spawnfile 0.1.25 → 0.1.27
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/README.md +1 -1
- package/dist/audit/readAudit.d.ts +2 -2
- package/dist/auth/trainingAuth.js +1 -1
- package/dist/cli/runCli.d.ts +2 -2
- package/dist/cli/runCli.js +2 -2
- package/dist/cli/trainCommand.js +4 -5
- package/dist/cli/{paideiaDelegation.d.ts → trainingExecution.d.ts} +3 -4
- package/dist/cli/trainingExecution.js +151 -0
- package/dist/compiler/containerDaimonGrokWorkerRender.d.ts +2 -2
- package/dist/compiler/containerDaimonGrokWorkerRender.js +3 -2
- package/dist/compiler/containerDaimonUidEntrypointRender.js +4 -3
- package/dist/compiler/containerPackageOverrides.d.ts +1 -1
- package/dist/compiler/training/broker/declaration.d.ts +2 -2
- package/dist/compiler/training/broker/declaration.js +3 -2
- package/dist/compiler/training/broker/entrypoint.d.ts +3 -3
- package/dist/compiler/training/broker/entrypoint.js +7 -6
- package/dist/compiler/training/broker/evaluatorSecrets.js +4 -2
- package/dist/compiler/training/broker/paths.d.ts +15 -15
- package/dist/compiler/training/broker/paths.js +23 -23
- package/dist/compiler/training/broker/projection.d.ts +4 -4
- package/dist/compiler/training/broker/projection.js +9 -7
- package/dist/compiler/training/broker/provisioning.d.ts +1 -1
- package/dist/compiler/training/broker/provisioning.js +3 -3
- package/dist/compiler/training/broker/registration.d.ts +1 -1
- package/dist/compiler/training/broker/registration.js +1 -1
- package/dist/compiler/training/broker/supervisor.d.ts +1 -1
- package/dist/compiler/training/container/contract.js +2 -1
- package/dist/compiler/training/container/evaluators.d.ts +3 -3
- package/dist/compiler/training/container/evaluators.js +6 -6
- package/dist/compiler/training/container/launch.js +15 -7
- package/dist/compiler/training/container/security.js +3 -3
- package/dist/compiler/training/container/subjectIsolation.js +2 -1
- package/dist/compiler/training/execution.d.ts +8 -10
- package/dist/compiler/training/execution.js +1 -1
- package/dist/compiler/training/moments.js +2 -2
- package/dist/compiler/training/plan/contract.d.ts +7 -7
- package/dist/compiler/training/plan/load.js +2 -0
- package/dist/compiler/training/preparation/assets/Dockerfile +8 -17
- package/dist/compiler/training/preparation/assets/package-lock.json +2012 -316
- package/dist/compiler/training/preparation/broker.js +2 -1
- package/dist/compiler/training/preparation/contract.d.ts +5 -9
- package/dist/compiler/training/preparation/contract.js +4 -3
- package/dist/compiler/training/preparation/image.js +2 -4
- package/dist/compiler/training/preparation/prepare.js +3 -3
- package/dist/compiler/training/repair/compatibility.js +17 -6
- package/dist/compiler/training/repair/contract.d.ts +14 -22
- package/dist/compiler/training/repair/contract.js +3 -2
- package/dist/compiler/training/repair/prepare.d.ts +5 -9
- package/dist/compiler/training/repair/prepare.js +10 -4
- package/dist/compiler/training/repair/witness.d.ts +4 -8
- package/dist/compiler/training/repair/witness.js +16 -2
- package/dist/compiler/training/rewind/contract.d.ts +9 -9
- package/dist/compiler/training/subjectDeclaration.d.ts +2 -1
- package/dist/compiler/training/subjectDeclaration.js +4 -2
- package/dist/compiler/training/subjectLease.d.ts +9 -8
- package/dist/compiler/training/subjectLease.js +17 -13
- package/dist/distribution/imageRuntimeAuth.js +33 -24
- package/dist/manifest/executionSchemas.d.ts +9 -9
- package/dist/manifest/mcpSchemas.d.ts +3 -3
- package/dist/manifest/schemas.d.ts +71 -71
- package/dist/manifest/teamNetworkAccessSchemas.d.ts +1 -1
- package/dist/manifest/teamNetworkAuthSchemas.d.ts +1 -1
- package/dist/manifest/teamNetworkSchemas.d.ts +2 -2
- package/dist/manifest/teamNetworkServerSchemas.d.ts +2 -2
- package/dist/manifest/trainingDataSchemas.d.ts +2 -2
- package/dist/runtime/codex-exec/training/adapter.d.ts +13 -0
- package/dist/runtime/codex-exec/training/adapter.js +101 -0
- package/dist/runtime/codex-exec/training/adapter.js.map +1 -0
- package/dist/runtime/codex-exec/training/environment.d.ts +3 -0
- package/dist/runtime/codex-exec/training/environment.js +26 -0
- package/dist/runtime/codex-exec/training/environment.js.map +1 -0
- package/dist/runtime/codex-exec/training/index.d.ts +2 -0
- package/dist/runtime/codex-exec/training/index.js +3 -0
- package/dist/runtime/codex-exec/training/index.js.map +1 -0
- package/dist/runtime/codex-exec/training/protocol.d.ts +15 -0
- package/dist/runtime/codex-exec/training/protocol.js +102 -0
- package/dist/runtime/codex-exec/training/protocol.js.map +1 -0
- package/dist/runtime/codex-exec/training/providers/codex-subscription/index.d.ts +1 -0
- package/dist/runtime/codex-exec/training/providers/codex-subscription/index.js +2 -0
- package/dist/runtime/codex-exec/training/providers/codex-subscription/index.js.map +1 -0
- package/dist/runtime/codex-exec/training/providers/codex-subscription/protocol.d.ts +17 -0
- package/dist/runtime/codex-exec/training/providers/codex-subscription/protocol.js +80 -0
- package/dist/runtime/codex-exec/training/providers/codex-subscription/protocol.js.map +1 -0
- package/dist/runtime/codex-exec/training/providers/codex-subscription/provider.d.ts +12 -0
- package/dist/runtime/codex-exec/training/providers/codex-subscription/provider.js +125 -0
- package/dist/runtime/codex-exec/training/providers/codex-subscription/provider.js.map +1 -0
- package/dist/runtime/codex-exec/training/providers/codex-subscription/strict-schema.d.ts +6 -0
- package/dist/runtime/codex-exec/training/providers/codex-subscription/strict-schema.js +219 -0
- package/dist/runtime/codex-exec/training/providers/codex-subscription/strict-schema.js.map +1 -0
- package/dist/runtime/container.js +1 -1
- package/dist/runtime/containerPackageOverrides.d.ts +2 -2
- package/dist/runtime/containerPackageOverrides.js +1 -1
- package/dist/runtime/daimon/adapter.js +9 -9
- package/dist/runtime/daimon/config.d.ts +2 -1
- package/dist/runtime/daimon/config.js +25 -25
- package/dist/runtime/daimon/contract-manifest.json +1 -1
- package/dist/runtime/daimon/contract-manifest.sha256 +1 -1
- package/dist/runtime/daimon/contractManifest.d.ts +206 -202
- package/dist/runtime/daimon/contractManifest.js +48 -145
- package/dist/runtime/daimon/contractManifest.js.map +1 -0
- package/dist/runtime/daimon/contractManifestSchema.d.ts +1455 -0
- package/dist/runtime/daimon/contractManifestSchema.js +19 -0
- package/dist/runtime/daimon/contractManifestSchema.js.map +1 -0
- package/dist/runtime/daimon/engineDescriptorSchema.d.ts +1826 -0
- package/dist/runtime/daimon/engineDescriptorSchema.js +107 -0
- package/dist/runtime/daimon/engineDescriptorSchema.js.map +1 -0
- package/dist/runtime/daimon/engineDescriptors.generated.d.ts +477 -0
- package/dist/runtime/daimon/engineDescriptors.generated.js +5 -0
- package/dist/runtime/daimon/engineDescriptors.generated.js.map +1 -0
- package/dist/runtime/daimon/engines.d.ts +664 -0
- package/dist/runtime/daimon/engines.js +25 -0
- package/dist/runtime/daimon/engines.js.map +1 -0
- package/dist/runtime/daimon/grokModel.d.ts +2 -1
- package/dist/runtime/daimon/grokModel.js +10 -7
- package/dist/runtime/daimon/grokSkillsSupport.js +2 -1
- package/dist/runtime/daimon/grokTurnLimits.js +3 -3
- package/dist/runtime/daimon/grokWorkerContract.d.ts +2 -2
- package/dist/runtime/daimon/grokWorkerContract.js +10 -9
- package/dist/runtime/daimon/index.d.ts +4 -0
- package/dist/runtime/daimon/index.js +5 -0
- package/dist/runtime/daimon/index.js.map +1 -0
- package/dist/runtime/daimon/packageName.d.ts +2 -0
- package/dist/runtime/daimon/packageName.js +3 -0
- package/dist/runtime/daimon/packageName.js.map +1 -0
- package/dist/runtime/daimon/runAuth.d.ts +4 -4
- package/dist/runtime/daimon/runAuth.js +71 -53
- package/dist/runtime/daimon/training/index.d.ts +2 -0
- package/dist/runtime/daimon/training/index.js +7 -0
- package/dist/runtime/daimon/training/index.js.map +1 -0
- package/dist/runtime/daimon/training/legacy/adapter.d.ts +11 -0
- package/dist/runtime/daimon/training/legacy/adapter.js +175 -0
- package/dist/runtime/daimon/training/legacy/adapter.js.map +1 -0
- package/dist/runtime/daimon/training/legacy/index.d.ts +2 -0
- package/dist/runtime/daimon/training/legacy/index.js +3 -0
- package/dist/runtime/daimon/training/legacy/index.js.map +1 -0
- package/dist/runtime/daimon/training/legacy/public.d.ts +49 -0
- package/dist/runtime/daimon/training/legacy/public.js +16 -0
- package/dist/runtime/daimon/training/legacy/public.js.map +1 -0
- package/dist/runtime/daimon/training/native/adapter.d.ts +18 -0
- package/dist/runtime/daimon/training/native/adapter.js +231 -0
- package/dist/runtime/daimon/training/native/adapter.js.map +1 -0
- package/dist/runtime/daimon/training/native/admission.d.ts +7 -0
- package/dist/runtime/daimon/training/native/admission.js +52 -0
- package/dist/runtime/daimon/training/native/admission.js.map +1 -0
- package/dist/runtime/daimon/training/native/child/arguments.d.ts +16 -0
- package/dist/runtime/daimon/training/native/child/arguments.js +69 -0
- package/dist/runtime/daimon/training/native/child/arguments.js.map +1 -0
- package/dist/runtime/daimon/training/native/child/deferred-trial.d.ts +24 -0
- package/dist/runtime/daimon/training/native/child/deferred-trial.js +63 -0
- package/dist/runtime/daimon/training/native/child/deferred-trial.js.map +1 -0
- package/dist/runtime/daimon/training/native/child/diagnostics.d.ts +58 -0
- package/dist/runtime/daimon/training/native/child/diagnostics.js +215 -0
- package/dist/runtime/daimon/training/native/child/diagnostics.js.map +1 -0
- package/dist/runtime/daimon/training/native/child/fixture.d.ts +16 -0
- package/dist/runtime/daimon/training/native/child/fixture.js +157 -0
- package/dist/runtime/daimon/training/native/child/fixture.js.map +1 -0
- package/dist/runtime/daimon/training/native/child/grok.d.ts +14 -0
- package/dist/runtime/daimon/training/native/child/grok.js +47 -0
- package/dist/runtime/daimon/training/native/child/grok.js.map +1 -0
- package/dist/runtime/daimon/training/native/child/index.d.ts +7 -0
- package/dist/runtime/daimon/training/native/child/index.js +5 -0
- package/dist/runtime/daimon/training/native/child/index.js.map +1 -0
- package/dist/runtime/daimon/training/native/child/lifecycle.d.ts +40 -0
- package/dist/runtime/daimon/training/native/child/lifecycle.js +391 -0
- package/dist/runtime/daimon/training/native/child/lifecycle.js.map +1 -0
- package/dist/runtime/daimon/training/native/child/probe.d.ts +2 -0
- package/dist/runtime/daimon/training/native/child/probe.js +11 -0
- package/dist/runtime/daimon/training/native/child/probe.js.map +1 -0
- package/dist/runtime/daimon/training/native/child/sandbox.d.ts +4 -0
- package/dist/runtime/daimon/training/native/child/sandbox.js +76 -0
- package/dist/runtime/daimon/training/native/child/sandbox.js.map +1 -0
- package/dist/runtime/daimon/training/native/child/types.d.ts +15 -0
- package/dist/runtime/daimon/training/native/child/types.js +2 -0
- package/dist/runtime/daimon/training/native/child/types.js.map +1 -0
- package/dist/runtime/daimon/training/native/container/docker.d.ts +29 -0
- package/dist/runtime/daimon/training/native/container/docker.js +105 -0
- package/dist/runtime/daimon/training/native/container/docker.js.map +1 -0
- package/dist/runtime/daimon/training/native/container/grok-slot.d.ts +22 -0
- package/dist/runtime/daimon/training/native/container/grok-slot.js +38 -0
- package/dist/runtime/daimon/training/native/container/grok-slot.js.map +1 -0
- package/dist/runtime/daimon/training/native/container/index.d.ts +6 -0
- package/dist/runtime/daimon/training/native/container/index.js +6 -0
- package/dist/runtime/daimon/training/native/container/index.js.map +1 -0
- package/dist/runtime/daimon/training/native/container/lifecycle.d.ts +8 -0
- package/dist/runtime/daimon/training/native/container/lifecycle.js +110 -0
- package/dist/runtime/daimon/training/native/container/lifecycle.js.map +1 -0
- package/dist/runtime/daimon/training/native/container/manifest.d.ts +17 -0
- package/dist/runtime/daimon/training/native/container/manifest.js +144 -0
- package/dist/runtime/daimon/training/native/container/manifest.js.map +1 -0
- package/dist/runtime/daimon/training/native/container/native-preflight.d.ts +8 -0
- package/dist/runtime/daimon/training/native/container/native-preflight.js +40 -0
- package/dist/runtime/daimon/training/native/container/native-preflight.js.map +1 -0
- package/dist/runtime/daimon/training/native/container/preflight.d.ts +15 -0
- package/dist/runtime/daimon/training/native/container/preflight.js +79 -0
- package/dist/runtime/daimon/training/native/container/preflight.js.map +1 -0
- package/dist/runtime/daimon/training/native/container/slot.d.ts +4 -0
- package/dist/runtime/daimon/training/native/container/slot.js +58 -0
- package/dist/runtime/daimon/training/native/container/slot.js.map +1 -0
- package/dist/runtime/daimon/training/native/external.d.ts +37 -0
- package/dist/runtime/daimon/training/native/external.js +75 -0
