@tangle-network/agent-eval 0.95.0 → 0.96.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +17 -0
- package/dist/adapters/http.d.ts +17 -10
- package/dist/adapters/langchain.d.ts +14 -7
- package/dist/adapters/otel.d.ts +25 -13
- package/dist/{rl/adversarial.d.ts → adversarial-DIVcDoI_.d.ts} +7 -6
- package/dist/analyst/index.d.ts +236 -28
- package/dist/analyst/index.js +1 -1
- package/dist/{trace-analyst/analyst.d.ts → analyst-C8HHvfJp.d.ts} +14 -12
- package/dist/{contract/analyze-runs.d.ts → analyze-runs-DtT6F_6T.d.ts} +10 -9
- package/dist/authenticity/index.d.ts +16 -15
- package/dist/{baseline.d.ts → baseline-Bbid3WoO.d.ts} +44 -8
- package/dist/belief-state/index.d.ts +605 -14
- package/dist/benchmarks/index.d.ts +5 -23
- package/dist/builder-eval/index.d.ts +250 -5
- package/dist/calibration-BPmzuVPk.d.ts +101 -0
- package/dist/campaign/index.d.ts +1452 -38
- package/dist/{chunk-AQ5WQAIV.js → chunk-3NHEO6ZC.js} +2 -2
- package/dist/chunk-3NHEO6ZC.js.map +1 -0
- package/dist/cli.d.ts +0 -2
- package/dist/{contract/intake/code-agent-session.d.ts → code-agent-session-CPHRCb4-.d.ts} +17 -15
- package/dist/contract/index.d.ts +824 -107
- package/dist/contract/index.js +145 -1
- package/dist/contract/index.js.map +1 -1
- package/dist/control-Doncu-B_.d.ts +259 -0
- package/dist/{control-runtime.d.ts → control-runtime-Acf9CGhw.d.ts} +26 -23
- package/dist/control.d.ts +10 -11
- package/dist/corpus-D4YW9UoJ.d.ts +560 -0
- package/dist/{cost-ledger.d.ts → cost-ledger-DuSqlw5B.d.ts} +11 -10
- package/dist/{counterfactual.d.ts → counterfactual-DlOz8PBx.d.ts} +13 -12
- package/dist/{dataset.d.ts → dataset-BbGkaN2I.d.ts} +14 -11
- package/dist/{analyst/registry.d.ts → default-registry-GyE8X5SP.d.ts} +37 -8
- package/dist/diagnose.d.ts +252 -1
- package/dist/{trace/emitter.d.ts → emitter-C2rqGH_l.d.ts} +12 -9
- package/dist/{errors.d.ts → errors-CzMUYo7b.d.ts} +11 -10
- package/dist/failure-cluster-DH9Flgcf.d.ts +76 -0
- package/dist/{feedback-trajectory.d.ts → feedback-trajectory-BxY0cKfs.d.ts} +38 -36
- package/dist/fuzz.d.ts +547 -1
- package/dist/gepa-Dprxvz8r.d.ts +414 -0
- package/dist/governance/index.d.ts +135 -5
- package/dist/harness-optimizer-mOl9XX_O.d.ts +106 -0
- package/dist/hosted/index.d.ts +239 -10
- package/dist/index-_Y4oNOOb.d.ts +159 -0
- package/dist/index.d.ts +5660 -277
- package/dist/index.js +1 -1
- package/dist/{contract/insight-report.d.ts → insight-report-BnRjTibG.d.ts} +19 -16
- package/dist/{trace/integrity.d.ts → integrity-D2t12mMw.d.ts} +14 -11
- package/dist/{judge-calibration.d.ts → judge-calibration-0p2QcWNE.d.ts} +17 -16
- package/dist/{analyst/kind-factory.d.ts → kind-factory-X3eDYbKn.d.ts} +61 -10
- package/dist/knowledge/index.d.ts +103 -3
- package/dist/{llm-client.d.ts → llm-client-Bj7g0rqu.d.ts} +23 -21
- package/dist/matrix/index.d.ts +30 -12
- package/dist/meta-eval/index.d.ts +182 -6
- package/dist/{multi-layer-verifier.d.ts → multi-layer-verifier-DUZXrPDA.d.ts} +15 -12
- package/dist/multishot/index.d.ts +290 -7
- package/dist/{rl/off-policy.d.ts → off-policy-DiwuKKg7.d.ts} +9 -8
- package/dist/openapi.json +1 -1
- package/dist/{meta-eval/outcome-store.d.ts → outcome-store-rnXLEqSn.d.ts} +8 -7
- package/dist/{pareto.d.ts → pareto-E-pembql.d.ts} +10 -9
- package/dist/perf/index.d.ts +119 -13
- package/dist/pipelines/index.d.ts +173 -8
- package/dist/pre-registration-nfUdc9EQ.d.ts +483 -0
- package/dist/prm/index.d.ts +104 -5
- package/dist/provenance-CIxfBnkl.d.ts +426 -0
- package/dist/query-B7GGjRox.d.ts +32 -0
- package/dist/{trace/raw-provider-sink.d.ts → raw-provider-sink-C46HDghv.d.ts} +14 -13
- package/dist/{red-team.d.ts → red-team-BWdoyleI.d.ts} +16 -14
- package/dist/{trace/redact.d.ts → redact-B40YG2M_.d.ts} +8 -7
- package/dist/release-report-pidWUMZ2.d.ts +233 -0
- package/dist/reporting.d.ts +16 -15
- package/dist/{eval-campaign.d.ts → researcher-Jr8ME1dZ.d.ts} +156 -34
- package/dist/rl.d.ts +1193 -1
- package/dist/{prm/rubric.d.ts → rubric-Cc6UHvUb.d.ts} +13 -10
- package/dist/{meta-eval/rubric-predictive-validity.d.ts → rubric-predictive-validity-C2hDKM8Z.d.ts} +11 -8
- package/dist/run-critic-CmMf05uV.d.ts +56 -0
- package/dist/{run-record.d.ts → run-record-CP2ObebC.d.ts} +117 -15
- package/dist/runtime-trajectory-BOUUjI0y.d.ts +49 -0
- package/dist/{trace/schema.d.ts → schema-m0gsnbt3.d.ts} +30 -29
- package/dist/semantic-concept-judge-DSBB2Cfp.d.ts +624 -0
- package/dist/{sequential.d.ts → sequential-5iSVfzl2.d.ts} +10 -9
- package/dist/{series-convergence.d.ts → series-convergence-D5OWMBg6.d.ts} +5 -4
- package/dist/sink-fetch-B1Yg4Til.d.ts +101 -0
- package/dist/{statistics.d.ts → statistics-CCJpTGOS.d.ts} +48 -46
- package/dist/{trace/store.d.ts → store-BcFXE6LG.d.ts} +12 -21
- package/dist/{trace-analyst/types.d.ts → store-C1YxJDEK.d.ts} +74 -18
- package/dist/storyboard/index.d.ts +81 -16
- package/dist/{summary-report.d.ts → summary-report-CInXwsza.d.ts} +160 -23
- package/dist/telemetry/{sink-file.d.ts → file.d.ts} +7 -5
- package/dist/telemetry/index.d.ts +35 -17
- package/dist/{sandbox-harness.d.ts → test-graded-scenario-DeODGLra.d.ts} +60 -15
- package/dist/{locked-jsonl-appender.d.ts → testing-C21CHsq2.d.ts} +4 -3
- package/dist/testing.d.ts +1 -5
- package/dist/traces.d.ts +976 -4
- package/dist/{trajectory.d.ts → trajectory-2TkpSEVh.d.ts} +9 -8
- package/dist/{analyst/types.d.ts → types-B5x54y6n.d.ts} +112 -19
- package/dist/{campaign/types.d.ts → types-BMahhhio.d.ts} +47 -44
- package/dist/{matrix/types.d.ts → types-BUxNaJ8c.d.ts} +11 -9
- package/dist/{types.d.ts → types-C7DGg5ex.d.ts} +33 -31
- package/dist/{verdict.d.ts → verdict-C9MlYujm.d.ts} +3 -2
- package/dist/wire/index.d.ts +570 -13
- package/dist/workflow/index.d.ts +496 -22
- package/package.json +2 -2
- package/dist/action-policy.d.ts +0 -24
- package/dist/action-policy.d.ts.map +0 -1
- package/dist/action-policy.test.d.ts +0 -2
- package/dist/action-policy.test.d.ts.map +0 -1
- package/dist/active-learning.d.ts +0 -41
- package/dist/active-learning.d.ts.map +0 -1
- package/dist/adapters/http.d.ts.map +0 -1
- package/dist/adapters/langchain.d.ts.map +0 -1
- package/dist/adapters/otel.d.ts.map +0 -1
- package/dist/agent-profile-cell.d.ts +0 -101
- package/dist/agent-profile-cell.d.ts.map +0 -1
- package/dist/agent-profile.d.ts +0 -27
- package/dist/agent-profile.d.ts.map +0 -1
- package/dist/agent-profile.test.d.ts +0 -2
- package/dist/agent-profile.test.d.ts.map +0 -1
- package/dist/analyst/adapters.d.ts +0 -62
- package/dist/analyst/adapters.d.ts.map +0 -1
- package/dist/analyst/analyst.test.d.ts +0 -2
- package/dist/analyst/analyst.test.d.ts.map +0 -1
- package/dist/analyst/ax-service.d.ts +0 -27
- package/dist/analyst/ax-service.d.ts.map +0 -1
- package/dist/analyst/behavioral-analyst.d.ts +0 -28
- package/dist/analyst/behavioral-analyst.d.ts.map +0 -1
- package/dist/analyst/chat-client.d.ts +0 -91
- package/dist/analyst/chat-client.d.ts.map +0 -1
- package/dist/analyst/default-registry.d.ts +0 -27
- package/dist/analyst/default-registry.d.ts.map +0 -1
- package/dist/analyst/default-registry.test.d.ts +0 -2
- package/dist/analyst/default-registry.test.d.ts.map +0 -1
- package/dist/analyst/finding-signature.d.ts +0 -48
- package/dist/analyst/finding-signature.d.ts.map +0 -1
- package/dist/analyst/finding-subject.d.ts +0 -146
- package/dist/analyst/finding-subject.d.ts.map +0 -1
- package/dist/analyst/finding-subject.test.d.ts +0 -2
- package/dist/analyst/finding-subject.test.d.ts.map +0 -1
- package/dist/analyst/findings-store.d.ts +0 -75
- package/dist/analyst/findings-store.d.ts.map +0 -1
- package/dist/analyst/index.d.ts.map +0 -1
- package/dist/analyst/kind-factory.d.ts.map +0 -1
- package/dist/analyst/kinds/failure-mode.d.ts +0 -19
- package/dist/analyst/kinds/failure-mode.d.ts.map +0 -1
- package/dist/analyst/kinds/improvement.d.ts +0 -23
- package/dist/analyst/kinds/improvement.d.ts.map +0 -1
- package/dist/analyst/kinds/index.d.ts +0 -22
- package/dist/analyst/kinds/index.d.ts.map +0 -1
- package/dist/analyst/kinds/kinds.test.d.ts +0 -2
- package/dist/analyst/kinds/kinds.test.d.ts.map +0 -1
- package/dist/analyst/kinds/knowledge-gap.d.ts +0 -28
- package/dist/analyst/kinds/knowledge-gap.d.ts.map +0 -1
- package/dist/analyst/kinds/knowledge-poisoning.d.ts +0 -22
- package/dist/analyst/kinds/knowledge-poisoning.d.ts.map +0 -1
- package/dist/analyst/kinds/skill-usage.d.ts +0 -84
- package/dist/analyst/kinds/skill-usage.d.ts.map +0 -1
- package/dist/analyst/kinds/skill-usage.test.d.ts +0 -2
- package/dist/analyst/kinds/skill-usage.test.d.ts.map +0 -1
- package/dist/analyst/parse-tolerant.d.ts +0 -26
- package/dist/analyst/parse-tolerant.d.ts.map +0 -1
- package/dist/analyst/parse-tolerant.test.d.ts +0 -2
- package/dist/analyst/parse-tolerant.test.d.ts.map +0 -1
- package/dist/analyst/registry.budget.test.d.ts +0 -2
- package/dist/analyst/registry.budget.test.d.ts.map +0 -1
- package/dist/analyst/registry.d.ts.map +0 -1
- package/dist/analyst/steer-firewall.d.ts +0 -35
- package/dist/analyst/steer-firewall.d.ts.map +0 -1
- package/dist/analyst/steer-firewall.test.d.ts +0 -2
- package/dist/analyst/steer-firewall.test.d.ts.map +0 -1
- package/dist/analyst/structure-findings.d.ts +0 -37
- package/dist/analyst/structure-findings.d.ts.map +0 -1
- package/dist/analyst/structure-findings.test.d.ts +0 -2
- package/dist/analyst/structure-findings.test.d.ts.map +0 -1
- package/dist/analyst/tool-groups.d.ts +0 -34
- package/dist/analyst/tool-groups.d.ts.map +0 -1
- package/dist/analyst/types.d.ts.map +0 -1
- package/dist/anti-slop.d.ts +0 -59
- package/dist/anti-slop.d.ts.map +0 -1
- package/dist/artifact-validator.d.ts +0 -74
- package/dist/artifact-validator.d.ts.map +0 -1
- package/dist/attestation.d.ts +0 -63
- package/dist/attestation.d.ts.map +0 -1
- package/dist/attestation.test.d.ts +0 -2
- package/dist/attestation.test.d.ts.map +0 -1
- package/dist/authenticity/index.d.ts.map +0 -1
- package/dist/authenticity/index.test.d.ts +0 -2
- package/dist/authenticity/index.test.d.ts.map +0 -1
- package/dist/auto-pr.d.ts +0 -120
- package/dist/auto-pr.d.ts.map +0 -1
- package/dist/baseline.d.ts.map +0 -1
- package/dist/behavior-dsl.d.ts +0 -73
- package/dist/behavior-dsl.d.ts.map +0 -1
- package/dist/belief-state/calibration.d.ts +0 -10
- package/dist/belief-state/calibration.d.ts.map +0 -1
- package/dist/belief-state/calibration.test.d.ts +0 -2
- package/dist/belief-state/calibration.test.d.ts.map +0 -1
- package/dist/belief-state/code-agent-corpus.d.ts +0 -66
- package/dist/belief-state/code-agent-corpus.d.ts.map +0 -1
- package/dist/belief-state/code-agent-corpus.test.d.ts +0 -2
- package/dist/belief-state/code-agent-corpus.test.d.ts.map +0 -1
- package/dist/belief-state/code-agent-evidence.d.ts +0 -22
- package/dist/belief-state/code-agent-evidence.d.ts.map +0 -1
- package/dist/belief-state/code-agent-evidence.test.d.ts +0 -2
- package/dist/belief-state/code-agent-evidence.test.d.ts.map +0 -1
- package/dist/belief-state/extract.d.ts +0 -7
- package/dist/belief-state/extract.d.ts.map +0 -1
- package/dist/belief-state/extract.test.d.ts +0 -2
- package/dist/belief-state/extract.test.d.ts.map +0 -1
- package/dist/belief-state/index.d.ts.map +0 -1
- package/dist/belief-state/ope.d.ts +0 -17
- package/dist/belief-state/ope.d.ts.map +0 -1
- package/dist/belief-state/ope.test.d.ts +0 -2
- package/dist/belief-state/ope.test.d.ts.map +0 -1
- package/dist/belief-state/phase0-measurement.d.ts +0 -55
- package/dist/belief-state/phase0-measurement.d.ts.map +0 -1
- package/dist/belief-state/report.d.ts +0 -17
- package/dist/belief-state/report.d.ts.map +0 -1
- package/dist/belief-state/report.test.d.ts +0 -2
- package/dist/belief-state/report.test.d.ts.map +0 -1
- package/dist/belief-state/research-evidence.d.ts +0 -23
