataAcquisitionPlan[];
|
|
26
|
-
//# sourceMappingURL=readiness.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"readiness.d.ts","sourceRoot":"","sources":["../../src/knowledge/readiness.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,iBAAiB,EAAiB,MAAM,oBAAoB,CAAA;AAC1E,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAA;AACpD,OAAO,KAAK,EACV,mBAAmB,EAInB,wBAAwB,EAExB,oBAAoB,EACpB,YAAY,EACb,MAAM,SAAS,CAAA;AAEhB,MAAM,WAAW,8BAA8B;IAC7C,MAAM,EAAE,MAAM,CAAA;IACd,YAAY,EAAE,oBAAoB,EAAE,CAAA;IACpC,WAAW,CAAC,EAAE,MAAM,EAAE,CAAA;IACtB,QAAQ,CAAC,EAAE,MAAM,EAAE,CAAA;IACnB,WAAW,CAAC,EAAE,MAAM,EAAE,CAAA;IACtB,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IACpC,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAClC,GAAG,CAAC,EAAE,IAAI,CAAA;CACX;AAED,wBAAgB,uBAAuB,CACrC,OAAO,EAAE,8BAA8B,GACtC,wBAAwB,CA6C1B;AAED,wBAAgB,qBAAqB,CACnC,MAAM,EAAE,wBAAwB,EAChC,OAAO,GAAE;IAAE,EAAE,CAAC,EAAE,MAAM,CAAC;IAAC,YAAY,CAAC,EAAE,MAAM,CAAC;IAAC,OAAO,CAAC,EAAE,YAAY,CAAA;CAAO,GAC3E,iBAAiB,CAqBnB;AAED,wBAAgB,8BAA8B,CAC5C,MAAM,EAAE,wBAAwB,EAChC,OAAO,GAAE;IAAE,MAAM,CAAC,EAAE,OAAO,CAAC;IAAC,YAAY,CAAC,EAAE,MAAM,CAAA;CAAO,GACxD,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAazB;AAED,wBAAgB,6BAA6B,CAAC,IAAI,EAAE,oBAAoB,EAAE,GAAG,YAAY,EAAE,CAY1F;AAED,wBAAgB,gCAAgC,CAC9C,IAAI,EAAE,oBAAoB,EAAE,GAC3B,mBAAmB,EAAE,CAiBvB"}
|
|
@@ -1,75 +0,0 @@
|
|
|
1
|
-
import type { ControlSeverity } from '../control-runtime';
|
|
2
|
-
export type KnowledgeRequirementCategory = 'user_specific' | 'company_specific' | 'domain_specific' | 'codebase_specific' | 'market_specific' | 'regulatory' | 'tool_api' | 'credential_or_secret' | 'runtime_environment' | 'preference' | 'historical_context';
|
|
3
|
-
export type KnowledgeAcquisitionMode = 'ask_user' | 'search_web' | 'query_connector' | 'inspect_repo' | 'run_command' | 'infer_low_confidence' | 'not_available';
|
|
4
|
-
export type KnowledgeImportance = 'blocking' | 'high' | 'medium' | 'low';
|
|
5
|
-
export type KnowledgeFreshness = 'static' | 'monthly' | 'weekly' | 'daily' | 'realtime';
|
|
6
|
-
export type KnowledgeSensitivity = 'public' | 'private' | 'secret';
|
|
7
|
-
export type KnowledgeFallbackPolicy = 'block' | 'ask' | 'continue_with_caveat' | 'use_default';
|
|
8
|
-
export interface KnowledgeRequirement {
|
|
9
|
-
id: string;
|
|
10
|
-
description: string;
|
|
11
|
-
requiredFor: string[];
|
|
12
|
-
category: KnowledgeRequirementCategory;
|
|
13
|
-
acquisitionMode: KnowledgeAcquisitionMode;
|
|
14
|
-
importance: KnowledgeImportance;
|
|
15
|
-
freshness: KnowledgeFreshness;
|
|
16
|
-
sensitivity: KnowledgeSensitivity;
|
|
17
|
-
confidenceNeeded: number;
|
|
18
|
-
currentConfidence: number;
|
|
19
|
-
evidenceIds: string[];
|
|
20
|
-
fallbackPolicy: KnowledgeFallbackPolicy;
|
|
21
|
-
/**
|
|
22
|
-
* ISO timestamp after which this requirement must be treated as stale.
|
|
23
|
-
* Stale requirements score as missing even when they still have evidence.
|
|
24
|
-
*/
|
|
25
|
-
validUntil?: string;
|
|
26
|
-
/** ISO timestamp for the last source-grounding or human verification pass. */
|
|
27
|
-
lastVerifiedAt?: string;
|
|
28
|
-
metadata?: Record<string, unknown>;
|
|
29
|
-
}
|
|
30
|
-
export interface KnowledgeBundle {
|
|
31
|
-
taskId: string;
|
|
32
|
-
requirements: KnowledgeRequirement[];
|
|
33
|
-
evidenceIds: string[];
|
|
34
|
-
claimIds: string[];
|
|
35
|
-
wikiPageIds: string[];
|
|
36
|
-
userAnswers: Record<string, string>;
|
|
37
|
-
missing: KnowledgeRequirement[];
|
|
38
|
-
readinessScore: number;
|
|
39
|
-
metadata?: Record<string, unknown>;
|
|
40
|
-
}
|
|
41
|
-
export type KnowledgeRecommendedAction = 'run_agent' | 'ask_user' | 'collect_web_data' | 'query_connectors' | 'inspect_repo' | 'build_domain_wiki' | 'continue_with_caveat' | 'abort_or_rescope';
|
|
42
|
-
export interface KnowledgeReadinessReport {
|
|
43
|
-
taskId: string;
|
|
44
|
-
readinessScore: number;
|
|
45
|
-
blockingMissingRequirements: KnowledgeRequirement[];
|
|
46
|
-
nonBlockingGaps: KnowledgeRequirement[];
|
|
47
|
-
recommendedAction: KnowledgeRecommendedAction;
|
|
48
|
-
bundle: KnowledgeBundle;
|
|
49
|
-
severity: ControlSeverity;
|
|
50
|
-
reason: string;
|
|
51
|
-
}
|
|
52
|
-
export interface UserQuestion {
|
|
53
|
-
id: string;
|
|
54
|
-
question: string;
|
|
55
|
-
reason: string;
|
|
56
|
-
requirementId: string;
|
|
57
|
-
importance: KnowledgeImportance;
|
|
58
|
-
answerType: 'free_text' | 'select_one' | 'multi_select' | 'file_upload' | 'credential' | 'url';