- package/dist/runtime/daimon/training/native/external.js.map +1 -0
- package/dist/runtime/daimon/training/native/grants/authority.d.ts +36 -0
- package/dist/runtime/daimon/training/native/grants/authority.js +49 -0
- package/dist/runtime/daimon/training/native/grants/authority.js.map +1 -0
- package/dist/runtime/daimon/training/native/grants/index.d.ts +2 -0
- package/dist/runtime/daimon/training/native/grants/index.js +3 -0
- package/dist/runtime/daimon/training/native/grants/index.js.map +1 -0
- package/dist/runtime/daimon/training/native/grants/pins.d.ts +30 -0
- package/dist/runtime/daimon/training/native/grants/pins.js +20 -0
- package/dist/runtime/daimon/training/native/grants/pins.js.map +1 -0
- package/dist/runtime/daimon/training/native/grants/test-socket.d.ts +6 -0
- package/dist/runtime/daimon/training/native/grants/test-socket.js +35 -0
- package/dist/runtime/daimon/training/native/grants/test-socket.js.map +1 -0
- package/dist/runtime/daimon/training/native/grok-fixtures.d.ts +34 -0
- package/dist/runtime/daimon/training/native/grok-fixtures.js +179 -0
- package/dist/runtime/daimon/training/native/grok-fixtures.js.map +1 -0
- package/dist/runtime/daimon/training/native/index.d.ts +38 -0
- package/dist/runtime/daimon/training/native/index.js +27 -0
- package/dist/runtime/daimon/training/native/index.js.map +1 -0
- package/dist/runtime/daimon/training/native/live-checkout.d.ts +2 -0
- package/dist/runtime/daimon/training/native/live-checkout.js +15 -0
- package/dist/runtime/daimon/training/native/live-checkout.js.map +1 -0
- package/dist/runtime/daimon/training/native/observations.d.ts +51 -0
- package/dist/runtime/daimon/training/native/observations.js +217 -0
- package/dist/runtime/daimon/training/native/observations.js.map +1 -0
- package/dist/runtime/daimon/training/native/outcome.d.ts +13 -0
- package/dist/runtime/daimon/training/native/outcome.js +37 -0
- package/dist/runtime/daimon/training/native/outcome.js.map +1 -0
- package/dist/runtime/daimon/training/native/preparation/engine-config.d.ts +21 -0
- package/dist/runtime/daimon/training/native/preparation/engine-config.js +30 -0
- package/dist/runtime/daimon/training/native/preparation/engine-config.js.map +1 -0
- package/dist/runtime/daimon/training/native/preparation/fixtures.d.ts +17 -0
- package/dist/runtime/daimon/training/native/preparation/fixtures.js +50 -0
- package/dist/runtime/daimon/training/native/preparation/fixtures.js.map +1 -0
- package/dist/runtime/daimon/training/native/preparation/index.d.ts +7 -0
- package/dist/runtime/daimon/training/native/preparation/index.js +5 -0
- package/dist/runtime/daimon/training/native/preparation/index.js.map +1 -0
- package/dist/runtime/daimon/training/native/preparation/layout.d.ts +66 -0
- package/dist/runtime/daimon/training/native/preparation/layout.js +87 -0
- package/dist/runtime/daimon/training/native/preparation/layout.js.map +1 -0
- package/dist/runtime/daimon/training/native/preparation/subject.d.ts +14 -0
- package/dist/runtime/daimon/training/native/preparation/subject.js +67 -0
- package/dist/runtime/daimon/training/native/preparation/subject.js.map +1 -0
- package/dist/runtime/daimon/training/native/preparation/types.d.ts +69 -0
- package/dist/runtime/daimon/training/native/preparation/types.js +2 -0
- package/dist/runtime/daimon/training/native/preparation/types.js.map +1 -0
- package/dist/runtime/daimon/training/native/receipt/artifacts.d.ts +25 -0
- package/dist/runtime/daimon/training/native/receipt/artifacts.js +32 -0
- package/dist/runtime/daimon/training/native/receipt/artifacts.js.map +1 -0
- package/dist/runtime/daimon/training/native/receipt/grok-clocked-pins.d.ts +129 -0
- package/dist/runtime/daimon/training/native/receipt/grok-clocked-pins.js +45 -0
- package/dist/runtime/daimon/training/native/receipt/grok-clocked-pins.js.map +1 -0
- package/dist/runtime/daimon/training/native/receipt/grok.d.ts +246 -0
- package/dist/runtime/daimon/training/native/receipt/grok.js +195 -0
- package/dist/runtime/daimon/training/native/receipt/grok.js.map +1 -0
- package/dist/runtime/daimon/training/native/receipt/index.d.ts +10 -0
- package/dist/runtime/daimon/training/native/receipt/index.js +7 -0
- package/dist/runtime/daimon/training/native/receipt/index.js.map +1 -0
- package/dist/runtime/daimon/training/native/receipt/paths.d.ts +3 -0
- package/dist/runtime/daimon/training/native/receipt/paths.js +18 -0
- package/dist/runtime/daimon/training/native/receipt/paths.js.map +1 -0
- package/dist/runtime/daimon/training/native/receipt/sandbox.d.ts +42 -0
- package/dist/runtime/daimon/training/native/receipt/sandbox.js +35 -0
- package/dist/runtime/daimon/training/native/receipt/sandbox.js.map +1 -0
- package/dist/runtime/daimon/training/native/receipt/schema.d.ts +486 -0
- package/dist/runtime/daimon/training/native/receipt/schema.js +138 -0
- package/dist/runtime/daimon/training/native/receipt/schema.js.map +1 -0
- package/dist/runtime/daimon/training/native/receipt/task-clock.d.ts +19 -0
- package/dist/runtime/daimon/training/native/receipt/task-clock.js +49 -0
- package/dist/runtime/daimon/training/native/receipt/task-clock.js.map +1 -0
- package/dist/runtime/daimon/training/native/sealed-usage.d.ts +218 -0
- package/dist/runtime/daimon/training/native/sealed-usage.js +261 -0
- package/dist/runtime/daimon/training/native/sealed-usage.js.map +1 -0
- package/dist/runtime/daimon/training/native/test-fixtures.d.ts +27 -0
- package/dist/runtime/daimon/training/native/test-fixtures.js +158 -0
- package/dist/runtime/daimon/training/native/test-fixtures.js.map +1 -0
- package/dist/runtime/daimon/training/native/transcript.d.ts +96 -0
- package/dist/runtime/daimon/training/native/transcript.js +239 -0
- package/dist/runtime/daimon/training/native/transcript.js.map +1 -0
- package/dist/runtime/daimon/training/native/transport/delivery.d.ts +27 -0
- package/dist/runtime/daimon/training/native/transport/delivery.js +98 -0
- package/dist/runtime/daimon/training/native/transport/delivery.js.map +1 -0
- package/dist/runtime/daimon/training/native/transport/http.d.ts +3 -0
- package/dist/runtime/daimon/training/native/transport/http.js +81 -0
- package/dist/runtime/daimon/training/native/transport/http.js.map +1 -0
- package/dist/runtime/daimon/training/native/transport/index.d.ts +4 -0
- package/dist/runtime/daimon/training/native/transport/index.js +3 -0
- package/dist/runtime/daimon/training/native/transport/index.js.map +1 -0
- package/dist/runtime/daimon/training/native/transport/types.d.ts +15 -0
- package/dist/runtime/daimon/training/native/transport/types.js +2 -0
- package/dist/runtime/daimon/training/native/transport/types.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/cli.d.ts +1 -0
- package/dist/runtime/daimon/training/native/worker/cli.js +17 -0
- package/dist/runtime/daimon/training/native/worker/cli.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/control.d.ts +20 -0
- package/dist/runtime/daimon/training/native/worker/control.js +266 -0
- package/dist/runtime/daimon/training/native/worker/control.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/deliveries.d.ts +38 -0
- package/dist/runtime/daimon/training/native/worker/deliveries.js +91 -0
- package/dist/runtime/daimon/training/native/worker/deliveries.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/evidence/collect.d.ts +54 -0
- package/dist/runtime/daimon/training/native/worker/evidence/collect.js +278 -0
- package/dist/runtime/daimon/training/native/worker/evidence/collect.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/evidence/execution.d.ts +102 -0
- package/dist/runtime/daimon/training/native/worker/evidence/execution.js +191 -0
- package/dist/runtime/daimon/training/native/worker/evidence/execution.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/evidence/index.d.ts +3 -0
- package/dist/runtime/daimon/training/native/worker/evidence/index.js +4 -0
- package/dist/runtime/daimon/training/native/worker/evidence/index.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/evidence/io.d.ts +20 -0
- package/dist/runtime/daimon/training/native/worker/evidence/io.js +130 -0
- package/dist/runtime/daimon/training/native/worker/evidence/io.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/evidence/seal.d.ts +3 -0
- package/dist/runtime/daimon/training/native/worker/evidence/seal.js +67 -0
- package/dist/runtime/daimon/training/native/worker/evidence/seal.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/index.d.ts +11 -0
- package/dist/runtime/daimon/training/native/worker/index.js +8 -0
- package/dist/runtime/daimon/training/native/worker/index.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/profile.d.ts +23 -0
- package/dist/runtime/daimon/training/native/worker/profile.js +149 -0
- package/dist/runtime/daimon/training/native/worker/profile.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/public.d.ts +41 -0
- package/dist/runtime/daimon/training/native/worker/public.js +22 -0
- package/dist/runtime/daimon/training/native/worker/public.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/sandbox.d.ts +4 -0
- package/dist/runtime/daimon/training/native/worker/sandbox.js +52 -0
- package/dist/runtime/daimon/training/native/worker/sandbox.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/server.d.ts +4 -0
- package/dist/runtime/daimon/training/native/worker/server.js +102 -0
- package/dist/runtime/daimon/training/native/worker/server.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/types.d.ts +318 -0
- package/dist/runtime/daimon/training/native/worker/types.js +2 -0
- package/dist/runtime/daimon/training/native/worker/types.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/index.d.ts +4 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/index.js +4 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/index.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/source.d.ts +3 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/source.js +16 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/source.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/terminal.d.ts +4 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/terminal.js +52 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/terminal.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/transport.d.ts +21 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/transport.js +198 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/transport.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/types.d.ts +91 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/types.js +2 -0
- package/dist/runtime/daimon/training/native/worker/wake-delivery/types.js.map +1 -0
- package/dist/runtime/daimon/training/native/worker/worker.d.ts +1 -0
- package/dist/runtime/daimon/training/native/worker/worker.js +103 -0
- package/dist/runtime/daimon/training/native/worker/worker.js.map +1 -0
- package/dist/runtime/daimon/training/providers/grok-broker/broker.d.ts +20 -0
- package/dist/runtime/daimon/training/providers/grok-broker/broker.js +55 -0
- package/dist/runtime/daimon/training/providers/grok-broker/broker.js.map +1 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grant-fixtures.d.ts +15 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grant-fixtures.js +29 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grant-fixtures.js.map +1 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grant-home.d.ts +16 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grant-home.js +60 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grant-home.js.map +1 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grant-pool.d.ts +81 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grant-pool.js +221 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grant-pool.js.map +1 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grant-releases.d.ts +31 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grant-releases.js +78 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grant-releases.js.map +1 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grants.d.ts +63 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grants.js +20 -0
- package/dist/runtime/daimon/training/providers/grok-broker/grants.js.map +1 -0
- package/dist/runtime/daimon/training/providers/grok-broker/index.d.ts +3 -0
- package/dist/runtime/daimon/training/providers/grok-broker/index.js +4 -0
- package/dist/runtime/daimon/training/providers/grok-broker/index.js.map +1 -0
- package/dist/runtime/daimon/training/providers/grok-subscription/cli.d.ts +21 -0
- package/dist/runtime/daimon/training/providers/grok-subscription/cli.js +81 -0
- package/dist/runtime/daimon/training/providers/grok-subscription/cli.js.map +1 -0
- package/dist/runtime/daimon/training/providers/grok-subscription/index.d.ts +4 -0
- package/dist/runtime/daimon/training/providers/grok-subscription/index.js +5 -0
- package/dist/runtime/daimon/training/providers/grok-subscription/index.js.map +1 -0
- package/dist/runtime/daimon/training/providers/grok-subscription/protocol.d.ts +26 -0
- package/dist/runtime/daimon/training/providers/grok-subscription/protocol.js +102 -0