- package/dist/belief-state/research-evidence.d.ts.map +0 -1
- package/dist/belief-state/research-evidence.test.d.ts +0 -2
- package/dist/belief-state/research-evidence.test.d.ts.map +0 -1
- package/dist/belief-state/runtime-benchmark-corpus.d.ts +0 -32
- package/dist/belief-state/runtime-benchmark-corpus.d.ts.map +0 -1
- package/dist/belief-state/runtime-hooks.d.ts +0 -87
- package/dist/belief-state/runtime-hooks.d.ts.map +0 -1
- package/dist/belief-state/runtime-hooks.test.d.ts +0 -2
- package/dist/belief-state/runtime-hooks.test.d.ts.map +0 -1
- package/dist/belief-state/selective.d.ts +0 -15
- package/dist/belief-state/selective.d.ts.map +0 -1
- package/dist/belief-state/selective.test.d.ts +0 -2
- package/dist/belief-state/selective.test.d.ts.map +0 -1
- package/dist/belief-state/shadow-probe.d.ts +0 -80
- package/dist/belief-state/shadow-probe.d.ts.map +0 -1
- package/dist/belief-state/shadow-probe.test.d.ts +0 -2
- package/dist/belief-state/shadow-probe.test.d.ts.map +0 -1
- package/dist/belief-state/types.d.ts +0 -195
- package/dist/belief-state/types.d.ts.map +0 -1
- package/dist/belief-state/types.test.d.ts +0 -2
- package/dist/belief-state/types.test.d.ts.map +0 -1
- package/dist/benchmark.d.ts +0 -14
- package/dist/benchmark.d.ts.map +0 -1
- package/dist/benchmarks/index.d.ts.map +0 -1
- package/dist/benchmarks/routing/dataset.d.ts +0 -34
- package/dist/benchmarks/routing/dataset.d.ts.map +0 -1
- package/dist/benchmarks/routing/index.d.ts +0 -34
- package/dist/benchmarks/routing/index.d.ts.map +0 -1
- package/dist/benchmarks/types.d.ts +0 -49
- package/dist/benchmarks/types.d.ts.map +0 -1
- package/dist/bisector.d.ts +0 -81
- package/dist/bisector.d.ts.map +0 -1
- package/dist/budget-guard.d.ts +0 -31
- package/dist/budget-guard.d.ts.map +0 -1
- package/dist/builder-eval/builder-session.d.ts +0 -111
- package/dist/builder-eval/builder-session.d.ts.map +0 -1
- package/dist/builder-eval/correlation.d.ts +0 -32
- package/dist/builder-eval/correlation.d.ts.map +0 -1
- package/dist/builder-eval/index.d.ts.map +0 -1
- package/dist/builder-eval/project-registry.d.ts +0 -51
- package/dist/builder-eval/project-registry.d.ts.map +0 -1
- package/dist/builder-eval/three-layer-eval.d.ts +0 -55
- package/dist/builder-eval/three-layer-eval.d.ts.map +0 -1
- package/dist/campaign/analyst-surface.d.ts +0 -108
- package/dist/campaign/analyst-surface.d.ts.map +0 -1
- package/dist/campaign/analyst-surface.test.d.ts +0 -2
- package/dist/campaign/analyst-surface.test.d.ts.map +0 -1
- package/dist/campaign/auto-pr.d.ts +0 -46
- package/dist/campaign/auto-pr.d.ts.map +0 -1
- package/dist/campaign/distillation/agreement-judge.d.ts +0 -69
- package/dist/campaign/distillation/agreement-judge.d.ts.map +0 -1
- package/dist/campaign/distillation/cli.d.ts +0 -35
- package/dist/campaign/distillation/cli.d.ts.map +0 -1
- package/dist/campaign/distillation/distillation.test.d.ts +0 -2
- package/dist/campaign/distillation/distillation.test.d.ts.map +0 -1
- package/dist/campaign/distillation/gold-scenarios.d.ts +0 -54
- package/dist/campaign/distillation/gold-scenarios.d.ts.map +0 -1
- package/dist/campaign/distillation/run-distillation.d.ts +0 -119
- package/dist/campaign/distillation/run-distillation.d.ts.map +0 -1
- package/dist/campaign/gates/compose.d.ts +0 -12
- package/dist/campaign/gates/compose.d.ts.map +0 -1
- package/dist/campaign/gates/default-production-gate.d.ts +0 -58
- package/dist/campaign/gates/default-production-gate.d.ts.map +0 -1
- package/dist/campaign/gates/heldout-gate.d.ts +0 -12
- package/dist/campaign/gates/heldout-gate.d.ts.map +0 -1
- package/dist/campaign/gates/promotion-policy.d.ts +0 -125
- package/dist/campaign/gates/promotion-policy.d.ts.map +0 -1
- package/dist/campaign/gates/promotion-policy.test.d.ts +0 -2
- package/dist/campaign/gates/promotion-policy.test.d.ts.map +0 -1
- package/dist/campaign/gates/sequential.d.ts +0 -146
- package/dist/campaign/gates/sequential.d.ts.map +0 -1
- package/dist/campaign/gates/sequential.test.d.ts +0 -2
- package/dist/campaign/gates/sequential.test.d.ts.map +0 -1
- package/dist/campaign/gates/statistical-heldout.d.ts +0 -99
- package/dist/campaign/gates/statistical-heldout.d.ts.map +0 -1
- package/dist/campaign/gates/statistical-heldout.test.d.ts +0 -2
- package/dist/campaign/gates/statistical-heldout.test.d.ts.map +0 -1
- package/dist/campaign/index.d.ts.map +0 -1
- package/dist/campaign/labeled-store/fs-adapter.d.ts +0 -59
- package/dist/campaign/labeled-store/fs-adapter.d.ts.map +0 -1
- package/dist/campaign/presets/compare-proposers.d.ts +0 -146
- package/dist/campaign/presets/compare-proposers.d.ts.map +0 -1
- package/dist/campaign/presets/playback.d.ts +0 -120
- package/dist/campaign/presets/playback.d.ts.map +0 -1
- package/dist/campaign/presets/playback.test.d.ts +0 -2
- package/dist/campaign/presets/playback.test.d.ts.map +0 -1
- package/dist/campaign/presets/run-eval.d.ts +0 -14
- package/dist/campaign/presets/run-eval.d.ts.map +0 -1
- package/dist/campaign/presets/run-improvement-loop.d.ts +0 -63
- package/dist/campaign/presets/run-improvement-loop.d.ts.map +0 -1
- package/dist/campaign/presets/run-improvement-loop.test.d.ts +0 -2
- package/dist/campaign/presets/run-improvement-loop.test.d.ts.map +0 -1
- package/dist/campaign/presets/run-optimization.d.ts +0 -92
- package/dist/campaign/presets/run-optimization.d.ts.map +0 -1
- package/dist/campaign/presets/run-profile-matrix.d.ts +0 -151
- package/dist/campaign/presets/run-profile-matrix.d.ts.map +0 -1
- package/dist/campaign/presets/run-skill-opt.d.ts +0 -96
- package/dist/campaign/presets/run-skill-opt.d.ts.map +0 -1
- package/dist/campaign/proposers/_findings-text.d.ts +0 -22
- package/dist/campaign/proposers/_findings-text.d.ts.map +0 -1
- package/dist/campaign/proposers/ace.d.ts +0 -33
- package/dist/campaign/proposers/ace.d.ts.map +0 -1
- package/dist/campaign/proposers/ace.test.d.ts +0 -2
- package/dist/campaign/proposers/ace.test.d.ts.map +0 -1
- package/dist/campaign/proposers/analysis-edit.d.ts +0 -32
- package/dist/campaign/proposers/analysis-edit.d.ts.map +0 -1
- package/dist/campaign/proposers/evolutionary.d.ts +0 -20
- package/dist/campaign/proposers/evolutionary.d.ts.map +0 -1
- package/dist/campaign/proposers/fapo.d.ts +0 -120
- package/dist/campaign/proposers/fapo.d.ts.map +0 -1
- package/dist/campaign/proposers/gepa.d.ts +0 -86
- package/dist/campaign/proposers/gepa.d.ts.map +0 -1
- package/dist/campaign/proposers/halo.d.ts +0 -44
- package/dist/campaign/proposers/halo.d.ts.map +0 -1
- package/dist/campaign/proposers/halo.test.d.ts +0 -2
- package/dist/campaign/proposers/halo.test.d.ts.map +0 -1
- package/dist/campaign/proposers/memory.d.ts +0 -47
- package/dist/campaign/proposers/memory.d.ts.map +0 -1
- package/dist/campaign/proposers/memory.test.d.ts +0 -2
- package/dist/campaign/proposers/memory.test.d.ts.map +0 -1
- package/dist/campaign/proposers/skill-opt.d.ts +0 -88
- package/dist/campaign/proposers/skill-opt.d.ts.map +0 -1
- package/dist/campaign/proposers/trace-analyst.d.ts +0 -48
- package/dist/campaign/proposers/trace-analyst.d.ts.map +0 -1
- package/dist/campaign/proposers/trace-analyst.test.d.ts +0 -2
- package/dist/campaign/proposers/trace-analyst.test.d.ts.map +0 -1
- package/dist/campaign/provenance.d.ts +0 -185
- package/dist/campaign/provenance.d.ts.map +0 -1
- package/dist/campaign/run-campaign.d.ts +0 -90
- package/dist/campaign/run-campaign.d.ts.map +0 -1
- package/dist/campaign/score-utils.d.ts +0 -26
- package/dist/campaign/score-utils.d.ts.map +0 -1
- package/dist/campaign/skill-patch.d.ts +0 -62
- package/dist/campaign/skill-patch.d.ts.map +0 -1
- package/dist/campaign/storage.d.ts +0 -38
- package/dist/campaign/storage.d.ts.map +0 -1
- package/dist/campaign/types.d.ts.map +0 -1
- package/dist/campaign/worktree/index.d.ts +0 -53
- package/dist/campaign/worktree/index.d.ts.map +0 -1
- package/dist/canary.d.ts +0 -101
- package/dist/canary.d.ts.map +0 -1
- package/dist/causal-attribution.d.ts +0 -45
- package/dist/causal-attribution.d.ts.map +0 -1
- package/dist/chunk-AQ5WQAIV.js.map +0 -1
- package/dist/ci-gate.d.ts +0 -44
- package/dist/ci-gate.d.ts.map +0 -1
- package/dist/cli.d.ts.map +0 -1
- package/dist/client.d.ts +0 -77
- package/dist/client.d.ts.map +0 -1
- package/dist/client.test.d.ts +0 -2
- package/dist/client.test.d.ts.map +0 -1
- package/dist/command-runner.d.ts +0 -74
- package/dist/command-runner.d.ts.map +0 -1
- package/dist/command-runner.test.d.ts +0 -2
- package/dist/command-runner.test.d.ts.map +0 -1
- package/dist/completion-verifier.d.ts +0 -147
- package/dist/completion-verifier.d.ts.map +0 -1
- package/dist/completion-verifier.test.d.ts +0 -9
- package/dist/completion-verifier.test.d.ts.map +0 -1
- package/dist/concurrency.d.ts +0 -23
- package/dist/concurrency.d.ts.map +0 -1
- package/dist/contamination-guard.d.ts +0 -81
- package/dist/contamination-guard.d.ts.map +0 -1
- package/dist/contract/analyze-runs.d.ts.map +0 -1
- package/dist/contract/define-agent-eval.d.ts +0 -52
- package/dist/contract/define-agent-eval.d.ts.map +0 -1
- package/dist/contract/diff.d.ts +0 -114
- package/dist/contract/diff.d.ts.map +0 -1
- package/dist/contract/index.d.ts.map +0 -1
- package/dist/contract/insight-report.d.ts.map +0 -1
- package/dist/contract/insight-types-fwd.d.ts +0 -7
- package/dist/contract/insight-types-fwd.d.ts.map +0 -1
- package/dist/contract/intake/agent-trace.d.ts +0 -97
- package/dist/contract/intake/agent-trace.d.ts.map +0 -1
- package/dist/contract/intake/code-agent-session.d.ts.map +0 -1
- package/dist/contract/intake/feedback-table.d.ts +0 -87
- package/dist/contract/intake/feedback-table.d.ts.map +0 -1
- package/dist/contract/intake/index.d.ts +0 -22
- package/dist/contract/intake/index.d.ts.map +0 -1
- package/dist/contract/intake/otel-spans.d.ts +0 -33
- package/dist/contract/intake/otel-spans.d.ts.map +0 -1
- package/dist/contract/self-improve.d.ts +0 -284
- package/dist/contract/self-improve.d.ts.map +0 -1
- package/dist/control-runtime.d.ts.map +0 -1
- package/dist/control-runtime.test.d.ts +0 -2
- package/dist/control-runtime.test.d.ts.map +0 -1
- package/dist/control.d.ts.map +0 -1
- package/dist/convergence.d.ts +0 -29
- package/dist/convergence.d.ts.map +0 -1
- package/dist/cost-ledger.d.ts.map +0 -1
- package/dist/cost-ledger.test.d.ts +0 -2
- package/dist/cost-ledger.test.d.ts.map +0 -1
- package/dist/cost-report.d.ts +0 -43
- package/dist/cost-report.d.ts.map +0 -1
- package/dist/cost-report.test.d.ts +0 -2
- package/dist/cost-report.test.d.ts.map +0 -1
- package/dist/cost-tracker.d.ts +0 -76
- package/dist/cost-tracker.d.ts.map +0 -1
- package/dist/counterfactual.d.ts.map +0 -1
- package/dist/cross-trace-diff.d.ts +0 -56
- package/dist/cross-trace-diff.d.ts.map +0 -1
- package/dist/dataset.d.ts.map +0 -1
- package/dist/deploy-gate-layer.d.ts +0 -125
- package/dist/deploy-gate-layer.d.ts.map +0 -1
- package/dist/deploy-gate-layer.test.d.ts +0 -2
- package/dist/deploy-gate-layer.test.d.ts.map +0 -1
- package/dist/description-length-gate.d.ts +0 -119
- package/dist/description-length-gate.d.ts.map +0 -1
- package/dist/detectors/edge.test.d.ts +0 -2
- package/dist/detectors/edge.test.d.ts.map +0 -1
- package/dist/detectors/index.d.ts +0 -81
- package/dist/detectors/index.d.ts.map +0 -1
- package/dist/detectors/index.test.d.ts +0 -2
- package/dist/detectors/index.test.d.ts.map +0 -1
- package/dist/diagnose/causal-sweep.d.ts +0 -100
- package/dist/diagnose/causal-sweep.d.ts.map +0 -1
- package/dist/diagnose/index.d.ts +0 -36
- package/dist/diagnose/index.d.ts.map +0 -1
- package/dist/diagnose/remediation.d.ts +0 -68
- package/dist/diagnose/remediation.d.ts.map +0 -1
- package/dist/diagnose/repair.d.ts +0 -77
