@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
|
@@ -1,146 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Anytime-valid sequential promotion gate — an e-process (betting
|
|
3
|
-
* test-martingale, see `eProcess` in `statistics.ts`) over paired
|
|
4
|
-
* per-scenario deltas, so a campaign stops the MOMENT evidence decides
|
|
5
|
-
* instead of burning a fixed-n budget. Decisions remain valid at any
|
|
6
|
-
* data-dependent stopping time (Ville's inequality), which is exactly what
|
|
7
|
-
* fixed-n machinery cannot offer: peeking at a bootstrap CI after every
|
|
8
|
-
* observation and stopping on the first significant peek inflates type-I
|
|
9
|
-
* error far beyond alpha.
|
|
10
|
-
*
|
|
11
|
-
* REPLACES, never layers on, a fixed-n gate. Running `heldoutSignificance`
|
|
12
|
-
* or `paretoSignificanceGate` repeatedly on a growing sample and stopping
|
|
13
|
-
* early is optional stopping no matter how it is dressed up; this gate is
|
|
14
|
-
* the valid way to stop early. Use one or the other per evidence stream.
|
|
15
|
-
*
|
|
16
|
-
* Pre-registration binding: anytime validity holds only for the
|
|
17
|
-
* PRE-REGISTERED statistic. When a `SignedManifest` is bound, the gate takes
|
|
18
|
-
* alpha from `manifest.alpha`, the observation budget from
|
|
19
|
-
* `manifest.preRegisteredN`, orients deltas by `manifest.direction`, and
|
|
20
|
-
* shifts the null boundary by `manifest.minEffect` — re-deciding the same
|
|
21
|
-
* stream under different parameters after seeing data would reopen optional
|
|
22
|
-
* stopping under a fancier name. The manifest's content hash is verified at
|
|
23
|
-
* construction (sync, same `sha256-content` scheme as `signManifest`).
|
|
24
|
-
*
|
|
25
|
-
* Non-iid caveat (stated honestly): the supermartingale guarantee needs each
|
|
26
|
-
* delta's conditional mean under H0 to stay ≤ the null boundary given the
|
|
27
|
-
* past — exchangeable scenario deltas suffice. Scenario streams ordered by
|
|
28
|
-
* difficulty or by scenario family violate this; `decide(ctx)` therefore
|
|
29
|
-
* shuffles the paired deltas with a SEEDED permutation by default (the
|
|
30
|
-
* permutation is data-independent, so bet predictability is preserved).
|
|
31
|
-
* Stratified betting (per-stratum λ) is future work, not implemented here.
|
|
32
|
-
*/
|
|
33
|
-
import { type SignedManifest } from '../../pre-registration';
|
|
34
|
-
import { type EProcessState } from '../../statistics';
|
|
35
|
-
import type { Gate, GenerationRecord, Scenario } from '../types';
|
|
36
|
-
export type SequentialDecision = 'promote' | 'continue' | 'undecided-at-maxN';
|
|
37
|
-
export interface SequentialObservation {
|
|
38
|
-
decision: SequentialDecision;
|
|
39
|
-
/** Current e-value (the betting wealth) against H0. */
|
|
40
|
-
eValue: number;
|
|
41
|
-
/** Paired deltas consumed so far. */
|
|
42
|
-
n: number;
|
|
43
|
-
/** Names the decision basis. For 'undecided-at-maxN' it states explicitly
|
|
44
|
-
* that exhausting the budget is NOT evidence of no effect. */
|
|
45
|
-
reason: string;
|
|
46
|
-
}
|
|
47
|
-
export interface SequentialPairedGateOptions {
|
|
48
|
-
/** Type-I budget. With `preRegistration` bound this MUST match
|
|
49
|
-
* `manifest.alpha` (conflict throws). Default 0.05. */
|
|
50
|
-
alpha?: number;
|
|
51
|
-
/** Minimum paired deltas before a promote may fire. The stopping rule is
|
|
52
|
-
* "first n ≥ minN with e-value ≥ 1/alpha" — still a valid stopping time.
|
|
53
|
-
* Default 5. */
|
|
54
|
-
minN?: number;
|
|
55
|
-
/** Pre-registered observation budget. Required unless `preRegistration`
|
|
56
|
-
* supplies it via `preRegisteredN` (conflict throws). */
|
|
57
|
-
maxN?: number;
|
|
58
|
-
/** Bet truncation forwarded to `eProcess`. Default 0.5. */
|
|
59
|
-
maxBet?: number;
|
|
60
|
-
/** Bound on |delta| in the judge's native scale; deltas are mapped to
|
|
61
|
-
* x = (d/scale + 1)/2 ∈ [0,1]. A delta outside ±scale throws (use
|
|
62
|
-
* `detectScale` to pick 1 vs 100 BEFORE streaming). Default 1. */
|
|
63
|
-
scale?: number;
|
|
64
|
-
/** Seed for the data-independent shuffle of paired deltas in `decide(ctx)`
|
|
65
|
-
* (exchangeability guard). Default 1337. */
|
|
66
|
-
shuffleSeed?: number;
|
|
67
|
-
/** Bind the pre-registered hypothesis. Verified (content hash) at
|
|
68
|
-
* construction; alpha/maxN/direction/minEffect come FROM the manifest. */
|
|
69
|
-
preRegistration?: SignedManifest;
|
|
70
|
-
/** Override the gate name in reports. */
|
|
71
|
-
name?: string;
|
|
72
|
-
}
|
|
73
|
-
export interface SequentialPairedGate<TArtifact = unknown, TScenario extends Scenario = Scenario> extends Gate<TArtifact, TScenario> {
|
|
74
|
-
/** Streaming entry point: feed one paired per-scenario delta
|
|
75
|
-
* (candidate − baseline, native scale). Each gate instance carries ONE
|
|
76
|
-
* observe-stream; `decide(ctx)` runs on its own fresh stream and never
|
|
77
|
-
* consumes or advances this one. 'promote' is sticky; observing past the
|
|
78
|
-
* pre-registered maxN throws (extending a finished stream after seeing
|
|
79
|
-
* the result reopens optional stopping — start a NEW pre-registered
|
|
80
|
-
* test). */
|
|
81
|
-
observe(delta: number): SequentialObservation;
|
|
82
|
-
/** Read-only snapshot of the observe-stream. */
|
|
83
|
-
state(): EProcessState & {
|
|
84
|
-
decision: SequentialDecision;
|
|
85
|
-
};
|
|
86
|
-
}
|
|
87
|
-
/**
|
|
88
|
-
* Anytime-valid sequential paired gate. Conforms to the existing `Gate`
|
|
89
|
-
* contract (`decide(ctx)` consumes candidate vs baseline judge scores via
|
|
90
|
-
* `pairHoldout` — same pairing granularity as the fixed-n gates: full cellId,
|
|
91
|
-
* never scenarioId) and adds a streaming `observe(delta)` entry for campaigns
|
|
92
|
-
* that score cells incrementally and want to stop mid-stream.
|
|
93
|
-
*
|
|
94
|
-
* Decision mapping onto the substrate's five-valued `GateDecision`:
|
|
95
|
-
* - 'promote' → 'ship'
|
|
96
|
-
* - 'continue' → 'need_more_work' (stream ended before maxN with
|
|
97
|
-
* the e-value undecided — more reps could decide)
|
|
98
|
-
* - 'undecided-at-maxN' → 'hold', with the reason stating it is NOT
|
|
99
|
-
* evidence of no effect (never a silent default)
|
|
100
|
-
*/
|
|
101
|
-
export declare function sequentialPairedGate<TArtifact = unknown, TScenario extends Scenario = Scenario>(options: SequentialPairedGateOptions): SequentialPairedGate<TArtifact, TScenario>;
|
|
102
|
-
export interface SequentialDecideOptions {
|
|
103
|
-
/** Type-I budget for the early-stop evidence. Default 0.05. */
|
|
104
|
-
alpha?: number;
|
|
105
|
-
/** Minimum paired deltas before a stop may fire. Default 5. */
|
|
106
|
-
minN?: number;
|
|
107
|
-
/** Bet truncation forwarded to `eProcess`. Default 0.5. */
|
|
108
|
-
maxBet?: number;
|
|
109
|
-
/** Bound on |per-scenario composite delta|. Default 1. */
|
|
110
|
-
scale?: number;
|
|
111
|
-
}
|
|
112
|
-
export interface SequentialDecideFn {
|
|
113
|
-
(args: {
|
|
114
|
-
history: GenerationRecord[];
|
|
115
|
-
}): {
|
|
116
|
-
stop: boolean;
|
|
117
|
-
reason?: string;
|
|
118
|
-
};
|
|
119
|
-
/** Read-only snapshot of the accumulated e-process (observability + tests). */
|
|
120
|
-
state(): EProcessState;
|
|
121
|
-
}
|
|
122
|
-
/**
|
|
123
|
-
* `SurfaceProposer.decide` adapter — stops the optimization loop the moment
|
|
124
|
-
* the e-process decides the loop has produced a real improvement, instead of
|
|
125
|
-
* always running `maxGenerations`.