- package/dist/runtime/daimon/training/providers/grok-subscription/protocol.js.map +1 -0
- package/dist/runtime/daimon/training/providers/grok-subscription/provider.d.ts +9 -0
- package/dist/runtime/daimon/training/providers/grok-subscription/provider.js +38 -0
- package/dist/runtime/daimon/training/providers/grok-subscription/provider.js.map +1 -0
- package/dist/runtime/daimon/training/services/freshness.d.ts +15 -0
- package/dist/runtime/daimon/training/services/freshness.js +19 -0
- package/dist/runtime/daimon/training/services/freshness.js.map +1 -0
- package/dist/runtime/daimon/training/services/index.d.ts +3 -0
- package/dist/runtime/daimon/training/services/index.js +3 -0
- package/dist/runtime/daimon/training/services/index.js.map +1 -0
- package/dist/runtime/daimon/training/services/slot-freshness.d.ts +34 -0
- package/dist/runtime/daimon/training/services/slot-freshness.js +178 -0
- package/dist/runtime/daimon/training/services/slot-freshness.js.map +1 -0
- package/dist/runtime/daimon/wakeMoments.js +1 -0
- package/dist/runtime/daimon/wakeMoments.js.map +1 -0
- package/dist/runtime/pi/appPreludeSource.js +3 -2
- package/dist/runtime/pi/appTemplate.d.ts +1 -1
- package/dist/runtime/pi/appTemplate.js +2 -1
- package/dist/runtime/registry.d.ts +6 -0
- package/dist/runtime/registry.js +2 -0
- package/dist/runtime/trainingCli.d.ts +5 -0
- package/dist/runtime/trainingCli.js +23 -0
- package/dist/runtime/trainingGraderProviders.d.ts +4 -0
- package/dist/runtime/trainingGraderProviders.js +55 -0
- package/dist/runtime/trainingProviders.d.ts +4 -0
- package/dist/runtime/trainingProviders.js +89 -0
- package/dist/runtime/trainingSubjects.d.ts +129 -0
- package/dist/runtime/trainingSubjects.js +32 -0
- package/dist/shared/daimonCodexDocker.js +9 -10
- package/dist/shared/errors.js +1 -0
- package/dist/shared/errors.js.map +1 -0
- package/dist/shared/index.d.ts +1 -0
- package/dist/shared/index.js +1 -0
- package/dist/shared/trainingLegacy/index.d.ts +5 -0
- package/dist/shared/trainingLegacy/index.js +6 -0
- package/dist/shared/trainingLegacy/index.js.map +1 -0
- package/dist/shared/trainingLegacy/trainingLegacyEnvironment.d.ts +6 -0
- package/dist/shared/trainingLegacy/trainingLegacyEnvironment.js +40 -0
- package/dist/shared/trainingLegacy/trainingLegacyEnvironment.js.map +1 -0
- package/dist/shared/trainingLegacy/trainingLegacyFiles.d.ts +4 -0
- package/dist/shared/trainingLegacy/trainingLegacyFiles.js +34 -0
- package/dist/shared/trainingLegacy/trainingLegacyFiles.js.map +1 -0
- package/dist/shared/trainingLegacy/trainingLegacyGraderProviders.d.ts +5 -0
- package/dist/shared/trainingLegacy/trainingLegacyGraderProviders.js +6 -0
- package/dist/shared/trainingLegacy/trainingLegacyGraderProviders.js.map +1 -0
- package/dist/shared/trainingLegacy/trainingLegacyHashDomains.d.ts +39 -0
- package/dist/shared/trainingLegacy/trainingLegacyHashDomains.js +41 -0
- package/dist/shared/trainingLegacy/trainingLegacyHashDomains.js.map +1 -0
- package/dist/shared/trainingLegacy/trainingLegacyWire.d.ts +15 -0
- package/dist/shared/trainingLegacy/trainingLegacyWire.js +21 -0
- package/dist/shared/trainingLegacy/trainingLegacyWire.js.map +1 -0
- package/dist/shared/trainingLegacy/trainingLegacyWireAliases.json +90 -0
- package/dist/target/contracts.d.ts +2 -2
- package/dist/training/candidates/authored.d.ts +3 -0
- package/dist/training/candidates/authored.js +27 -0
- package/dist/training/candidates/authored.js.map +1 -0
- package/dist/training/candidates/derive.d.ts +3 -0
- package/dist/training/candidates/derive.js +66 -0
- package/dist/training/candidates/derive.js.map +1 -0
- package/dist/training/candidates/filesystem.d.ts +7 -0
- package/dist/training/candidates/filesystem.js +41 -0
- package/dist/training/candidates/filesystem.js.map +1 -0
- package/dist/training/candidates/headings.d.ts +9 -0
- package/dist/training/candidates/headings.js +47 -0
- package/dist/training/candidates/headings.js.map +1 -0
- package/dist/training/candidates/identity.d.ts +6 -0
- package/dist/training/candidates/identity.js +29 -0
- package/dist/training/candidates/identity.js.map +1 -0
- package/dist/training/candidates/index.d.ts +6 -0
- package/dist/training/candidates/index.js +6 -0
- package/dist/training/candidates/index.js.map +1 -0
- package/dist/training/candidates/materialize.d.ts +2 -0
- package/dist/training/candidates/materialize.js +39 -0
- package/dist/training/candidates/materialize.js.map +1 -0
- package/dist/training/candidates/snapshot.d.ts +5 -0
- package/dist/training/candidates/snapshot.js +22 -0
- package/dist/training/candidates/snapshot.js.map +1 -0
- package/dist/training/candidates/types.d.ts +41 -0
- package/dist/training/candidates/types.js +2 -0
- package/dist/training/candidates/types.js.map +1 -0
- package/dist/training/candidates/validation.d.ts +4 -0
- package/dist/training/candidates/validation.js +89 -0
- package/dist/training/candidates/validation.js.map +1 -0
- package/dist/training/capture/index.d.ts +2 -0
- package/dist/training/capture/index.js +2 -0
- package/dist/training/capture/index.js.map +1 -0
- package/dist/training/capture/trace.d.ts +16 -0
- package/dist/training/capture/trace.js +74 -0
- package/dist/training/capture/trace.js.map +1 -0
- package/dist/training/cases/builtin-evaluators.d.ts +2 -0
- package/dist/training/cases/builtin-evaluators.js +120 -0
- package/dist/training/cases/builtin-evaluators.js.map +1 -0
- package/dist/training/cases/check-schema.d.ts +2 -0
- package/dist/training/cases/check-schema.js +60 -0
- package/dist/training/cases/check-schema.js.map +1 -0
- package/dist/training/cases/cost-projection.d.ts +27 -0
- package/dist/training/cases/cost-projection.js +29 -0
- package/dist/training/cases/cost-projection.js.map +1 -0
- package/dist/training/cases/fixtures.d.ts +5 -0
- package/dist/training/cases/fixtures.js +134 -0
- package/dist/training/cases/fixtures.js.map +1 -0
- package/dist/training/cases/grader.d.ts +14 -0
- package/dist/training/cases/grader.js +122 -0
- package/dist/training/cases/grader.js.map +1 -0
- package/dist/training/cases/index.d.ts +12 -0
- package/dist/training/cases/index.js +11 -0
- package/dist/training/cases/index.js.map +1 -0
- package/dist/training/cases/json-schema.d.ts +3 -0
- package/dist/training/cases/json-schema.js +6 -0
- package/dist/training/cases/json-schema.js.map +1 -0
- package/dist/training/cases/parser.d.ts +6 -0
- package/dist/training/cases/parser.js +45 -0
- package/dist/training/cases/parser.js.map +1 -0
- package/dist/training/cases/plans.d.ts +14 -0
- package/dist/training/cases/plans.js +186 -0
- package/dist/training/cases/plans.js.map +1 -0
- package/dist/training/cases/resolver.d.ts +5 -0
- package/dist/training/cases/resolver.js +90 -0
- package/dist/training/cases/resolver.js.map +1 -0
- package/dist/training/cases/schema.d.ts +5 -0
- package/dist/training/cases/schema.js +154 -0
- package/dist/training/cases/schema.js.map +1 -0
- package/dist/training/cases/splits.d.ts +16 -0
- package/dist/training/cases/splits.js +71 -0
- package/dist/training/cases/splits.js.map +1 -0
- package/dist/training/cases/text-checks.d.ts +3 -0
- package/dist/training/cases/text-checks.js +50 -0
- package/dist/training/cases/text-checks.js.map +1 -0
- package/dist/training/cases/training/identity.d.ts +6 -0
- package/dist/training/cases/training/identity.js +26 -0
- package/dist/training/cases/training/identity.js.map +1 -0
- package/dist/training/cases/training/index.d.ts +2 -0
- package/dist/training/cases/training/index.js +3 -0
- package/dist/training/cases/training/index.js.map +1 -0
- package/dist/training/cases/training/schema.d.ts +4 -0
- package/dist/training/cases/training/schema.js +207 -0
- package/dist/training/cases/training/schema.js.map +1 -0
- package/dist/training/cases/turn-bindings.d.ts +3 -0
- package/dist/training/cases/turn-bindings.js +61 -0
- package/dist/training/cases/turn-bindings.js.map +1 -0
- package/dist/training/cases/validation.d.ts +17 -0
- package/dist/training/cases/validation.js +92 -0
- package/dist/training/cases/validation.js.map +1 -0
- package/dist/training/cli/arguments.d.ts +18 -0
- package/dist/training/cli/arguments.js +165 -0
- package/dist/training/cli/arguments.js.map +1 -0
- package/dist/training/cli/commands.d.ts +3 -0
- package/dist/training/cli/commands.js +255 -0
- package/dist/training/cli/commands.js.map +1 -0
- package/dist/training/cli/datasets.d.ts +30 -0
- package/dist/training/cli/datasets.js +116 -0
- package/dist/training/cli/datasets.js.map +1 -0
- package/dist/training/cli/dry-run.d.ts +7 -0
- package/dist/training/cli/dry-run.js +37 -0
- package/dist/training/cli/dry-run.js.map +1 -0
- package/dist/training/cli/grader/atomic/decompose.d.ts +43 -0
- package/dist/training/cli/grader/atomic/decompose.js +208 -0
- package/dist/training/cli/grader/atomic/decompose.js.map +1 -0
- package/dist/training/cli/grader/atomic/files.d.ts +159 -0
- package/dist/training/cli/grader/atomic/files.js +97 -0
- package/dist/training/cli/grader/atomic/files.js.map +1 -0
- package/dist/training/cli/grader/atomic/index.d.ts +8 -0
- package/dist/training/cli/grader/atomic/index.js +34 -0
- package/dist/training/cli/grader/atomic/index.js.map +1 -0
- package/dist/training/cli/grader/atomic/reference.d.ts +131 -0
- package/dist/training/cli/grader/atomic/reference.js +133 -0
- package/dist/training/cli/grader/atomic/reference.js.map +1 -0
- package/dist/training/cli/grader/atomic/runtime.d.ts +3 -0
- package/dist/training/cli/grader/atomic/runtime.js +91 -0
- package/dist/training/cli/grader/atomic/runtime.js.map +1 -0
- package/dist/training/cli/grader/atomic/traps.d.ts +15 -0
- package/dist/training/cli/grader/atomic/traps.js +38 -0
- package/dist/training/cli/grader/atomic/traps.js.map +1 -0
- package/dist/training/cli/grader/backend.d.ts +19 -0
- package/dist/training/cli/grader/backend.js +39 -0
- package/dist/training/cli/grader/backend.js.map +1 -0
- package/dist/training/cli/grader/cache.d.ts +2 -0
- package/dist/training/cli/grader/cache.js +75 -0
- package/dist/training/cli/grader/cache.js.map +1 -0
- package/dist/training/cli/grader/calibrate.d.ts +19 -0
- package/dist/training/cli/grader/calibrate.js +237 -0
- package/dist/training/cli/grader/calibrate.js.map +1 -0
- package/dist/training/cli/grader/cost.d.ts +24 -0
- package/dist/training/cli/grader/cost.js +54 -0
- package/dist/training/cli/grader/cost.js.map +1 -0
- package/dist/training/cli/grader/exclusions.d.ts +12 -0
- package/dist/training/cli/grader/exclusions.js +49 -0
- package/dist/training/cli/grader/exclusions.js.map +1 -0
- package/dist/training/cli/grader/grade.d.ts +26 -0
- package/dist/training/cli/grader/grade.js +33 -0
- package/dist/training/cli/grader/grade.js.map +1 -0
- package/dist/training/cli/grader/index.d.ts +10 -0
- package/dist/training/cli/grader/index.js +73 -0
- package/dist/training/cli/grader/index.js.map +1 -0
- package/dist/training/cli/grader/label-call.d.ts +13 -0
- package/dist/training/cli/grader/label-call.js +67 -0
- package/dist/training/cli/grader/label-call.js.map +1 -0
- package/dist/training/cli/grader/label-import.d.ts +16 -0
- package/dist/training/cli/grader/label-import.js +84 -0
- package/dist/training/cli/grader/label-import.js.map +1 -0
- package/dist/training/cli/grader/label-inputs.d.ts +9 -0
- package/dist/training/cli/grader/label-inputs.js +56 -0
- package/dist/training/cli/grader/label-inputs.js.map +1 -0
- package/dist/training/cli/grader/label.d.ts +25 -0
- package/dist/training/cli/grader/label.js +172 -0
- package/dist/training/cli/grader/label.js.map +1 -0
- package/dist/training/cli/grader/records.d.ts +20 -0
- package/dist/training/cli/grader/records.js +65 -0
- package/dist/training/cli/grader/records.js.map +1 -0
- package/dist/training/cli/grader/report.d.ts +26 -0
- package/dist/training/cli/grader/report.js +42 -0
- package/dist/training/cli/grader/report.js.map +1 -0
- package/dist/training/cli/grader/runtime.d.ts +5 -0
- package/dist/training/cli/grader/runtime.js +64 -0
- package/dist/training/cli/grader/runtime.js.map +1 -0
- package/dist/training/cli/grader/validate.d.ts +5 -0
- package/dist/training/cli/grader/validate.js +42 -0
- package/dist/training/cli/grader/validate.js.map +1 -0
- package/dist/training/cli/index.d.ts +6 -0
- package/dist/training/cli/index.js +4 -0
- package/dist/training/cli/index.js.map +1 -0
- package/dist/training/cli/launch.d.ts +3 -0
- package/dist/training/cli/launch.js +17 -0
- package/dist/training/cli/launch.js.map +1 -0
- package/dist/training/cli/native.d.ts +235 -0
- package/dist/training/cli/native.js +83 -0
- package/dist/training/cli/native.js.map +1 -0
- package/dist/training/cli/output.d.ts +21 -0
- package/dist/training/cli/output.js +95 -0
- package/dist/training/cli/output.js.map +1 -0
- package/dist/training/cli/progress/activity.d.ts +11 -0
- package/dist/training/cli/progress/activity.js +45 -0
- package/dist/training/cli/progress/activity.js.map +1 -0
- package/dist/training/cli/progress/failure.d.ts +37 -0
- package/dist/training/cli/progress/failure.js +48 -0