- package/dist/diagnose/repair.d.ts.map +0 -1
- package/dist/discover-personas.d.ts +0 -35
- package/dist/discover-personas.d.ts.map +0 -1
- package/dist/driver.d.ts +0 -95
- package/dist/driver.d.ts.map +0 -1
- package/dist/driver.test.d.ts +0 -8
- package/dist/driver.test.d.ts.map +0 -1
- package/dist/dual-agent-bench.d.ts +0 -81
- package/dist/dual-agent-bench.d.ts.map +0 -1
- package/dist/error-count-extractor.d.ts +0 -47
- package/dist/error-count-extractor.d.ts.map +0 -1
- package/dist/error-count-extractor.test.d.ts +0 -2
- package/dist/error-count-extractor.test.d.ts.map +0 -1
- package/dist/errors.d.ts.map +0 -1
- package/dist/eval-campaign.d.ts.map +0 -1
- package/dist/eval-campaign.test.d.ts +0 -2
- package/dist/eval-campaign.test.d.ts.map +0 -1
- package/dist/eval-tools.d.ts +0 -55
- package/dist/eval-tools.d.ts.map +0 -1
- package/dist/eval-trace-store.d.ts +0 -107
- package/dist/eval-trace-store.d.ts.map +0 -1
- package/dist/eval-trace-store.test.d.ts +0 -2
- package/dist/eval-trace-store.test.d.ts.map +0 -1
- package/dist/executor.d.ts +0 -38
- package/dist/executor.d.ts.map +0 -1
- package/dist/executor.test.d.ts +0 -10
- package/dist/executor.test.d.ts.map +0 -1
- package/dist/experiment-tracker.d.ts +0 -178
- package/dist/experiment-tracker.d.ts.map +0 -1
- package/dist/experiment-tracker.test.d.ts +0 -2
- package/dist/experiment-tracker.test.d.ts.map +0 -1
- package/dist/failure-taxonomy.d.ts +0 -38
- package/dist/failure-taxonomy.d.ts.map +0 -1
- package/dist/feedback-trajectory.d.ts.map +0 -1
- package/dist/feedback-trajectory.test.d.ts +0 -2
- package/dist/feedback-trajectory.test.d.ts.map +0 -1
- package/dist/flow-layer.d.ts +0 -90
- package/dist/flow-layer.d.ts.map +0 -1
- package/dist/flow-layer.test.d.ts +0 -2
- package/dist/flow-layer.test.d.ts.map +0 -1
- package/dist/fuzz/capsule.d.ts +0 -46
- package/dist/fuzz/capsule.d.ts.map +0 -1
- package/dist/fuzz/cube.d.ts +0 -36
- package/dist/fuzz/cube.d.ts.map +0 -1
- package/dist/fuzz/explorer-cost.test.d.ts +0 -2
- package/dist/fuzz/explorer-cost.test.d.ts.map +0 -1
- package/dist/fuzz/explorer.d.ts +0 -64
- package/dist/fuzz/explorer.d.ts.map +0 -1
- package/dist/fuzz/fuzz-agent.d.ts +0 -16
- package/dist/fuzz/fuzz-agent.d.ts.map +0 -1
- package/dist/fuzz/fuzz-agent.test.d.ts +0 -2
- package/dist/fuzz/fuzz-agent.test.d.ts.map +0 -1
- package/dist/fuzz/gates.d.ts +0 -33
- package/dist/fuzz/gates.d.ts.map +0 -1
- package/dist/fuzz/index.d.ts +0 -26
- package/dist/fuzz/index.d.ts.map +0 -1
- package/dist/fuzz/policies.d.ts +0 -28
- package/dist/fuzz/policies.d.ts.map +0 -1
- package/dist/fuzz/tools.d.ts +0 -20
- package/dist/fuzz/tools.d.ts.map +0 -1
- package/dist/fuzz/types.d.ts +0 -307
- package/dist/fuzz/types.d.ts.map +0 -1
- package/dist/golden-matcher.d.ts +0 -71
- package/dist/golden-matcher.d.ts.map +0 -1
- package/dist/governance/eu-ai-act.d.ts +0 -37
- package/dist/governance/eu-ai-act.d.ts.map +0 -1
- package/dist/governance/index.d.ts.map +0 -1
- package/dist/governance/nist-ai-rmf.d.ts +0 -15
- package/dist/governance/nist-ai-rmf.d.ts.map +0 -1
- package/dist/governance/soc2.d.ts +0 -12
- package/dist/governance/soc2.d.ts.map +0 -1
- package/dist/governance/types.d.ts +0 -66
- package/dist/governance/types.d.ts.map +0 -1
- package/dist/harness-optimizer.d.ts +0 -82
- package/dist/harness-optimizer.d.ts.map +0 -1
- package/dist/held-out-gate.d.ts +0 -135
- package/dist/held-out-gate.d.ts.map +0 -1
- package/dist/hosted/client.d.ts +0 -73
- package/dist/hosted/client.d.ts.map +0 -1
- package/dist/hosted/from-env.test.d.ts +0 -8
- package/dist/hosted/from-env.test.d.ts.map +0 -1
- package/dist/hosted/index.d.ts.map +0 -1
- package/dist/hosted/types.d.ts +0 -159
- package/dist/hosted/types.d.ts.map +0 -1
- package/dist/index.d.ts.map +0 -1
- package/dist/integrity/backend-integrity.d.ts +0 -71
- package/dist/integrity/backend-integrity.d.ts.map +0 -1
- package/dist/integrity/preflight.d.ts +0 -72
- package/dist/integrity/preflight.d.ts.map +0 -1
- package/dist/integrity/preflight.test.d.ts +0 -2
- package/dist/integrity/preflight.test.d.ts.map +0 -1
- package/dist/integrity/single-backend.d.ts +0 -67
- package/dist/integrity/single-backend.d.ts.map +0 -1
- package/dist/intent-match-judge.d.ts +0 -69
- package/dist/intent-match-judge.d.ts.map +0 -1
- package/dist/intent-match-judge.test.d.ts +0 -2
- package/dist/intent-match-judge.test.d.ts.map +0 -1
- package/dist/judge-calibration.d.ts.map +0 -1
- package/dist/judge-ensemble.d.ts +0 -66
- package/dist/judge-ensemble.d.ts.map +0 -1
- package/dist/judge-ensemble.test.d.ts +0 -8
- package/dist/judge-ensemble.test.d.ts.map +0 -1
- package/dist/judge-families.d.ts +0 -38
- package/dist/judge-families.d.ts.map +0 -1
- package/dist/judge-panel.d.ts +0 -65
- package/dist/judge-panel.d.ts.map +0 -1
- package/dist/judge-retry.d.ts +0 -70
- package/dist/judge-retry.d.ts.map +0 -1
- package/dist/judge-runner.d.ts +0 -36
- package/dist/judge-runner.d.ts.map +0 -1
- package/dist/judge-runner.test.d.ts +0 -2
- package/dist/judge-runner.test.d.ts.map +0 -1
- package/dist/judges.d.ts +0 -74
- package/dist/judges.d.ts.map +0 -1
- package/dist/keyword-coverage-judge.d.ts +0 -89
- package/dist/keyword-coverage-judge.d.ts.map +0 -1
- package/dist/keyword-coverage-judge.test.d.ts +0 -2
- package/dist/keyword-coverage-judge.test.d.ts.map +0 -1
- package/dist/knowledge/index.d.ts.map +0 -1
- package/dist/knowledge/readiness.d.ts +0 -26
- package/dist/knowledge/readiness.d.ts.map +0 -1
- package/dist/knowledge/types.d.ts +0 -75
- package/dist/knowledge/types.d.ts.map +0 -1
- package/dist/live-proof.d.ts +0 -62
- package/dist/live-proof.d.ts.map +0 -1
- package/dist/llm-client.d.ts.map +0 -1
- package/dist/llm-client.test.d.ts +0 -2
- package/dist/llm-client.test.d.ts.map +0 -1
- package/dist/locked-jsonl-appender.d.ts.map +0 -1
- package/dist/matrix/aggregation.d.ts +0 -16
- package/dist/matrix/aggregation.d.ts.map +0 -1
- package/dist/matrix/index.d.ts.map +0 -1
- package/dist/matrix/runner.d.ts +0 -15
- package/dist/matrix/runner.d.ts.map +0 -1
- package/dist/matrix/types.d.ts.map +0 -1
- package/dist/meta-eval/calibration.d.ts +0 -47
- package/dist/meta-eval/calibration.d.ts.map +0 -1
- package/dist/meta-eval/correlation-study.d.ts +0 -53
- package/dist/meta-eval/correlation-study.d.ts.map +0 -1
- package/dist/meta-eval/index.d.ts.map +0 -1
- package/dist/meta-eval/outcome-store.d.ts.map +0 -1
- package/dist/meta-eval/rubric-predictive-validity.d.ts.map +0 -1
- package/dist/meta-eval/sentinel.d.ts +0 -169
- package/dist/meta-eval/sentinel.d.ts.map +0 -1
- package/dist/metrics.d.ts +0 -63
- package/dist/metrics.d.ts.map +0 -1
- package/dist/model-seats.d.ts +0 -71
- package/dist/model-seats.d.ts.map +0 -1
- package/dist/model-seats.test.d.ts +0 -2
- package/dist/model-seats.test.d.ts.map +0 -1
- package/dist/muffled-gate-scanner.d.ts +0 -102
- package/dist/muffled-gate-scanner.d.ts.map +0 -1
- package/dist/multi-layer-verifier.d.ts.map +0 -1
- package/dist/multi-layer-verifier.test.d.ts +0 -2
- package/dist/multi-layer-verifier.test.d.ts.map +0 -1
- package/dist/multi-toolchain-layer.d.ts +0 -80
- package/dist/multi-toolchain-layer.d.ts.map +0 -1
- package/dist/multi-toolchain-layer.test.d.ts +0 -2
- package/dist/multi-toolchain-layer.test.d.ts.map +0 -1
- package/dist/multishot/default-tools.d.ts +0 -34
- package/dist/multishot/default-tools.d.ts.map +0 -1
- package/dist/multishot/index.d.ts.map +0 -1
- package/dist/multishot/judges.d.ts +0 -32
- package/dist/multishot/judges.d.ts.map +0 -1
- package/dist/multishot/matrix.d.ts +0 -107
- package/dist/multishot/matrix.d.ts.map +0 -1
- package/dist/multishot/multishot.d.ts +0 -23
- package/dist/multishot/multishot.d.ts.map +0 -1
- package/dist/multishot/router.d.ts +0 -37
- package/dist/multishot/router.d.ts.map +0 -1
- package/dist/multishot/types.d.ts +0 -60
- package/dist/multishot/types.d.ts.map +0 -1
- package/dist/observability.d.ts +0 -71
- package/dist/observability.d.ts.map +0 -1
- package/dist/oracle.d.ts +0 -55
- package/dist/oracle.d.ts.map +0 -1
- package/dist/orthogonality.d.ts +0 -35
- package/dist/orthogonality.d.ts.map +0 -1
- package/dist/otel-pipeline.d.ts +0 -31
- package/dist/otel-pipeline.d.ts.map +0 -1
- package/dist/paraphrase.d.ts +0 -107
- package/dist/paraphrase.d.ts.map +0 -1
- package/dist/pareto.d.ts.map +0 -1
- package/dist/partition-held-out.d.ts +0 -70
- package/dist/partition-held-out.d.ts.map +0 -1
- package/dist/partition-held-out.test.d.ts +0 -2
- package/dist/partition-held-out.test.d.ts.map +0 -1
- package/dist/perf/index.d.ts.map +0 -1
- package/dist/perf/integrity.d.ts +0 -30
- package/dist/perf/integrity.d.ts.map +0 -1
- package/dist/perf/journey.d.ts +0 -45
- package/dist/perf/journey.d.ts.map +0 -1
- package/dist/perf/ratchet.d.ts +0 -47
- package/dist/perf/ratchet.d.ts.map +0 -1
- package/dist/pipelines/budget-breach.d.ts +0 -31
- package/dist/pipelines/budget-breach.d.ts.map +0 -1
- package/dist/pipelines/budget-breach.test.d.ts +0 -2
- package/dist/pipelines/budget-breach.test.d.ts.map +0 -1
- package/dist/pipelines/failure-cluster.d.ts +0 -38
- package/dist/pipelines/failure-cluster.d.ts.map +0 -1
- package/dist/pipelines/failure-cluster.test.d.ts +0 -2
- package/dist/pipelines/failure-cluster.test.d.ts.map +0 -1
- package/dist/pipelines/first-divergence.d.ts +0 -26
- package/dist/pipelines/first-divergence.d.ts.map +0 -1
- package/dist/pipelines/first-divergence.test.d.ts +0 -2
- package/dist/pipelines/first-divergence.test.d.ts.map +0 -1
- package/dist/pipelines/index.d.ts.map +0 -1
- package/dist/pipelines/judge-agreement.d.ts +0 -26
- package/dist/pipelines/judge-agreement.d.ts.map +0 -1
- package/dist/pipelines/judge-agreement.test.d.ts +0 -2
- package/dist/pipelines/judge-agreement.test.d.ts.map +0 -1
- package/dist/pipelines/regression.d.ts +0 -23
- package/dist/pipelines/regression.d.ts.map +0 -1
- package/dist/pipelines/regression.test.d.ts +0 -2
- package/dist/pipelines/regression.test.d.ts.map +0 -1
- package/dist/pipelines/stuck-loop.d.ts +0 -32
- package/dist/pipelines/stuck-loop.d.ts.map +0 -1
- package/dist/pipelines/stuck-loop.test.d.ts +0 -2
- package/dist/pipelines/stuck-loop.test.d.ts.map +0 -1
- package/dist/pipelines/tool-waste.d.ts +0 -34
- package/dist/pipelines/tool-waste.d.ts.map +0 -1
- package/dist/pipelines/tool-waste.test.d.ts +0 -2
- package/dist/pipelines/tool-waste.test.d.ts.map +0 -1
- package/dist/playbook.d.ts +0 -16
- package/dist/playbook.d.ts.map +0 -1
- package/dist/pr-review-benchmark.d.ts +0 -88
- package/dist/pr-review-benchmark.d.ts.map +0 -1
- package/dist/pr-review-benchmark.test.d.ts +0 -2
- package/dist/pr-review-benchmark.test.d.ts.map +0 -1
- package/dist/pre-registration.d.ts +0 -125
- package/dist/pre-registration.d.ts.map +0 -1
- package/dist/prm/builtin-rubrics.d.ts +0 -33
- package/dist/prm/builtin-rubrics.d.ts.map +0 -1
- package/dist/prm/index.d.ts.map +0 -1
- package/dist/prm/inference.d.ts +0 -29
- package/dist/prm/inference.d.ts.map +0 -1
- package/dist/prm/inference.test.d.ts +0 -2
- package/dist/prm/inference.test.d.ts.map +0 -1
- package/dist/prm/rubric.d.ts.map +0 -1
- package/dist/prm/training-export.d.ts +0 -38
- package/dist/prm/training-export.d.ts.map +0 -1
- package/dist/produced-state.d.ts +0 -63
- package/dist/produced-state.d.ts.map +0 -1
- package/dist/produced-state.test.d.ts +0 -8
- package/dist/produced-state.test.d.ts.map +0 -1
- package/dist/profile/baselines.d.ts +0 -37
- package/dist/profile/baselines.d.ts.map +0 -1
- package/dist/profile/index.d.ts +0 -105
- package/dist/profile/index.d.ts.map +0 -1
- package/dist/promotion-gate.d.ts +0 -93
- package/dist/promotion-gate.d.ts.map +0 -1
- package/dist/prompt-registry.d.ts +0 -41