|
|
59
|
-
defaultIfSkipped?: string;
|
|
60
|
-
impactIfUnknown: string;
|
|
61
|
-
options?: string[];
|
|
62
|
-
metadata?: Record<string, unknown>;
|
|
63
|
-
}
|
|
64
|
-
export interface DataAcquisitionPlan {
|
|
65
|
-
id: string;
|
|
66
|
-
requirementIds: string[];
|
|
67
|
-
mode: Exclude<KnowledgeAcquisitionMode, 'not_available' | 'infer_low_confidence'> | 'build_domain_wiki';
|
|
68
|
-
description: string;
|
|
69
|
-
priority: KnowledgeImportance;
|
|
70
|
-
expectedEvidenceIds?: string[];
|
|
71
|
-
questions?: UserQuestion[];
|
|
72
|
-
metadata?: Record<string, unknown>;
|
|
73
|
-
}
|
|
74
|
-
export type KnowledgeResponsibleSurface = 'knowledge-requirements' | 'data-acquisition' | 'retrieval-policy' | 'user-question-policy';
|
|
75
|
-
//# sourceMappingURL=types.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/knowledge/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,oBAAoB,CAAA;AAEzD,MAAM,MAAM,4BAA4B,GACpC,eAAe,GACf,kBAAkB,GAClB,iBAAiB,GACjB,mBAAmB,GACnB,iBAAiB,GACjB,YAAY,GACZ,UAAU,GACV,sBAAsB,GACtB,qBAAqB,GACrB,YAAY,GACZ,oBAAoB,CAAA;AAExB,MAAM,MAAM,wBAAwB,GAChC,UAAU,GACV,YAAY,GACZ,iBAAiB,GACjB,cAAc,GACd,aAAa,GACb,sBAAsB,GACtB,eAAe,CAAA;AAEnB,MAAM,MAAM,mBAAmB,GAAG,UAAU,GAAG,MAAM,GAAG,QAAQ,GAAG,KAAK,CAAA;AACxE,MAAM,MAAM,kBAAkB,GAAG,QAAQ,GAAG,SAAS,GAAG,QAAQ,GAAG,OAAO,GAAG,UAAU,CAAA;AACvF,MAAM,MAAM,oBAAoB,GAAG,QAAQ,GAAG,SAAS,GAAG,QAAQ,CAAA;AAClE,MAAM,MAAM,uBAAuB,GAAG,OAAO,GAAG,KAAK,GAAG,sBAAsB,GAAG,aAAa,CAAA;AAE9F,MAAM,WAAW,oBAAoB;IACnC,EAAE,EAAE,MAAM,CAAA;IACV,WAAW,EAAE,MAAM,CAAA;IACnB,WAAW,EAAE,MAAM,EAAE,CAAA;IACrB,QAAQ,EAAE,4BAA4B,CAAA;IACtC,eAAe,EAAE,wBAAwB,CAAA;IACzC,UAAU,EAAE,mBAAmB,CAAA;IAC/B,SAAS,EAAE,kBAAkB,CAAA;IAC7B,WAAW,EAAE,oBAAoB,CAAA;IACjC,gBAAgB,EAAE,MAAM,CAAA;IACxB,iBAAiB,EAAE,MAAM,CAAA;IACzB,WAAW,EAAE,MAAM,EAAE,CAAA;IACrB,cAAc,EAAE,uBAAuB,CAAA;IACvC;;;OAGG;IACH,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,8EAA8E;IAC9E,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,eAAe;IAC9B,MAAM,EAAE,MAAM,CAAA;IACd,YAAY,EAAE,oBAAoB,EAAE,CAAA;IACpC,WAAW,EAAE,MAAM,EAAE,CAAA;IACrB,QAAQ,EAAE,MAAM,EAAE,CAAA;IAClB,WAAW,EAAE,MAAM,EAAE,CAAA;IACrB,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IACnC,OAAO,EAAE,oBAAoB,EAAE,CAAA;IAC/B,cAAc,EAAE,MAAM,CAAA;IACtB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,MAAM,0BAA0B,GAClC,WAAW,GACX,UAAU,GACV,kBAAkB,GAClB,kBAAkB,GAClB,cAAc,GACd,mBAAmB,GACnB,sBAAsB,GACtB,kBAAkB,CAAA;AAEtB,MAAM,WAAW,wBAAwB;IACvC,MAAM,EAAE,MAAM,CAAA;IACd,cAAc,EAAE,MAAM,CAAA;IACtB,2BAA2B,EAAE,oBAAoB,EAAE,CAAA;IACnD,eAAe,EAAE,oBAAoB,EAAE,CAAA;IACvC,iBAAiB,EAAE,0BAA0B,CAAA;IAC7C,MAAM,EAAE,eAAe,CAAA;IACvB,QAAQ,EAAE,eAAe,CAAA;IACzB,MAAM,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,YAAY;IAC3B,EAAE,EAAE,MAAM,CAAA;IACV,QAAQ,EAAE,MAAM,CAAA;IAChB,MAAM,EAAE,MAAM,CAAA;IACd,aAAa,EAAE,MAAM,CAAA;IACrB,UAAU,EAAE,mBAAmB,CAAA;IAC/B,UAAU,EAAE,WAAW,GAAG,YAAY,GAAG,cAAc,GAAG,aAAa,GAAG,YAAY,GAAG,KAAK,CAAA;IAC9F,gBAAgB,CAAC,EAAE,MAAM,CAAA;IACzB,eAAe,EAAE,MAAM,CAAA;IACvB,OAAO,CAAC,EAAE,MAAM,EAAE,CAAA;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,mBAAmB;IAClC,EAAE,EAAE,MAAM,CAAA;IACV,cAAc,EAAE,MAAM,EAAE,CAAA;IACxB,IAAI,EACA,OAAO,CAAC,wBAAwB,EAAE,eAAe,GAAG,sBAAsB,CAAC,GAC3E,mBAAmB,CAAA;IACvB,WAAW,EAAE,MAAM,CAAA;IACnB,QAAQ,EAAE,mBAAmB,CAAA;IAC7B,mBAAmB,CAAC,EAAE,MAAM,EAAE,CAAA;IAC9B,SAAS,CAAC,EAAE,YAAY,EAAE,CAAA;IAC1B,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,MAAM,2BAA2B,GACnC,wBAAwB,GACxB,kBAAkB,GAClB,kBAAkB,GAClB,sBAAsB,CAAA"}
|
package/dist/live-proof.d.ts
DELETED
|
@@ -1,62 +0,0 @@
|
|
|
1
|
-
import { type FeedbackLabel, type FeedbackTrajectory, type FeedbackTrajectoryStore } from './feedback-trajectory';
|
|
2
|
-
import type { ReleaseConfidenceScorecard, ReleaseConfidenceThresholds } from './release-confidence';
|
|
3
|
-
import type { CheckResult, TestResult } from './types';
|
|
4
|
-
export interface LiveProofArtifact {
|
|
5
|
-
kind: string;
|
|
6
|
-
id?: string;
|
|
7
|
-
path?: string;
|
|
8
|
-
url?: string;
|
|
9
|
-
metadata?: Record<string, unknown>;
|
|
10
|
-
}
|
|
11
|
-
export interface LiveProofContext {
|
|
12
|
-
projectId: string;
|
|
13
|
-
scenarioId: string;
|
|
14
|
-
task: string;
|
|
15
|
-
checks: CheckResult[];
|
|
16
|
-
artifacts: LiveProofArtifact[];
|
|
17
|
-
labels: FeedbackLabel[];
|
|
18
|
-
metadata: Record<string, unknown>;
|
|
19
|
-
transcript: {
|
|
20
|
-
role: 'user' | 'assistant' | 'system' | 'tool';
|
|
21
|
-
content: string;
|
|
22
|
-
at: string;
|
|
23
|
-
}[];
|
|