- package/dist/training/cli/progress/failure.js.map +1 -0
- package/dist/training/cli/progress/generations.d.ts +11 -0
- package/dist/training/cli/progress/generations.js +50 -0
- package/dist/training/cli/progress/generations.js.map +1 -0
- package/dist/training/cli/progress/index.d.ts +6 -0
- package/dist/training/cli/progress/index.js +5 -0
- package/dist/training/cli/progress/index.js.map +1 -0
- package/dist/training/cli/progress/journal.d.ts +38 -0
- package/dist/training/cli/progress/journal.js +101 -0
- package/dist/training/cli/progress/journal.js.map +1 -0
- package/dist/training/cli/report/diff.d.ts +2 -0
- package/dist/training/cli/report/diff.js +24 -0
- package/dist/training/cli/report/diff.js.map +1 -0
- package/dist/training/cli/report/export.d.ts +9 -0
- package/dist/training/cli/report/export.js +81 -0
- package/dist/training/cli/report/export.js.map +1 -0
- package/dist/training/cli/report/fixture.d.ts +9 -0
- package/dist/training/cli/report/fixture.js +65 -0
- package/dist/training/cli/report/fixture.js.map +1 -0
- package/dist/training/cli/report/index.d.ts +1 -0
- package/dist/training/cli/report/index.js +2 -0
- package/dist/training/cli/report/index.js.map +1 -0
- package/dist/training/cli/report/markdown.d.ts +2 -0
- package/dist/training/cli/report/markdown.js +60 -0
- package/dist/training/cli/report/markdown.js.map +1 -0
- package/dist/training/cli/report/summary.d.ts +130 -0
- package/dist/training/cli/report/summary.js +165 -0
- package/dist/training/cli/report/summary.js.map +1 -0
- package/dist/training/cli/spawnfile/context.d.ts +75 -0
- package/dist/training/cli/spawnfile/context.js +78 -0
- package/dist/training/cli/spawnfile/context.js.map +1 -0
- package/dist/training/cli/spawnfile/index.d.ts +5 -0
- package/dist/training/cli/spawnfile/index.js +4 -0
- package/dist/training/cli/spawnfile/index.js.map +1 -0
- package/dist/training/cli/spawnfile/moments.d.ts +44 -0
- package/dist/training/cli/spawnfile/moments.js +71 -0
- package/dist/training/cli/spawnfile/moments.js.map +1 -0
- package/dist/training/cli/spawnfile/selection.d.ts +83 -0
- package/dist/training/cli/spawnfile/selection.js +54 -0
- package/dist/training/cli/spawnfile/selection.js.map +1 -0
- package/dist/training/cli/training/cache.d.ts +4 -0
- package/dist/training/cli/training/cache.js +43 -0
- package/dist/training/cli/training/cache.js.map +1 -0
- package/dist/training/cli/training/checkpoint.d.ts +19 -0
- package/dist/training/cli/training/checkpoint.js +62 -0
- package/dist/training/cli/training/checkpoint.js.map +1 -0
- package/dist/training/cli/training/comparison.d.ts +8 -0
- package/dist/training/cli/training/comparison.js +22 -0
- package/dist/training/cli/training/comparison.js.map +1 -0
- package/dist/training/cli/training/configuration.d.ts +10 -0
- package/dist/training/cli/training/configuration.js +81 -0
- package/dist/training/cli/training/configuration.js.map +1 -0
- package/dist/training/cli/training/deadlines.d.ts +7 -0
- package/dist/training/cli/training/deadlines.js +14 -0
- package/dist/training/cli/training/deadlines.js.map +1 -0
- package/dist/training/cli/training/index.d.ts +8 -0
- package/dist/training/cli/training/index.js +9 -0
- package/dist/training/cli/training/index.js.map +1 -0
- package/dist/training/cli/training/inference.d.ts +48 -0
- package/dist/training/cli/training/inference.js +96 -0
- package/dist/training/cli/training/inference.js.map +1 -0
- package/dist/training/cli/training/optimizer.d.ts +33 -0
- package/dist/training/cli/training/optimizer.js +92 -0
- package/dist/training/cli/training/optimizer.js.map +1 -0
- package/dist/training/cli/training/output-lock.d.ts +2 -0
- package/dist/training/cli/training/output-lock.js +30 -0
- package/dist/training/cli/training/output-lock.js.map +1 -0
- package/dist/training/cli/types.d.ts +101 -0
- package/dist/training/cli/types.js +2 -0
- package/dist/training/cli/types.js.map +1 -0
- package/dist/training/cli/viewer.d.ts +40 -0
- package/dist/training/cli/viewer.js +210 -0
- package/dist/training/cli/viewer.js.map +1 -0
- package/dist/training/contracts/atomic-feedback.d.ts +2 -0
- package/dist/training/contracts/atomic-feedback.js +2 -0
- package/dist/training/contracts/atomic-feedback.js.map +1 -0
- package/dist/training/contracts/dataset.d.ts +174 -0
- package/dist/training/contracts/dataset.js +2 -0
- package/dist/training/contracts/dataset.js.map +1 -0
- package/dist/training/contracts/evaluation.d.ts +256 -0
- package/dist/training/contracts/evaluation.js +41 -0
- package/dist/training/contracts/evaluation.js.map +1 -0
- package/dist/training/contracts/execution.d.ts +363 -0
- package/dist/training/contracts/execution.js +15 -0
- package/dist/training/contracts/execution.js.map +1 -0
- package/dist/training/contracts/grader-providers.d.ts +40 -0
- package/dist/training/contracts/grader-providers.js +27 -0
- package/dist/training/contracts/grader-providers.js.map +1 -0
- package/dist/training/contracts/grading.d.ts +74 -0
- package/dist/training/contracts/grading.js +2 -0
- package/dist/training/contracts/grading.js.map +1 -0
- package/dist/training/contracts/index.d.ts +16 -0
- package/dist/training/contracts/index.js +8 -0
- package/dist/training/contracts/index.js.map +1 -0
- package/dist/training/contracts/inference-grants.d.ts +34 -0
- package/dist/training/contracts/inference-grants.js +2 -0
- package/dist/training/contracts/inference-grants.js.map +1 -0
- package/dist/training/contracts/inference-registry.d.ts +38 -0
- package/dist/training/contracts/inference-registry.js +7 -0
- package/dist/training/contracts/inference-registry.js.map +1 -0
- package/dist/training/contracts/training-config.d.ts +95 -0
- package/dist/training/contracts/training-config.js +15 -0
- package/dist/training/contracts/training-config.js.map +1 -0
- package/dist/training/contracts/trial-slot.d.ts +12 -0
- package/dist/training/contracts/trial-slot.js +2 -0
- package/dist/training/contracts/trial-slot.js.map +1 -0
- package/dist/training/dspy/AGENTS.md +29 -0
- package/dist/training/dspy/README.md +206 -0
- package/dist/training/dspy/protocol.md +311 -0
- package/dist/training/dspy/pyproject.toml +36 -0
- package/dist/training/dspy/requirements.lock +69 -0
- package/dist/training/dspy/spawnfile_training_dspy/AGENTS.md +63 -0
- package/dist/training/dspy/spawnfile_training_dspy/__init__.py +6 -0
- package/dist/training/dspy/spawnfile_training_dspy/checkpoint.py +146 -0
- package/dist/training/dspy/spawnfile_training_dspy/legacy_wire_aliases.json +90 -0
- package/dist/training/dspy/spawnfile_training_dspy/main.py +36 -0
- package/dist/training/dspy/spawnfile_training_dspy/native.py +89 -0
- package/dist/training/dspy/spawnfile_training_dspy/optimizer.py +74 -0
- package/dist/training/dspy/spawnfile_training_dspy/proposal_history.py +275 -0
- package/dist/training/dspy/spawnfile_training_dspy/protocol.py +149 -0
- package/dist/training/dspy/spawnfile_training_dspy/reflection.py +199 -0
- package/dist/training/dspy/spawnfile_training_dspy/search/AGENTS.md +41 -0
- package/dist/training/dspy/spawnfile_training_dspy/search/__init__.py +1 -0
- package/dist/training/dspy/spawnfile_training_dspy/search/behavior_change.py +152 -0
- package/dist/training/dspy/spawnfile_training_dspy/search/boundaries.py +38 -0
- package/dist/training/dspy/spawnfile_training_dspy/search/instruction_text.py +47 -0
- package/dist/training/dspy/spawnfile_training_dspy/search/population.py +213 -0
- package/dist/training/dspy/spawnfile_training_dspy/search/selection.py +24 -0
- package/dist/training/dspy/spawnfile_training_dspy/search/variants.py +83 -0
- package/dist/training/dspy/spawnfile_training_dspy/session.py +116 -0
- package/dist/training/dspy/spawnfile_training_dspy/transport.py +129 -0
- package/dist/training/evaluation/atomic-grader.d.ts +5 -0
- package/dist/training/evaluation/atomic-grader.js +55 -0
- package/dist/training/evaluation/atomic-grader.js.map +1 -0
- package/dist/training/evaluation/embedding/index.d.ts +1 -0
- package/dist/training/evaluation/embedding/index.js +2 -0
- package/dist/training/evaluation/embedding/index.js.map +1 -0
- package/dist/training/evaluation/embedding/similarity.d.ts +4 -0
- package/dist/training/evaluation/embedding/similarity.js +38 -0
- package/dist/training/evaluation/embedding/similarity.js.map +1 -0
- package/dist/training/evaluation/engine.d.ts +6 -0
- package/dist/training/evaluation/engine.js +191 -0
- package/dist/training/evaluation/engine.js.map +1 -0
- package/dist/training/evaluation/grader.d.ts +4 -0
- package/dist/training/evaluation/grader.js +67 -0
- package/dist/training/evaluation/grader.js.map +1 -0
- package/dist/training/evaluation/index.d.ts +10 -0
- package/dist/training/evaluation/index.js +9 -0
- package/dist/training/evaluation/index.js.map +1 -0
- package/dist/training/evaluation/logic/index.d.ts +1 -0
- package/dist/training/evaluation/logic/index.js +2 -0
- package/dist/training/evaluation/logic/index.js.map +1 -0
- package/dist/training/evaluation/logic/results.d.ts +7 -0
- package/dist/training/evaluation/logic/results.js +70 -0
- package/dist/training/evaluation/logic/results.js.map +1 -0
- package/dist/training/evaluation/mechanical/index.d.ts +6 -0
- package/dist/training/evaluation/mechanical/index.js +7 -0
- package/dist/training/evaluation/mechanical/index.js.map +1 -0
- package/dist/training/evaluation/mechanical/predicates.d.ts +2 -0
- package/dist/training/evaluation/mechanical/predicates.js +27 -0
- package/dist/training/evaluation/mechanical/predicates.js.map +1 -0
- package/dist/training/evaluation/mechanical/sequence.d.ts +3 -0
- package/dist/training/evaluation/mechanical/sequence.js +75 -0
- package/dist/training/evaluation/mechanical/sequence.js.map +1 -0
- package/dist/training/evaluation/mechanical/text-checks.d.ts +3 -0
- package/dist/training/evaluation/mechanical/text-checks.js +65 -0
- package/dist/training/evaluation/mechanical/text-checks.js.map +1 -0
- package/dist/training/evaluation/mechanical/text-values.d.ts +7 -0
- package/dist/training/evaluation/mechanical/text-values.js +28 -0
- package/dist/training/evaluation/mechanical/text-values.js.map +1 -0
- package/dist/training/evaluation/mechanical/text.d.ts +3 -0
- package/dist/training/evaluation/mechanical/text.js +40 -0
- package/dist/training/evaluation/mechanical/text.js.map +1 -0
- package/dist/training/evaluation/mechanical/token-efficiency.d.ts +4 -0
- package/dist/training/evaluation/mechanical/token-efficiency.js +28 -0
- package/dist/training/evaluation/mechanical/token-efficiency.js.map +1 -0
- package/dist/training/evaluation/mechanical/tools.d.ts +3 -0
- package/dist/training/evaluation/mechanical/tools.js +39 -0
- package/dist/training/evaluation/mechanical/tools.js.map +1 -0
- package/dist/training/evaluation/registry.d.ts +4 -0
- package/dist/training/evaluation/registry.js +64 -0
- package/dist/training/evaluation/registry.js.map +1 -0
- package/dist/training/evaluation/repair.d.ts +14 -0
- package/dist/training/evaluation/repair.js +73 -0
- package/dist/training/evaluation/repair.js.map +1 -0
- package/dist/training/evaluation/schedule.d.ts +21 -0
- package/dist/training/evaluation/schedule.js +98 -0
- package/dist/training/evaluation/schedule.js.map +1 -0
- package/dist/training/evaluation/selectors.d.ts +22 -0
- package/dist/training/evaluation/selectors.js +61 -0
- package/dist/training/evaluation/selectors.js.map +1 -0
- package/dist/training/evaluation/semantic/attempts.d.ts +12 -0
- package/dist/training/evaluation/semantic/attempts.js +71 -0
- package/dist/training/evaluation/semantic/attempts.js.map +1 -0
- package/dist/training/evaluation/semantic/cache.d.ts +9 -0
- package/dist/training/evaluation/semantic/cache.js +40 -0
- package/dist/training/evaluation/semantic/cache.js.map +1 -0
- package/dist/training/evaluation/semantic/evidence-units.d.ts +14 -0
- package/dist/training/evaluation/semantic/evidence-units.js +58 -0
- package/dist/training/evaluation/semantic/evidence-units.js.map +1 -0
- package/dist/training/evaluation/semantic/failure.d.ts +30 -0
- package/dist/training/evaluation/semantic/failure.js +79 -0
- package/dist/training/evaluation/semantic/failure.js.map +1 -0
- package/dist/training/evaluation/semantic/index.d.ts +5 -0
- package/dist/training/evaluation/semantic/index.js +5 -0
- package/dist/training/evaluation/semantic/index.js.map +1 -0
- package/dist/training/evaluation/semantic/judgment.d.ts +21 -0
- package/dist/training/evaluation/semantic/judgment.js +50 -0
- package/dist/training/evaluation/semantic/judgment.js.map +1 -0
- package/dist/training/evaluation/semantic/protocol.d.ts +118 -0
- package/dist/training/evaluation/semantic/protocol.js +48 -0
- package/dist/training/evaluation/semantic/protocol.js.map +1 -0
- package/dist/training/evaluation/semantic/rubric.d.ts +3 -0
- package/dist/training/evaluation/semantic/rubric.js +202 -0
- package/dist/training/evaluation/semantic/rubric.js.map +1 -0
- package/dist/training/evaluation/semantic/templates.d.ts +7 -0
- package/dist/training/evaluation/semantic/templates.js +75 -0