- package/dist/prompt-registry.d.ts.map +0 -1
- package/dist/propose-review-control.d.ts +0 -49
- package/dist/propose-review-control.d.ts.map +0 -1
- package/dist/propose-review-control.test.d.ts +0 -2
- package/dist/propose-review-control.test.d.ts.map +0 -1
- package/dist/propose-review.d.ts +0 -155
- package/dist/propose-review.d.ts.map +0 -1
- package/dist/red-team.d.ts.map +0 -1
- package/dist/reference-replay-steering.d.ts +0 -11
- package/dist/reference-replay-steering.d.ts.map +0 -1
- package/dist/reference-replay.d.ts +0 -176
- package/dist/reference-replay.d.ts.map +0 -1
- package/dist/reflective-mutation.d.ts +0 -79
- package/dist/reflective-mutation.d.ts.map +0 -1
- package/dist/registry.d.ts +0 -31
- package/dist/registry.d.ts.map +0 -1
- package/dist/release-confidence.d.ts +0 -128
- package/dist/release-confidence.d.ts.map +0 -1
- package/dist/release-report.d.ts +0 -11
- package/dist/release-report.d.ts.map +0 -1
- package/dist/replay.d.ts +0 -120
- package/dist/replay.d.ts.map +0 -1
- package/dist/reporter.d.ts +0 -14
- package/dist/reporter.d.ts.map +0 -1
- package/dist/reporting.d.ts.map +0 -1
- package/dist/researcher.d.ts +0 -140
- package/dist/researcher.d.ts.map +0 -1
- package/dist/reviewer.d.ts +0 -118
- package/dist/reviewer.d.ts.map +0 -1
- package/dist/reviewer.test.d.ts +0 -2
- package/dist/reviewer.test.d.ts.map +0 -1
- package/dist/reward-model-export.d.ts +0 -60
- package/dist/reward-model-export.d.ts.map +0 -1
- package/dist/rl/active-curriculum.d.ts +0 -110
- package/dist/rl/active-curriculum.d.ts.map +0 -1
- package/dist/rl/adaptation-eval.d.ts +0 -109
- package/dist/rl/adaptation-eval.d.ts.map +0 -1
- package/dist/rl/adversarial.d.ts.map +0 -1
- package/dist/rl/compute-curves.d.ts +0 -127
- package/dist/rl/compute-curves.d.ts.map +0 -1
- package/dist/rl/contamination.d.ts +0 -117
- package/dist/rl/contamination.d.ts.map +0 -1
- package/dist/rl/corpus.d.ts +0 -55
- package/dist/rl/corpus.d.ts.map +0 -1
- package/dist/rl/corpus.test.d.ts +0 -2
- package/dist/rl/corpus.test.d.ts.map +0 -1
- package/dist/rl/dataset.d.ts +0 -102
- package/dist/rl/dataset.d.ts.map +0 -1
- package/dist/rl/dataset.test.d.ts +0 -2
- package/dist/rl/dataset.test.d.ts.map +0 -1
- package/dist/rl/exporters.d.ts +0 -141
- package/dist/rl/exporters.d.ts.map +0 -1
- package/dist/rl/index.d.ts +0 -49
- package/dist/rl/index.d.ts.map +0 -1
- package/dist/rl/off-policy.d.ts.map +0 -1
- package/dist/rl/predictive-validity-researcher.d.ts +0 -69
- package/dist/rl/predictive-validity-researcher.d.ts.map +0 -1
- package/dist/rl/preferences.d.ts +0 -141
- package/dist/rl/preferences.d.ts.map +0 -1
- package/dist/rl/process-reward.d.ts +0 -122
- package/dist/rl/process-reward.d.ts.map +0 -1
- package/dist/rl/reward-hacking.d.ts +0 -104
- package/dist/rl/reward-hacking.d.ts.map +0 -1
- package/dist/rl/rl-campaign.d.ts +0 -85
- package/dist/rl/rl-campaign.d.ts.map +0 -1
- package/dist/rl/run-record-adapters.d.ts +0 -56
- package/dist/rl/run-record-adapters.d.ts.map +0 -1
- package/dist/rl/sim-fidelity.d.ts +0 -166
- package/dist/rl/sim-fidelity.d.ts.map +0 -1
- package/dist/rl/sim-fidelity.test.d.ts +0 -2
- package/dist/rl/sim-fidelity.test.d.ts.map +0 -1
- package/dist/rl/tournament.d.ts +0 -115
- package/dist/rl/tournament.d.ts.map +0 -1
- package/dist/rl/verifiable-reward.d.ts +0 -124
- package/dist/rl/verifiable-reward.d.ts.map +0 -1
- package/dist/run-critic.d.ts +0 -23
- package/dist/run-critic.d.ts.map +0 -1
- package/dist/run-evidence.d.ts +0 -32
- package/dist/run-evidence.d.ts.map +0 -1
- package/dist/run-record.d.ts.map +0 -1
- package/dist/run-record.test.d.ts +0 -2
- package/dist/run-record.test.d.ts.map +0 -1
- package/dist/run-score.d.ts +0 -31
- package/dist/run-score.d.ts.map +0 -1
- package/dist/runtime-trajectory.d.ts +0 -47
- package/dist/runtime-trajectory.d.ts.map +0 -1
- package/dist/sandbox-harness.d.ts.map +0 -1
- package/dist/sandbox-harness.test.d.ts +0 -2
- package/dist/sandbox-harness.test.d.ts.map +0 -1
- package/dist/sandbox-pool.d.ts +0 -74
- package/dist/sandbox-pool.d.ts.map +0 -1
- package/dist/sandbox-pool.test.d.ts +0 -2
- package/dist/sandbox-pool.test.d.ts.map +0 -1
- package/dist/scorecard.d.ts +0 -133
- package/dist/scorecard.d.ts.map +0 -1
- package/dist/scorecard.test.d.ts +0 -2
- package/dist/scorecard.test.d.ts.map +0 -1
- package/dist/self-play.d.ts +0 -69
- package/dist/self-play.d.ts.map +0 -1
- package/dist/semantic-concept-judge.d.ts +0 -135
- package/dist/semantic-concept-judge.d.ts.map +0 -1
- package/dist/semantic-concept-judge.test.d.ts +0 -2
- package/dist/semantic-concept-judge.test.d.ts.map +0 -1
- package/dist/sequential.d.ts.map +0 -1
- package/dist/series-convergence.d.ts.map +0 -1
- package/dist/slo.d.ts +0 -48
- package/dist/slo.d.ts.map +0 -1
- package/dist/state-continuity.d.ts +0 -47
- package/dist/state-continuity.d.ts.map +0 -1
- package/dist/statistics.d.ts.map +0 -1
- package/dist/statistics.test.d.ts +0 -2
- package/dist/statistics.test.d.ts.map +0 -1
- package/dist/steering-optimizer.d.ts +0 -58
- package/dist/steering-optimizer.d.ts.map +0 -1
- package/dist/steering.d.ts +0 -24
- package/dist/steering.d.ts.map +0 -1
- package/dist/storyboard/code-edit.d.ts +0 -64
- package/dist/storyboard/code-edit.d.ts.map +0 -1
- package/dist/storyboard/code-edit.test.d.ts +0 -2
- package/dist/storyboard/code-edit.test.d.ts.map +0 -1
- package/dist/storyboard/index.d.ts.map +0 -1
- package/dist/storyboard/index.test.d.ts +0 -2
- package/dist/storyboard/index.test.d.ts.map +0 -1
- package/dist/summary-report.d.ts.map +0 -1
- package/dist/telemetry/client.d.ts +0 -35
- package/dist/telemetry/client.d.ts.map +0 -1
- package/dist/telemetry/index.d.ts.map +0 -1
- package/dist/telemetry/schema.d.ts +0 -61
- package/dist/telemetry/schema.d.ts.map +0 -1
- package/dist/telemetry/sink-fetch.d.ts +0 -39
- package/dist/telemetry/sink-fetch.d.ts.map +0 -1
- package/dist/telemetry/sink-file.d.ts.map +0 -1
- package/dist/test-graded-scenario.d.ts +0 -42
- package/dist/test-graded-scenario.d.ts.map +0 -1
- package/dist/testing.d.ts.map +0 -1
- package/dist/tool-use-metrics.d.ts +0 -35
- package/dist/tool-use-metrics.d.ts.map +0 -1
- package/dist/trace/capture-fetch.d.ts +0 -48
- package/dist/trace/capture-fetch.d.ts.map +0 -1
- package/dist/trace/capture-fetch.test.d.ts +0 -2
- package/dist/trace/capture-fetch.test.d.ts.map +0 -1
- package/dist/trace/emitter.d.ts.map +0 -1
- package/dist/trace/extract-usage.d.ts +0 -46
- package/dist/trace/extract-usage.d.ts.map +0 -1
- package/dist/trace/extract-usage.test.d.ts +0 -2
- package/dist/trace/extract-usage.test.d.ts.map +0 -1
- package/dist/trace/index.d.ts +0 -15
- package/dist/trace/index.d.ts.map +0 -1
- package/dist/trace/integrity.d.ts.map +0 -1
- package/dist/trace/otel-bridge.d.ts +0 -29
- package/dist/trace/otel-bridge.d.ts.map +0 -1
- package/dist/trace/otel-export.d.ts +0 -52
- package/dist/trace/otel-export.d.ts.map +0 -1
- package/dist/trace/otel.d.ts +0 -57
- package/dist/trace/otel.d.ts.map +0 -1
- package/dist/trace/otlp-attributes.d.ts +0 -17
- package/dist/trace/otlp-attributes.d.ts.map +0 -1
- package/dist/trace/query.d.ts +0 -29
- package/dist/trace/query.d.ts.map +0 -1
- package/dist/trace/query.test.d.ts +0 -2
- package/dist/trace/query.test.d.ts.map +0 -1
- package/dist/trace/raw-provider-sink.d.ts.map +0 -1
- package/dist/trace/redact.d.ts.map +0 -1
- package/dist/trace/schema.d.ts.map +0 -1
- package/dist/trace/store-to-otlp.d.ts +0 -72
- package/dist/trace/store-to-otlp.d.ts.map +0 -1
- package/dist/trace/store-to-otlp.test.d.ts +0 -2
- package/dist/trace/store-to-otlp.test.d.ts.map +0 -1
- package/dist/trace/store.d.ts.map +0 -1
- package/dist/trace/store.test.d.ts +0 -2
- package/dist/trace/store.test.d.ts.map +0 -1
- package/dist/trace-analyst/analyst.d.ts.map +0 -1
- package/dist/trace-analyst/analyst.test.d.ts +0 -2
- package/dist/trace-analyst/analyst.test.d.ts.map +0 -1
- package/dist/trace-analyst/behavioral-metrics.d.ts +0 -40
- package/dist/trace-analyst/behavioral-metrics.d.ts.map +0 -1
- package/dist/trace-analyst/behavioral-metrics.test.d.ts +0 -2
- package/dist/trace-analyst/behavioral-metrics.test.d.ts.map +0 -1
- package/dist/trace-analyst/hook.d.ts +0 -55
- package/dist/trace-analyst/hook.d.ts.map +0 -1
- package/dist/trace-analyst/index.d.ts +0 -18
- package/dist/trace-analyst/index.d.ts.map +0 -1
- package/dist/trace-analyst/insights.d.ts +0 -71
- package/dist/trace-analyst/insights.d.ts.map +0 -1
- package/dist/trace-analyst/insights.test.d.ts +0 -2
- package/dist/trace-analyst/insights.test.d.ts.map +0 -1
- package/dist/trace-analyst/otlp-flatten.d.ts +0 -42
- package/dist/trace-analyst/otlp-flatten.d.ts.map +0 -1
- package/dist/trace-analyst/otlp-span.d.ts +0 -85
- package/dist/trace-analyst/otlp-span.d.ts.map +0 -1
- package/dist/trace-analyst/otlp-span.test.d.ts +0 -2
- package/dist/trace-analyst/otlp-span.test.d.ts.map +0 -1
- package/dist/trace-analyst/otlp-to-run-records.d.ts +0 -115
- package/dist/trace-analyst/otlp-to-run-records.d.ts.map +0 -1
- package/dist/trace-analyst/otlp-to-run-records.test.d.ts +0 -2
- package/dist/trace-analyst/otlp-to-run-records.test.d.ts.map +0 -1
- package/dist/trace-analyst/otlp-to-run-records.timestamps.test.d.ts +0 -2
- package/dist/trace-analyst/otlp-to-run-records.timestamps.test.d.ts.map +0 -1
- package/dist/trace-analyst/prompts.d.ts +0 -6
- package/dist/trace-analyst/prompts.d.ts.map +0 -1
- package/dist/trace-analyst/store-otlp.d.ts +0 -126
- package/dist/trace-analyst/store-otlp.d.ts.map +0 -1
- package/dist/trace-analyst/store-otlp.test.d.ts +0 -8
- package/dist/trace-analyst/store-otlp.test.d.ts.map +0 -1
- package/dist/trace-analyst/store-otlp.timestamps.test.d.ts +0 -2
- package/dist/trace-analyst/store-otlp.timestamps.test.d.ts.map +0 -1
- package/dist/trace-analyst/store.d.ts +0 -63
- package/dist/trace-analyst/store.d.ts.map +0 -1
- package/dist/trace-analyst/tools.d.ts +0 -44
- package/dist/trace-analyst/tools.d.ts.map +0 -1
- package/dist/trace-analyst/tools.test.d.ts +0 -10
- package/dist/trace-analyst/tools.test.d.ts.map +0 -1
- package/dist/trace-analyst/types.d.ts.map +0 -1
- package/dist/trace-contracts.d.ts +0 -180
- package/dist/trace-contracts.d.ts.map +0 -1
- package/dist/traced-analyst.d.ts +0 -26
- package/dist/traced-analyst.d.ts.map +0 -1
- package/dist/traced-judges.d.ts +0 -27
- package/dist/traced-judges.d.ts.map +0 -1
- package/dist/traces.d.ts.map +0 -1
- package/dist/trajectory.d.ts.map +0 -1
- package/dist/types.d.ts.map +0 -1
- package/dist/ui-finding.d.ts +0 -104
- package/dist/ui-finding.d.ts.map +0 -1
- package/dist/verdict-cache.d.ts +0 -78
- package/dist/verdict-cache.d.ts.map +0 -1
- package/dist/verdict-cache.test.d.ts +0 -2
- package/dist/verdict-cache.test.d.ts.map +0 -1
- package/dist/verdict.d.ts.map +0 -1
- package/dist/visual-diff.d.ts +0 -32
- package/dist/visual-diff.d.ts.map +0 -1
- package/dist/wire/handlers.d.ts +0 -54
- package/dist/wire/handlers.d.ts.map +0 -1
- package/dist/wire/index.d.ts.map +0 -1
- package/dist/wire/openapi.d.ts +0 -3
- package/dist/wire/openapi.d.ts.map +0 -1
- package/dist/wire/rpc.d.ts +0 -21
- package/dist/wire/rpc.d.ts.map +0 -1
- package/dist/wire/rubrics.d.ts +0 -34
- package/dist/wire/rubrics.d.ts.map +0 -1
- package/dist/wire/schemas.d.ts +0 -410
- package/dist/wire/schemas.d.ts.map +0 -1
- package/dist/wire/server.d.ts +0 -60
- package/dist/wire/server.d.ts.map +0 -1
- package/dist/workflow/event-schema.d.ts +0 -5
- package/dist/workflow/event-schema.d.ts.map +0 -1
- package/dist/workflow/feedback-pack.d.ts +0 -99
- package/dist/workflow/feedback-pack.d.ts.map +0 -1
- package/dist/workflow/index.d.ts.map +0 -1
- package/dist/workflow/intelligence-export.d.ts +0 -62