|
|
126
|
-
*
|
|
127
|
-
* Stream: for each generation g ≥ 1, the per-scenario composite deltas of
|
|
128
|
-
* generation g's top candidate vs the generation-0 top candidate (the
|
|
129
|
-
* incumbent the loop set out to beat), paired by scenarioId. H0: no proposed
|
|
130
|
-
* surface improves any scenario's expected composite over the incumbent —
|
|
131
|
-
* under it every delta has conditional mean ≤ 0 and the e-process is valid.
|
|
132
|
-
* Once wealth ≥ 1/alpha the loop stops and hands the winner to the promotion
|
|
133
|
-
* gate (which re-scores on HELD-OUT data — this adapter only spends the
|
|
134
|
-
* exploration budget, it never promotes).
|
|
135
|
-
*
|
|
136
|
-
* Honesty caveats: (1) the incumbent's scores are measured once and shared
|
|
137
|
-
* across all generations' deltas, so type-I control is exact only insofar as
|
|
138
|
-
* those scores approximate the incumbent's true per-scenario means (more reps
|
|
139
|
-
* → tighter); (2) an UNDECIDED process never stops the loop — absence of a
|
|
140
|
-
* crossing is NOT evidence of no effect, so the loop simply runs its normal
|
|
141
|
-
* course. Calling the adapter repeatedly with a growing history consumes each
|
|
142
|
-
* generation exactly once (re-feeding an already-seen record would double-count
|
|
143
|
-
* evidence).
|
|
144
|
-
*/
|
|
145
|
-
export declare function sequentialDecide(options?: SequentialDecideOptions): SequentialDecideFn;
|
|
146
|
-
//# sourceMappingURL=sequential.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"sequential.d.ts","sourceRoot":"","sources":["../../../src/campaign/gates/sequential.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA+BG;AAGH,OAAO,EAAgB,KAAK,cAAc,EAAE,MAAM,wBAAwB,CAAA;AAC1E,OAAO,EAAE,KAAK,aAAa,EAAwB,MAAM,kBAAkB,CAAA;AAC3E,OAAO,KAAK,EAAE,IAAI,EAA2B,gBAAgB,EAAE,QAAQ,EAAE,MAAM,UAAU,CAAA;AAGzF,MAAM,MAAM,kBAAkB,GAAG,SAAS,GAAG,UAAU,GAAG,mBAAmB,CAAA;AAE7E,MAAM,WAAW,qBAAqB;IACpC,QAAQ,EAAE,kBAAkB,CAAA;IAC5B,uDAAuD;IACvD,MAAM,EAAE,MAAM,CAAA;IACd,qCAAqC;IACrC,CAAC,EAAE,MAAM,CAAA;IACT;mEAC+D;IAC/D,MAAM,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,2BAA2B;IAC1C;4DACwD;IACxD,KAAK,CAAC,EAAE,MAAM,CAAA;IACd;;qBAEiB;IACjB,IAAI,CAAC,EAAE,MAAM,CAAA;IACb;8DAC0D;IAC1D,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,2DAA2D;IAC3D,MAAM,CAAC,EAAE,MAAM,CAAA;IACf;;uEAEmE;IACnE,KAAK,CAAC,EAAE,MAAM,CAAA;IACd;iDAC6C;IAC7C,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB;+EAC2E;IAC3E,eAAe,CAAC,EAAE,cAAc,CAAA;IAChC,yCAAyC;IACzC,IAAI,CAAC,EAAE,MAAM,CAAA;CACd;AAED,MAAM,WAAW,oBAAoB,CAAC,SAAS,GAAG,OAAO,EAAE,SAAS,SAAS,QAAQ,GAAG,QAAQ,CAC9F,SAAQ,IAAI,CAAC,SAAS,EAAE,SAAS,CAAC;IAClC;;;;;;iBAMa;IACb,OAAO,CAAC,KAAK,EAAE,MAAM,GAAG,qBAAqB,CAAA;IAC7C,gDAAgD;IAChD,KAAK,IAAI,aAAa,GAAG;QAAE,QAAQ,EAAE,kBAAkB,CAAA;KAAE,CAAA;CAC1D;AA2LD;;;;;;;;;;;;;GAaG;AACH,wBAAgB,oBAAoB,CAAC,SAAS,GAAG,OAAO,EAAE,SAAS,SAAS,QAAQ,GAAG,QAAQ,EAC7F,OAAO,EAAE,2BAA2B,GACnC,oBAAoB,CAAC,SAAS,EAAE,SAAS,CAAC,CA0E5C;AAED,MAAM,WAAW,uBAAuB;IACtC,+DAA+D;IAC/D,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,+DAA+D;IAC/D,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,2DAA2D;IAC3D,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,0DAA0D;IAC1D,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,kBAAkB;IACjC,CAAC,IAAI,EAAE;QAAE,OAAO,EAAE,gBAAgB,EAAE,CAAA;KAAE,GAAG;QAAE,IAAI,EAAE,OAAO,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,CAAA;IAC3E,+EAA+E;IAC/E,KAAK,IAAI,aAAa,CAAA;CACvB;AAED;;;;;;;;;;;;;;;;;;;;;;GAsBG;AACH,wBAAgB,gBAAgB,CAAC,OAAO,GAAE,uBAA4B,GAAG,kBAAkB,CAuE1F"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"sequential.test.d.ts","sourceRoot":"","sources":["../../../src/campaign/gates/sequential.test.ts"],"names":[],"mappings":""}
|
|
@@ -1,99 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Statistical held-out promotion machinery — the trustworthy core the
|
|
3
|
-
* point-estimate `heldout-delta` gate lacked.
|
|
4
|
-
*
|
|
5
|
-
* The shipped false positive it prevents: a winner re-scored against the
|
|
6
|
-
* baseline on the holdout read run-to-run model NOISE (e.g. 91 vs 95) as a
|
|
7
|
-
* "+4 lift" and shipped, because the gate compared point estimates with no
|
|
8
|
-
* confidence interval. Here we pair candidate vs baseline holdout observations
|
|
9
|
-
* and bootstrap a CI on the paired delta — a candidate ships only when the CI
|
|
10
|
-
* lower bound clears the effect-size threshold (the gain is real at the
|
|
11
|
-
* confidence level, not noise), and is blocked when a critical dimension
|
|
12
|
-
* (e.g. `hallucination_free` for a legal agent) significantly regresses even if
|
|
13
|
-
* the net composite rose (anti-Goodhart).
|
|
14
|
-
*
|
|
15
|
-
* Two traps this module is built around (both produce a NEW false positive if
|
|
16
|
-
* gotten wrong):
|
|
17
|
-
* 1. PAIRING GRANULARITY — pairs by FULL `cellId` (`scenario:rep`), never by
|
|
18
|
-
* `scenarioId` (which averages reps away and destroys the within-pair
|
|
19
|
-
* variance reduction that makes a paired bootstrap tighter than unpaired).
|
|
20
|
-
* One paired observation per cell ⇒ reps multiply n.
|
|
21
|
-
* 2. SCALE — a judge may emit composites/dimensions on [0,1] or 0-100. The
|
|
22
|
-
* threshold + tolerance are interpreted in the judge's NATIVE scale; the
|
|
23
|
-
* per-dimension tolerance auto-scales off the observed baseline magnitudes
|
|
24
|
-
* so `-0.10` on [0,1] doesn't silently become a no-op on a 0-100 dimension.
|
|
25
|
-
*/
|
|
26
|
-
import { type PairedBootstrapResult } from '../../statistics';
|
|
27
|
-
import type { JudgeScore } from '../types';
|
|
28
|
-
export interface PairedHoldout {
|
|
29
|
-
/** Baseline scalar per paired cell (same order as `after`/`cellIds`). */
|
|
30
|
-
before: number[];
|
|
31
|
-
/** Candidate scalar per paired cell. */
|
|
32
|
-
after: number[];
|
|
33
|
-
/** The full cellIds (`scenario:rep`) that paired, in order. */
|
|
34
|
-
cellIds: string[];
|
|
35
|
-
}
|
|
36
|
-
/**
|
|
37
|
-
* Pair candidate vs baseline holdout observations by FULL cellId. `select`
|
|
38
|
-
* pulls the scalar from a cell's judge reports (composite, or a named
|
|
39
|
-
* dimension); a cell contributes the mean of `select` across its judges. Cells
|
|
40
|
-
* whose scenario is not in `scenarioIds`, or where `select` is undefined for
|
|
41
|
-
* every judge on either side, are skipped on BOTH sides so the arrays stay
|
|
42
|
-
* paired. Throws when the two maps disagree on which holdout cells exist — a
|
|
43
|
-
* load-bearing invariant: the baseline + winner holdout campaigns run the same
|
|
44
|
-
* scenarios with the same seed base, so their cellIds MUST align; a mismatch
|
|
45
|
-
* means a silent pairing bug, not a soft fallback.