- package/dist/training/evaluation/semantic/templates.js.map +1 -0
- package/dist/training/experiments/comparison/index.d.ts +2 -0
- package/dist/training/experiments/comparison/index.js +2 -0
- package/dist/training/experiments/comparison/index.js.map +1 -0
- package/dist/training/experiments/comparison/run.d.ts +3 -0
- package/dist/training/experiments/comparison/run.js +92 -0
- package/dist/training/experiments/comparison/run.js.map +1 -0
- package/dist/training/experiments/comparison/types.d.ts +39 -0
- package/dist/training/experiments/comparison/types.js +2 -0
- package/dist/training/experiments/comparison/types.js.map +1 -0
- package/dist/training/experiments/concurrency/index.d.ts +2 -0
- package/dist/training/experiments/concurrency/index.js +2 -0
- package/dist/training/experiments/concurrency/index.js.map +1 -0
- package/dist/training/experiments/concurrency/slots.d.ts +50 -0
- package/dist/training/experiments/concurrency/slots.js +120 -0
- package/dist/training/experiments/concurrency/slots.js.map +1 -0
- package/dist/training/experiments/costs/assumptions.d.ts +4 -0
- package/dist/training/experiments/costs/assumptions.js +18 -0
- package/dist/training/experiments/costs/assumptions.js.map +1 -0
- package/dist/training/experiments/costs/catalog.d.ts +3 -0
- package/dist/training/experiments/costs/catalog.js +24 -0
- package/dist/training/experiments/costs/catalog.js.map +1 -0
- package/dist/training/experiments/costs/config.d.ts +6 -0
- package/dist/training/experiments/costs/config.js +22 -0
- package/dist/training/experiments/costs/config.js.map +1 -0
- package/dist/training/experiments/costs/decisions.d.ts +5 -0
- package/dist/training/experiments/costs/decisions.js +48 -0
- package/dist/training/experiments/costs/decisions.js.map +1 -0
- package/dist/training/experiments/costs/index.d.ts +9 -0
- package/dist/training/experiments/costs/index.js +7 -0
- package/dist/training/experiments/costs/index.js.map +1 -0
- package/dist/training/experiments/costs/planning.d.ts +71 -0
- package/dist/training/experiments/costs/planning.js +140 -0
- package/dist/training/experiments/costs/planning.js.map +1 -0
- package/dist/training/experiments/costs/report.d.ts +71 -0
- package/dist/training/experiments/costs/report.js +269 -0
- package/dist/training/experiments/costs/report.js.map +1 -0
- package/dist/training/experiments/costs/types.d.ts +53 -0
- package/dist/training/experiments/costs/types.js +2 -0
- package/dist/training/experiments/costs/types.js.map +1 -0
- package/dist/training/experiments/costs/valuation.d.ts +3 -0
- package/dist/training/experiments/costs/valuation.js +124 -0
- package/dist/training/experiments/costs/valuation.js.map +1 -0
- package/dist/training/experiments/index.d.ts +12 -0
- package/dist/training/experiments/index.js +12 -0
- package/dist/training/experiments/index.js.map +1 -0
- package/dist/training/experiments/repair/import.d.ts +21 -0
- package/dist/training/experiments/repair/import.js +104 -0
- package/dist/training/experiments/repair/import.js.map +1 -0
- package/dist/training/experiments/repair/index.d.ts +6 -0
- package/dist/training/experiments/repair/index.js +5 -0
- package/dist/training/experiments/repair/index.js.map +1 -0
- package/dist/training/experiments/repair/measurement.d.ts +14 -0
- package/dist/training/experiments/repair/measurement.js +101 -0
- package/dist/training/experiments/repair/measurement.js.map +1 -0
- package/dist/training/experiments/repair/projection.d.ts +3 -0
- package/dist/training/experiments/repair/projection.js +63 -0
- package/dist/training/experiments/repair/projection.js.map +1 -0
- package/dist/training/experiments/repair/receipt.d.ts +43 -0
- package/dist/training/experiments/repair/receipt.js +52 -0
- package/dist/training/experiments/repair/receipt.js.map +1 -0
- package/dist/training/experiments/repair/record.d.ts +4 -0
- package/dist/training/experiments/repair/record.js +24 -0
- package/dist/training/experiments/repair/record.js.map +1 -0
- package/dist/training/experiments/replay/fixture.d.ts +10 -0
- package/dist/training/experiments/replay/fixture.js +53 -0
- package/dist/training/experiments/replay/fixture.js.map +1 -0
- package/dist/training/experiments/replay/index.d.ts +3 -0
- package/dist/training/experiments/replay/index.js +3 -0
- package/dist/training/experiments/replay/index.js.map +1 -0
- package/dist/training/experiments/replay/rescore.d.ts +3 -0
- package/dist/training/experiments/replay/rescore.js +63 -0
- package/dist/training/experiments/replay/rescore.js.map +1 -0
- package/dist/training/experiments/replay/schema.d.ts +408 -0
- package/dist/training/experiments/replay/schema.js +126 -0
- package/dist/training/experiments/replay/schema.js.map +1 -0
- package/dist/training/experiments/replay/source.d.ts +6 -0
- package/dist/training/experiments/replay/source.js +35 -0
- package/dist/training/experiments/replay/source.js.map +1 -0
- package/dist/training/experiments/replay/subject-usage.d.ts +3 -0
- package/dist/training/experiments/replay/subject-usage.js +18 -0
- package/dist/training/experiments/replay/subject-usage.js.map +1 -0
- package/dist/training/experiments/replay/types.d.ts +63 -0
- package/dist/training/experiments/replay/types.js +2 -0
- package/dist/training/experiments/replay/types.js.map +1 -0
- package/dist/training/experiments/training/checkpoint.d.ts +51 -0
- package/dist/training/experiments/training/checkpoint.js +200 -0
- package/dist/training/experiments/training/checkpoint.js.map +1 -0
- package/dist/training/experiments/training/criteria.d.ts +3 -0
- package/dist/training/experiments/training/criteria.js +23 -0
- package/dist/training/experiments/training/criteria.js.map +1 -0
- package/dist/training/experiments/training/feedback.d.ts +15 -0
- package/dist/training/experiments/training/feedback.js +133 -0
- package/dist/training/experiments/training/feedback.js.map +1 -0
- package/dist/training/experiments/training/held-out.d.ts +24 -0
- package/dist/training/experiments/training/held-out.js +41 -0
- package/dist/training/experiments/training/held-out.js.map +1 -0
- package/dist/training/experiments/training/index.d.ts +7 -0
- package/dist/training/experiments/training/index.js +7 -0
- package/dist/training/experiments/training/index.js.map +1 -0
- package/dist/training/experiments/training/partition.d.ts +20 -0
- package/dist/training/experiments/training/partition.js +24 -0
- package/dist/training/experiments/training/partition.js.map +1 -0
- package/dist/training/experiments/training/recovery.d.ts +17 -0
- package/dist/training/experiments/training/recovery.js +54 -0
- package/dist/training/experiments/training/recovery.js.map +1 -0
- package/dist/training/experiments/training/retry.d.ts +14 -0
- package/dist/training/experiments/training/retry.js +16 -0
- package/dist/training/experiments/training/retry.js.map +1 -0
- package/dist/training/experiments/training/scoring.d.ts +4 -0
- package/dist/training/experiments/training/scoring.js +42 -0
- package/dist/training/experiments/training/scoring.js.map +1 -0
- package/dist/training/experiments/training/training.d.ts +85 -0
- package/dist/training/experiments/training/training.js +208 -0
- package/dist/training/experiments/training/training.js.map +1 -0
- package/dist/training/experiments/trials/bindings.d.ts +4 -0
- package/dist/training/experiments/trials/bindings.js +46 -0
- package/dist/training/experiments/trials/bindings.js.map +1 -0
- package/dist/training/experiments/trials/closure.d.ts +2 -0
- package/dist/training/experiments/trials/closure.js +26 -0
- package/dist/training/experiments/trials/closure.js.map +1 -0
- package/dist/training/experiments/trials/control.d.ts +13 -0
- package/dist/training/experiments/trials/control.js +74 -0
- package/dist/training/experiments/trials/control.js.map +1 -0
- package/dist/training/experiments/trials/diagnostics.d.ts +25 -0
- package/dist/training/experiments/trials/diagnostics.js +99 -0
- package/dist/training/experiments/trials/diagnostics.js.map +1 -0
- package/dist/training/experiments/trials/files.d.ts +8 -0
- package/dist/training/experiments/trials/files.js +113 -0
- package/dist/training/experiments/trials/files.js.map +1 -0
- package/dist/training/experiments/trials/fixture.d.ts +13 -0
- package/dist/training/experiments/trials/fixture.js +40 -0
- package/dist/training/experiments/trials/fixture.js.map +1 -0
- package/dist/training/experiments/trials/index.d.ts +5 -0
- package/dist/training/experiments/trials/index.js +5 -0
- package/dist/training/experiments/trials/index.js.map +1 -0
- package/dist/training/experiments/trials/live.d.ts +20 -0
- package/dist/training/experiments/trials/live.js +68 -0
- package/dist/training/experiments/trials/live.js.map +1 -0
- package/dist/training/experiments/trials/paths.d.ts +5 -0
- package/dist/training/experiments/trials/paths.js +90 -0
- package/dist/training/experiments/trials/paths.js.map +1 -0
- package/dist/training/experiments/trials/preflight.d.ts +3 -0
- package/dist/training/experiments/trials/preflight.js +23 -0
- package/dist/training/experiments/trials/preflight.js.map +1 -0
- package/dist/training/experiments/trials/stored.d.ts +23 -0
- package/dist/training/experiments/trials/stored.js +46 -0
- package/dist/training/experiments/trials/stored.js.map +1 -0
- package/dist/training/experiments/trials/subject-usage.d.ts +7 -0
- package/dist/training/experiments/trials/subject-usage.js +66 -0
- package/dist/training/experiments/trials/subject-usage.js.map +1 -0
- package/dist/training/experiments/trials/trial.d.ts +11 -0
- package/dist/training/experiments/trials/trial.js +361 -0
- package/dist/training/experiments/trials/trial.js.map +1 -0
- package/dist/training/experiments/trials/types.d.ts +124 -0
- package/dist/training/experiments/trials/types.js +2 -0
- package/dist/training/experiments/trials/types.js.map +1 -0
- package/dist/training/graders/admission.d.ts +11 -0
- package/dist/training/graders/admission.js +22 -0
- package/dist/training/graders/admission.js.map +1 -0
- package/dist/training/graders/aggregation.d.ts +30 -0
- package/dist/training/graders/aggregation.js +71 -0
- package/dist/training/graders/aggregation.js.map +1 -0
- package/dist/training/graders/atomic/aggregate.d.ts +10 -0
- package/dist/training/graders/atomic/aggregate.js +31 -0
- package/dist/training/graders/atomic/aggregate.js.map +1 -0
- package/dist/training/graders/atomic/feedback.d.ts +3 -0
- package/dist/training/graders/atomic/feedback.js +16 -0
- package/dist/training/graders/atomic/feedback.js.map +1 -0
- package/dist/training/graders/atomic/format.d.ts +61 -0
- package/dist/training/graders/atomic/format.js +53 -0
- package/dist/training/graders/atomic/format.js.map +1 -0
- package/dist/training/graders/atomic/index.d.ts +7 -0
- package/dist/training/graders/atomic/index.js +6 -0
- package/dist/training/graders/atomic/index.js.map +1 -0
- package/dist/training/graders/atomic/runtime.d.ts +4 -0
- package/dist/training/graders/atomic/runtime.js +94 -0
- package/dist/training/graders/atomic/runtime.js.map +1 -0
- package/dist/training/graders/atomic/schema.d.ts +5 -0
- package/dist/training/graders/atomic/schema.js +37 -0
- package/dist/training/graders/atomic/schema.js.map +1 -0
- package/dist/training/graders/atomic/types.d.ts +68 -0
- package/dist/training/graders/atomic/types.js +2 -0
- package/dist/training/graders/atomic/types.js.map +1 -0
- package/dist/training/graders/calibration/escalation.d.ts +50 -0
- package/dist/training/graders/calibration/escalation.js +145 -0
- package/dist/training/graders/calibration/escalation.js.map +1 -0
- package/dist/training/graders/calibration/fitting.d.ts +28 -0
- package/dist/training/graders/calibration/fitting.js +121 -0
- package/dist/training/graders/calibration/fitting.js.map +1 -0
- package/dist/training/graders/calibration/index.d.ts +7 -0
- package/dist/training/graders/calibration/index.js +5 -0
- package/dist/training/graders/calibration/index.js.map +1 -0
- package/dist/training/graders/calibration/metrics.d.ts +31 -0
- package/dist/training/graders/calibration/metrics.js +145 -0
- package/dist/training/graders/calibration/metrics.js.map +1 -0
- package/dist/training/graders/calibration/selection.d.ts +13 -0
- package/dist/training/graders/calibration/selection.js +30 -0
- package/dist/training/graders/calibration/selection.js.map +1 -0
- package/dist/training/graders/calibration/types.d.ts +68 -0
- package/dist/training/graders/calibration/types.js +2 -0
- package/dist/training/graders/calibration/types.js.map +1 -0
- package/dist/training/graders/consensus.d.ts +11 -0
- package/dist/training/graders/consensus.js +29 -0
- package/dist/training/graders/consensus.js.map +1 -0
- package/dist/training/graders/decomposition/evidence.d.ts +21 -0
- package/dist/training/graders/decomposition/evidence.js +107 -0
- package/dist/training/graders/decomposition/evidence.js.map +1 -0
- package/dist/training/graders/decomposition/format.d.ts +265 -0
- package/dist/training/graders/decomposition/format.js +98 -0
- package/dist/training/graders/decomposition/format.js.map +1 -0