24
|
-
addCheck(check: CheckResult): void;
|
|
25
|
-
addArtifact(artifact: LiveProofArtifact): void;
|
|
26
|
-
addLabel(label: Omit<FeedbackLabel, 'createdAt'> & {
|
|
27
|
-
createdAt?: string;
|
|
28
|
-
}): void;
|
|
29
|
-
addTurn(turn: {
|
|
30
|
-
role: 'user' | 'assistant' | 'system' | 'tool';
|
|
31
|
-
content: string;
|
|
32
|
-
at?: string;
|
|
33
|
-
}): void;
|
|
34
|
-
}
|
|
35
|
-
export interface LiveProofConfig {
|
|
36
|
-
projectId: string;
|
|
37
|
-
scenarioId: string;
|
|
38
|
-
task: string;
|
|
39
|
-
drive(context: LiveProofContext): Promise<void> | void;
|
|
40
|
-
validate?(context: LiveProofContext): Promise<CheckResult[] | undefined> | CheckResult[] | undefined;
|
|
41
|
-
requiredArtifacts?: string[];
|
|
42
|
-
minPassRate?: number;
|
|
43
|
-
trajectoryStore?: FeedbackTrajectoryStore;
|
|
44
|
-
releaseConfidence?: {
|
|
45
|
-
target: string;
|
|
46
|
-
candidateId?: string;
|
|
47
|
-
baselineId?: string;
|
|
48
|
-
thresholds?: ReleaseConfidenceThresholds;
|
|
49
|
-
};
|
|
50
|
-
metadata?: Record<string, unknown>;
|
|
51
|
-
}
|
|
52
|
-
export interface LiveProofResult extends TestResult {
|
|
53
|
-
projectId: string;
|
|
54
|
-
scenarioId: string;
|
|
55
|
-
artifacts: LiveProofArtifact[];
|
|
56
|
-
labels: FeedbackLabel[];
|
|
57
|
-
transcript: LiveProofContext['transcript'];
|
|
58
|
-
trajectory: FeedbackTrajectory;
|
|
59
|
-
releaseConfidence?: ReleaseConfidenceScorecard;
|
|
60
|
-
}
|
|
61
|
-
export declare function runLiveProof(config: LiveProofConfig): Promise<LiveProofResult>;
|
|
62
|
-
//# sourceMappingURL=live-proof.d.ts.map
|
package/dist/live-proof.d.ts.map
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"live-proof.d.ts","sourceRoot":"","sources":["../src/live-proof.ts"],"names":[],"mappings":"AAAA,OAAO,EAEL,KAAK,aAAa,EAClB,KAAK,kBAAkB,EACvB,KAAK,uBAAuB,EAC7B,MAAM,uBAAuB,CAAA;AAC9B,OAAO,KAAK,EACV,0BAA0B,EAC1B,2BAA2B,EAE5B,MAAM,sBAAsB,CAAA;AAE7B,OAAO,KAAK,EAAE,WAAW,EAAE,UAAU,EAAE,MAAM,SAAS,CAAA;AAEtD,MAAM,WAAW,iBAAiB;IAChC,IAAI,EAAE,MAAM,CAAA;IACZ,EAAE,CAAC,EAAE,MAAM,CAAA;IACX,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,GAAG,CAAC,EAAE,MAAM,CAAA;IACZ,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,gBAAgB;IAC/B,SAAS,EAAE,MAAM,CAAA;IACjB,UAAU,EAAE,MAAM,CAAA;IAClB,IAAI,EAAE,MAAM,CAAA;IACZ,MAAM,EAAE,WAAW,EAAE,CAAA;IACrB,SAAS,EAAE,iBAAiB,EAAE,CAAA;IAC9B,MAAM,EAAE,aAAa,EAAE,CAAA;IACvB,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IACjC,UAAU,EAAE;QAAE,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,QAAQ,GAAG,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,EAAE,EAAE,MAAM,CAAA;KAAE,EAAE,CAAA;IAC7F,QAAQ,CAAC,KAAK,EAAE,WAAW,GAAG,IAAI,CAAA;IAClC,WAAW,CAAC,QAAQ,EAAE,iBAAiB,GAAG,IAAI,CAAA;IAC9C,QAAQ,CAAC,KAAK,EAAE,IAAI,CAAC,aAAa,EAAE,WAAW,CAAC,GAAG;QAAE,SAAS,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAA;IAChF,OAAO,CAAC,IAAI,EAAE;QACZ,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,QAAQ,GAAG,MAAM,CAAA;QAC9C,OAAO,EAAE,MAAM,CAAA;QACf,EAAE,CAAC,EAAE,MAAM,CAAA;KACZ,GAAG,IAAI,CAAA;CACT;AAED,MAAM,WAAW,eAAe;IAC9B,SAAS,EAAE,MAAM,CAAA;IACjB,UAAU,EAAE,MAAM,CAAA;IAClB,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAAC,IAAI,CAAC,GAAG,IAAI,CAAA;IACtD,QAAQ,CAAC,CACP,OAAO,EAAE,gBAAgB,GACxB,OAAO,CAAC,WAAW,EAAE,GAAG,SAAS,CAAC,GAAG,WAAW,EAAE,GAAG,SAAS,CAAA;IACjE,iBAAiB,CAAC,EAAE,MAAM,EAAE,CAAA;IAC5B,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,eAAe,CAAC,EAAE,uBAAuB,CAAA;IACzC,iBAAiB,CAAC,EAAE;QAClB,MAAM,EAAE,MAAM,CAAA;QACd,WAAW,CAAC,EAAE,MAAM,CAAA;QACpB,UAAU,CAAC,EAAE,MAAM,CAAA;QACnB,UAAU,CAAC,EAAE,2BAA2B,CAAA;KACzC,CAAA;IACD,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,eAAgB,SAAQ,UAAU;IACjD,SAAS,EAAE,MAAM,CAAA;IACjB,UAAU,EAAE,MAAM,CAAA;IAClB,SAAS,EAAE,iBAAiB,EAAE,CAAA;IAC9B,MAAM,EAAE,aAAa,EAAE,CAAA;IACvB,UAAU,EAAE,gBAAgB,CAAC,YAAY,CAAC,CAAA;IAC1C,UAAU,EAAE,kBAAkB,CAAA;IAC9B,iBAAiB,CAAC,EAAE,0BAA0B,CAAA;CAC/C;AAED,wBAAsB,YAAY,CAAC,MAAM,EAAE,eAAe,GAAG,OAAO,CAAC,eAAe,CAAC,CA4GpF"}
|
package/dist/llm-client.d.ts.map