|
|
46
|
-
*/
|
|
47
|
-
export declare function pairHoldout(candidate: Map<string, Record<string, JudgeScore>>, baseline: Map<string, Record<string, JudgeScore>>, scenarioIds: Set<string>, select: (s: JudgeScore) => number | undefined): PairedHoldout;
|
|
48
|
-
export interface HeldoutSignificance {
|
|
49
|
-
paired: PairedHoldout;
|
|
50
|
-
bootstrap: PairedBootstrapResult;
|
|
51
|
-
/** n paired observations. */
|
|
52
|
-
n: number;
|
|
53
|
-
/** True iff n >= minProductiveRuns AND the CI lower bound clears the threshold. */
|
|
54
|
-
significant: boolean;
|
|
55
|
-
/** Set when n < minProductiveRuns — too little evidence to claim significance. */
|
|
56
|
-
fewRuns: boolean;
|
|
57
|
-
}
|
|
58
|
-
export interface HeldoutSignificanceOptions {
|
|
59
|
-
deltaThreshold?: number;
|
|
60
|
-
minProductiveRuns?: number;
|
|
61
|
-
confidence?: number;
|
|
62
|
-
resamples?: number;
|
|
63
|
-
/** Fixed by default for a deterministic, reproducible gate verdict. */
|
|
64
|
-
seed?: number;
|
|
65
|
-
statistic?: 'mean' | 'median';
|
|
66
|
-
}
|
|
67
|
-
/** Significance of the held-out composite lift: ship only when the paired
|
|
68
|
-
* bootstrap CI lower bound on (candidate − baseline) exceeds `deltaThreshold`
|
|
69
|
-
* (default 0 ⇒ "confidently positive"). Below `minProductiveRuns` paired
|
|
70
|
-
* observations there is not enough evidence to claim significance → not
|
|
71
|
-
* significant (`fewRuns`). Interpret `deltaThreshold` in the judge's native
|
|
72
|
-
* composite scale. */
|
|
73
|
-
export declare function heldoutSignificance(paired: PairedHoldout, opts?: HeldoutSignificanceOptions): HeldoutSignificance;
|
|
74
|
-
export interface DimensionRegression {
|
|
75
|
-
dimension: string;
|
|
76
|
-
bootstrap: PairedBootstrapResult;
|
|
77
|
-
/** True iff the CI lower bound on (candidate − baseline) is below −tolerance:
|
|
78
|
-
* the candidate may have regressed this dimension by more than tolerance. */
|
|
79
|
-
regressed: boolean;
|
|
80
|
-
tolerance: number;
|
|
81
|
-
n: number;
|
|
82
|
-
}
|
|
83
|
-
/** Detect the native scale of a set of scores: 0-100 when any magnitude clears
|
|
84
|
-
* 1.5, else [0,1]. Used to auto-scale the regression tolerance so a default
|
|
85
|
-
* expressed for [0,1] is not silently a no-op on a 0-100 dimension. */
|
|
86
|
-
export declare function detectScale(values: number[]): 1 | 100;
|
|
87
|
-
/** Per-critical-dimension regression guard. For each dimension, pair the
|
|
88
|
-
* candidate vs baseline values by full cellId and bootstrap the paired delta;
|
|
89
|
-
* a dimension is "regressed" when the CI lower bound < −tolerance (conservative
|
|
90
|
-
* — blocks if the credible worst case exceeds tolerance, which is the right
|
|
91
|
-
* posture for safety dimensions like `hallucination_free`). When `tolerance`
|
|
92
|
-
* is omitted it auto-scales: 0.05 on [0,1], 5 on 0-100. */
|
|
93
|
-
export declare function dimensionRegressions(candidate: Map<string, Record<string, JudgeScore>>, baseline: Map<string, Record<string, JudgeScore>>, scenarioIds: Set<string>, criticalDimensions: string[], opts?: {
|
|
94
|
-
tolerance?: number;
|
|
95
|
-
confidence?: number;
|
|
96
|
-
resamples?: number;
|
|
97
|
-
seed?: number;
|
|
98
|
-
}): DimensionRegression[];
|
|
99
|
-
//# sourceMappingURL=statistical-heldout.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"statistical-heldout.d.ts","sourceRoot":"","sources":["../../../src/campaign/gates/statistical-heldout.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AAEH,OAAO,EAAE,KAAK,qBAAqB,EAAmB,MAAM,kBAAkB,CAAA;AAC9E,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,UAAU,CAAA;AAE1C,MAAM,WAAW,aAAa;IAC5B,yEAAyE;IACzE,MAAM,EAAE,MAAM,EAAE,CAAA;IAChB,wCAAwC;IACxC,KAAK,EAAE,MAAM,EAAE,CAAA;IACf,+DAA+D;IAC/D,OAAO,EAAE,MAAM,EAAE,CAAA;CAClB;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,WAAW,CACzB,SAAS,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC,EAClD,QAAQ,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC,EACjD,WAAW,EAAE,GAAG,CAAC,MAAM,CAAC,EACxB,MAAM,EAAE,CAAC,CAAC,EAAE,UAAU,KAAK,MAAM,GAAG,SAAS,GAC5C,aAAa,CA2Cf;AAED,MAAM,WAAW,mBAAmB;IAClC,MAAM,EAAE,aAAa,CAAA;IACrB,SAAS,EAAE,qBAAqB,CAAA;IAChC,6BAA6B;IAC7B,CAAC,EAAE,MAAM,CAAA;IACT,mFAAmF;IACnF,WAAW,EAAE,OAAO,CAAA;IACpB,kFAAkF;IAClF,OAAO,EAAE,OAAO,CAAA;CACjB;AAED,MAAM,WAAW,0BAA0B;IACzC,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,iBAAiB,CAAC,EAAE,MAAM,CAAA;IAC1B,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,uEAAuE;IACvE,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,SAAS,CAAC,EAAE,MAAM,GAAG,QAAQ,CAAA;CAC9B;AAED;;;;;uBAKuB;AACvB,wBAAgB,mBAAmB,CACjC,MAAM,EAAE,aAAa,EACrB,IAAI,GAAE,0BAA+B,GACpC,mBAAmB,CAarB;AAED,MAAM,WAAW,mBAAmB;IAClC,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,EAAE,qBAAqB,CAAA;IAChC;kFAC8E;IAC9E,SAAS,EAAE,OAAO,CAAA;IAClB,SAAS,EAAE,MAAM,CAAA;IACjB,CAAC,EAAE,MAAM,CAAA;CACV;AAED;;wEAEwE;AACxE,wBAAgB,WAAW,CAAC,MAAM,EAAE,MAAM,EAAE,GAAG,CAAC,GAAG,GAAG,CAErD;AAED;;;;;4DAK4D;AAC5D,wBAAgB,oBAAoB,CAClC,SAAS,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC,EAClD,QAAQ,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC,EACjD,WAAW,EAAE,GAAG,CAAC,MAAM,CAAC,EACxB,kBAAkB,EAAE,MAAM,EAAE,EAC5B,IAAI,GAAE;IAAE,SAAS,CAAC,EAAE,MAAM,CAAC;IAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,MAAM,CAAC;IAAC,IAAI,CAAC,EAAE,MAAM,CAAA;CAAO,GACxF,mBAAmB,EAAE,CAqBvB"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"statistical-heldout.test.d.ts","sourceRoot":"","sources":["../../../src/campaign/gates/statistical-heldout.test.ts"],"names":[],"mappings":""}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/campaign/index.