- package/dist/training/graders/decomposition/index.d.ts +10 -0
- package/dist/training/graders/decomposition/index.js +6 -0
- package/dist/training/graders/decomposition/index.js.map +1 -0
- package/dist/training/graders/decomposition/instantiate.d.ts +26 -0
- package/dist/training/graders/decomposition/instantiate.js +129 -0
- package/dist/training/graders/decomposition/instantiate.js.map +1 -0
- package/dist/training/graders/decomposition/location.d.ts +20 -0
- package/dist/training/graders/decomposition/location.js +75 -0
- package/dist/training/graders/decomposition/location.js.map +1 -0
- package/dist/training/graders/decomposition/metadata.d.ts +8 -0
- package/dist/training/graders/decomposition/metadata.js +14 -0
- package/dist/training/graders/decomposition/metadata.js.map +1 -0
- package/dist/training/graders/decomposition/spans.d.ts +6 -0
- package/dist/training/graders/decomposition/spans.js +18 -0
- package/dist/training/graders/decomposition/spans.js.map +1 -0
- package/dist/training/graders/decomposition/transport.d.ts +76 -0
- package/dist/training/graders/decomposition/transport.js +14 -0
- package/dist/training/graders/decomposition/transport.js.map +1 -0
- package/dist/training/graders/format/binding.d.ts +3 -0
- package/dist/training/graders/format/binding.js +37 -0
- package/dist/training/graders/format/binding.js.map +1 -0
- package/dist/training/graders/format/common.d.ts +13 -0
- package/dist/training/graders/format/common.js +77 -0
- package/dist/training/graders/format/common.js.map +1 -0
- package/dist/training/graders/format/grader.d.ts +236 -0
- package/dist/training/graders/format/grader.js +161 -0
- package/dist/training/graders/format/grader.js.map +1 -0
- package/dist/training/graders/format/index.d.ts +11 -0
- package/dist/training/graders/format/index.js +8 -0
- package/dist/training/graders/format/index.js.map +1 -0
- package/dist/training/graders/format/label-import.d.ts +42 -0
- package/dist/training/graders/format/label-import.js +44 -0
- package/dist/training/graders/format/label-import.js.map +1 -0
- package/dist/training/graders/format/labels.d.ts +45 -0
- package/dist/training/graders/format/labels.js +109 -0
- package/dist/training/graders/format/labels.js.map +1 -0
- package/dist/training/graders/format/questions.d.ts +67 -0
- package/dist/training/graders/format/questions.js +104 -0
- package/dist/training/graders/format/questions.js.map +1 -0
- package/dist/training/graders/format/row-regex.d.ts +2 -0
- package/dist/training/graders/format/row-regex.js +56 -0
- package/dist/training/graders/format/row-regex.js.map +1 -0
- package/dist/training/graders/format/rows.d.ts +15 -0
- package/dist/training/graders/format/rows.js +31 -0
- package/dist/training/graders/format/rows.js.map +1 -0
- package/dist/training/graders/format/sentences.d.ts +2 -0
- package/dist/training/graders/format/sentences.js +30 -0
- package/dist/training/graders/format/sentences.js.map +1 -0
- package/dist/training/graders/identity-calibration.d.ts +4 -0
- package/dist/training/graders/identity-calibration.js +15 -0
- package/dist/training/graders/identity-calibration.js.map +1 -0
- package/dist/training/graders/identity.d.ts +119 -0
- package/dist/training/graders/identity.js +119 -0
- package/dist/training/graders/identity.js.map +1 -0
- package/dist/training/graders/index.d.ts +23 -0
- package/dist/training/graders/index.js +14 -0
- package/dist/training/graders/index.js.map +1 -0
- package/dist/training/graders/instances.d.ts +21 -0
- package/dist/training/graders/instances.js +121 -0
- package/dist/training/graders/instances.js.map +1 -0
- package/dist/training/graders/labeling/import.d.ts +8 -0
- package/dist/training/graders/labeling/import.js +51 -0
- package/dist/training/graders/labeling/import.js.map +1 -0
- package/dist/training/graders/labeling/index.d.ts +3 -0
- package/dist/training/graders/labeling/index.js +3 -0
- package/dist/training/graders/labeling/index.js.map +1 -0
- package/dist/training/graders/labeling/planning.d.ts +39 -0
- package/dist/training/graders/labeling/planning.js +117 -0
- package/dist/training/graders/labeling/planning.js.map +1 -0
- package/dist/training/graders/references/council.d.ts +2 -0
- package/dist/training/graders/references/council.js +182 -0
- package/dist/training/graders/references/council.js.map +1 -0
- package/dist/training/graders/references/index.d.ts +5 -0
- package/dist/training/graders/references/index.js +5 -0
- package/dist/training/graders/references/index.js.map +1 -0
- package/dist/training/graders/references/metrics.d.ts +133 -0
- package/dist/training/graders/references/metrics.js +113 -0
- package/dist/training/graders/references/metrics.js.map +1 -0
- package/dist/training/graders/references/policy.d.ts +10 -0
- package/dist/training/graders/references/policy.js +16 -0
- package/dist/training/graders/references/policy.js.map +1 -0
- package/dist/training/graders/references/protocol.d.ts +11 -0
- package/dist/training/graders/references/protocol.js +63 -0
- package/dist/training/graders/references/protocol.js.map +1 -0
- package/dist/training/graders/references/types.d.ts +92 -0
- package/dist/training/graders/references/types.js +2 -0
- package/dist/training/graders/references/types.js.map +1 -0
- package/dist/training/graders/rows.d.ts +14 -0
- package/dist/training/graders/rows.js +39 -0
- package/dist/training/graders/rows.js.map +1 -0
- package/dist/training/graders/runtime/accounting.d.ts +10 -0
- package/dist/training/graders/runtime/accounting.js +43 -0
- package/dist/training/graders/runtime/accounting.js.map +1 -0
- package/dist/training/graders/runtime/grade.d.ts +3 -0
- package/dist/training/graders/runtime/grade.js +161 -0
- package/dist/training/graders/runtime/grade.js.map +1 -0
- package/dist/training/graders/runtime/identity.d.ts +6 -0
- package/dist/training/graders/runtime/identity.js +24 -0
- package/dist/training/graders/runtime/identity.js.map +1 -0
- package/dist/training/graders/runtime/index.d.ts +5 -0
- package/dist/training/graders/runtime/index.js +5 -0
- package/dist/training/graders/runtime/index.js.map +1 -0
- package/dist/training/graders/runtime/schema.d.ts +29 -0
- package/dist/training/graders/runtime/schema.js +27 -0
- package/dist/training/graders/runtime/schema.js.map +1 -0
- package/dist/training/graders/runtime/teacher.d.ts +8 -0
- package/dist/training/graders/runtime/teacher.js +43 -0
- package/dist/training/graders/runtime/teacher.js.map +1 -0
- package/dist/training/graders/runtime/types.d.ts +48 -0
- package/dist/training/graders/runtime/types.js +2 -0
- package/dist/training/graders/runtime/types.js.map +1 -0
- package/dist/training/graders/traps/generate.d.ts +6 -0
- package/dist/training/graders/traps/generate.js +94 -0
- package/dist/training/graders/traps/generate.js.map +1 -0
- package/dist/training/graders/traps/index.d.ts +3 -0
- package/dist/training/graders/traps/index.js +3 -0
- package/dist/training/graders/traps/index.js.map +1 -0
- package/dist/training/graders/traps/mutations.d.ts +10 -0
- package/dist/training/graders/traps/mutations.js +48 -0
- package/dist/training/graders/traps/mutations.js.map +1 -0
- package/dist/training/graders/traps/types.d.ts +37 -0
- package/dist/training/graders/traps/types.js +2 -0
- package/dist/training/graders/traps/types.js.map +1 -0
- package/dist/training/graders/traps/validation.d.ts +4 -0
- package/dist/training/graders/traps/validation.js +118 -0
- package/dist/training/graders/traps/validation.js.map +1 -0
- package/dist/training/index.d.ts +17 -0
- package/dist/training/index.js +16 -0
- package/dist/training/index.js.map +1 -0
- package/dist/training/inference/claude/index.d.ts +1 -0
- package/dist/training/inference/claude/index.js +2 -0
- package/dist/training/inference/claude/index.js.map +1 -0
- package/dist/training/inference/claude/protocol.d.ts +16 -0
- package/dist/training/inference/claude/protocol.js +72 -0
- package/dist/training/inference/claude/protocol.js.map +1 -0
- package/dist/training/inference/claude/provider.d.ts +9 -0
- package/dist/training/inference/claude/provider.js +153 -0
- package/dist/training/inference/claude/provider.js.map +1 -0
- package/dist/training/inference/claude/refusal.d.ts +2 -0
- package/dist/training/inference/claude/refusal.js +37 -0
- package/dist/training/inference/claude/refusal.js.map +1 -0
- package/dist/training/inference/decision/http.d.ts +14 -0
- package/dist/training/inference/decision/http.js +57 -0
- package/dist/training/inference/decision/http.js.map +1 -0
- package/dist/training/inference/decision/index.d.ts +6 -0
- package/dist/training/inference/decision/index.js +6 -0
- package/dist/training/inference/decision/index.js.map +1 -0
- package/dist/training/inference/decision/jev.d.ts +4 -0
- package/dist/training/inference/decision/jev.js +163 -0
- package/dist/training/inference/decision/jev.js.map +1 -0
- package/dist/training/inference/decision/ollama.d.ts +9 -0
- package/dist/training/inference/decision/ollama.js +101 -0
- package/dist/training/inference/decision/ollama.js.map +1 -0
- package/dist/training/inference/decision/prompt.d.ts +6 -0
- package/dist/training/inference/decision/prompt.js +19 -0
- package/dist/training/inference/decision/prompt.js.map +1 -0
- package/dist/training/inference/decision/teacher.d.ts +7 -0
- package/dist/training/inference/decision/teacher.js +61 -0
- package/dist/training/inference/decision/teacher.js.map +1 -0
- package/dist/training/inference/decision/typesafe/index.d.ts +9 -0
- package/dist/training/inference/decision/typesafe/index.js +82 -0
- package/dist/training/inference/decision/typesafe/index.js.map +1 -0
- package/dist/training/inference/decision/typesafe/transport.d.ts +44 -0
- package/dist/training/inference/decision/typesafe/transport.js +151 -0
- package/dist/training/inference/decision/typesafe/transport.js.map +1 -0
- package/dist/training/inference/decision/typesafe/wire.d.ts +25 -0
- package/dist/training/inference/decision/typesafe/wire.js +49 -0
- package/dist/training/inference/decision/typesafe/wire.js.map +1 -0
- package/dist/training/inference/diagnostic.d.ts +3 -0
- package/dist/training/inference/diagnostic.js +14 -0
- package/dist/training/inference/diagnostic.js.map +1 -0
- package/dist/training/inference/evaluator/index.d.ts +3 -0
- package/dist/training/inference/evaluator/index.js +4 -0
- package/dist/training/inference/evaluator/index.js.map +1 -0
- package/dist/training/inference/evaluator/ledger.d.ts +29 -0
- package/dist/training/inference/evaluator/ledger.js +70 -0
- package/dist/training/inference/evaluator/ledger.js.map +1 -0
- package/dist/training/inference/evaluator/policy.d.ts +39 -0
- package/dist/training/inference/evaluator/policy.js +67 -0
- package/dist/training/inference/evaluator/policy.js.map +1 -0
- package/dist/training/inference/evaluator/provider.d.ts +6 -0
- package/dist/training/inference/evaluator/provider.js +45 -0
- package/dist/training/inference/evaluator/provider.js.map +1 -0
- package/dist/training/inference/failure.d.ts +27 -0
- package/dist/training/inference/failure.js +36 -0
- package/dist/training/inference/failure.js.map +1 -0
- package/dist/training/inference/index.d.ts +8 -0
- package/dist/training/inference/index.js +9 -0
- package/dist/training/inference/index.js.map +1 -0
- package/dist/training/inference/ollama/context.d.ts +2 -0
- package/dist/training/inference/ollama/context.js +28 -0
- package/dist/training/inference/ollama/context.js.map +1 -0
- package/dist/training/inference/ollama/index.d.ts +2 -0
- package/dist/training/inference/ollama/index.js +3 -0
- package/dist/training/inference/ollama/index.js.map +1 -0
- package/dist/training/inference/ollama/provider.d.ts +12 -0
- package/dist/training/inference/ollama/provider.js +100 -0
- package/dist/training/inference/ollama/provider.js.map +1 -0
- package/dist/training/inference/request-cap.d.ts +7 -0
- package/dist/training/inference/request-cap.js +22 -0
- package/dist/training/inference/request-cap.js.map +1 -0
- package/dist/training/inference/response-schema.d.ts +9 -0
- package/dist/training/inference/response-schema.js +54 -0
- package/dist/training/inference/response-schema.js.map +1 -0
- package/dist/training/inference/routing.d.ts +17 -0
- package/dist/training/inference/routing.js +152 -0
- package/dist/training/inference/routing.js.map +1 -0
- package/dist/training/optimization/boundaries.d.ts +8 -0
- package/dist/training/optimization/boundaries.js +31 -0
- package/dist/training/optimization/boundaries.js.map +1 -0
- package/dist/training/optimization/bridge-fixtures.d.ts +13 -0
- package/dist/training/optimization/bridge-fixtures.js +40 -0
- package/dist/training/optimization/bridge-fixtures.js.map +1 -0
- package/dist/training/optimization/bridge-identity.d.ts +8 -0
- package/dist/training/optimization/bridge-identity.js +110 -0
- package/dist/training/optimization/bridge-identity.js.map +1 -0
- package/dist/training/optimization/checkpoint.d.ts +35 -0
- package/dist/training/optimization/checkpoint.js +187 -0
- package/dist/training/optimization/checkpoint.js.map +1 -0