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"llm-client.d.ts","sourceRoot":"","sources":["../src/llm-client.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAEH,OAAO,EAAE,cAAc,EAAE,qBAAqB,EAAE,MAAM,UAAU,CAAA;AAChE,OAAO,EAEL,KAAK,gBAAgB,EAGrB,KAAK,eAAe,EACrB,MAAM,2BAA2B,CAAA;AAIlC,MAAM,WAAW,UAAU;IACzB,IAAI,EAAE,QAAQ,GAAG,MAAM,GAAG,WAAW,CAAA;IACrC;;;OAGG;IACH,OAAO,EACH,MAAM,GACN,KAAK,CACD;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,GAC9B;QAAE,IAAI,EAAE,WAAW,CAAC;QAAC,SAAS,EAAE;YAAE,GAAG,EAAE,MAAM,CAAC;YAAC,MAAM,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,CAAA;SAAE,CAAA;KAAE,CACtF,CAAA;CACN;AAED,MAAM,WAAW,cAAc;IAC7B,KAAK,EAAE,MAAM,CAAA;IACb,QAAQ,EAAE,UAAU,EAAE,CAAA;IACtB,yEAAyE;IACzE,QAAQ,CAAC,EAAE,OAAO,CAAA;IAClB,oFAAoF;IACpF,UAAU,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;KAAE,CAAA;IAC9D,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,sCAAsC;IACtC,SAAS,CAAC,EAAE,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,QAAQ;IACvB,YAAY,EAAE,MAAM,CAAA;IACpB,gBAAgB,EAAE,MAAM,CAAA;IACxB,WAAW,EAAE,MAAM,CAAA;IACnB,uDAAuD;IACvD,kBAAkB,CAAC,EAAE,MAAM,CAAA;CAC5B;AAED,MAAM,WAAW,aAAa;IAC5B,kEAAkE;IAClE,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,EAAE,QAAQ,CAAA;IACf;;;OAGG;IACH,OAAO,EAAE,MAAM,GAAG,IAAI,CAAA;IACtB,uDAAuD;IACvD,KAAK,EAAE,MAAM,CAAA;IACb,uEAAuE;IACvE,UAAU,EAAE,MAAM,CAAA;IAClB;;;;;;;OAOG;IACH,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAA;IAC5B;;;;;OAKG;IACH,YAAY,CAAC,EAAE,OAAO,CAAA;IACtB,yBAAyB;IACzB,GAAG,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAC7B;AAED,qBAAa,YAAa,SAAQ,cAAc;aAG5B,MAAM,EAAE,MAAM;aACd,IAAI,EAAE,MAAM;aACZ,KAAK,EAAE,MAAM;gBAH7B,OAAO,EAAE,MAAM,EACC,MAAM,EAAE,MAAM,EACd,IAAI,EAAE,MAAM,EACZ,KAAK,EAAE,MAAM;CAIhC;AAED,MAAM,WAAW,gBAAgB;IAC/B,uEAAuE;IACvE,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,wFAAwF;IACxF,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,yGAAyG;IACzG,UAAU,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAA;IAC5C,oDAAoD;IACpD,gBAAgB,CAAC,EAAE,MAAM,CAAA;IACzB;;;;;OAKG;IACH,MAAM,CAAC,EAAE,WAAW,CAAA;IACpB;;;;;;OAMG;IACH,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,iFAAiF;IACjF,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,qGAAqG;IACrG,KAAK,CAAC,EAAE,OAAO,KAAK,CAAA;IACpB;;;;;;OAMG;IACH,OAAO,CAAC,EAAE,eAAe,CAAA;IACzB;;;OAGG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,gFAAgF;IAChF,YAAY,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,CAAA;IAClD,4FAA4F;IAC5F,QAAQ,CAAC,EAAE,gBAAgB,CAAA;CAC5B;AA2CD;;;;;;;;;;;GAWG;AACH,wBAAgB,mBAAmB,CAAC,GAAG,EAAE,OAAO,GAAG,OAAO,CAEzD;AA6BD,uFAAuF;AACvF,wBAAgB,SAAS,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAEjD;AAoFD;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAInD;AAED,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAwBtD;AAsCD;;;;GAIG;AACH,wBAAsB,OAAO,CAC3B,GAAG,EAAE,cAAc,EACnB,IAAI,GAAE,gBAAqB,GAC1B,OAAO,CAAC,aAAa,CAAC,CA8OxB;AA6BD;;;;;GAKG;AACH,wBAAsB,WAAW,CAAC,CAAC,GAAG,OAAO,EAC3C,GAAG,EAAE,cAAc,EACnB,IAAI,GAAE,gBAAqB,GAC1B,OAAO,CAAC;IAAE,KAAK,EAAE,CAAC,CAAC;IAAC,MAAM,EAAE,aAAa,CAAA;CAAE,CAAC,CAe9C;AAiBD,MAAM,MAAM,uBAAuB,GAC/B,sBAAsB,GACtB,kBAAkB,GAClB,sBAAsB,GACtB,SAAS,GACT,gBAAgB,CAAA;AAEpB,qBAAa,sBAAuB,SAAQ,qBAAqB;aAG7C,MAAM,EAAE,uBAAuB;aAC/B,OAAO,EAAE,MAAM;gBAF/B,OAAO,EAAE,MAAM,EACC,MAAM,EAAE,uBAAuB,EAC/B,OAAO,EAAE,MAAM;CAIlC;AAED,MAAM,WAAW,oBAAoB;IACnC;;;;;OAKG;IACH,sBAAsB,CAAC,EAAE,OAAO,CAAA;IAChC;;;OAGG;IACH,eAAe,CAAC,EAAE,KAAK,CAAC,MAAM,GAAG,MAAM,CAAC,CAAA;IACxC,8DAA8D;IAC9D,eAAe,CAAC,EAAE,KAAK,CAAC,MAAM,GAAG,MAAM,CAAC,CAAA;IACxC,uDAAuD;IACvD,WAAW,CAAC,EAAE,OAAO,CAAA;IACrB;;;OAGG;IACH,gBAAgB,CAAC,EAAE,MAAM,CAAA;CAC1B;AAED;;;;;;;;GAQG;AACH,wBAAgB,cAAc,CAAC,IAAI,EAAE,gBAAgB,EAAE,GAAG,GAAE,oBAAyB,GAAG,IAAI,CAiD3F;AAWD;;;;;;;;;;GAUG;AACH,wBAAsB,QAAQ,CAC5B,KAAK,EAAE,MAAM,EACb,IAAI,GAAE,gBAAgB,GAAG;IAAE,SAAS,CAAC,EAAE,MAAM,CAAA;CAAO,GACnD,OAAO,CAAC;IAAE,EAAE,EAAE,OAAO,CAAC;IAAC,SAAS,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAoBnE;AAED;;;;GAIG;AACH,qBAAa,SAAS;IACR,OAAO,CAAC,QAAQ,CAAC,IAAI;gBAAJ,IAAI,GAAE,gBAAqB;IAExD,IAAI,CAAC,GAAG,EAAE,cAAc,EAAE,GAAG,CAAC,EAAE,gBAAgB,GAAG,OAAO,CAAC,aAAa,CAAC;IAIzE,QAAQ,CAAC,CAAC,GAAG,OAAO,EAClB,GAAG,EAAE,cAAc,EACnB,GAAG,CAAC,EAAE,gBAAgB,GACrB,OAAO,CAAC;QAAE,KAAK,EAAE,CAAC,CAAC;QAAC,MAAM,EAAE,aAAa,CAAA;KAAE,CAAC;CAGhD"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"llm-client.test.d.ts","sourceRoot":"","sources":["../src/llm-client.test.ts"],"names":[],"mappings":""}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"locked-jsonl-appender.d.ts","sourceRoot":"","sources":["../src/locked-jsonl-appender.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAiBH,qBAAa,mBAAmB;aAEF,IAAI,EAAE,MAAM;IADxC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAO;gBACD,IAAI,EAAE,MAAM;IAOlC,MAAM,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,CAAC,IAAI,CAAC;CAM5C;AAED,mDAAmD;AACnD,wBAAgB,8BAA8B,IAAI,IAAI,CAErD"}
|
|
@@ -1,16 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Per-axis aggregation of cell runs into `AxisSummary` rows.