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,EACL,KAAK,eAAe,EACpB,KAAK,eAAe,EACpB,KAAK,kCAAkC,EACvC,2BAA2B,EAC3B,KAAK,6BAA6B,EAClC,sBAAsB,GACvB,MAAM,mBAAmB,CAAA;AAE1B,OAAO,EACL,KAAK,iBAAiB,EACtB,KAAK,gBAAgB,EACrB,UAAU,GACX,MAAM,WAAW,CAAA;AAElB,OAAO,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAA;AAC7C,OAAO,EACL,KAAK,4BAA4B,EACjC,qBAAqB,GACtB,MAAM,iCAAiC,CAAA;AACxC,OAAO,EAAE,KAAK,kBAAkB,EAAE,WAAW,EAAE,MAAM,sBAAsB,CAAA;AAC3E,OAAO,EACL,KAAK,YAAY,EACjB,KAAK,WAAW,EAChB,KAAK,0BAA0B,EAC/B,mBAAmB,EACnB,KAAK,cAAc,EACnB,KAAK,eAAe,EACpB,KAAK,6BAA6B,EAClC,KAAK,kBAAkB,EACvB,KAAK,eAAe,EACpB,YAAY,EACZ,sBAAsB,GACvB,MAAM,0BAA0B,CAAA;AACjC,OAAO,EACL,KAAK,kBAAkB,EACvB,KAAK,uBAAuB,EAC5B,KAAK,kBAAkB,EACvB,KAAK,qBAAqB,EAC1B,KAAK,oBAAoB,EACzB,KAAK,2BAA2B,EAChC,gBAAgB,EAChB,oBAAoB,GACrB,MAAM,oBAAoB,CAAA;AAC3B,OAAO,EACL,KAAK,mBAAmB,EACxB,WAAW,EACX,oBAAoB,EACpB,KAAK,mBAAmB,EACxB,KAAK,0BAA0B,EAC/B,mBAAmB,EACnB,KAAK,aAAa,EAClB,WAAW,GACZ,MAAM,6BAA6B,CAAA;AACpC,OAAO,EACL,sBAAsB,EACtB,KAAK,6BAA6B,EAClC,yBAAyB,GAC1B,MAAM,4BAA4B,CAAA;AAEnC,OAAO,EACL,KAAK,uBAAuB,EAC5B,gBAAgB,EAChB,KAAK,eAAe,EACpB,mBAAmB,EACnB,eAAe,EACf,mBAAmB,EACnB,KAAK,oBAAoB,EACzB,KAAK,kBAAkB,EACvB,KAAK,aAAa,EAClB,KAAK,gBAAgB,EACrB,KAAK,aAAa,EAClB,aAAa,GACd,MAAM,6BAA6B,CAAA;AACpC,OAAO,EACL,oBAAoB,EACpB,KAAK,eAAe,EACpB,KAAK,cAAc,EACnB,KAAK,YAAY,EACjB,wBAAwB,EACxB,KAAK,uBAAuB,EAC5B,KAAK,aAAa,EAClB,KAAK,iBAAiB,EACtB,iBAAiB,EACjB,cAAc,EACd,KAAK,SAAS,EACd,KAAK,gBAAgB,EACrB,mBAAmB,GACpB,MAAM,oBAAoB,CAAA;AAC3B,OAAO,EAAE,KAAK,cAAc,EAAE,OAAO,EAAE,MAAM,oBAAoB,CAAA;AACjE,OAAO,EACL,iBAAiB,EACjB,KAAK,yBAAyB,EAC9B,KAAK,wBAAwB,EAC7B,kBAAkB,GACnB,MAAM,gCAAgC,CAAA;AACvC,OAAO,EACL,KAAK,sBAAsB,EAC3B,KAAK,qBAAqB,EAC1B,eAAe,EACf,WAAW,GACZ,MAAM,4BAA4B,CAAA;AACnC,OAAO,EACL,KAAK,iBAAiB,EACtB,kBAAkB,EAClB,KAAK,cAAc,EACnB,KAAK,uBAAuB,EAC5B,KAAK,sBAAsB,EAC3B,gBAAgB,EAChB,KAAK,cAAc,GACpB,MAAM,8BAA8B,CAAA;AACrC,OAAO,EACL,KAAK,YAAY,EACjB,KAAK,kBAAkB,EACvB,KAAK,iBAAiB,EACtB,WAAW,EACX,KAAK,mBAAmB,GACzB,MAAM,yBAAyB,CAAA;AAChC,OAAO,EAAE,KAAK,kBAAkB,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAA;AACtE,OAAO,EAAE,KAAK,2BAA2B,EAAE,oBAAoB,EAAE,MAAM,0BAA0B,CAAA;AACjG,OAAO,EACL,6BAA6B,EAC7B,KAAK,sBAAsB,EAC3B,KAAK,kBAAkB,EACvB,KAAK,qBAAqB,EAC1B,KAAK,mBAAmB,EACxB,KAAK,eAAe,EACpB,KAAK,eAAe,EACpB,KAAK,gBAAgB,EACrB,KAAK,iBAAiB,EACtB,YAAY,EACZ,KAAK,aAAa,EAClB,KAAK,SAAS,EACd,KAAK,kBAAkB,EACvB,KAAK,eAAe,EACpB,KAAK,6BAA6B,EAClC,sBAAsB,GACvB,MAAM,kBAAkB,CAAA;AACzB,OAAO,EACL,kBAAkB,EAClB,iBAAiB,EACjB,KAAK,uBAAuB,EAC5B,KAAK,mBAAmB,EACxB,YAAY,GACb,MAAM,kBAAkB,CAAA;AAEzB,OAAO,EAAE,KAAK,mBAAmB,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAA;AACzE,OAAO,EAAE,KAAK,6BAA6B,EAAE,sBAAsB,EAAE,MAAM,oBAAoB,CAAA;AAC/F,OAAO,EACL,KAAK,kBAAkB,EACvB,uBAAuB,EACvB,KAAK,YAAY,EACjB,KAAK,gBAAgB,EACrB,KAAK,gBAAgB,EACrB,KAAK,uBAAuB,EAC5B,oBAAoB,EACpB,gBAAgB,GACjB,MAAM,uBAAuB,CAAA;AAC9B,OAAO,EAAE,KAAK,2BAA2B,EAAE,oBAAoB,EAAE,MAAM,2BAA2B,CAAA;AAElG,OAAO,EACL,KAAK,uBAAuB,EAC5B,yBAAyB,EACzB,KAAK,sBAAsB,EAC3B,KAAK,wBAAwB,EAC7B,kBAAkB,EAClB,KAAK,qBAAqB,EAC1B,KAAK,uBAAuB,EAC5B,KAAK,oBAAoB,EACzB,mBAAmB,EACnB,oBAAoB,EACpB,mBAAmB,EACnB,kBAAkB,GACnB,MAAM,cAAc,CAAA;AACrB,OAAO,EAAE,KAAK,kBAAkB,EAAE,WAAW,EAAE,MAAM,gBAAgB,CAAA;AACrE,OAAO,EAAE,KAAK,iBAAiB,EAAE,iBAAiB,EAAE,qBAAqB,EAAE,MAAM,eAAe,CAAA;AAChG,OAAO,EACL,KAAK,qBAAqB,EAC1B,eAAe,EACf,cAAc,EACd,KAAK,UAAU,EACf,KAAK,YAAY,EACjB,KAAK,mBAAmB,GACzB,MAAM,eAAe,CAAA;AACtB,OAAO,EAAE,KAAK,eAAe,EAAE,iBAAiB,EAAE,uBAAuB,EAAE,MAAM,WAAW,CAAA;AAC5F,YAAY,EACV,kBAAkB,EAClB,sBAAsB,EACtB,kBAAkB,EAClB,iBAAiB,EACjB,cAAc,EACd,kBAAkB,EAClB,mBAAmB,EACnB,WAAW,EACX,eAAe,EACf,UAAU,EACV,IAAI,EACJ,WAAW,EACX,YAAY,EACZ,UAAU,EACV,mBAAmB,EACnB,gBAAgB,EAChB,cAAc,EACd,WAAW,EACX,cAAc,EACd,UAAU,EACV,qBAAqB,EACrB,yBAAyB,EACzB,qBAAqB,EACrB,oBAAoB,EACpB,oBAAoB,EACpB,UAAU,EACV,cAAc,EACd,OAAO,EACP,oBAAoB,EACpB,eAAe,EACf,YAAY,EACZ,cAAc,EACd,iBAAiB,EACjB,eAAe,EACf,QAAQ,EACR,iBAAiB,EACjB,aAAa,EACb,eAAe,EACf,SAAS,GACV,MAAM,SAAS,CAAA;AAChB,OAAO,EAAE,mBAAmB,EAAE,cAAc,EAAE,MAAM,SAAS,CAAA;AAE7D,OAAO,EACL,KAAK,yBAAyB,EAC9B,kBAAkB,EAClB,mBAAmB,EACnB,KAAK,QAAQ,EACb,KAAK,eAAe,EACpB,oBAAoB,GACrB,MAAM,YAAY,CAAA"}
|
|
@@ -1,59 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Filesystem `LabeledScenarioStore` adapter. The default capture sink for
|
|
3
|
-
* traces + eval artifacts. Production deployments typically swap for a
|
|
4
|
-
* Turso/SQLite adapter (same interface).
|
|
5
|
-
*
|
|
6
|
-
* Records land as one JSONL file per source under `<root>/<source>.jsonl`.
|
|
7
|
-
* Each line is a `LabeledScenarioRecord`. Append-only — no in-place edits.
|
|
8
|
-
*
|
|
9
|
-
* Safety properties enforced at write-time:
|
|
10
|
-
*
|
|
11
|
-
* - **Provenance required**: writes without `source`, `sourceVersionHash`,
|
|
12
|
-
* `capturedAt`, `redactionStatus` are rejected. Closes the alignment
|
|
13
|
-
* reviewer's data-poisoning gap.
|
|
14
|
-
* - **Per-source rate limits**: optional `rateLimitBucket` + `maxWritesPerMinute`
|
|
15
|
-
* stops a single tenant/source from flooding the store.
|
|
16
|
-
*
|
|
17
|
-
* Safety properties enforced at sample-time:
|
|
18
|
-
*
|
|
19
|
-
* - **Required split + capturedBefore**: substrate refuses to sample without
|
|
20
|
-
* an explicit `split` ('train' | 'test') AND a temporal cutoff. Eliminates
|
|
21
|
-
* accidental train/test contamination.
|
|
22
|
-
* - **Default training-source filter**: when the store is sampled with
|
|
23
|
-
* `split: 'train'`, production-trace records are EXCLUDED unless the
|
|
24
|
-
* caller passes `filter.source: 'production-trace'` explicitly. Closes
|
|
25
|
-
* the contamination-by-default gap flagged by the senior eval engineer.