- package/dist/workflow/intelligence-export.d.ts.map +0 -1
- package/dist/workflow/partner-report.d.ts +0 -49
- package/dist/workflow/partner-report.d.ts.map +0 -1
- package/dist/workflow/phase-graph.d.ts +0 -43
- package/dist/workflow/phase-graph.d.ts.map +0 -1
- package/dist/workflow/promotion-gate.d.ts +0 -61
- package/dist/workflow/promotion-gate.d.ts.map +0 -1
- package/dist/workflow/run-record.d.ts +0 -12
- package/dist/workflow/run-record.d.ts.map +0 -1
- package/dist/workflow/runtime-adapter.d.ts +0 -20
- package/dist/workflow/runtime-adapter.d.ts.map +0 -1
- package/dist/workflow/sanitize.d.ts +0 -21
- package/dist/workflow/sanitize.d.ts.map +0 -1
- package/dist/workflow/schema.d.ts +0 -5
- package/dist/workflow/schema.d.ts.map +0 -1
- package/dist/workflow/summary.d.ts +0 -43
- package/dist/workflow/summary.d.ts.map +0 -1
- package/dist/workflow/trace-event-fields.d.ts +0 -6
- package/dist/workflow/trace-event-fields.d.ts.map +0 -1
- package/dist/workflow/trajectory.d.ts +0 -15
- package/dist/workflow/trajectory.d.ts.map +0 -1
- package/dist/workflow/types.d.ts +0 -68
- package/dist/workflow/types.d.ts.map +0 -1
- package/dist/workspace-inspector.d.ts +0 -67
- package/dist/workspace-inspector.d.ts.map +0 -1
- package/dist/wrangler-deploy-runner.test.d.ts +0 -2
- package/dist/wrangler-deploy-runner.test.d.ts.map +0 -1
|
@@ -1,49 +0,0 @@
|
|
|
1
|
-
import { type ControlRunResult, type ControlRuntimeConfig } from './control-runtime';
|
|
2
|
-
import { type ProposeFn, type Review, type ReviewFn, type ReviewMemoryEntry, type ReviewMemoryStore, type Verification, type VerifyFn } from './propose-review';
|
|
3
|
-
import type { FailureClass } from './trace/schema';
|
|
4
|
-
import type { TraceStore } from './trace/store';
|
|
5
|
-
export interface ProposeReviewControlState<State, Summary = unknown> {
|
|
6
|
-
shot: number;
|
|
7
|
-
state: State;
|
|
8
|
-
priorReview: Review | null;
|
|
9
|
-
verification: Verification;
|
|
10
|
-
traceSummary?: Summary;
|
|
11
|
-
memory: ReviewMemoryEntry[];
|
|
12
|
-
completed: boolean;
|
|
13
|
-
reviewAvailable: boolean;
|
|
14
|
-
reviewError?: string;
|
|
15
|
-
}
|
|
16
|
-
export interface ProposeReviewControlAction {
|
|
17
|
-
type: 'propose-review-shot';
|
|
18
|
-
shot: number;
|
|
19
|
-
}
|
|
20
|
-
export interface ProposeReviewControlResult<State, Summary = unknown> {
|
|
21
|
-
state: State;
|
|
22
|
-
verification: Verification;
|
|
23
|
-
traceSummary?: Summary;
|
|
24
|
-
review: Review | null;
|
|
25
|
-
reviewAvailable: boolean;
|
|
26
|
-
reviewError?: string;
|
|
27
|
-
}
|
|
28
|
-
export interface ProposeReviewControlConfig<State, Summary = unknown> {
|
|
29
|
-
goal: string;
|
|
30
|
-
initialState: State;
|
|
31
|
-
propose: ProposeFn<State, Summary>;
|
|
32
|
-
verify: VerifyFn<State>;
|
|
33
|
-
review: ReviewFn<State, Summary>;
|
|
34
|
-
maxShots?: number;
|
|
35
|
-
maxWallMs?: number;
|
|
36
|
-
memory?: ReviewMemoryStore;
|
|
37
|
-
store?: TraceStore;
|
|
38
|
-
scenarioId?: string;
|
|
39
|
-
projectId?: string;
|
|
40
|
-
variantId?: string;
|
|
41
|
-
fallbackInstruction?: string;
|
|
42
|
-
confidenceFloor?: number;
|
|
43
|
-
confidenceFloorWindow?: number;
|
|
44
|
-
failureClassFromVerification?: (verification: Verification) => FailureClass | undefined;
|
|
45
|
-
actionFailure?: ControlRuntimeConfig<ProposeReviewControlState<State, Summary>, ProposeReviewControlAction, ProposeReviewControlResult<State, Summary>>['actionFailure'];
|
|
46
|
-
}
|
|
47
|
-
export declare function runProposeReviewAsControlLoop<State, Summary = unknown>(config: ProposeReviewControlConfig<State, Summary>): Promise<ControlRunResult<ProposeReviewControlState<State, Summary>, ProposeReviewControlAction, ProposeReviewControlResult<State, Summary>>>;
|
|
48
|
-
export declare function controlFailureClassFromVerification(verification: Verification): FailureClass | undefined;
|
|
49
|
-
//# sourceMappingURL=propose-review-control.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"propose-review-control.d.ts","sourceRoot":"","sources":["../src/propose-review-control.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,KAAK,gBAAgB,EACrB,KAAK,oBAAoB,EAG1B,MAAM,mBAAmB,CAAA;AAC1B,OAAO,EAEL,KAAK,SAAS,EAEd,KAAK,MAAM,EACX,KAAK,QAAQ,EACb,KAAK,iBAAiB,EACtB,KAAK,iBAAiB,EACtB,KAAK,YAAY,EACjB,KAAK,QAAQ,EACd,MAAM,kBAAkB,CAAA;AACzB,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAA;AAClD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,eAAe,CAAA;AAE/C,MAAM,WAAW,yBAAyB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IACjE,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,KAAK,CAAA;IACZ,WAAW,EAAE,MAAM,GAAG,IAAI,CAAA;IAC1B,YAAY,EAAE,YAAY,CAAA;IAC1B,YAAY,CAAC,EAAE,OAAO,CAAA;IACtB,MAAM,EAAE,iBAAiB,EAAE,CAAA;IAC3B,SAAS,EAAE,OAAO,CAAA;IAClB,eAAe,EAAE,OAAO,CAAA;IACxB,WAAW,CAAC,EAAE,MAAM,CAAA;CACrB;AAED,MAAM,WAAW,0BAA0B;IACzC,IAAI,EAAE,qBAAqB,CAAA;IAC3B,IAAI,EAAE,MAAM,CAAA;CACb;AAED,MAAM,WAAW,0BAA0B,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IAClE,KAAK,EAAE,KAAK,CAAA;IACZ,YAAY,EAAE,YAAY,CAAA;IAC1B,YAAY,CAAC,EAAE,OAAO,CAAA;IACtB,MAAM,EAAE,MAAM,GAAG,IAAI,CAAA;IACrB,eAAe,EAAE,OAAO,CAAA;IACxB,WAAW,CAAC,EAAE,MAAM,CAAA;CACrB;AAED,MAAM,WAAW,0BAA0B,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IAClE,IAAI,EAAE,MAAM,CAAA;IACZ,YAAY,EAAE,KAAK,CAAA;IACnB,OAAO,EAAE,SAAS,CAAC,KAAK,EAAE,OAAO,CAAC,CAAA;IAClC,MAAM,EAAE,QAAQ,CAAC,KAAK,CAAC,CAAA;IACvB,MAAM,EAAE,QAAQ,CAAC,KAAK,EAAE,OAAO,CAAC,CAAA;IAChC,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,MAAM,CAAC,EAAE,iBAAiB,CAAA;IAC1B,KAAK,CAAC,EAAE,UAAU,CAAA;IAClB,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,mBAAmB,CAAC,EAAE,MAAM,CAAA;IAC5B,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,qBAAqB,CAAC,EAAE,MAAM,CAAA;IAC9B,4BAA4B,CAAC,EAAE,CAAC,YAAY,EAAE,YAAY,KAAK,YAAY,GAAG,SAAS,CAAA;IACvF,aAAa,CAAC,EAAE,oBAAoB,CAClC,yBAAyB,CAAC,KAAK,EAAE,OAAO,CAAC,EACzC,0BAA0B,EAC1B,0BAA0B,CAAC,KAAK,EAAE,OAAO,CAAC,CAC3C,CAAC,eAAe,CAAC,CAAA;CACnB;AAKD,wBAAsB,6BAA6B,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,EAC1E,MAAM,EAAE,0BAA0B,CAAC,KAAK,EAAE,OAAO,CAAC,GACjD,OAAO,CACR,gBAAgB,CACd,yBAAyB,CAAC,KAAK,EAAE,OAAO,CAAC,EACzC,0BAA0B,EAC1B,0BAA0B,CAAC,KAAK,EAAE,OAAO,CAAC,CAC3C,CACF,CAoKA;AAED,wBAAgB,mCAAmC,CACjD,YAAY,EAAE,YAAY,GACzB,YAAY,GAAG,SAAS,CAG1B"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"propose-review-control.test.d.ts","sourceRoot":"","sources":["../src/propose-review-control.test.ts"],"names":[],"mappings":""}
|
package/dist/propose-review.d.ts
DELETED
|
@@ -1,155 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Propose / Verify / Review — the core multi-shot primitive.
|
|
3
|
-
*
|
|
4
|
-
* shot N: propose(state, priorReview) → new state
|
|
5
|
-
* verify(state) → pass/fail, optional layers
|
|
6
|
-
* review(state, verification, memory) → observations + next-shot
|
|
7
|
-
* instruction + shouldContinue
|
|
8
|
-
* memory.append(entry)
|
|
9
|
-
*
|
|
10
|
-
* Roles are strictly separated:
|
|
11
|
-
*
|
|
12
|
-
* - The WORKER is whatever the caller wraps in `propose`. It is
|
|
13
|
-
* stateful — caller owns its resume/session mechanism.
|
|
14
|
-
* - The VERIFIER grades the state. It produces the ground truth.
|
|
15
|
-
* The reviewer cannot overturn or downgrade a verification layer.
|
|
16
|
-
* - The REVIEWER is stateless per call. Its continuity is the
|
|
17
|
-
* `ReviewMemoryStore` — durable JSONL by default, or any store
|
|
18
|
-
* implementing the interface. It reads memory + trace summary +
|
|
19
|
-
* verification and directs the NEXT proposer shot.
|
|
20
|
-
*
|
|
21
|
-
* This shape is load-bearing. The reviewer never grades; the verifier
|
|
22
|
-
* never directs. Two processes, two prompts, two concerns — which is
|
|
23
|
-
* what keeps the loop from confirmation-biasing itself into "all
|
|
24
|
-
* passed" when it didn't.
|
|
25
|
-
*
|
|
26
|
-
* Short-circuits and soft-fails are both first-class:
|
|
27
|
-
* - verify.pass === true → reviewer LLM call is skipped, memory
|
|
28
|
-
* records a success entry, loop exits.
|
|
29
|
-
* - review throws → the shot still counts; the loop uses the
|
|
30
|
-
* last-known instruction (or `fallbackInstruction`) for the next
|
|
31
|
-
* propose call. A transient reviewer failure must NEVER abort a
|
|
32
|
-
* valid arc.
|
|
33
|
-
*
|
|
34
|
-
* Composable: `propose` itself can be another `runProposeReview` call.
|
|
35
|
-
* That's the dogfooding path — a harness built on this primitive is in
|
|
36
|
-
* turn evaluable by it.
|
|
37
|
-
*/
|
|
38
|
-
import { TraceEmitter } from './trace/emitter';
|
|
39
|
-
import type { FailureClass } from './trace/schema';
|
|
40
|
-
import type { TraceStore } from './trace/store';
|
|
41
|
-
export interface Verification {
|
|
42
|
-
pass: boolean;
|
|
43
|
-
score?: number;
|
|
44
|
-
failingLayers?: string[];
|
|
45
|
-
details?: unknown;
|
|
46
|
-
}
|
|
47
|
-
export interface Review {
|
|
48
|
-
observations: string;
|
|
49
|
-
diagnosis: string;
|
|
50
|
-
nextShotInstruction: string;
|
|
51
|
-
shouldContinue: boolean;
|
|
52
|
-
confidence: number;
|
|
53
|
-
}
|
|
54
|
-
export interface ReviewMemoryEntry extends Review {
|
|
55
|
-
shot: number;
|
|
56
|
-
timestamp: number;
|
|
57
|
-
verification: {
|
|
58
|
-
pass: boolean;
|
|
59
|
-
score?: number;
|
|
60
|
-
failingLayers?: string[];
|
|
61
|
-
};
|
|
62
|
-
}
|
|
63
|
-
export interface ProposeInput<State> {
|
|
64
|
-
shot: number;
|
|
65
|
-
goal: string;
|
|
66
|
-
state: State;
|
|
67
|
-
priorReview: Review | null;
|
|
68
|
-
abortSignal: AbortSignal;
|
|
69
|
-
emitter?: TraceEmitter;
|
|
70
|
-
}
|
|
71
|
-
export interface ProposeOutput<State, Summary = unknown> {
|
|
72
|
-
state: State;
|
|
73
|
-
traceSummary?: Summary;
|
|
74
|
-
}
|
|
75
|
-
export interface ReviewInput<State, Summary = unknown> {
|
|
76
|
-
shot: number;
|
|
77
|
-
goal: string;
|
|
78
|
-
state: State;
|
|
79
|
-
verification: Verification;
|
|
80
|
-
traceSummary: Summary | undefined;
|
|
81
|
-
memory: ReviewMemoryEntry[];
|
|
82
|
-
}
|
|
83
|
-
export type ProposeFn<State, Summary = unknown> = (input: ProposeInput<State>) => Promise<ProposeOutput<State, Summary>>;
|
|
84
|
-
export type VerifyFn<State> = (state: State) => Promise<Verification>;
|
|
85
|
-
export type ReviewFn<State, Summary = unknown> = (input: ReviewInput<State, Summary>) => Promise<Review>;
|
|
86
|
-
export interface ReviewMemoryStore {
|
|
87
|
-
load(): Promise<ReviewMemoryEntry[]>;
|
|
88
|
-
append(entry: ReviewMemoryEntry): Promise<void>;
|
|
89
|
-
}
|
|
90
|
-
export interface ProposeReviewConfig<State, Summary = unknown> {
|
|
91
|
-
goal: string;
|
|
92
|
-
initialState: State;
|
|
93
|
-
propose: ProposeFn<State, Summary>;
|
|
94
|
-
verify: VerifyFn<State>;
|
|
95
|
-
review: ReviewFn<State, Summary>;
|
|
96
|
-
/** Hard shot cap. Default 10. */
|
|
97
|
-
maxShots?: number;
|
|
98
|
-
/** Wall-clock cap in ms. Default 10 min. */
|
|
99
|
-
maxWallMs?: number;
|
|
100
|
-
/**
|
|
101
|
-
* If the reviewer returns confidence ≤ floor on `confidenceFloorWindow`
|
|
102
|
-
* consecutive shots, terminate early. Default floor 0.3, window 2.
|
|
103
|
-
* Set window to 0 or floor to <0 to disable.
|
|
104
|
-
*/
|
|
105
|
-
confidenceFloor?: number;
|
|
106
|
-
confidenceFloorWindow?: number;
|
|
107
|
-
/** Defaults to an in-memory store if omitted. */
|
|
108
|
-
memory?: ReviewMemoryStore;
|
|
109
|
-
/** If provided, emit a Run + per-shot spans. */
|
|
110
|
-
store?: TraceStore;
|
|
111
|
-
scenarioId?: string;
|
|
112
|
-
projectId?: string;
|
|
113
|
-
variantId?: string;
|
|
114
|
-
/**
|
|
115
|
-
* Used when the reviewer soft-fails on shot 1 (no prior instruction to
|
|
116
|
-
* fall back to). Default is a generic "inspect failures and fix".