|
|
3
|
-
*
|
|
4
|
-
* Pure: consumes the final `cells: [{cell, runs}]` array and returns the
|
|
5
|
-
* `byAxis` table. Error runs contribute 0 to passRate and meanScore. Cost
|
|
6
|
-
* and duration always count — the budget was spent regardless.
|
|
7
|
-
*/
|
|
8
|
-
import type { AxisSummary, CellResult, MatrixAxis, MatrixCell, MatrixResult } from './types';
|
|
9
|
-
interface Row<Output> {
|
|
10
|
-
cell: MatrixCell;
|
|
11
|
-
result: CellResult<Output>;
|
|
12
|
-
}
|
|
13
|
-
export declare function summariseRows<Output>(rows: Row<Output>[], axisName: string, axisValue: string): AxisSummary;
|
|
14
|
-
export declare function buildByAxis<Output>(cells: MatrixResult<Output>['cells'], axes: MatrixAxis<unknown>[], aggregateBy: string[]): Record<string, Record<string, AxisSummary>>;
|
|
15
|
-
export {};
|
|
16
|
-
//# sourceMappingURL=aggregation.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"aggregation.d.ts","sourceRoot":"","sources":["../../src/matrix/aggregation.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,UAAU,EAAE,UAAU,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,SAAS,CAAA;AAE5F,UAAU,GAAG,CAAC,MAAM;IAClB,IAAI,EAAE,UAAU,CAAA;IAChB,MAAM,EAAE,UAAU,CAAC,MAAM,CAAC,CAAA;CAC3B;AAqBD,wBAAgB,aAAa,CAAC,MAAM,EAClC,IAAI,EAAE,GAAG,CAAC,MAAM,CAAC,EAAE,EACnB,QAAQ,EAAE,MAAM,EAChB,SAAS,EAAE,MAAM,GAChB,WAAW,CAyCb;AA2BD,wBAAgB,WAAW,CAAC,MAAM,EAChC,KAAK,EAAE,YAAY,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC,EACpC,IAAI,EAAE,UAAU,CAAC,OAAO,CAAC,EAAE,EAC3B,WAAW,EAAE,MAAM,EAAE,GACpB,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC,CAe7C"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/matrix/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAE,WAAW,EAAE,aAAa,EAAE,MAAM,eAAe,CAAA;AAC1D,OAAO,EAAE,cAAc,EAAE,MAAM,UAAU,CAAA;AACzC,YAAY,EACV,WAAW,EACX,UAAU,EACV,cAAc,EACd,UAAU,EACV,UAAU,EACV,YAAY,EACZ,qBAAqB,GACtB,MAAM,SAAS,CAAA"}
|
package/dist/matrix/runner.d.ts
DELETED
|
@@ -1,15 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* N-axis cartesian runner.
|
|
3
|
-
*
|
|
4
|
-
* Expansion order: cartesian over `axes` in declared order, then `reps` as the
|
|
5
|
-
* inner-most dim → `ordinal = (cartIdx * reps) + rep`. The returned
|
|
6
|
-
* `cells[]` is sorted by `ordinal` so concurrent execution does not reorder
|
|
7
|
-
* the output.
|
|
8
|
-
*
|
|
9
|
-
* Scheduling is a sliding window of in-flight promises capped at
|
|
10
|
-
* `maxConcurrency`. The window stops admitting new cells when the cost
|
|
11
|
-
* ceiling trips or the abort signal fires; in-flight cells finish.