|
|
26
|
-
*/
|
|
27
|
-
import type { LabeledScenarioRecord, LabeledScenarioSampleArgs, LabeledScenarioStore, LabeledScenarioWrite, LabelTrust } from '../types';
|
|
28
|
-
export interface FsLabeledScenarioStoreOptions {
|
|
29
|
-
/** Root directory for JSONL files. Created if missing. */
|
|
30
|
-
root: string;
|
|
31
|
-
/** Per-source rate limit. When set, writes exceeding the cap are rejected
|
|
32
|
-
* with a typed error. Default: no limit. */
|
|
33
|
-
maxWritesPerMinutePerBucket?: number;
|
|
34
|
-
/** Test seam — override `Date.now()` for deterministic tests. */
|
|
35
|
-
now?: () => number;
|
|
36
|
-
}
|
|
37
|
-
export declare class LabeledScenarioStoreError extends Error {
|
|
38
|
-
readonly code: string;
|
|
39
|
-
constructor(code: string, message: string);
|
|
40
|
-
}
|
|
41
|
-
export declare class FsLabeledScenarioStore implements LabeledScenarioStore {
|
|
42
|
-
private readonly options;
|
|
43
|
-
private readonly now;
|
|
44
|
-
private readonly rateLimits;
|
|
45
|
-
constructor(options: FsLabeledScenarioStoreOptions);
|
|
46
|
-
observe(write: LabeledScenarioWrite): Promise<void>;
|
|
47
|
-
sample(args: LabeledScenarioSampleArgs): Promise<LabeledScenarioRecord[]>;
|
|
48
|
-
size(): Promise<{
|
|
49
|
-
train: number;
|
|
50
|
-
test: number;
|
|
51
|
-
bySource: Record<string, number>;
|
|
52
|
-
byTrust: Record<LabelTrust, number>;
|
|
53
|
-
}>;
|
|
54
|
-
private assertProvenance;
|
|
55
|
-
private assertRateLimit;
|
|
56
|
-
private toRecord;
|
|
57
|
-
private pathForSource;
|
|
58
|
-
}
|
|
59
|
-
//# sourceMappingURL=fs-adapter.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"fs-adapter.d.ts","sourceRoot":"","sources":["../../../src/campaign/labeled-store/fs-adapter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AAKH,OAAO,KAAK,EACV,qBAAqB,EACrB,yBAAyB,EAEzB,oBAAoB,EACpB,oBAAoB,EACpB,UAAU,EACX,MAAM,UAAU,CAAA;AAGjB,MAAM,WAAW,6BAA6B;IAC5C,0DAA0D;IAC1D,IAAI,EAAE,MAAM,CAAA;IACZ;iDAC6C;IAC7C,2BAA2B,CAAC,EAAE,MAAM,CAAA;IACpC,iEAAiE;IACjE,GAAG,CAAC,EAAE,MAAM,MAAM,CAAA;CACnB;AAED,qBAAa,yBAA0B,SAAQ,KAAK;aAEhC,IAAI,EAAE,MAAM;gBAAZ,IAAI,EAAE,MAAM,EAC5B,OAAO,EAAE,MAAM;CAKlB;AAQD,qBAAa,sBAAuB,YAAW,oBAAoB;IAIrD,OAAO,CAAC,QAAQ,CAAC,OAAO;IAHpC,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAc;IAClC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAoC;gBAElC,OAAO,EAAE,6BAA6B;IAK7D,OAAO,CAAC,KAAK,EAAE,oBAAoB,GAAG,OAAO,CAAC,IAAI,CAAC;IAYnD,MAAM,CAAC,IAAI,EAAE,yBAAyB,GAAG,OAAO,CAAC,qBAAqB,EAAE,CAAC;IA8CzE,IAAI,IAAI,OAAO,CAAC;QACpB,KAAK,EAAE,MAAM,CAAA;QACb,IAAI,EAAE,MAAM,CAAA;QACZ,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;QAChC,OAAO,EAAE,MAAM,CAAC,UAAU,EAAE,MAAM,CAAC,CAAA;KACpC,CAAC;IAiCF,OAAO,CAAC,gBAAgB;IAiCxB,OAAO,CAAC,eAAe;IAmBvB,OAAO,CAAC,QAAQ;IAkBhB,OAAO,CAAC,aAAa;CAGtB"}
|
|
@@ -1,146 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* `compareProposers` — a head-to-head lift benchmark across surface proposers
|
|
3
|
-
* on ONE corpus. This is the forcing function: optimizer quality (GEPA
|
|
4
|
-
* reflection vs GEPA+Pareto vs SkillOpt) becomes a NUMBER with a confidence
|
|
5
|
-
* interval, so a proposer regression — or shipping a simplified proposer and
|
|
6
|
-
* calling it the real one — turns a build red instead of going
|
|
7
|
-
* measurement-invisible.
|
|
8
|
-
*
|
|
9
|
-
* Every entrant is scored the SAME way: each proposer returns the surface it
|
|
10
|
-
* promoted, then the benchmark scores the baseline + every winner on the
|
|
11
|
-
* SAME held-out scenarios with the SAME judges. Apples-to-apples by
|
|
12
|
-
* construction — the comparison never depends on how a proposer measured itself.
|
|
13
|
-
* The per-scenario held-out composites feed a paired bootstrap (`statistics.ts`)
|
|
14
|
-
* for each proposer's lift CI and for the pairwise "which proposer wins" CI.
|
|
15
|
-
*/
|
|
16
|
-
import type { LlmClientOptions } from '../../llm-client';
|
|
17
|
-
import { type FapoProposerOptions, type ParameterCandidate } from '../proposers/fapo';
|
|
18
|
-
import { type RunCampaignOptions } from '../run-campaign';
|
|
19
|
-
import type { DispatchContext, JudgeConfig, MutableSurface, Scenario, SurfaceProposer } from '../types';
|
|
20
|
-
import { type RunImprovementLoopOptions } from './run-improvement-loop';
|
|
21
|
-
/** What an optimizer produced: the surface it promoted + what it cost to get
|
|
22
|
-
* there. The comparison does the held-out scoring itself, so an entry only
|
|
23
|
-
* needs to run its loop and hand back the winner. */
|
|
24
|
-
export interface ProposerEntry {
|
|
25
|
-
name: string;
|
|
26
|
-
optimize: () => Promise<{
|
|
27
|
-
winnerSurface: MutableSurface;
|
|
28
|
-
costUsd: number;
|
|
29
|
-
durationMs?: number;
|
|
30
|
-
}>;
|
|
31
|
-
}
|
|
32
|
-
export interface ProposerScore {
|
|
33
|
-
name: string;
|
|
34
|
-
/** Mean held-out composite of the baseline (identical across proposers). */
|
|
35
|
-
baselineComposite: number;
|
|
36
|
-
/** Mean held-out composite of this proposer's promoted surface. */
|
|
37
|
-
winnerComposite: number;
|
|
38
|
-
/** Mean per-scenario held-out lift (winner − baseline). */
|
|
39
|
-
lift: number;
|
|
40
|
-
/** Paired-bootstrap CI of the per-scenario lift. `low > 0` ⇒ a real gain. */
|
|
41
|
-
liftCi: {
|
|
42
|
-
low: number;
|
|
43
|
-
high: number;
|
|
44
|
-
};
|
|
45
|
-
costUsd: number;
|
|
46
|
-
durationMs?: number;
|
|
47
|
-
winnerSurface: MutableSurface;
|
|
48
|
-
/** 1-based, by descending lift. */
|
|
49
|
-
rank: number;
|
|
50
|
-
}
|
|
51
|
-
export interface ProposerPairwise {
|
|
52
|
-
/** Higher-ranked proposer. */
|
|
53
|
-
a: string;
|
|
54
|
-
b: string;
|
|
55
|
-
/** Mean per-scenario held-out delta (a − b). */
|
|
56
|
-
deltaMean: number;
|
|
57
|
-
low: number;
|
|
58
|
-
high: number;
|
|
59
|
-
/** `a` if the CI clears 0, `b` if it is entirely negative, else `'tie'`. */
|
|
60
|
-
favored: string;
|
|
61
|
-
}
|
|
62
|
-
export interface ProposerComparison {
|
|
63
|
-
/** Sorted by descending lift; `rank` set accordingly. */
|
|
64
|
-
scores: ProposerScore[];
|
|
65
|
-
best: ProposerScore;
|
|
66
|
-
/** Best vs each other proposer, paired-bootstrap on the held-out winners. */
|
|
67
|
-
pairwise: ProposerPairwise[];
|
|
68
|
-
holdoutScenarioIds: string[];
|
|
69
|
-
}
|
|
70
|
-
export interface CompareProposersOptions<TScenario extends Scenario, TArtifact> extends Omit<RunCampaignOptions<TScenario, TArtifact>, 'dispatch' | 'scenarios'> {
|
|
71
|
-
proposers: ProposerEntry[];
|
|
72
|
-
baselineSurface: MutableSurface;
|
|
73
|
-