|
|
117
|
-
*/
|
|
118
|
-
fallbackInstruction?: string;
|
|
119
|
-
}
|
|
120
|
-
export interface ProposeReviewShot<State, Summary = unknown> {
|
|
121
|
-
shot: number;
|
|
122
|
-
state: State;
|
|
123
|
-
verification: Verification;
|
|
124
|
-
traceSummary: Summary | undefined;
|
|
125
|
-
review: Review;
|
|
126
|
-
reviewAvailable: boolean;
|
|
127
|
-
reviewError?: string;
|
|
128
|
-
durationMs: number;
|
|
129
|
-
}
|
|
130
|
-
export interface ProposeReviewReport<State, Summary = unknown> {
|
|
131
|
-
runId: string | null;
|
|
132
|
-
completed: boolean;
|
|
133
|
-
shots: ProposeReviewShot<State, Summary>[];
|
|
134
|
-
finalState: State;
|
|
135
|
-
finalVerification: Verification;
|
|
136
|
-
failureClass?: FailureClass;
|
|
137
|
-
wallMs: number;
|
|
138
|
-
score: number;
|
|
139
|
-
}
|
|
140
|
-
export declare function inMemoryReviewStore(initial?: ReviewMemoryEntry[]): ReviewMemoryStore;
|
|
141
|
-
export declare function jsonlReviewStore(path: string): ReviewMemoryStore;
|
|
142
|
-
export declare function runProposeReview<State, Summary = unknown>(config: ProposeReviewConfig<State, Summary>): Promise<ProposeReviewReport<State, Summary>>;
|
|
143
|
-
export type LlmJsonCall = (req: {
|
|
144
|
-
system: string;
|
|
145
|
-
user: string;
|
|
146
|
-
}) => Promise<unknown>;
|
|
147
|
-
export interface LlmReviewerConfig<State, Summary = unknown> {
|
|
148
|
-
callJson: LlmJsonCall;
|
|
149
|
-
renderState?: (state: State) => string;
|
|
150
|
-
renderTraceSummary?: (summary: Summary | undefined) => string;
|
|
151
|
-
/** Appended to the default system prompt. */
|
|
152
|
-
systemPromptAddendum?: string;
|
|
153
|
-
}
|
|
154
|
-
export declare function createLlmReviewer<State, Summary = unknown>(cfg: LlmReviewerConfig<State, Summary>): ReviewFn<State, Summary>;
|
|
155
|
-
//# sourceMappingURL=propose-review.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"propose-review.d.ts","sourceRoot":"","sources":["../src/propose-review.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoCG;AAIH,OAAO,EAAmB,YAAY,EAAE,MAAM,iBAAiB,CAAA;AAC/D,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAA;AAClD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,eAAe,CAAA;AAI/C,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,OAAO,CAAA;IACb,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,aAAa,CAAC,EAAE,MAAM,EAAE,CAAA;IACxB,OAAO,CAAC,EAAE,OAAO,CAAA;CAClB;AAED,MAAM,WAAW,MAAM;IACrB,YAAY,EAAE,MAAM,CAAA;IACpB,SAAS,EAAE,MAAM,CAAA;IACjB,mBAAmB,EAAE,MAAM,CAAA;IAC3B,cAAc,EAAE,OAAO,CAAA;IACvB,UAAU,EAAE,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,iBAAkB,SAAQ,MAAM;IAC/C,IAAI,EAAE,MAAM,CAAA;IACZ,SAAS,EAAE,MAAM,CAAA;IACjB,YAAY,EAAE;QACZ,IAAI,EAAE,OAAO,CAAA;QACb,KAAK,CAAC,EAAE,MAAM,CAAA;QACd,aAAa,CAAC,EAAE,MAAM,EAAE,CAAA;KACzB,CAAA;CACF;AAED,MAAM,WAAW,YAAY,CAAC,KAAK;IACjC,IAAI,EAAE,MAAM,CAAA;IACZ,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,KAAK,CAAA;IACZ,WAAW,EAAE,MAAM,GAAG,IAAI,CAAA;IAC1B,WAAW,EAAE,WAAW,CAAA;IACxB,OAAO,CAAC,EAAE,YAAY,CAAA;CACvB;AAED,MAAM,WAAW,aAAa,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IACrD,KAAK,EAAE,KAAK,CAAA;IACZ,YAAY,CAAC,EAAE,OAAO,CAAA;CACvB;AAED,MAAM,WAAW,WAAW,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IACnD,IAAI,EAAE,MAAM,CAAA;IACZ,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,KAAK,CAAA;IACZ,YAAY,EAAE,YAAY,CAAA;IAC1B,YAAY,EAAE,OAAO,GAAG,SAAS,CAAA;IACjC,MAAM,EAAE,iBAAiB,EAAE,CAAA;CAC5B;AAED,MAAM,MAAM,SAAS,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,IAAI,CAChD,KAAK,EAAE,YAAY,CAAC,KAAK,CAAC,KACvB,OAAO,CAAC,aAAa,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC,CAAA;AAE3C,MAAM,MAAM,QAAQ,CAAC,KAAK,IAAI,CAAC,KAAK,EAAE,KAAK,KAAK,OAAO,CAAC,YAAY,CAAC,CAAA;AAErE,MAAM,MAAM,QAAQ,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,IAAI,CAC/C,KAAK,EAAE,WAAW,CAAC,KAAK,EAAE,OAAO,CAAC,KAC/B,OAAO,CAAC,MAAM,CAAC,CAAA;AAEpB,MAAM,WAAW,iBAAiB;IAChC,IAAI,IAAI,OAAO,CAAC,iBAAiB,EAAE,CAAC,CAAA;IACpC,MAAM,CAAC,KAAK,EAAE,iBAAiB,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;CAChD;AAED,MAAM,WAAW,mBAAmB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IAC3D,IAAI,EAAE,MAAM,CAAA;IACZ,YAAY,EAAE,KAAK,CAAA;IACnB,OAAO,EAAE,SAAS,CAAC,KAAK,EAAE,OAAO,CAAC,CAAA;IAClC,MAAM,EAAE,QAAQ,CAAC,KAAK,CAAC,CAAA;IACvB,MAAM,EAAE,QAAQ,CAAC,KAAK,EAAE,OAAO,CAAC,CAAA;IAChC,iCAAiC;IACjC,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,4CAA4C;IAC5C,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB;;;;OAIG;IACH,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,qBAAqB,CAAC,EAAE,MAAM,CAAA;IAC9B,iDAAiD;IACjD,MAAM,CAAC,EAAE,iBAAiB,CAAA;IAC1B,gDAAgD;IAChD,KAAK,CAAC,EAAE,UAAU,CAAA;IAClB,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB;;;OAGG;IACH,mBAAmB,CAAC,EAAE,MAAM,CAAA;CAC7B;AAED,MAAM,WAAW,iBAAiB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IACzD,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,KAAK,CAAA;IACZ,YAAY,EAAE,YAAY,CAAA;IAC1B,YAAY,EAAE,OAAO,GAAG,SAAS,CAAA;IACjC,MAAM,EAAE,MAAM,CAAA;IACd,eAAe,EAAE,OAAO,CAAA;IACxB,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,UAAU,EAAE,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,mBAAmB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IAC3D,KAAK,EAAE,MAAM,GAAG,IAAI,CAAA;IACpB,SAAS,EAAE,OAAO,CAAA;IAClB,KAAK,EAAE,iBAAiB,CAAC,KAAK,EAAE,OAAO,CAAC,EAAE,CAAA;IAC1C,UAAU,EAAE,KAAK,CAAA;IACjB,iBAAiB,EAAE,YAAY,CAAA;IAC/B,YAAY,CAAC,EAAE,YAAY,CAAA;IAC3B,MAAM,EAAE,MAAM,CAAA;IACd,KAAK,EAAE,MAAM,CAAA;CACd;AAID,wBAAgB,mBAAmB,CAAC,OAAO,GAAE,iBAAiB,EAAO,GAAG,iBAAiB,CAUxF;AAED,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,MAAM,GAAG,iBAAiB,CAuBhE;AAOD,wBAAsB,gBAAgB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,EAC7D,MAAM,EAAE,mBAAmB,CAAC,KAAK,EAAE,OAAO,CAAC,GAC1C,OAAO,CAAC,mBAAmB,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC,CA6M9C;AAID,MAAM,MAAM,WAAW,GAAG,CAAC,GAAG,EAAE;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAA;CAAE,KAAK,OAAO,CAAC,OAAO,CAAC,CAAA;AAErF,MAAM,WAAW,iBAAiB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IACzD,QAAQ,EAAE,WAAW,CAAA;IACrB,WAAW,CAAC,EAAE,CAAC,KAAK,EAAE,KAAK,KAAK,MAAM,CAAA;IACtC,kBAAkB,CAAC,EAAE,CAAC,OAAO,EAAE,OAAO,GAAG,SAAS,KAAK,MAAM,CAAA;IAC7D,6CAA6C;IAC7C,oBAAoB,CAAC,EAAE,MAAM,CAAA;CAC9B;AAOD,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,EACxD,GAAG,EAAE,iBAAiB,CAAC,KAAK,EAAE,OAAO,CAAC,GACrC,QAAQ,CAAC,KAAK,EAAE,OAAO,CAAC,CA6D1B"}
|
package/dist/red-team.d.ts.map
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"red-team.d.ts","sourceRoot":"","sources":["../src/red-team.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,WAAW,CAAA;AAChD,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAA;AAGnC,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,eAAe,CAAA;AAE/C,MAAM,MAAM,eAAe,GACvB,yBAAyB,GACzB,2BAA2B,GAC3B,mBAAmB,GACnB,eAAe,GACf,UAAU,GACV,uBAAuB,GACvB,YAAY,GACZ,iBAAiB,CAAA;AAErB,MAAM,WAAW,cAAc;IAC7B,QAAQ,EAAE,eAAe,CAAA;IACzB,wCAAwC;IACxC,KAAK,EAAE,MAAM,CAAA;IACb;;;OAGG;IACH,QAAQ,EAAE,QAAQ,GAAG,QAAQ,CAAA;IAC7B,mDAAmD;IACnD,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAA;IAC3B,4DAA4D;IAC5D,cAAc,CAAC,EAAE,MAAM,EAAE,CAAA;CAC1B;AAED,MAAM,WAAW,WAAY,SAAQ,eAAe;IAClD,OAAO,EAAE,cAAc,CAAA;CACxB;AAED,MAAM,WAAW,cAAc;IAC7B,UAAU,EAAE,MAAM,CAAA;IAClB,QAAQ,EAAE,eAAe,CAAA;IACzB,MAAM,EAAE,OAAO,CAAA;IACf,MAAM,EAAE,MAAM,CAAA;IACd,QAAQ,CAAC,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,WAAW,aAAa;IAC5B,QAAQ,EAAE,cAAc,EAAE,CAAA;IAC1B,kBAAkB,EAAE,MAAM,CAAC,eAAe,EAAE,MAAM,CAAC,CAAA;IACnD,eAAe,EAAE,MAAM,CAAA;CACxB;AAUD,2EAA2E;AAC3E,eAAO,MAAM,uBAAuB,EAAE,WAAW,EA2FhD,CAAA;AAED,wBAAgB,cAAc,CAAC,UAAU,GAAE,WAAW,EAAO,GAAG,OAAO,CAYtE;AAED;;;GAGG;AACH,wBAAgB,kBAAkB,CAChC,MAAM,EAAE,MAAM,EACd,SAAS,EAAE,MAAM,EAAE,EACnB,MAAM,EAAE,WAAW,GAClB,cAAc,CA8DhB;AAED,gEAAgE;AAChE,wBAAgB,aAAa,CAAC,QAAQ,EAAE,cAAc,EAAE,GAAG,aAAa,CAevE;AAED;;;GAGG;AACH,wBAAsB,eAAe,CAAC,KAAK,EAAE,UAAU,EAAE,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC,CAGzF"}
|
|
@@ -1,11 +0,0 @@
|
|
|
1
|
-
import type { ReferenceReplayCaseRun, ReferenceReplayRun, ReferenceReplayScenarioScore } from './reference-replay';
|
|
2
|
-
import type { RunScore } from './run-score';
|
|
3
|
-
import type { SteeringBundle } from './steering';
|
|
4
|
-
import type { SteeringOptimizationRow } from './steering-optimizer';
|
|
5
|
-
export interface ReferenceReplaySteeringRowsOptions<Input = unknown> {
|
|
6
|
-
bundleForRun?: (run: ReferenceReplayRun<Input>) => SteeringBundle;
|
|
7
|
-
scoreForCase?: (caseRun: ReferenceReplayCaseRun<Input>, run: ReferenceReplayRun<Input>) => RunScore;
|
|
8
|
-
}
|
|
9
|
-
export declare function referenceReplayRunsToSteeringRows<Input = unknown>(runs: ReferenceReplayRun<Input>[], options?: ReferenceReplaySteeringRowsOptions<Input>): SteeringOptimizationRow[];
|
|
10
|
-
export declare function referenceReplayScenarioToRunScore(scenarioScore: ReferenceReplayScenarioScore, durationMs?: number): RunScore;
|
|
11
|
-
//# sourceMappingURL=reference-replay-steering.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"reference-replay-steering.d.ts","sourceRoot":"","sources":["../src/reference-replay-steering.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,sBAAsB,EACtB,kBAAkB,EAClB,4BAA4B,EAC7B,MAAM,oBAAoB,CAAA;AAC3B,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAA;AAC3C,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,YAAY,CAAA;AAChD,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,sBAAsB,CAAA;AAEnE,MAAM,WAAW,kCAAkC,CAAC,KAAK,GAAG,OAAO;IACjE,YAAY,CAAC,EAAE,CAAC,GAAG,EAAE,kBAAkB,CAAC,KAAK,CAAC,KAAK,cAAc,CAAA;IACjE,YAAY,CAAC,EAAE,CACb,OAAO,EAAE,sBAAsB,CAAC,KAAK,CAAC,EACtC,GAAG,EAAE,kBAAkB,CAAC,KAAK,CAAC,KAC3B,QAAQ,CAAA;CACd;AAED,wBAAgB,iCAAiC,CAAC,KAAK,GAAG,OAAO,EAC/D,IAAI,EAAE,kBAAkB,CAAC,KAAK,CAAC,EAAE,EACjC,OAAO,GAAE,kCAAkC,CAAC,KAAK,CAAM,GACtD,uBAAuB,EAAE,CAoC3B;AAED,wBAAgB,iCAAiC,CAC/C,aAAa,EAAE,4BAA4B,EAC3C,UAAU,SAAI,GACb,QAAQ,CAuBV"}
|
|
@@ -1,176 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Reference replay — score an agent against withheld historical outcomes.
|
|
3
|
-
*
|
|
4
|
-
* This is the generic version of the public-audit replay pattern:
|
|
5
|
-
* run a candidate system on an old task, keep the reference answers hidden
|
|
6
|
-
* until after execution, then score recall/precision and gate promotion
|
|
7
|
-
* across train/dev/test/holdout splits.