- package/dist/training/optimization/evidence.d.ts +4 -0
- package/dist/training/optimization/evidence.js +65 -0
- package/dist/training/optimization/evidence.js.map +1 -0
- package/dist/training/optimization/host.d.ts +27 -0
- package/dist/training/optimization/host.js +144 -0
- package/dist/training/optimization/host.js.map +1 -0
- package/dist/training/optimization/index.d.ts +8 -0
- package/dist/training/optimization/index.js +6 -0
- package/dist/training/optimization/index.js.map +1 -0
- package/dist/training/optimization/inference.d.ts +4 -0
- package/dist/training/optimization/inference.js +33 -0
- package/dist/training/optimization/inference.js.map +1 -0
- package/dist/training/optimization/population-evidence.d.ts +4 -0
- package/dist/training/optimization/population-evidence.js +152 -0
- package/dist/training/optimization/population-evidence.js.map +1 -0
- package/dist/training/optimization/protocol.d.ts +532 -0
- package/dist/training/optimization/protocol.js +146 -0
- package/dist/training/optimization/protocol.js.map +1 -0
- package/dist/training/optimization/runner.d.ts +5 -0
- package/dist/training/optimization/runner.js +129 -0
- package/dist/training/optimization/runner.js.map +1 -0
- package/dist/training/optimization/testSupport.d.ts +9 -0
- package/dist/training/optimization/testSupport.js +29 -0
- package/dist/training/optimization/testSupport.js.map +1 -0
- package/dist/training/optimization/types.d.ts +73 -0
- package/dist/training/optimization/types.js +11 -0
- package/dist/training/optimization/types.js.map +1 -0
- package/dist/training/platform/index.d.ts +1 -0
- package/dist/training/platform/index.js +2 -0
- package/dist/training/platform/index.js.map +1 -0
- package/dist/training/platform/mutations/index.d.ts +5 -0
- package/dist/training/platform/mutations/index.js +10 -0
- package/dist/training/platform/mutations/index.js.map +1 -0
- package/dist/training/platform/process/index.d.ts +1 -0
- package/dist/training/platform/process/index.js +2 -0
- package/dist/training/platform/process/index.js.map +1 -0
- package/dist/training/platform/process/runner.d.ts +32 -0
- package/dist/training/platform/process/runner.js +190 -0
- package/dist/training/platform/process/runner.js.map +1 -0
- package/dist/training/providers/index.d.ts +2 -0
- package/dist/training/providers/index.js +3 -0
- package/dist/training/providers/index.js.map +1 -0
- package/dist/training/providers/policy.d.ts +23 -0
- package/dist/training/providers/policy.js +27 -0
- package/dist/training/providers/policy.js.map +1 -0
- package/dist/training/providers/registry.d.ts +2 -0
- package/dist/training/providers/registry.js +62 -0
- package/dist/training/providers/registry.js.map +1 -0
- package/dist/training/storage/cache/index.d.ts +2 -0
- package/dist/training/storage/cache/index.js +2 -0
- package/dist/training/storage/cache/index.js.map +1 -0
- package/dist/training/storage/cache/persistent.d.ts +12 -0
- package/dist/training/storage/cache/persistent.js +124 -0
- package/dist/training/storage/cache/persistent.js.map +1 -0
- package/dist/training/storage/cache/schema.d.ts +2 -0
- package/dist/training/storage/cache/schema.js +62 -0
- package/dist/training/storage/cache/schema.js.map +1 -0
- package/dist/training/storage/content.d.ts +13 -0
- package/dist/training/storage/content.js +44 -0
- package/dist/training/storage/content.js.map +1 -0
- package/dist/training/storage/evaluation-json.d.ts +12 -0
- package/dist/training/storage/evaluation-json.js +46 -0
- package/dist/training/storage/evaluation-json.js.map +1 -0
- package/dist/training/storage/filesystem.d.ts +6 -0
- package/dist/training/storage/filesystem.js +95 -0
- package/dist/training/storage/filesystem.js.map +1 -0
- package/dist/training/storage/index.d.ts +6 -0
- package/dist/training/storage/index.js +5 -0
- package/dist/training/storage/index.js.map +1 -0
- package/dist/training/storage/journal.d.ts +11 -0
- package/dist/training/storage/journal.js +81 -0
- package/dist/training/storage/journal.js.map +1 -0
- package/dist/training/storage/permissions.d.ts +5 -0
- package/dist/training/storage/permissions.js +40 -0
- package/dist/training/storage/permissions.js.map +1 -0
- package/dist/training/storage/store.d.ts +2 -0
- package/dist/training/storage/store.js +162 -0
- package/dist/training/storage/store.js.map +1 -0
- package/dist/training/storage/types.d.ts +67 -0
- package/dist/training/storage/types.js +2 -0
- package/dist/training/storage/types.js.map +1 -0
- package/dist/training/subjects/environment.d.ts +2 -0
- package/dist/training/subjects/environment.js +14 -0
- package/dist/training/subjects/environment.js.map +1 -0
- package/dist/training/subjects/index.d.ts +3 -0
- package/dist/training/subjects/index.js +3 -0
- package/dist/training/subjects/index.js.map +1 -0
- package/dist/training/subjects/registry.d.ts +15 -0
- package/dist/training/subjects/registry.js +13 -0
- package/dist/training/subjects/registry.js.map +1 -0
- package/dist/training/testing/evidence.d.ts +6 -0
- package/dist/training/testing/evidence.js +21 -0
- package/dist/training/testing/evidence.js.map +1 -0
- package/dist/training/testing/index.d.ts +1 -0
- package/dist/training/testing/index.js +2 -0
- package/dist/training/testing/index.js.map +1 -0
- package/dist/training/viewer/cockpit/activity.d.ts +2 -0
- package/dist/training/viewer/cockpit/activity.js +14 -0
- package/dist/training/viewer/cockpit/activity.js.map +1 -0
- package/dist/training/viewer/cockpit/evidence.d.ts +4 -0
- package/dist/training/viewer/cockpit/evidence.js +58 -0
- package/dist/training/viewer/cockpit/evidence.js.map +1 -0
- package/dist/training/viewer/cockpit/format.d.ts +11 -0
- package/dist/training/viewer/cockpit/format.js +27 -0
- package/dist/training/viewer/cockpit/format.js.map +1 -0
- package/dist/training/viewer/cockpit/index.d.ts +4 -0
- package/dist/training/viewer/cockpit/index.js +5 -0
- package/dist/training/viewer/cockpit/index.js.map +1 -0
- package/dist/training/viewer/cockpit/interaction.d.ts +2 -0
- package/dist/training/viewer/cockpit/interaction.js +103 -0
- package/dist/training/viewer/cockpit/interaction.js.map +1 -0
- package/dist/training/viewer/cockpit/overview.d.ts +4 -0
- package/dist/training/viewer/cockpit/overview.js +62 -0
- package/dist/training/viewer/cockpit/overview.js.map +1 -0
- package/dist/training/viewer/cockpit/page.d.ts +2 -0
- package/dist/training/viewer/cockpit/page.js +39 -0
- package/dist/training/viewer/cockpit/page.js.map +1 -0
- package/dist/training/viewer/cockpit/plots.d.ts +6 -0
- package/dist/training/viewer/cockpit/plots.js +67 -0
- package/dist/training/viewer/cockpit/plots.js.map +1 -0
- package/dist/training/viewer/cockpit/responsive.d.ts +2 -0
- package/dist/training/viewer/cockpit/responsive.js +86 -0
- package/dist/training/viewer/cockpit/responsive.js.map +1 -0
- package/dist/training/viewer/cockpit/styles.d.ts +2 -0
- package/dist/training/viewer/cockpit/styles.js +349 -0
- package/dist/training/viewer/cockpit/styles.js.map +1 -0
- package/dist/training/viewer/costs.d.ts +4 -0
- package/dist/training/viewer/costs.js +81 -0
- package/dist/training/viewer/costs.js.map +1 -0
- package/dist/training/viewer/from-trials.d.ts +33 -0
- package/dist/training/viewer/from-trials.js +96 -0
- package/dist/training/viewer/from-trials.js.map +1 -0
- package/dist/training/viewer/index.d.ts +4 -0
- package/dist/training/viewer/index.js +4 -0
- package/dist/training/viewer/index.js.map +1 -0
- package/dist/training/viewer/metrics/candidates.d.ts +10 -0
- package/dist/training/viewer/metrics/candidates.js +56 -0
- package/dist/training/viewer/metrics/candidates.js.map +1 -0
- package/dist/training/viewer/metrics/index.d.ts +2 -0
- package/dist/training/viewer/metrics/index.js +2 -0
- package/dist/training/viewer/metrics/index.js.map +1 -0
- package/dist/training/viewer/metrics/model.d.ts +4 -0
- package/dist/training/viewer/metrics/model.js +11 -0
- package/dist/training/viewer/metrics/model.js.map +1 -0
- package/dist/training/viewer/metrics/recorded-costs.d.ts +4 -0
- package/dist/training/viewer/metrics/recorded-costs.js +55 -0
- package/dist/training/viewer/metrics/recorded-costs.js.map +1 -0
- package/dist/training/viewer/metrics/scores.d.ts +11 -0
- package/dist/training/viewer/metrics/scores.js +139 -0
- package/dist/training/viewer/metrics/scores.js.map +1 -0
- package/dist/training/viewer/metrics/types.d.ts +95 -0
- package/dist/training/viewer/metrics/types.js +2 -0
- package/dist/training/viewer/metrics/types.js.map +1 -0
- package/dist/training/viewer/page.d.ts +4 -0
- package/dist/training/viewer/page.js +7 -0
- package/dist/training/viewer/page.js.map +1 -0
- package/dist/training/viewer/projection.d.ts +3 -0
- package/dist/training/viewer/projection.js +98 -0
- package/dist/training/viewer/projection.js.map +1 -0
- package/dist/training/viewer/server.d.ts +3 -0
- package/dist/training/viewer/server.js +80 -0
- package/dist/training/viewer/server.js.map +1 -0
- package/dist/training/viewer/stream.d.ts +7 -0
- package/dist/training/viewer/stream.js +70 -0
- package/dist/training/viewer/stream.js.map +1 -0
- package/dist/training/viewer/types.d.ts +172 -0
- package/dist/training/viewer/types.js +2 -0
- package/dist/training/viewer/types.js.map +1 -0
- package/examples/training/AGENTS.md +19 -0
- package/examples/training/fixed-questions/AGENTS.md +7 -0
- package/examples/training/fixed-questions/README.md +19 -0
- package/examples/training/fixed-questions/evals/AGENTS.md +7 -0
- package/examples/training/fixed-questions/evals/delivery.spawnfile-training.yaml +33 -0
- package/examples/training/fixed-questions/evals/item.grader.yaml +41 -0
- package/examples/training/native-formatting/AGENTS.md +10 -0
- package/examples/training/native-formatting/PROMPT.md +4 -0
- package/examples/training/native-formatting/README.md +63 -0
- package/examples/training/native-formatting/SOUL.md +4 -0
- package/examples/training/native-formatting/evals/AGENTS.md +6 -0
- package/examples/training/native-formatting/evals/test.spawnfile-training.yaml +22 -0
- package/examples/training/native-formatting/evals/train.spawnfile-training.yaml +32 -0
- package/examples/training/newsroom-reporting/.spawnignore +4 -0
- package/examples/training/newsroom-reporting/AGENTS.md +24 -0
- package/examples/training/newsroom-reporting/README.md +255 -0
- package/examples/training/newsroom-reporting/Spawnfile +248 -0
- package/examples/training/newsroom-reporting/environment/AGENTS.md +14 -0
- package/examples/training/newsroom-reporting/environment/scribe/AGENTS.md +28 -0
- package/examples/training/newsroom-reporting/environment/scribe/RUNBOOK.md +50 -0
- package/examples/training/newsroom-reporting/environment/scribe/SOUL.md +6 -0
- package/examples/training/newsroom-reporting/evals/AGENTS.md +14 -0
- package/examples/training/newsroom-reporting/evals/test.spawnfile-training.yaml +165 -0
- package/examples/training/newsroom-reporting/evals/train.spawnfile-training.yaml +195 -0
- package/package.json +19 -6
- package/runtimes.yaml +6 -6
- package/dist/cli/paideiaDelegation.js +0 -211
- package/dist/cli/paideiaSupervisor.d.ts +0 -17
- package/dist/cli/paideiaSupervisor.js +0 -48
|
@@ -0,0 +1,145 @@
|
|
|
1
|
+
import { planCallAdmission, escalationCallKey as callKey } from '../admission.js';
|
|
2
|
+
import { deriveConsensus } from '../consensus.js';
|
|
3
|
+
import { evidenceCounts, teacherAccuracyFailures, wilson90 } from './metrics.js';
|
|
4
|
+
export function inEscalationBand(p, band) { return p >= band[0] && p <= band[1]; }
|
|
5
|
+
/** Retained group means a lineage with at least one retained (non-escalated) instance. */
|
|
6
|
+
function retainedCoverage(rows, band) {
|
|
7
|
+
const total = new Set(rows.map(row => row.group)).size;
|
|
8
|
+
return total ? new Set(rows.filter(row => !row.missing && !inEscalationBand(row.p, band)).map(row => row.group)).size / total : 0;
|
|
9
|
+
}
|
|
10
|
+
function retainedAccuracy(rows, band) {
|
|
11
|
+
const retained = rows.filter(row => !row.missing && !inEscalationBand(row.p, band));
|
|
12
|
+
return retained.length ? retained.filter(row => (row.p >= 0.5) === row.good).length / retained.length : 0;
|
|
13
|
+
}
|
|
14
|
+
/** Remove the escalation teacher before deriving strict option consensus, never binary majority. */
|
|
15
|
+
function independentGood(row, teacherId) {
|
|
16
|
+
if (row.votes && row.labelers && row.expected && teacherId !== undefined) {
|
|
17
|
+
const votes = Object.fromEntries(Object.entries(row.votes).filter(([name]) => name !== teacherId));
|
|
18
|
+
const asked = [...row.asked ?? Object.keys(row.votes)].filter(name => name !== teacherId);
|
|
19
|
+
const label = { id: 'cascade', lineage: row.group, output: null, context: {}, votes: { answer: votes }, asked, provenance: {} };
|
|
20
|
+
const answer = deriveConsensus(label, row.labelers).answer.answer;
|
|
21
|
+
return answer === undefined ? undefined : row.expected.includes(answer);
|
|
22
|
+
}
|
|
23
|
+
// Human/editor truth is independent; model-derived truth requires original votes.