/** The held-out scenarios every winner is scored on. */
|
|
74
|
-
holdoutScenarios: TScenario[];
|
|
75
|
-
/** Scores a surface on a scenario — the same dispatcher the proposers used. */
|
|
76
|
-
dispatchWithSurface: (surface: MutableSurface, scenario: TScenario, ctx: DispatchContext) => Promise<TArtifact>;
|
|
77
|
-
/** Bootstrap resamples for the lift CIs. Default 2000. */
|
|
78
|
-
resamples?: number;
|
|
79
|
-
/** CI confidence. Default 0.95. */
|
|
80
|
-
confidence?: number;
|
|
81
|
-
}
|
|
82
|
-
export declare function compareProposers<TScenario extends Scenario, TArtifact>(opts: CompareProposersOptions<TScenario, TArtifact>): Promise<ProposerComparison>;
|
|
83
|
-
/** Shared corpus + transport for the three built-in optimizer entries. */
|
|
84
|
-
export interface OptimizerEntryConfig<TScenario extends Scenario, TArtifact> {
|
|
85
|
-
baselineSurface: string;
|
|
86
|
-
/** Training scenarios the proposers reflect on. */
|
|
87
|
-
trainScenarios: TScenario[];
|
|
88
|
-
/** Held-out scenarios (the gate axis + the benchmark scoring axis). */
|
|
89
|
-
holdoutScenarios: TScenario[];
|
|
90
|
-
dispatchWithSurface: (surface: MutableSurface, scenario: TScenario, ctx: DispatchContext) => Promise<TArtifact>;
|
|
91
|
-
judges: JudgeConfig<TArtifact, TScenario>[];
|
|
92
|
-
llm: LlmClientOptions;
|
|
93
|
-
model: string;
|
|
94
|
-
target: string;
|
|
95
|
-
runDir: string;
|
|
96
|
-
seed?: number;
|
|
97
|
-
/** GEPA population per generation. Default 2. */
|
|
98
|
-
populationSize?: number;
|
|
99
|
-
/** GEPA generations. Default 3. */
|
|
100
|
-
maxGenerations?: number;
|
|
101
|
-
/** SkillOpt epochs. Default 6. */
|
|
102
|
-
maxEpochs?: number;
|
|
103
|
-
mutationPrimitives?: string[];
|
|
104
|
-
/** Static findings seed forwarded to each GEPA proposer's `propose()` as
|
|
105
|
-
* `ctx.findings`. Forwarded by `gepaReflectionEntry` / `gepaParetoEntry`;
|
|
106
|
-
* `skillOptEntry` runs without findings (see its doc). */
|
|
107
|
-
findings?: unknown[];
|
|
108
|
-
/** Per-generation findings producer: after each generation scores, this
|
|
109
|
-
* re-diagnoses and REPLACES `ctx.findings` for the
|
|
110
|
-
* next generation's `propose()`. Reuses the `runOptimization` field type so
|
|
111
|
-
* it cannot drift. GEPA entries only. */
|
|
112
|
-
analyzeGeneration?: RunImprovementLoopOptions<TScenario, TArtifact>['analyzeGeneration'];
|
|
113
|
-
/** Optional analysis report forwarded to `propose()` as `ctx.report`. */
|
|
114
|
-
report?: unknown;
|
|
115
|
-
}
|
|
116
|
-
/** GEPA, reflection-only (single-parent, no Pareto combine). */
|
|
117
|
-
export declare function gepaReflectionEntry<TScenario extends Scenario, TArtifact>(config: OptimizerEntryConfig<TScenario, TArtifact>, name?: string): ProposerEntry;
|
|
118
|
-
/** GEPA with the Pareto frontier + combine-complementary-lessons. */
|
|
119
|
-
export declare function gepaParetoEntry<TScenario extends Scenario, TArtifact>(config: OptimizerEntryConfig<TScenario, TArtifact>, name?: string): ProposerEntry;
|
|
120
|
-
/** SkillOpt patch-mode hill-climb. Runs findings-BLIND: `runSkillOpt` owns its
|
|
121
|
-
* own epoch acceptance/budget loop and does not thread `analyzeGeneration`, so
|
|
122
|
-
* `config.findings` is intentionally NOT forwarded here. In a findings-fed
|
|
123
|
-
* comparison this entry is the blind control — do not read its result as
|
|
124
|
-
* findings-fed. (Threading findings into the SkillOpt epoch loop is a separate
|
|
125
|
-
* refactor, deferred not faked.) */
|
|
126
|
-
export declare function skillOptEntry<TScenario extends Scenario, TArtifact>(config: OptimizerEntryConfig<TScenario, TArtifact>, name?: string): ProposerEntry;
|
|
127
|
-
/** FAPO reviewed-escalation policy. This is an orchestration layer over
|
|
128
|
-
* level-specific proposers, not a new mutation operator:
|
|
129
|
-
* prompt -> parameter -> structural, with scope + reviewer + plateau rules in
|
|
130
|
-
* `fapoProposer`. The prompt proposer defaults to GEPA+Pareto because that is
|
|
131
|
-
* the package's strongest prompt-tier proposer; parameter/structural proposers
|
|
132
|
-
* are opt-in so we do not fake code-generation inside agent-eval. */
|
|
133
|
-
export interface FapoEntryConfig<TScenario extends Scenario, TArtifact> extends OptimizerEntryConfig<TScenario, TArtifact> {
|
|
134
|
-
/** Override the prompt-level proposer. Default: `gepaProposer({ combineParents: true })`. */
|
|
135
|
-
promptProposer?: SurfaceProposer;
|
|
136
|
-
/** Parameter/config-level proposer. If omitted, `parameterCandidates` builds one. */
|
|
137
|
-
parameterProposer?: SurfaceProposer;
|
|
138
|
-
/** Structural/code-level proposer, typically supplied by agent-runtime. */
|
|
139
|
-
structuralProposer?: SurfaceProposer;
|
|
140
|
-
/** Convenience: build a `parameterSweepProposer` from these candidates. */
|
|
141
|
-
parameterCandidates?: readonly ParameterCandidate[];
|
|
142
|
-
/** FAPO policy knobs: scope, reviewer, plateau thresholds. */
|
|
143
|
-
fapo?: Omit<FapoProposerOptions, 'proposers' | 'promptProposer' | 'parameterProposer' | 'structuralProposer'>;
|
|
144
|
-
}
|
|
145
|
-
export declare function fapoEscalationEntry<TScenario extends Scenario, TArtifact>(config: FapoEntryConfig<TScenario, TArtifact>, name?: string): ProposerEntry;
|
|
146
|
-
//# sourceMappingURL=compare-proposers.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"compare-proposers.d.ts","sourceRoot":"","sources":["../../../src/campaign/presets/compare-proposers.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAA;AAGxD,OAAO,EACL,KAAK,mBAAmB,EAExB,KAAK,kBAAkB,EAExB,MAAM,mBAAmB,CAAA;AAG1B,OAAO,EAAE,KAAK,kBAAkB,EAAe,MAAM,iBAAiB,CAAA;AAEtE,OAAO,KAAK,EAEV,eAAe,EACf,WAAW,EACX,cAAc,EACd,QAAQ,EACR,eAAe,EAChB,MAAM,UAAU,CAAA;AACjB,OAAO,EAAE,KAAK,yBAAyB,EAAsB,MAAM,wBAAwB,CAAA;AAG3F;;sDAEsD;AACtD,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,MAAM,CAAA;IACZ,QAAQ,EAAE,MAAM,OAAO,CAAC;QAAE,aAAa,EAAE,cAAc,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,UAAU,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;CACjG;AAED,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,MAAM,CAAA;IACZ,4EAA4E;IAC5E,iBAAiB,EAAE,MAAM,CAAA;IACzB,mEAAmE;IACnE,eAAe,EAAE,MAAM,CAAA;IACvB,2DAA2D;IAC3D,IAAI,EAAE,MAAM,CAAA;IACZ,6EAA6E;IAC7E,MAAM,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAA;IACrC,OAAO,EAAE,MAAM,CAAA;IACf,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,aAAa,EAAE,cAAc,CAAA;IAC7B,mCAAmC;IACnC,IAAI,EAAE,MAAM,CAAA;CACb;AAED,MAAM,WAAW,gBAAgB;IAC/B,8BAA8B;IAC9B,CAAC,EAAE,MAAM,CAAA;IACT,CAAC