|
|
12
|
-
*/
|
|
13
|
-
import type { MatrixResult, RunAgentMatrixOptions } from './types';
|
|
14
|
-
export declare function runAgentMatrix<Output>(opts: RunAgentMatrixOptions<Output>): Promise<MatrixResult<Output>>;
|
|
15
|
-
//# sourceMappingURL=runner.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/matrix/runner.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAGH,OAAO,KAAK,EAIV,YAAY,EACZ,qBAAqB,EACtB,MAAM,SAAS,CAAA;AA8DhB,wBAAsB,cAAc,CAAC,MAAM,EACzC,IAAI,EAAE,qBAAqB,CAAC,MAAM,CAAC,GAClC,OAAO,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC,CAwJ/B"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/matrix/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,YAAY,CAAA;AAEhD,YAAY,EAAE,cAAc,EAAE,MAAM,YAAY,CAAA;AAEhD;uEACuE;AACvE,MAAM,WAAW,UAAU,CAAC,CAAC;IAC3B,2DAA2D;IAC3D,IAAI,EAAE,MAAM,CAAA;IACZ,kEAAkE;IAClE,MAAM,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,CAAC,CAAA;KAAE,CAAC,CAAA;IACvC;4DACwD;IACxD,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,CAAC,EAAE,EAAE,EAAE,MAAM,KAAK,MAAM,CAAA;CACzC;AAED,0EAA0E;AAC1E,MAAM,WAAW,UAAU;IACzB,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,OAAO,CAAA;KAAE,CAAC,CAAA;IACpD,gEAAgE;IAChE,GAAG,EAAE,MAAM,CAAA;IACX,+EAA+E;IAC/E,OAAO,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,UAAU,CAAC,MAAM;IAChC,MAAM,EAAE,MAAM,CAAA;IACd,OAAO,EAAE,cAAc,CAAA;IACvB,OAAO,EAAE,MAAM,CAAA;IACf,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,CAAC,EAAE,MAAM,CAAA;IACd;6CACyC;IACzC,KAAK,CAAC,EAAE;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAA;CAC1C;AAED,MAAM,WAAW,WAAW;IAC1B,QAAQ,EAAE,MAAM,CAAA;IAChB,SAAS,EAAE,MAAM,CAAA;IACjB,KAAK,EAAE,MAAM,CAAA;IACb,QAAQ,EAAE,MAAM,CAAA;IAChB,SAAS,EAAE,MAAM,CAAA;IACjB,QAAQ,EAAE,MAAM,CAAA;IAChB,QAAQ,EAAE,MAAM,CAAA;IAChB,YAAY,EAAE,MAAM,CAAA;IACpB,cAAc,EAAE,MAAM,CAAA;CACvB;AAED,MAAM,WAAW,YAAY,CAAC,MAAM;IAClC,KAAK,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,UAAU,CAAC;QAAC,IAAI,EAAE,UAAU,CAAC,MAAM,CAAC,EAAE,CAAA;KAAE,CAAC,CAAA;IAC9D;mEAC+D;IAC/D,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC,CAAA;IACnD,OAAO,EAAE;QACP,UAAU,EAAE,MAAM,CAAA;QAClB,YAAY,EAAE,MAAM,CAAA;QACpB;+CACuC;QACvC,YAAY,EAAE,MAAM,CAAA;QACpB,eAAe,EAAE,MAAM,CAAA;QACvB,gBAAgB,EAAE,MAAM,CAAA;QACxB,YAAY,EAAE,MAAM,CAAA;QACpB,UAAU,EAAE,MAAM,CAAA;KACnB,CAAA;IACD,6DAA6D;IAC7D,QAAQ,EAAE,MAAM,CAAA;CACjB;AAED,MAAM,WAAW,qBAAqB,CAAC,MAAM;IAC3C,IAAI,EAAE,UAAU,CAAC,OAAO,CAAC,EAAE,CAAA;IAC3B;4CACwC;IACxC,OAAO,EAAE,CAAC,IAAI,EAAE,UAAU,KAAK,OAAO,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAA;IAC1D,sCAAsC;IACtC,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,2DAA2D;IAC3D,MAAM,CAAC,EAAE,CAAC,IAAI,EAAE,IAAI,CAAC,UAAU,EAAE,KAAK,GAAG,SAAS,CAAC,KAAK,OAAO,CAAA;IAC/D,sEAAsE;IACtE,WAAW,CAAC,EAAE,MAAM,EAAE,CAAA;IACtB,iEAAiE;IACjE,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB;;sDAEkD;IAClD,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,+DAA+D;IAC/D,cAAc,CAAC,EAAE,CAAC,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,CAAC,MAAM,CAAC,KAAK,IAAI,CAAA;IACvE;iDAC6C;IAC7C,MAAM,CAAC,EAAE,WAAW,CAAA;CACrB"}
|
|
@@ -1,47 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Calibration curve — binned "if eval says X, what does reality show?"
|
|
3
|
-
*
|
|
4
|
-
* Companion to correlationStudy. Raw correlation is a single number;
|
|
5
|
-
* the calibration curve shows *where* the eval is well-calibrated vs
|
|
6
|
-
* overconfident / underconfident. Buckets the eval metric, computes
|
|
7
|
-
* mean outcome per bucket, reports expected-calibration-error (ECE).
|
|
8
|
-
*/
|
|
9
|
-
import type { TraceStore } from '../trace/store';
|
|
10
|
-
import type { EvalMetricSpec } from './correlation-study';
|
|
11
|
-
import type { OutcomeStore } from './outcome-store';
|
|
12
|
-
export interface CalibrationBin {
|
|
13
|
-
lower: number;
|
|
14
|
-
upper: number;
|
|
15
|
-
n: number;
|
|
16
|
-
evalMean: number;
|
|
17
|
-
outcomeMean: number;
|
|
18
|
-
/** |outcomeMean − evalMean|; contributes to ECE weighted by n/total. */
|
|
19
|
-
gap: number;
|
|
20
|
-
}
|
|
21
|
-
export interface CalibrationReport {
|
|
22
|
-
evalMetric: string;
|
|
23
|
-
outcomeMetric: string;
|
|
24
|
-
n: number;
|
|
25
|
-
bins: CalibrationBin[];
|
|
26
|
-
/** Expected Calibration Error — Σ (n_i/N) × |outcomeMean_i − evalMean_i|. */
|
|
27
|
-
ece: number;
|
|
28
|
-
/** Max bin gap — upper bound on miscalibration. */
|
|
29
|
-
maxGap: number;
|
|
30
|
-
}
|
|
31
|
-
export interface CalibrationOptions {
|
|
32
|
-
bins?: number;
|
|
33
|
-
/** Equal-width (fixed bin edges) or equal-frequency (quantile bins). */
|
|
34
|
-
binning?: 'equal-width' | 'equal-frequency';
|
|
35
|
-
/** Clip eval values to [lo, hi] before binning. */
|
|
36
|
-
range?: {
|
|
37
|
-
lo: number;
|
|
38
|
-
hi: number;
|
|
39
|
-
};
|
|
40
|
-
}
|
|
41
|
-
export interface CalibrationPair {
|
|
42
|
-
evalScore: number;
|
|
43
|
-
outcome: number;
|
|
44
|
-
}
|
|
45
|
-
export declare function calibrationCurve(traceStore: TraceStore, outcomeStore: OutcomeStore, evalMetric: EvalMetricSpec, outcomeMetric: string, options?: CalibrationOptions): Promise<CalibrationReport | null>;
|
|
46
|
-
export declare function calibrationFromPairs(inputPairs: CalibrationPair[], evalMetric: string, outcomeMetric: string, options?: CalibrationOptions): CalibrationReport | null;
|
|
47
|
-
//# sourceMappingURL=calibration.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"calibration.d.ts","sourceRoot":"","sources":["../../src/meta-eval/calibration.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAGH,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,gBAAgB,CAAA;AAChD,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAA;AACzD,OAAO,KAAK,EAAqB,YAAY,EAAE,MAAM,iBAAiB,CAAA;AAEtE,MAAM,WAAW,cAAc;IAC7B,KAAK,EAAE,MAAM,CAAA;IACb,KAAK,EAAE,MAAM,CAAA;IACb,CAAC,EAAE,MAAM,CAAA;IACT,QAAQ,EAAE,MAAM,CAAA;IAChB,WAAW,EAAE,MAAM,CAAA;IACnB,wEAAwE;IACxE,GAAG,EAAE,MAAM,CAAA;CACZ;AAED,MAAM,WAAW,iBAAiB;IAChC,UAAU,EAAE,MAAM,CAAA;IAClB,aAAa,EAAE,MAAM,CAAA;IACrB,CAAC,EAAE,MAAM,CAAA;IACT,IAAI,EAAE,cAAc,EAAE,CAAA;IACtB,6EAA6E;IAC7E,GAAG,EAAE,MAAM,CAAA;IACX,mDAAmD;IACnD,MAAM,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,kBAAkB;IACjC,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,wEAAwE;IACxE,OAAO,CAAC,EAAE,aAAa,GAAG,iBAAiB,CAAA;IAC3C,mDAAmD;IACnD,KAAK,CAAC,EAAE;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,EAAE,EAAE,MAAM,CAAA;KAAE,CAAA;CACnC;AAED,MAAM,WAAW,eAAe;IAC9B,SAAS,EAAE,MAAM,CAAA;IACjB,OAAO,EAAE,MAAM,CAAA;CAChB;AAED,wBAAsB,gBAAgB,CACpC,UAAU,EAAE,UAAU,EACtB,YAAY,EAAE,YAAY,EAC1B,UAAU,EAAE,cAAc,EAC1B,aAAa,EAAE,MAAM,EACrB,OAAO,GAAE,kBAAuB,GAC/B,OAAO,CAAC,iBAAiB,GAAG,IAAI,CAAC,CA8BnC;AAED,wBAAgB,oBAAoB,CAClC,UAAU,EAAE,eAAe,EAAE,EAC7B,UAAU,EAAE,MAAM,EAClB,aAAa,EAAE,MAAM,EACrB,OAAO,GAAE,kBAAuB,GAC/B,iBAAiB,GAAG,IAAI,CAsC1B"}
|
|
@@ -1,53 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Correlation study — "does our eval score predict real-world outcomes?"