|
|
8
|
-
*/
|
|
9
|
-
export type ReferenceReplaySplit = 'train' | 'dev' | 'test' | 'holdout';
|
|
10
|
-
export type ReferenceReplayMatchStrategy = 'reference-order' | 'global-greedy';
|
|
11
|
-
export interface ReferenceReplayItem {
|
|
12
|
-
id: string;
|
|
13
|
-
title: string;
|
|
14
|
-
description?: string;
|
|
15
|
-
severity?: string;
|
|
16
|
-
tags?: string[];
|
|
17
|
-
weight?: number;
|
|
18
|
-
}
|
|
19
|
-
export interface ReferenceReplayCandidate {
|
|
20
|
-
id: string;
|
|
21
|
-
title: string;
|
|
22
|
-
description?: string;
|
|
23
|
-
severity?: string;
|
|
24
|
-
tags?: string[];
|
|
25
|
-
metadata?: Record<string, unknown>;
|
|
26
|
-
}
|
|
27
|
-
export interface ReferenceReplayScenario {
|
|
28
|
-
id: string;
|
|
29
|
-
split?: ReferenceReplaySplit;
|
|
30
|
-
references: ReferenceReplayItem[];
|
|
31
|
-
candidates: ReferenceReplayCandidate[];
|
|
32
|
-
metadata?: Record<string, unknown>;
|
|
33
|
-
}
|
|
34
|
-
export interface ReferenceReplayCase<Input = unknown> {
|
|
35
|
-
id: string;
|
|
36
|
-
split?: ReferenceReplaySplit;
|
|
37
|
-
input: Input;
|
|
38
|
-
references: ReferenceReplayItem[];
|
|
39
|
-
metadata?: Record<string, unknown>;
|
|
40
|
-
}
|
|
41
|
-
export interface ReferenceReplayExecutionScenario<Input = unknown> {
|
|
42
|
-
id: string;
|
|
43
|
-
split: ReferenceReplaySplit;
|
|
44
|
-
input: Input;
|
|
45
|
-
metadata?: Record<string, unknown>;
|
|
46
|
-
}
|
|
47
|
-
export interface ReferenceReplayRunContext {
|
|
48
|
-
runId: string;
|
|
49
|
-
caseIndex: number;
|
|
50
|
-
abortSignal?: AbortSignal;
|
|
51
|
-
}
|
|
52
|
-
export interface ReferenceReplayAdapter<Input = unknown> {
|
|
53
|
-
run(scenario: ReferenceReplayExecutionScenario<Input>, context: ReferenceReplayRunContext): Promise<ReferenceReplayCandidate[]>;
|
|
54
|
-
}
|
|
55
|
-
export type ReferenceReplayAdapterFn<Input = unknown> = (scenario: ReferenceReplayExecutionScenario<Input>, context: ReferenceReplayRunContext) => Promise<ReferenceReplayCandidate[]>;
|
|
56
|
-
export type ReferenceReplayAdapterLike<Input = unknown> = ReferenceReplayAdapter<Input> | ReferenceReplayAdapterFn<Input>;
|
|
57
|
-
export interface ReferenceReplayMatch {
|
|
58
|
-
scenarioId: string;
|
|
59
|
-
referenceId: string;
|
|
60
|
-
candidateId: string | null;
|
|
61
|
-
score: number;
|
|
62
|
-
matched: boolean;
|
|
63
|
-
weight: number;
|
|
64
|
-
reason: string;
|
|
65
|
-
}
|
|
66
|
-
export interface ReferenceReplayScenarioScore {
|
|
67
|
-
scenarioId: string;
|
|
68
|
-
split: ReferenceReplaySplit;
|
|
69
|
-
matched: number;
|
|
70
|
-
total: number;
|
|
71
|
-
falsePositives: number;
|
|
72
|
-
matchedWeight: number;
|
|
73
|
-
totalWeight: number;
|
|
74
|
-
precision: number;
|
|
75
|
-
recall: number;
|
|
76
|
-
f1: number;
|
|
77
|
-
matches: ReferenceReplayMatch[];
|
|
78
|
-
}
|
|
79
|
-
export interface ReferenceReplayAggregate {
|
|
80
|
-
matched: number;
|
|
81
|
-
total: number;
|
|
82
|
-
falsePositives: number;
|
|
83
|
-
matchedWeight: number;
|
|
84
|
-
totalWeight: number;
|
|
85
|
-
precision: number;
|
|
86
|
-
recall: number;
|
|
87
|
-
f1: number;
|
|
88
|
-
weightedRecall: number;
|
|
89
|
-
}
|
|
90
|
-
export interface ReferenceReplayScore {
|
|
91
|
-
scenarios: ReferenceReplayScenarioScore[];
|
|
92
|
-
aggregate: ReferenceReplayAggregate;
|
|
93
|
-
bySplit: Partial<Record<ReferenceReplaySplit, ReferenceReplayAggregate>>;
|
|
94
|
-
}
|
|
95
|
-
export interface ReferenceMatchResult {
|
|
96
|
-
score: number;
|
|
97
|
-
reason?: string;
|
|
98
|
-
}
|
|
99
|
-
export type ReferenceReplayMatcher = (reference: ReferenceReplayItem, candidate: ReferenceReplayCandidate, scenario: ReferenceReplayScenario) => ReferenceMatchResult;
|
|
100
|
-
export interface ReferenceReplayScoreOptions {
|
|
101
|
-
matcher?: ReferenceReplayMatcher;
|
|
102
|
-
matchThreshold?: number;
|
|
103
|
-
matchStrategy?: ReferenceReplayMatchStrategy;
|
|
104
|
-
includeHoldout?: boolean;
|
|
105
|
-
splits?: ReferenceReplaySplit[];
|
|
106
|
-
}
|
|
107
|
-
export interface ReferenceReplayPromotionPolicy {
|
|
108
|
-
/** Splits that must improve or stay flat. Default: ['dev', 'test']. */
|
|
109
|
-
requiredSplits?: ReferenceReplaySplit[];
|
|
110
|
-
/** Minimum aggregate F1 lift required on required splits. Default 0. */
|
|
111
|
-
minF1Delta?: number;
|
|
112
|
-
/** Maximum F1 drop allowed on any compared split. Default 0. */
|
|
113
|
-
maxRegression?: number;
|
|
114
|
-
/** If true, holdout must be present and must not regress. Default true. */
|
|
115
|
-
requireHoldoutNonRegression?: boolean;
|
|
116
|
-
}
|
|
117
|
-
export interface ReferenceReplaySplitComparison {
|
|
118
|
-
split: ReferenceReplaySplit;
|
|
119
|
-
baselineF1: number;
|
|
120
|
-
candidateF1: number;
|
|
121
|
-
f1Delta: number;
|
|
122
|
-
baselineRecall: number;
|
|
123
|
-
candidateRecall: number;
|
|
124
|
-
recallDelta: number;
|
|
125
|
-
}
|
|
126
|
-
export interface ReferenceReplayPromotionDecision {
|
|
127
|
-
promote: boolean;
|
|
128
|
-
reason: string;
|
|
129
|
-
aggregateDelta: number;
|
|
130
|
-
comparisons: ReferenceReplaySplitComparison[];
|
|
131
|
-
regressions: ReferenceReplaySplitComparison[];
|
|
132
|
-
}
|
|
133
|
-
export interface ReferenceReplayCaseRun<Input = unknown> {
|
|
134
|
-
caseId: string;
|
|
135
|
-
split: ReferenceReplaySplit;
|
|
136
|
-
input: Input;
|
|
137
|
-
metadata?: Record<string, unknown>;
|
|
138
|
-
references: ReferenceReplayItem[];
|
|
139
|
-
candidates: ReferenceReplayCandidate[];
|
|
140
|
-
score: ReferenceReplayScenarioScore;
|
|
141
|
-
durationMs: number;
|
|
142
|
-
error?: string;
|
|
143
|
-
}
|
|
144
|
-
export interface ReferenceReplayRun<Input = unknown> {
|
|
145
|
-
id: string;
|
|
146
|
-
variantId?: string;
|
|
147
|
-
startedAt: number;
|
|
148
|
-
completedAt: number;
|
|
149
|
-
durationMs: number;
|
|
150
|
-
cases: ReferenceReplayCaseRun<Input>[];
|
|
151
|
-
score: ReferenceReplayScore;
|
|
152
|
-
metadata?: Record<string, unknown>;
|
|
153
|
-
}
|
|
154
|
-
export interface ReferenceReplayRunOptions<Input = unknown> extends ReferenceReplayScoreOptions {
|
|
155
|
-
adapter: ReferenceReplayAdapterLike<Input>;
|
|
156
|
-
runId?: string;
|
|
157
|
-
variantId?: string;
|
|
158
|
-
metadata?: Record<string, unknown>;
|
|
159
|
-
store?: ReferenceReplayRunStore<Input>;
|
|
160
|
-
abortSignal?: AbortSignal;
|
|
161
|
-
continueOnError?: boolean;
|
|
162
|
-
now?: () => number;
|
|
163
|
-
}
|
|
164
|
-
export interface ReferenceReplayRunStore<Input = unknown> {
|
|
165
|
-
save(run: ReferenceReplayRun<Input>): Promise<void>;
|
|
166
|
-
list(): Promise<ReferenceReplayRun<Input>[]>;
|
|
167
|
-
}
|
|
168
|
-
export declare function runReferenceReplay<Input = unknown>(cases: ReferenceReplayCase<Input>[], options: ReferenceReplayRunOptions<Input>): Promise<ReferenceReplayRun<Input>>;
|
|
169
|
-
export declare function decideReferenceReplayRunPromotion(baseline: ReferenceReplayRun, candidate: ReferenceReplayRun, policy?: ReferenceReplayPromotionPolicy): ReferenceReplayPromotionDecision;
|
|
170
|
-
export declare function inMemoryReferenceReplayStore<Input = unknown>(initial?: ReferenceReplayRun<Input>[]): ReferenceReplayRunStore<Input>;
|
|
171
|
-
export declare function jsonlReferenceReplayStore<Input = unknown>(path: string): ReferenceReplayRunStore<Input>;
|
|
172
|
-
export declare function scoreReferenceReplay(scenarios: ReferenceReplayScenario[], options?: ReferenceReplayScoreOptions): ReferenceReplayScore;
|
|
173
|
-
export declare function compareReferenceReplay(baseline: ReferenceReplayScore, candidate: ReferenceReplayScore): ReferenceReplaySplitComparison[];
|
|
174
|
-
export declare function decideReferenceReplayPromotion(baseline: ReferenceReplayScore, candidate: ReferenceReplayScore, policy?: ReferenceReplayPromotionPolicy): ReferenceReplayPromotionDecision;
|
|
175
|
-
export declare function defaultReferenceReplayMatcher(reference: ReferenceReplayItem, candidate: ReferenceReplayCandidate): ReferenceMatchResult;
|
|
176
|
-
//# sourceMappingURL=reference-replay.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"reference-replay.d.ts","sourceRoot":"","sources":["../src/reference-replay.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAMH,MAAM,MAAM,oBAAoB,GAAG,OAAO,GAAG,KAAK,GAAG,MAAM,GAAG,SAAS,CAAA;AACvE,MAAM,MAAM,4BAA4B,GAAG,iBAAiB,GAAG,eAAe,CAAA;AAE9E,MAAM,WAAW,mBAAmB;IAClC,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,EAAE,MAAM,CAAA;IACb,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAA;IACf,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,wBAAwB;IACvC,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,EAAE,MAAM,CAAA;IACb,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAA;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,uBAAuB;IACtC,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,CAAC,EAAE,oBAAoB,CAAA;IAC5B,UAAU,EAAE,mBAAmB,EAAE,CAAA;IACjC,UAAU,EAAE,wBAAwB,EAAE,CAAA;IACtC,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,mBAAmB,CAAC,KAAK,GAAG,OAAO;IAClD,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,CAAC,EAAE,oBAAoB,CAAA;IAC5B,KAAK,EAAE,KAAK,CAAA;IACZ,UAAU,EAAE,mBAAmB,EAAE,CAAA;IACjC,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,gCAAgC,CAAC,KAAK,GAAG,OAAO;IAC/D,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,EAAE,oBAAoB,CAAA;IAC3B,KAAK,EAAE,KAAK,CAAA;IACZ,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,yBAAyB;IACxC,KAAK,EAAE,MAAM,CAAA;IACb,SAAS,EAAE,MAAM,CAAA;IACjB,WAAW,CAAC,EAAE,WAAW,CAAA;CAC1B;AAED,MAAM,WAAW,sBAAsB,CAAC,KAAK,GAAG,OAAO;IACrD,GAAG,CACD,QAAQ,EAAE,gCAAgC,CAAC,KAAK,CAAC,EACjD,OAAO,EAAE,yBAAyB,GACjC,OAAO,CAAC,wBAAwB,EAAE,CAAC,CAAA;CACvC;AAED,MAAM,MAAM,wBAAwB,CAAC,KAAK,GAAG,OAAO,IAAI,CACtD,QAAQ,EAAE,gCAAgC,CAAC,KAAK,CAAC,EACjD,OAAO,EAAE,yBAAyB,KAC/B,OAAO,CAAC,wBAAwB,EAAE,CAAC,CAAA;AAExC,MAAM,MAAM,0BAA0B,CAAC,KAAK,GAAG,OAAO,IAClD,sBAAsB,CAAC,KAAK,CAAC,GAC7B,wBAAwB,CAAC,KAAK,CAAC,CAAA;AAEnC,MAAM,WAAW,oBAAoB;IACnC,UAAU,EAAE,MAAM,CAAA;IAClB,WAAW,EAAE,MAAM,CAAA;IACnB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAA;IAC1B,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,EAAE,OAAO,CAAA;IAChB,MAAM,EAAE,MAAM,CAAA;IACd,MAAM,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,4BAA4B;IAC3C,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,EAAE,oBAAoB,CAAA;IAC3B,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,EAAE,MAAM,CAAA;IACb,cAAc,EAAE,MAAM,CAAA;IACtB,aAAa,EAAE,MAAM,CAAA;IACrB,WAAW,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,MAAM,EAAE,MAAM,CAAA;IACd,EAAE,EAAE,MAAM,CAAA;IACV,OAAO,EAAE,oBAAoB,EAAE,CAAA;CAChC;AAED,MAAM,WAAW,wBAAwB;IACvC,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,EAAE,MAAM,CAAA;IACb,cAAc,EAAE,MAAM,CAAA;IACtB,aAAa,EAAE,MAAM,CAAA;IACrB,WAAW,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,MAAM,EAAE,MAAM,CAAA;IACd,EAAE,EAAE,MAAM,CAAA;IACV,cAAc,EAAE,MAAM,CAAA;CACvB;AAED,MAAM,WAAW,oBAAoB;IACnC,SAAS,EAAE,4BAA4B,EAAE,CAAA;IACzC,SAAS,EAAE,wBAAwB,CAAA;IACnC,OAAO,EAAE,OAAO,CAAC,MAAM,CAAC,oBAAoB,EAAE,wBAAwB,CAAC,CAAC,CAAA;CACzE;AAED,MAAM,WAAW,oBAAoB;IACnC,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