,EAAE,MAAM,CAAA;IACT,gDAAgD;IAChD,SAAS,EAAE,MAAM,CAAA;IACjB,GAAG,EAAE,MAAM,CAAA;IACX,IAAI,EAAE,MAAM,CAAA;IACZ,4EAA4E;IAC5E,OAAO,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,kBAAkB;IACjC,yDAAyD;IACzD,MAAM,EAAE,aAAa,EAAE,CAAA;IACvB,IAAI,EAAE,aAAa,CAAA;IACnB,6EAA6E;IAC7E,QAAQ,EAAE,gBAAgB,EAAE,CAAA;IAC5B,kBAAkB,EAAE,MAAM,EAAE,CAAA;CAC7B;AAED,MAAM,WAAW,uBAAuB,CAAC,SAAS,SAAS,QAAQ,EAAE,SAAS,CAC5E,SAAQ,IAAI,CAAC,kBAAkB,CAAC,SAAS,EAAE,SAAS,CAAC,EAAE,UAAU,GAAG,WAAW,CAAC;IAChF,SAAS,EAAE,aAAa,EAAE,CAAA;IAC1B,eAAe,EAAE,cAAc,CAAA;IAC/B,wDAAwD;IACxD,gBAAgB,EAAE,SAAS,EAAE,CAAA;IAC7B,+EAA+E;IAC/E,mBAAmB,EAAE,CACnB,OAAO,EAAE,cAAc,EACvB,QAAQ,EAAE,SAAS,EACnB,GAAG,EAAE,eAAe,KACjB,OAAO,CAAC,SAAS,CAAC,CAAA;IACvB,0DAA0D;IAC1D,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,mCAAmC;IACnC,UAAU,CAAC,EAAE,MAAM,CAAA;CACpB;AAED,wBAAsB,gBAAgB,CAAC,SAAS,SAAS,QAAQ,EAAE,SAAS,EAC1E,IAAI,EAAE,uBAAuB,CAAC,SAAS,EAAE,SAAS,CAAC,GAClD,OAAO,CAAC,kBAAkB,CAAC,CAE7B;AAqID,0EAA0E;AAC1E,MAAM,WAAW,oBAAoB,CAAC,SAAS,SAAS,QAAQ,EAAE,SAAS;IACzE,eAAe,EAAE,MAAM,CAAA;IACvB,mDAAmD;IACnD,cAAc,EAAE,SAAS,EAAE,CAAA;IAC3B,uEAAuE;IACvE,gBAAgB,EAAE,SAAS,EAAE,CAAA;IAC7B,mBAAmB,EAAE,CACnB,OAAO,EAAE,cAAc,EACvB,QAAQ,EAAE,SAAS,EACnB,GAAG,EAAE,eAAe,KACjB,OAAO,CAAC,SAAS,CAAC,CAAA;IACvB,MAAM,EAAE,WAAW,CAAC,SAAS,EAAE,SAAS,CAAC,EAAE,CAAA;IAC3C,GAAG,EAAE,gBAAgB,CAAA;IACrB,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,MAAM,CAAA;IACd,MAAM,EAAE,MAAM,CAAA;IACd,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,iDAAiD;IACjD,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,mCAAmC;IACnC,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,kCAAkC;IAClC,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAA;IAC7B;;+DAE2D;IAC3D,QAAQ,CAAC,EAAE,OAAO,EAAE,CAAA;IACpB;;;8CAG0C;IAC1C,iBAAiB,CAAC,EAAE,yBAAyB,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC,mBAAmB,CAAC,CAAA;IACxF,yEAAyE;IACzE,MAAM,CAAC,EAAE,OAAO,CAAA;CACjB;AAED,gEAAgE;AAChE,wBAAgB,mBAAmB,CAAC,SAAS,SAAS,QAAQ,EAAE,SAAS,EACvE,MAAM,EAAE,oBAAoB,CAAC,SAAS,EAAE,SAAS,CAAC,EAClD,IAAI,SAAoB,GACvB,aAAa,CAEf;AAED,qEAAqE;AACrE,wBAAgB,eAAe,CAAC,SAAS,SAAS,QAAQ,EAAE,SAAS,EACnE,MAAM,EAAE,oBAAoB,CAAC,SAAS,EAAE,SAAS,CAAC,EAClD,IAAI,SAAgB,GACnB,aAAa,CAEf;AAqDD;;;;;qCAKqC;AACrC,wBAAgB,aAAa,CAAC,SAAS,SAAS,QAAQ,EAAE,SAAS,EACjE,MAAM,EAAE,oBAAoB,CAAC,SAAS,EAAE,SAAS,CAAC,EAClD,IAAI,SAAc,GACjB,aAAa,CA4Bf;AAED;;;;;qEAKqE;AACrE,MAAM,WAAW,eAAe,CAAC,SAAS,SAAS,QAAQ,EAAE,SAAS,CACpE,SAAQ,oBAAoB,CAAC,SAAS,EAAE,SAAS,CAAC;IAClD,6FAA6F;IAC7F,cAAc,CAAC,EAAE,eAAe,CAAA;IAChC,qFAAqF;IACrF,iBAAiB,CAAC,EAAE,eAAe,CAAA;IACnC,2EAA2E;IAC3E,kBAAkB,CAAC,EAAE,eAAe,CAAA;IACpC,2EAA2E;IAC3E,mBAAmB,CAAC,EAAE,SAAS,kBAAkB,EAAE,CAAA;IACnD,8DAA8D;IAC9D,IAAI,CAAC,EAAE,IAAI,CACT,mBAAmB,EACnB,WAAW,GAAG,gBAAgB,GAAG,mBAAmB,GAAG,oBAAoB,CAC5E,CAAA;CACF;AAED,wBAAgB,mBAAmB,CAAC,SAAS,SAAS,QAAQ,EAAE,SAAS,EACvE,MAAM,EAAE,eAAe,CAAC,SAAS,EAAE,SAAS,CAAC,EAC7C,IAAI,SAAoB,GACvB,aAAa,CAwDf"}
|
|
@@ -1,120 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Product-flow playback — drive the REAL product through a user story and
|
|
3
|
-
* score the produced state per requirement (the launch "Jira tick-off").
|
|
4
|
-
*
|
|
5
|
-
* This is the substrate adapter + contract only. It plugs a `PlaybackDriver`
|
|
6
|
-
* into the existing `runProfileMatrix` dispatch seam: a driver drives the real
|
|
7
|
-
* product (a Playwright UI session or a sandbox workspace) and returns the
|
|
8
|
-
* runtime event stream; `extractProducedState` + `verifyCompletion` then score
|
|
9
|
-
* each requirement PASS/FAIL. The concrete drivers live in consumers — they
|
|
10
|
-
* depend on browser / runtime infra the substrate must not import — so
|
|
11
|
-
* agent-eval owns the seam, the `UserStory` contract, and the scoreboard.
|
|
12
|
-
*/
|
|
13
|
-
import type { AgentProfile } from '@tangle-network/agent-interface';
|
|
14
|
-
import { type CompletionRequirement, type CompletionVerdict, type CorrectnessChecker, type ProducedState } from '../../completion-verifier';
|
|
15
|
-
import { type RuntimeEventLike } from '../../produced-state';
|
|
16
|
-
import type { DispatchContext, Scenario } from '../types';
|
|
17
|
-
import type { ProfileDispatchFn } from './run-profile-matrix';
|
|
18
|
-
/** One step of a user story — what the user does. The driver interprets
|
|
19
|
-
* `payload` (a Playwright selector + action, or a sandbox chat turn). */
|
|
20
|
-
export interface PlaybackStep {
|
|
21
|
-
/** Human-readable action, captured verbatim in the UX narrative. */
|
|
22
|
-
action: string;
|
|
23
|
-
/** Driver-specific payload (e.g. `{ selector, fill }` or `{ turn }`). */
|
|
24
|
-
payload?: Record<string, unknown>;
|
|
25
|
-
}
|
|
26
|
-
/**
|
|
27
|
-
* A user story = a runnable product journey plus the requirements that define
|
|
28
|
-
* "this story works". Each requirement is one Jira ticket line. Extends
|
|
29
|
-
* `Scenario` so a catalog drops straight into `runProfileMatrix({ scenarios })`.
|
|
30
|
-
*/
|
|
31
|
-
export interface UserStory extends Scenario {
|
|
32
|
-
/** Human-readable story title (the ticket headline). */
|
|
33
|
-
title: string;
|
|
34
|
-
/** Ordered steps the driver executes. */
|
|
35
|
-
steps: PlaybackStep[];
|
|
36
|
-
/** What must hold in the produced state for the story to pass. */
|
|
37
|
-
requirements: CompletionRequirement[];
|
|
38
|
-
}
|
|
39
|
-
/** Dispatch context plus the profile under test (which cheap model, etc.). */
|
|
40
|
-
export interface PlaybackContext extends DispatchContext {
|
|
41
|
-
profile: AgentProfile;
|
|
42
|
-
}
|
|
43
|
-
/**
|
|
44
|
-
* Drives the real product through a story and returns the runtime event stream
|
|
45
|
-
* `extractProducedState` consumes. Implemented by CONSUMERS —
|
|
46
|
-
* `SandboxPlaybackDriver` (real API / sandbox workspace) and
|
|
47
|
-
* `PlaywrightPlaybackDriver` (real UI) — because they depend on runtime /
|
|
48
|
-
* browser infra the substrate must not import. The driver MUST report LLM
|
|
49
|
-
* usage via `ctx.cost.observeTokens` so the backend-integrity guard sees real
|
|
50
|
-
* tokens (a run that never reports tokens reads as a stub).
|
|
51
|
-
*/
|
|
52
|
-
export interface PlaybackDriver<TStory extends UserStory = UserStory> {
|
|
53
|
-
run(story: TStory, ctx: PlaybackContext): Promise<readonly RuntimeEventLike[]>;
|
|
54
|
-
}
|
|
55
|
-
/**
|
|
56
|
-
* Adapt a `PlaybackDriver` into a `runProfileMatrix` dispatch. The artifact the
|
|
57
|
-
* matrix scores is the `ProducedState` extracted from the driver's event
|
|
58
|
-
* stream — grade it with `scoreUserStory` (or a judge wrapping it).