|
|
3
|
-
*
|
|
4
|
-
* This is the load-bearing signal. Takes a TraceStore + OutcomeStore,
|
|
5
|
-
* joins on runId, computes Pearson + Spearman + bootstrap CI for every
|
|
6
|
-
* (evalMetric, outcomeMetric) pair the caller declares.
|
|
7
|
-
*
|
|
8
|
-
* Without this number the framework is ornamental. With it and r > 0.6
|
|
9
|
-
* the framework is a moat — no other agent-eval tool publishes one.
|
|
10
|
-
*/
|
|
11
|
-
import type { Run } from '../trace/schema';
|
|
12
|
-
import type { TraceStore } from '../trace/store';
|
|
13
|
-
import type { OutcomeFilter, OutcomeStore } from './outcome-store';
|
|
14
|
-
export interface EvalMetricSpec {
|
|
15
|
-
id: string;
|
|
16
|
-
/** Extract a scalar from a run (defaults cover score/pass/durationMs/costUsd/tokens). */
|
|
17
|
-
extract?: (run: Run, store: TraceStore) => Promise<number | null>;
|
|
18
|
-
}
|
|
19
|
-
export interface OutcomePair {
|
|
20
|
-
evalMetric: string;
|
|
21
|
-
outcomeMetric: string;
|
|
22
|
-
}
|
|
23
|
-
export interface CorrelationResult {
|
|
24
|
-
evalMetric: string;
|
|
25
|
-
outcomeMetric: string;
|
|
26
|
-
n: number;
|
|
27
|
-
pearson: number;
|
|
28
|
-
spearman: number;
|
|
29
|
-
/** 95% bootstrap CI for Pearson. */
|
|
30
|
-
pearsonCi95: {
|
|
31
|
-
lower: number;
|
|
32
|
-
upper: number;
|
|
33
|
-
};
|
|
34
|
-
/** Rough verdict: 'strong' ≥ 0.7, 'moderate' ≥ 0.4, else 'weak'. */
|
|
35
|
-
verdict: 'strong' | 'moderate' | 'weak';
|
|
36
|
-
}
|
|
37
|
-
export interface CorrelationStudyResult {
|
|
38
|
-
pairs: CorrelationResult[];
|
|
39
|
-
joinedSamples: number;
|
|
40
|
-
skippedRuns: number;
|
|
41
|
-
}
|
|
42
|
-
export interface CorrelationStudyOptions {
|
|
43
|
-
/** Only join outcomes captured within this window after run.startedAt. */
|
|
44
|
-
maxCaptureLagMs?: number;
|
|
45
|
-
/** Restrict to a subset of outcomes (cohort, region, source). */
|
|
46
|
-
outcomeFilter?: OutcomeFilter;
|
|
47
|
-
/** Which outcome per run to use when multiple exist. Default 'latest'. */
|
|
48
|
-
reduction?: 'latest' | 'mean' | 'max';
|
|
49
|
-
/** Bootstrap iterations for the CI. Default 500. */
|
|
50
|
-
bootstrapIterations?: number;
|
|
51
|
-
}
|
|
52
|
-
export declare function correlationStudy(traceStore: TraceStore, outcomeStore: OutcomeStore, evalMetrics: EvalMetricSpec[], outcomeMetricNames: string[], options?: CorrelationStudyOptions): Promise<CorrelationStudyResult>;
|
|
53
|
-
//# sourceMappingURL=correlation-study.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"correlation-study.d.ts","sourceRoot":"","sources":["../../src/meta-eval/correlation-study.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAGH,OAAO,KAAK,EAAE,GAAG,EAAE,MAAM,iBAAiB,CAAA;AAC1C,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,gBAAgB,CAAA;AAChD,OAAO,KAAK,EAAqB,aAAa,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAA;AAErF,MAAM,WAAW,cAAc;IAC7B,EAAE,EAAE,MAAM,CAAA;IACV,yFAAyF;IACzF,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,EAAE,KAAK,EAAE,UAAU,KAAK,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,CAAA;CAClE;AAED,MAAM,WAAW,WAAW;IAC1B,UAAU,EAAE,MAAM,CAAA;IAClB,aAAa,EAAE,MAAM,CAAA;CACtB;AAED,MAAM,WAAW,iBAAiB;IAChC,UAAU,EAAE,MAAM,CAAA;IAClB,aAAa,EAAE,MAAM,CAAA;IACrB,CAAC,EAAE,MAAM,CAAA;IACT,OAAO,EAAE,MAAM,CAAA;IACf,QAAQ,EAAE,MAAM,CAAA;IAChB,oCAAoC;IACpC,WAAW,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAA;IAC7C,oEAAoE;IACpE,OAAO,EAAE,QAAQ,GAAG,UAAU,GAAG,MAAM,CAAA;CACxC;AAED,MAAM,WAAW,sBAAsB;IACrC,KAAK,EAAE,iBAAiB,EAAE,CAAA;IAC1B,aAAa,EAAE,MAAM,CAAA;IACrB,WAAW,EAAE,MAAM,CAAA;CACpB;AAED,MAAM,WAAW,uBAAuB;IACtC,0EAA0E;IAC1E,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,iEAAiE;IACjE,aAAa,CAAC,EAAE,aAAa,CAAA;IAC7B,0EAA0E;IAC1E,SAAS,CAAC,EAAE,QAAQ,GAAG,MAAM,GAAG,KAAK,CAAA;IACrC,oDAAoD;IACpD,mBAAmB,CAAC,EAAE,MAAM,CAAA;CAC7B;AAED,wBAAsB,gBAAgB,CACpC,UAAU,EAAE,UAAU,EACtB,YAAY,EAAE,YAAY,EAC1B,WAAW,EAAE,cAAc,EAAE,EAC7B,kBAAkB,EAAE,MAAM,EAAE,EAC5B,OAAO,GAAE,uBAA4B,GACpC,OAAO,CAAC,sBAAsB,CAAC,CA0EjC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/meta-eval/index.ts"],"names":[],"mappings":"AAAA,cAAc,eAAe,CAAA;AAC7B,cAAc,qBAAqB,CAAA;AACnC,cAAc,iBAAiB,CAAA;AAC/B,cAAc,8BAA8B,CAAA;AAC5C,cAAc,YAAY,CAAA"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"outcome-store.d.ts","sourceRoot":"","sources":["../../src/meta-eval/outcome-store.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,MAAM,WAAW,iBAAiB;IAChC,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,EAAE,MAAM,CAAA;IAClB,6EAA6E;IAC7E,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC/B,yEAAyE;IACzE,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC/B,8DAA8D;IAC9D,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,aAAa;IAC5B,MAAM,CAAC,EAAE,MAAM,EAAE,CAAA;IACjB,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,KAAK,CAAC,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAA;IACtC,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,YAAY;IAC3B,MAAM,CAAC,OAAO,EAAE,iBAAiB,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IACjD;iDAC6C;IAC7C,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,iBAAiB,EAAE,CAAC,CAAA;IACnD,IAAI,CAAC,MAAM,CAAC,EAAE,aAAa,GAAG,OAAO,CAAC,iBAAiB,EAAE,CAAC,CAAA;CAC3D;AAED,qBAAa,oBAAqB,YAAW,YAAY;IACvD,OAAO,CAAC,KAAK,CAA0B;IAEjC,MAAM,CAAC,OAAO,EAAE,iBAAiB,GAAG,OAAO,CAAC,IAAI,CAAC;IAIjD,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,iBAAiB,EAAE,CAAC;IAInD,IAAI,CAAC,MAAM,GAAE,aAAkB,GAAG,OAAO,CAAC,iBAAiB,EAAE,CAAC;CAGrE;AAED,MAAM,WAAW,6BAA6B;IAC5C,GAAG,EAAE,MAAM,CAAA;IACX,QAAQ,CAAC,EAAE,MAAM,CAAA;CAClB;AAED,qBAAa,sBAAuB,YAAW,YAAY;IACzD,OAAO,CAAC,GAAG,CAAQ;IACnB,OAAO,CAAC,QAAQ,CAAQ;IACxB,OAAO,CAAC,IAAI,CAAC,CAAsB;IACnC,OAAO,CAAC,MAAM,CAAQ;gBAEV,OAAO,EAAE,6BAA6B;YAKpC,SAAS;IAKjB,MAAM,CAAC,OAAO,EAAE,iBAAiB,GAAG,OAAO,CAAC,IAAI,CAAC;YAiBzC,IAAI;IAuBZ,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,iBAAiB,EAAE,CAAC;IAInD,IAAI,CAAC,MAAM,CAAC,EAAE,aAAa,GAAG,OAAO,CAAC,iBAAiB,EAAE,CAAC;CAGjE"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"rubric-predictive-validity.d.ts","sourceRoot":"","sources":["../../src/meta-eval/rubric-predictive-validity.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AAEH,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,eAAe,CAAA;AAC9C,OAAO,KAAK,EAAqB,YAAY,EAAE,MAAM,iBAAiB,CAAA;AAEtE,MAAM,WAAW,6BAA6B;IAC5C;;;OAGG;IACH,IAAI,EAAE,SAAS,EAAE,CAAA;IACjB,QAAQ,EAAE,YAAY,CAAA;IACtB;;;;OAIG;IACH,cAAc,EAAE,MAAM,EAAE,CAAA;IACxB;;;;OAIG;IACH,OAAO,CAAC,EAAE,MAAM,EAAE,CAAA;IAClB,wEAAwE;IACxE,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,+CAA+C;IAC/C,kBAAkB,CAAC,EAAE,MAAM,CAAA;IAC3B,+EAA+E;IAC/E,IAAI,CAAC,EAAE,MAAM,CAAA;IACb;;;OAGG;IACH,SAAS,CAAC,EAAE,QAAQ,GAAG,MAAM,GAAG,KAAK,CAAA;CACtC;AAED,MAAM,WAAW,iBAAiB;IAChC,MAAM,EAAE,MAAM,CAAA;IACd,OAAO,EAAE,MAAM,CAAA;IACf,CAAC,EAAE,MAAM,CAAA;IACT,OAAO,EAAE,MAAM,CAAA;IACf,QAAQ,EAAE,MAAM,CAAA;IAChB,IAAI,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAA;IACnC;;;;;OAKG;IACH,OAAO,EAAE,cAAc,GAAG,aAAa,GAAG,YAAY,CAAA;CACvD;AAED,MAAM,WAAW,aAAa;IAC5B,MAAM,EAAE,MAAM,CAAA;IACd,uDAAuD;IACvD,WAAW,EAAE,MAAM,CAAA;IACnB,QAAQ,EAAE,MAAM,CAAA;IAChB,OAAO,EAAE,MAAM,CAAA;IACf,CAAC,EAAE,MAAM,CAAA;IACT,OAAO,EAAE,iBAAiB,CAAC,SAAS,CAAC,CAAA;CACtC;AAED,MAAM,WAAW,8BAA8B;IAC7C,KAAK,EAAE,iBAAiB,EAAE,CAAA;IAC1B,6DAA6D;IAC7D,MAAM,EAAE,aAAa,EAAE,CAAA;IACvB,aAAa,EAAE,MAAM,CAAA;IACrB,WAAW,EAAE,MAAM,CAAA;IACnB,oEAAoE;IACpE,kBAAkB,EAAE,MAAM,EAAE,CAAA;CAC7B;AAED,wBAAsB,wBAAwB,CAC5C,KAAK,EAAE,6BAA6B,GACnC,OAAO,CAAC,8BAA8B,CAAC,CAyGzC"}
|