,MAAM,sBAAsB,GAAG,CACnC,SAAS,EAAE,mBAAmB,EAC9B,SAAS,EAAE,wBAAwB,EACnC,QAAQ,EAAE,uBAAuB,KAC9B,oBAAoB,CAAA;AAEzB,MAAM,WAAW,2BAA2B;IAC1C,OAAO,CAAC,EAAE,sBAAsB,CAAA;IAChC,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,aAAa,CAAC,EAAE,4BAA4B,CAAA;IAC5C,cAAc,CAAC,EAAE,OAAO,CAAA;IACxB,MAAM,CAAC,EAAE,oBAAoB,EAAE,CAAA;CAChC;AAED,MAAM,WAAW,8BAA8B;IAC7C,uEAAuE;IACvE,cAAc,CAAC,EAAE,oBAAoB,EAAE,CAAA;IACvC,wEAAwE;IACxE,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,gEAAgE;IAChE,aAAa,CAAC,EAAE,MAAM,CAAA;IACtB,2EAA2E;IAC3E,2BAA2B,CAAC,EAAE,OAAO,CAAA;CACtC;AAED,MAAM,WAAW,8BAA8B;IAC7C,KAAK,EAAE,oBAAoB,CAAA;IAC3B,UAAU,EAAE,MAAM,CAAA;IAClB,WAAW,EAAE,MAAM,CAAA;IACnB,OAAO,EAAE,MAAM,CAAA;IACf,cAAc,EAAE,MAAM,CAAA;IACtB,eAAe,EAAE,MAAM,CAAA;IACvB,WAAW,EAAE,MAAM,CAAA;CACpB;AAED,MAAM,WAAW,gCAAgC;IAC/C,OAAO,EAAE,OAAO,CAAA;IAChB,MAAM,EAAE,MAAM,CAAA;IACd,cAAc,EAAE,MAAM,CAAA;IACtB,WAAW,EAAE,8BAA8B,EAAE,CAAA;IAC7C,WAAW,EAAE,8BAA8B,EAAE,CAAA;CAC9C;AAED,MAAM,WAAW,sBAAsB,CAAC,KAAK,GAAG,OAAO;IACrD,MAAM,EAAE,MAAM,CAAA;IACd,KAAK,EAAE,oBAAoB,CAAA;IAC3B,KAAK,EAAE,KAAK,CAAA;IACZ,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAClC,UAAU,EAAE,mBAAmB,EAAE,CAAA;IACjC,UAAU,EAAE,wBAAwB,EAAE,CAAA;IACtC,KAAK,EAAE,4BAA4B,CAAA;IACnC,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,kBAAkB,CAAC,KAAK,GAAG,OAAO;IACjD,EAAE,EAAE,MAAM,CAAA;IACV,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,SAAS,EAAE,MAAM,CAAA;IACjB,WAAW,EAAE,MAAM,CAAA;IACnB,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,EAAE,sBAAsB,CAAC,KAAK,CAAC,EAAE,CAAA;IACtC,KAAK,EAAE,oBAAoB,CAAA;IAC3B,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,yBAAyB,CAAC,KAAK,GAAG,OAAO,CAAE,SAAQ,2BAA2B;IAC7F,OAAO,EAAE,0BAA0B,CAAC,KAAK,CAAC,CAAA;IAC1C,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAClC,KAAK,CAAC,EAAE,uBAAuB,CAAC,KAAK,CAAC,CAAA;IACtC,WAAW,CAAC,EAAE,WAAW,CAAA;IACzB,eAAe,CAAC,EAAE,OAAO,CAAA;IACzB,GAAG,CAAC,EAAE,MAAM,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,uBAAuB,CAAC,KAAK,GAAG,OAAO;IACtD,IAAI,CAAC,GAAG,EAAE,kBAAkB,CAAC,KAAK,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IACnD,IAAI,IAAI,OAAO,CAAC,kBAAkB,CAAC,KAAK,CAAC,EAAE,CAAC,CAAA;CAC7C;AAKD,wBAAsB,kBAAkB,CAAC,KAAK,GAAG,OAAO,EACtD,KAAK,EAAE,mBAAmB,CAAC,KAAK,CAAC,EAAE,EACnC,OAAO,EAAE,yBAAyB,CAAC,KAAK,CAAC,GACxC,OAAO,CAAC,kBAAkB,CAAC,KAAK,CAAC,CAAC,CA4FpC;AAED,wBAAgB,iCAAiC,CAC/C,QAAQ,EAAE,kBAAkB,EAC5B,SAAS,EAAE,kBAAkB,EAC7B,MAAM,GAAE,8BAAmC,GAC1C,gCAAgC,CAElC;AAED,wBAAgB,4BAA4B,CAAC,KAAK,GAAG,OAAO,EAC1D,OAAO,GAAE,kBAAkB,CAAC,KAAK,CAAC,EAAO,GACxC,uBAAuB,CAAC,KAAK,CAAC,CAUhC;AAkBD,wBAAgB,yBAAyB,CAAC,KAAK,GAAG,OAAO,EACvD,IAAI,EAAE,MAAM,GACX,uBAAuB,CAAC,KAAK,CAAC,CAgBhC;AAED,wBAAgB,oBAAoB,CAClC,SAAS,EAAE,uBAAuB,EAAE,EACpC,OAAO,GAAE,2BAAgC,GACxC,oBAAoB,CAkBtB;AAED,wBAAgB,sBAAsB,CACpC,QAAQ,EAAE,oBAAoB,EAC9B,SAAS,EAAE,oBAAoB,GAC9B,8BAA8B,EAAE,CAkBlC;AAED,wBAAgB,8BAA8B,CAC5C,QAAQ,EAAE,oBAAoB,EAC9B,SAAS,EAAE,oBAAoB,EAC/B,MAAM,GAAE,8BAAmC,GAC1C,gCAAgC,CAuElC;AAED,wBAAgB,6BAA6B,CAC3C,SAAS,EAAE,mBAAmB,EAC9B,SAAS,EAAE,wBAAwB,GAClC,oBAAoB,CAgBtB"}
|
|
@@ -1,79 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Reflective mutation — primitives for trace-conditioned prompt rewriting.
|
|
3
|
-
*
|
|
4
|
-
* Used by `prompt-evolution.ts` (and any consumer running iterative
|
|
5
|
-
* improvement). Given a parent prompt + concrete trace evidence (top trials,
|
|
6
|
-
* bottom trials, missed expectations), produce an LLM-ready prompt that
|
|
7
|
-
* proposes targeted mutations — not blind rephrasings.
|
|
8
|
-
*
|
|
9
|
-
* Why this lives outside `prompt-evolution.ts`: any consumer that wants to
|
|
10
|
-
* run reflective rewriting WITHOUT the population/Pareto machinery can
|
|
11
|
-
* import these primitives directly.
|
|
12
|
-
*
|
|
13
|
-
* Quality bar (vs. naive "mutate this prompt"):
|
|
14
|
-
* - Show parent ↔ children diff, not just one variant
|
|
15
|
-
* - Quote specific missed goldens with their match phrases
|
|
16
|
-
* - Surface the model's actual emitted output side-by-side with what was expected
|
|
17
|
-
* - Quote concrete mutation primitives so the model has a vocabulary
|
|
18
|
-
*/
|
|
19
|
-
export interface TrialTrace {
|
|
20
|
-
/** Stable id for the trial — surfaces in the prompt for grounding. */
|
|
21
|
-
id: string;
|
|
22
|
-
/** Score the trial received on its primary metric. */
|
|
23
|
-
score: number;
|
|
24
|
-
/** Candidate inputs the agent was given (e.g., the fixture or scenario). */
|
|
25
|
-
inputName?: string;
|
|
26
|
-
/**
|
|
27
|
-
* Goldens / expectations this trial was tested against, with whether each
|
|
28
|
-
* was matched. The reflection prompt quotes the missed ones specifically.
|
|
29
|
-
*/
|
|
30
|
-
expectations?: Array<{
|
|
31
|
-
id: string;
|
|
32
|
-
phrase: string;
|
|
33
|
-
matched: boolean;
|
|
34
|
-
}>;
|
|
35
|
-
/** Free-form text — what the agent actually emitted (e.g., findings, plan). */
|
|
36
|
-
emitted?: string;
|
|
37
|
-
/** The judge's free-form note on WHY this trial scored low — the failure
|
|
38
|
-
* evidence the reflection targets. Should be a generalizable pattern (which
|
|
39
|
-
* checks/lines/dimensions failed, and how), never case-specific ground
|
|
40
|
-
* truth (that would teach memorization, not capability). */
|
|
41
|
-
failureNote?: string;
|
|
42
|
-
/** Optional structured metrics (recall, precision, cost, latency). */
|
|
43
|
-
metrics?: Record<string, number>;
|
|
44
|
-
}
|
|
45
|
-
export interface ReflectionContext {
|
|
46
|
-
/** What is being mutated — appears in the system prompt for orientation. */
|
|
47
|
-
target: string;
|
|
48
|
-
/** Current variant's payload — JSON-serialised for the prompt. */
|
|
49
|
-
parentPayload: unknown;
|
|
50
|
-
/** Best-performing trials this generation. */
|
|
51
|
-
topTrials: TrialTrace[];
|
|
52
|
-
/** Worst-performing trials this generation — the missed-golden source. */
|
|
53
|
-
bottomTrials: TrialTrace[];
|
|
54
|
-
/** How many children the mutator should propose. */
|
|
55
|
-
childCount: number;
|
|
56
|
-
/** Optional: domain-specific mutation primitives the model can pick from. */
|
|
57
|
-
mutationPrimitives?: string[];
|
|
58
|
-
}
|
|
59
|
-
export declare const DEFAULT_MUTATION_PRIMITIVES: string[];
|
|
60
|
-
/**
|
|
61
|
-
* Build the LLM-ready reflection prompt. Output is plain text — pass it as
|
|
62
|
-
* the user message. The system message should be small and stable (e.g.
|
|
63
|
-
* "Output ONLY a JSON object matching the schema below.").
|
|
64
|
-
*/
|
|
65
|
-
export declare function buildReflectionPrompt(ctx: ReflectionContext): string;
|
|
66
|
-
/** Render analyst findings + an optional analysis report into a prompt evidence
|
|
67
|
-
* block. A findings producer diagnoses the generation's traces; this surfaces
|
|
68
|
-
* that diagnosis to the proposer so a mutation targets a NAMED root cause instead of blindly
|
|
69
|
-
* rephrasing. Duck-types the finding shape (claim / severity / area /
|
|
70
|
-
* recommended_action) so any findings producer's rows render. Returns null when
|
|
71
|
-
* there is nothing to add, so callers append nothing. */
|
|
72
|
-
export declare function renderAnalystEvidence(findings: unknown[], report?: unknown): string | null;
|
|
73
|
-
export interface ReflectionProposal {
|
|
74
|
-
label: string;
|
|
75
|
-
rationale: string;
|
|
76
|
-
payload: unknown;
|
|
77
|
-
}
|
|
78
|
-
export declare function parseReflectionResponse(raw: string, maxProposals?: number): ReflectionProposal[];
|
|
79
|
-
//# sourceMappingURL=reflective-mutation.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"reflective-mutation.d.ts","sourceRoot":"","sources":["../src/reflective-mutation.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAEH,MAAM,WAAW,UAAU;IACzB,sEAAsE;IACtE,EAAE,EAAE,MAAM,CAAA;IACV,sDAAsD;IACtD,KAAK,EAAE,MAAM,CAAA;IACb,4EAA4E;IAC5E,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB;;;OAGG;IACH,YAAY,CAAC,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,OAAO,CAAA;KAAE,CAAC,CAAA;IACtE,+EAA+E;IAC/E,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB;;;iEAG6D;IAC7D,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,sEAAsE;IACtE,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CACjC;AAED,MAAM,WAAW,iBAAiB;IAChC,4EAA4E;IAC5E,MAAM,EAAE,MAAM,CAAA;IACd,kEAAkE;IAClE,aAAa,EAAE,OAAO,CAAA;IACtB,8CAA8C;IAC9C,SAAS,EAAE,UAAU,EAAE,CAAA;IACvB,0EAA0E;IAC1E,YAAY,EAAE,UAAU,EAAE,CAAA;IAC1B,oDAAoD;IACpD,UAAU,EAAE,MAAM,CAAA;IAClB,6EAA6E;IAC7E,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAA;CAC9B;AAED,eAAO,MAAM,2BAA2B,EAAE,MAAM,EAO/C,CAAA;AAED;;;;GAIG;AACH,wBAAgB,qBAAqB,CAAC,GAAG,EAAE,iBAAiB,GAAG,MAAM,CAqFpE;AAED;;;;;0DAK0D;AAC1D,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,MAAM,CAAC,EAAE,OAAO,GAAG,MAAM,GAAG,IAAI,CAuC1F;AAWD,MAAM,WAAW,kBAAkB;IACjC,KAAK,EAAE,MAAM,CAAA;IACb,SAAS,EAAE,MAAM,CAAA;IACjB,OAAO,EAAE,OAAO,CAAA;CACjB;AA0DD,wBAAgB,uBAAuB,CAAC,GAAG,EAAE,MAAM,EAAE,YAAY,CAAC,EAAE,MAAM,GAAG,kBAAkB,EAAE,CA8EhG"}
|
package/dist/registry.d.ts
DELETED
|
@@ -1,31 +0,0 @@
|
|
|
1
|
-
import type { Scenario, ScenarioFile } from './types';
|
|
2
|
-
/**
|
|
3
|
-
* ScenarioRegistry — manages scenario discovery and filtering.
|
|
4
|
-
*
|
|
5
|
-
* Each agent registers its scenarios. The registry handles conversion
|
|
6
|
-
* from ScenarioFile format to the framework's Scenario type.
|
|
7
|
-
*/
|
|
8
|
-
export declare class ScenarioRegistry {
|
|
9
|
-
private scenarios;
|
|
10
|
-
private scenarioFiles;
|
|
11
|
-
/** Register scenarios from ScenarioFile format */
|
|
12
|
-
registerFiles(files: ScenarioFile[]): void;
|
|
13
|
-
/** Register pre-built Scenario objects directly */
|
|
14
|
-
register(scenarios: Scenario[]): void;
|
|
15
|
-
/** Get all scenarios */
|
|
16
|
-
all(): Scenario[];
|
|
17
|
-
/** Get scenarios filtered by category */
|
|
18
|
-
byCategory(category: string): Scenario[];
|
|
19
|
-
/** List all categories with counts */
|
|
20
|
-
listCategories(): {
|
|
21
|
-
category: string;
|
|
22
|
-
count: number;
|
|
23
|
-
}[];
|
|
24
|
-
/** Get scenarios filtered by persona */
|
|
25
|
-
byPersona(persona: string): Scenario[];
|
|
26
|
-
/** Get a single scenario by ID */
|
|
27
|
-
byId(id: string): Scenario | undefined;
|
|
28
|
-
/** Count total scenarios */
|
|
29
|
-
get count(): number;
|
|
30
|
-
}
|
|
31
|
-
//# sourceMappingURL=registry.d.ts.map
|
package/dist/registry.d.ts.map
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"registry.d.ts","sourceRoot":"","sources":["../src/registry.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,YAAY,EAAE,MAAM,SAAS,CAAA;AAErD;;;;;GAKG;AACH,qBAAa,gBAAgB;IAC3B,OAAO,CAAC,SAAS,CAAiB;IAClC,OAAO,CAAC,aAAa,CAAqB;IAE1C,kDAAkD;IAClD,aAAa,CAAC,KAAK,EAAE,YAAY,EAAE,GAAG,IAAI;IAK1C,mDAAmD;IACnD,QAAQ,CAAC,SAAS,EAAE,QAAQ,EAAE,GAAG,IAAI;IAIrC,wBAAwB;IACxB,GAAG,IAAI,QAAQ,EAAE;IAIjB,yCAAyC;IACzC,UAAU,CAAC,QAAQ,EAAE,MAAM,GAAG,QAAQ,EAAE;IAKxC,sCAAsC;IACtC,cAAc,IAAI;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,EAAE;IAQvD,wCAAwC;IACxC,SAAS,CAAC,OAAO,EAAE,MAAM,GAAG,QAAQ,EAAE;IAItC,kCAAkC;IAClC,IAAI,CAAC,EAAE,EAAE,MAAM,GAAG,QAAQ,GAAG,SAAS;IAItC,4BAA4B;IAC5B,IAAI,KAAK,IAAI,MAAM,CAElB;CACF"}
|