|
|
59
|
-
*/
|
|
60
|
-
export declare function makePlaybackDispatch<TStory extends UserStory>(driver: PlaybackDriver<TStory>): ProfileDispatchFn<TStory, ProducedState>;
|
|
61
|
-
/** A scored user story — the completion verdict plus its human title. */
|
|
62
|
-
export interface UserStoryVerdict extends CompletionVerdict {
|
|
63
|
-
title: string;
|
|
64
|
-
}
|
|
65
|
-
/**
|
|
66
|
-
* Score one story's produced state against its requirements. Thin wrapper over
|
|
67
|
-
* `verifyCompletion` that builds the gold from the story and returns a
|
|
68
|
-
* per-requirement PASS/FAIL verdict. `checkCorrectness` is injected — a
|
|
69
|
-
* deterministic stub in tests, `createLlmCorrectnessChecker` in production.
|
|
70
|
-
*/
|
|
71
|
-
export declare function scoreUserStory(story: UserStory, state: ProducedState, checkCorrectness: CorrectnessChecker): Promise<UserStoryVerdict>;
|
|
72
|
-
/** One row of the launch scoreboard — story × requirement → PASS/FAIL. */
|
|
73
|
-
export interface ScoreboardRow {
|
|
74
|
-
storyId: string;
|
|
75
|
-
storyTitle: string;
|
|
76
|
-
reqId: string;
|
|
77
|
-
reqTitle: string;
|
|
78
|
-
status: 'PASS' | 'FAIL';
|
|
79
|
-
evidence: string[];
|
|
80
|
-
}
|
|
81
|
-
/**
|
|
82
|
-
* Flatten story verdicts into the per-requirement scoreboard — the literal
|
|
83
|
-
* Jira tick-off: one row per (story, requirement) with PASS/FAIL and the
|
|
84
|
-
* evidence behind the verdict.
|
|
85
|
-
*/
|
|
86
|
-
export declare function userStoryScoreboard(verdicts: readonly UserStoryVerdict[]): ScoreboardRow[];
|
|
87
|
-
/** Launch-readiness headline counts rolled up from the per-requirement rows. */
|
|
88
|
-
export interface ScoreboardSummary {
|
|
89
|
-
/** Distinct user stories on the board. */
|
|
90
|
-
stories: number;
|
|
91
|
-
/** Stories whose every requirement passed. */
|
|
92
|
-
storiesFullyComplete: number;
|
|
93
|
-
/** Total (story, requirement) rows. */
|
|
94
|
-
requirements: number;
|
|
95
|
-
/** Rows with status PASS. */
|
|
96
|
-
passed: number;
|
|
97
|
-
/** Rows with status FAIL. */
|
|
98
|
-
failed: number;
|
|
99
|
-
/** passed / requirements; 0 when there are no rows. */
|
|
100
|
-
passRate: number;
|
|
101
|
-
}
|
|
102
|
-
/** Roll the per-requirement rows up into the launch headline counts. */
|
|
103
|
-
export declare function scoreboardSummary(rows: readonly ScoreboardRow[]): ScoreboardSummary;
|
|
104
|
-
export interface ScoreboardRenderOptions {
|
|
105
|
-
/** Document H1. Defaults to a generic playback title. */
|
|
106
|
-
title?: string;
|
|
107
|
-
/** Key/value run metadata rendered under the headline (runId, backend, model, date). */
|
|
108
|
-
meta?: Record<string, string>;
|
|
109
|
-
/** Max chars of joined evidence shown per row. Default 160. */
|
|
110
|
-
maxEvidenceChars?: number;
|
|
111
|
-
}
|
|
112
|
-
/**
|
|
113
|
-
* Render the scoreboard as a launch-readiness Markdown document — the literal
|
|
114
|
-
* "tick off every user story" artifact: a headline roll-up, the open tickets
|
|
115
|
-
* (FAIL rows) up top as the launch blockers, then a per-story table of
|
|
116
|
-
* requirement → PASS/FAIL with the evidence behind each verdict. Pure: same
|
|
117
|
-
* rows in, same bytes out (no clock/random), so it is safe to snapshot.
|
|
118
|
-
*/
|
|
119
|
-
export declare function renderScoreboardMarkdown(rows: readonly ScoreboardRow[], opts?: ScoreboardRenderOptions): string;
|
|
120
|
-
//# sourceMappingURL=playback.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"playback.d.ts","sourceRoot":"","sources":["../../../src/campaign/presets/playback.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,iCAAiC,CAAA;AACnE,OAAO,EACL,KAAK,qBAAqB,EAC1B,KAAK,iBAAiB,EACtB,KAAK,kBAAkB,EACvB,KAAK,aAAa,EAEnB,MAAM,2BAA2B,CAAA;AAClC,OAAO,EAAwB,KAAK,gBAAgB,EAAE,MAAM,sBAAsB,CAAA;AAClF,OAAO,KAAK,EAAE,eAAe,EAAE,QAAQ,EAAE,MAAM,UAAU,CAAA;AACzD,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,sBAAsB,CAAA;AAE7D;0EAC0E;AAC1E,MAAM,WAAW,YAAY;IAC3B,oEAAoE;IACpE,MAAM,EAAE,MAAM,CAAA;IACd,yEAAyE;IACzE,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAClC;AAED;;;;GAIG;AACH,MAAM,WAAW,SAAU,SAAQ,QAAQ;IACzC,wDAAwD;IACxD,KAAK,EAAE,MAAM,CAAA;IACb,yCAAyC;IACzC,KAAK,EAAE,YAAY,EAAE,CAAA;IACrB,kEAAkE;IAClE,YAAY,EAAE,qBAAqB,EAAE,CAAA;CACtC;AAED,8EAA8E;AAC9E,MAAM,WAAW,eAAgB,SAAQ,eAAe;IACtD,OAAO,EAAE,YAAY,CAAA;CACtB;AAED;;;;;;;;GAQG;AACH,MAAM,WAAW,cAAc,CAAC,MAAM,SAAS,SAAS,GAAG,SAAS;IAClE,GAAG,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,eAAe,GAAG,OAAO,CAAC,SAAS,gBAAgB,EAAE,CAAC,CAAA;CAC/E;AAED;;;;GAIG;AACH,wBAAgB,oBAAoB,CAAC,MAAM,SAAS,SAAS,EAC3D,MAAM,EAAE,cAAc,CAAC,MAAM,CAAC,GAC7B,iBAAiB,CAAC,MAAM,EAAE,aAAa,CAAC,CAK1C;AAED,yEAAyE;AACzE,MAAM,WAAW,gBAAiB,SAAQ,iBAAiB;IACzD,KAAK,EAAE,MAAM,CAAA;CACd;AAED;;;;;GAKG;AACH,wBAAsB,cAAc,CAClC,KAAK,EAAE,SAAS,EAChB,KAAK,EAAE,aAAa,EACpB,gBAAgB,EAAE,kBAAkB,GACnC,OAAO,CAAC,gBAAgB,CAAC,CAO3B;AAED,0EAA0E;AAC1E,MAAM,WAAW,aAAa;IAC5B,OAAO,EAAE,MAAM,CAAA;IACf,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,EAAE,MAAM,CAAA;IACb,QAAQ,EAAE,MAAM,CAAA;IAChB,MAAM,EAAE,MAAM,GAAG,MAAM,CAAA;IACvB,QAAQ,EAAE,MAAM,EAAE,CAAA;CACnB;AAED;;;;GAIG;AACH,wBAAgB,mBAAmB,CAAC,QAAQ,EAAE,SAAS,gBAAgB,EAAE,GAAG,aAAa,EAAE,CAe1F;AAED,gFAAgF;AAChF,MAAM,WAAW,iBAAiB;IAChC,0CAA0C;IAC1C,OAAO,EAAE,MAAM,CAAA;IACf,8CAA8C;IAC9C,oBAAoB,EAAE,MAAM,CAAA;IAC5B,uCAAuC;IACvC,YAAY,EAAE,MAAM,CAAA;IACpB,6BAA6B;IAC7B,MAAM,EAAE,MAAM,CAAA;IACd,6BAA6B;IAC7B,MAAM,EAAE,MAAM,CAAA;IACd,uDAAuD;IACvD,QAAQ,EAAE,MAAM,CAAA;CACjB;AAED,wEAAwE;AACxE,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,SAAS,aAAa,EAAE,GAAG,iBAAiB,CAsBnF;AAED,MAAM,WAAW,uBAAuB;IACtC,yDAAyD;IACzD,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,wFAAwF;IACxF,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC7B,+DAA+D;IAC/D,gBAAgB,CAAC,EAAE,MAAM,CAAA;CAC1B;AAUD;;;;;;GAMG;AACH,wBAAgB,wBAAwB,CACtC,IAAI,EAAE,SAAS,aAAa,EAAE,EAC9B,IAAI,GAAE,uBAA4B,GACjC,MAAM,CA6CR"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"playback.test.d.ts","sourceRoot":"","sources":["../../../src/campaign/presets/playback.test.ts"],"names":[],"mappings":""}
|