|
|
24
|
+
return row.source === 'human' || row.source === 'editor' ? row.good : undefined;
|
|
25
|
+
}
|
|
26
|
+
function teacherAnswer(row, teacherId) {
|
|
27
|
+
if (row.votes && row.expected && teacherId !== undefined) {
|
|
28
|
+
const answer = row.votes[teacherId];
|
|
29
|
+
return answer === undefined || answer === 'abstain' ? undefined : row.expected.includes(answer);
|
|
30
|
+
}
|
|
31
|
+
return row.teacherGood;
|
|
32
|
+
}
|
|
33
|
+
export function escalationTeacherReport(heldOut, teacherId) {
|
|
34
|
+
const independent = heldOut.flatMap(row => {
|
|
35
|
+
const good = independentGood(row, teacherId);
|
|
36
|
+
if (good === undefined)
|
|
37
|
+
return [];
|
|
38
|
+
let correct = teacherAnswer(row, teacherId) === good;
|
|
39
|
+
if (row.votes && row.labelers && row.expected) {
|
|
40
|
+
const votes = Object.fromEntries(Object.entries(row.votes).filter(([name]) => name !== teacherId));
|
|
41
|
+
const asked = [...row.asked ?? Object.keys(row.votes)].filter(name => name !== teacherId);
|
|
42
|
+
const answer = deriveConsensus({ id: 'teacher', lineage: row.group, output: null, context: {},
|
|
43
|
+
votes: { answer: votes }, asked, provenance: {} }, row.labelers).answer.answer;
|
|
44
|
+
// Exact options must agree; two different acceptable options are not agreement.
|
|
45
|
+
correct = row.votes[teacherId] === answer;
|
|
46
|
+
}
|
|
47
|
+
return [{ ...row, good, correct }];
|
|
48
|
+
});
|
|
49
|
+
const counts = evidenceCounts(independent);
|
|
50
|
+
const agreement = independent.length ? independent.filter(row => row.correct).length / independent.length : 0;
|
|
51
|
+
const source = independent.length && independent.every(row => row.source === 'human') ? 'human'
|
|
52
|
+
: independent.length && independent.every(row => row.source === 'editor') ? 'editor' : 'independent';
|
|
53
|
+
return { ...counts, source, agreement, failures: teacherAccuracyFailures({ ...counts, agreement }),
|
|
54
|
+
warnings: independent.length < heldOut.length ? ['Instances without independent golden answers are excluded from cascade verification.'] : [] };
|
|
55
|
+
}
|
|
56
|
+
function callFor(row) {
|
|
57
|
+
return row.outputId !== undefined && row.projectionDigest !== undefined
|
|
58
|
+
? { outputId: row.outputId, projectionDigest: row.projectionDigest } : undefined;
|
|
59
|
+
}
|
|
60
|
+
export function planEscalation(fit, heldOut, policy, options = {}) {
|
|
61
|
+
if ([...fit, ...heldOut].some(row => !row.missing && (!Number.isFinite(row.p) || row.p < 0 || row.p > 1))) {
|
|
62
|
+
throw new Error('p: expected finite probability');
|
|
63
|
+
}
|
|
64
|
+
const limit = options.maxCallsPerTrial ?? Infinity;
|
|
65
|
+
if (options.maxCallsPerTrial !== undefined && (!Number.isInteger(limit) || limit < 1)) {
|
|
66
|
+
throw new Error('maxCallsPerTrial: expected positive integer');
|
|
67
|
+
}
|
|
68
|
+
fit = fit.filter(row => !row.disputed);
|
|
69
|
+
const report = options.teacherId === undefined ? {} : { teacherReport: escalationTeacherReport(heldOut, options.teacherId) };
|
|
70
|
+
let band;
|
|
71
|
+
if ('between' in policy) {
|
|
72
|
+
const [lo, hi] = policy.between;
|
|
73
|
+
if (!(Number.isFinite(lo) && Number.isFinite(hi) && lo >= 0 && lo < hi && hi <= 1)) {
|
|
74
|
+
throw new Error('between: expected 0 <= lo < hi <= 1');
|
|
75
|
+
}
|
|
76
|
+
band = policy.between;
|
|
77
|
+
}
|
|
78
|
+
else {
|
|
79
|
+
const target = policy.target_accuracy;
|
|
80
|
+
if (!(Number.isFinite(target) && target > 0.5 && target <= 1))
|
|
81
|
+
throw new Error('target_accuracy: expected (0.5, 1]');
|
|
82
|
+
// Width-first traversal is deterministic; ties choose the lowest lower endpoint.
|
|
83
|
+
for (let width = 1; width <= 20 && !band; width++) {
|
|
84
|
+
for (let lower = 0; lower + width <= 20; lower++) {
|
|
85
|
+
const candidate = [lower / 20, (lower + width) / 20];
|
|
86
|
+
if (retainedCoverage(fit, candidate) >= 0.30 && retainedAccuracy(fit, candidate) >= target) {
|
|
87
|
+
band = candidate;
|
|
88
|
+
break;
|
|
89
|
+
}
|
|
90
|
+
}
|
|
91
|
+
}
|
|
92
|
+
}
|
|
93
|
+
if (!band && options.forceTeacher)
|
|
94
|
+
band = [0, 1];
|
|
95
|
+
if (!band)
|
|
96
|
+
return { status: 'escalation-infeasible', fitAccuracy: 0, heldOutAccuracy: 0, cascadeInterval: [0, 1],
|
|
97
|
+
independentGroups: 0, escalationRate: 0, coverage: 0, failures: ['fit.target_accuracy and retainedCoverage >= 0.30'], ...report };
|
|
98
|
+
const chosen = band;
|
|
99
|
+
const escalated = heldOut.filter(row => options.forceTeacher || !row.missing && inEscalationBand(row.p, chosen));
|
|
100
|
+
const calls = options.calls ?? escalated.flatMap(row => { const call = callFor(row); return call ? [call] : []; });
|
|
101
|
+
const allowed = new Set(planCallAdmission(calls, limit).filter(call => call.admitted).map(callKey));
|
|
102
|
+
const independent = heldOut.flatMap(row => {
|
|
103
|
+
const good = independentGood(row, options.teacherId);
|
|
104
|
+
if (good === undefined)
|
|
105
|
+
return [];
|
|
106
|
+
const call = callFor(row);
|
|
107
|
+
const permitted = limit === Infinity || (call !== undefined && allowed.has(callKey(call)));
|
|
108
|
+
const answer = options.forceTeacher || !row.missing && inEscalationBand(row.p, chosen)
|
|
109
|
+
? (permitted ? teacherAnswer(row, options.teacherId) : undefined) : row.missing ? undefined : row.p >= 0.5;
|
|
110
|
+
return [{ ...row, good, answer }];
|
|
111
|
+
});
|
|
112
|
+
const independentGroups = new Set(independent.map(row => row.group)).size;
|
|
113
|
+
const correct = independent.filter(row => row.answer !== undefined && row.answer === row.good).length;
|
|
114
|
+
const heldOutAccuracy = independent.length ? correct / independent.length : 0;
|
|
115
|
+
const cascadeInterval = wilson90(correct, independent.length);
|
|
116
|
+
const coverage = options.forceTeacher ? 0 : retainedCoverage(heldOut, chosen);
|
|
117
|
+
const failures = [];
|
|
118
|
+
if (independentGroups < 15)
|
|
119
|
+
failures.push('heldOut.independentGroups >= 15');
|
|
120
|
+
if (options.forceTeacher && report.teacherReport)
|
|
121
|
+
failures.push(...report.teacherReport.failures);
|
|
122
|
+
if ('target_accuracy' in policy) {
|
|
123
|
+
if (heldOutAccuracy < policy.target_accuracy)
|
|
124
|
+
failures.push('heldOut.cascadeAccuracy >= target_accuracy');
|
|
125
|
+
if (cascadeInterval[0] < policy.target_accuracy - 0.10)
|
|
126
|
+
failures.push('heldOut.cascade90Lower >= target_accuracy - 0.10');
|
|
127
|
+
}
|
|
128
|
+
if (!options.forceTeacher && coverage < 0.30)
|
|
129
|
+
failures.push('heldOut.retainedCoverage >= 0.30');
|
|
130
|
+
const bad = independent.filter(row => !row.good);
|
|
131
|
+
const falsePassCount = bad.filter(row => row.answer === true).length;
|
|
132
|
+
const falsePassRate = bad.length ? falsePassCount / bad.length : 0;
|
|
133
|
+
const falsePassInterval = wilson90(falsePassCount, bad.length);
|
|
134
|
+
if (options.gate) {
|
|
135
|
+
if (falsePassRate > 0.10)
|
|
136
|
+
failures.push('gate.falsePassRate <= 0.10');
|
|
137
|
+
if (falsePassInterval[1] > 0.20)
|
|
138
|
+
failures.push('gate.falsePass90Upper <= 0.20');
|
|
139
|
+
}
|
|
140
|
+
return { status: independentGroups < 15 ? 'unverified' : failures.length ? 'escalation-infeasible' : 'calibrated', band: chosen,
|
|
141
|
+
fitAccuracy: retainedAccuracy(fit, chosen), heldOutAccuracy, cascadeInterval, independentGroups,
|
|
142
|
+
...(options.gate ? { falsePassRate, falsePassInterval } : {}),
|
|
143
|
+
escalationRate: heldOut.length ? escalated.length / heldOut.length : 0, coverage, failures, ...report };
|
|
144
|
+
}
|
|
145
|
+
//# sourceMappingURL=escalation.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"escalation.js","sourceRoot":"","sources":["../../../../src/training/graders/calibration/escalation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAE,iBAAiB,IAAI,OAAO,EAAuB,MAAM,iBAAiB,CAAC;AAEvG,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAElD,OAAO,EAAE,cAAc,EAAE,uBAAuB,EAAE,QAAQ,EAAE,MAAM,cAAc,CAAC;AAqCjF,MAAM,UAAU,gBAAgB,CAAC,CAAS,EAAE,IAAoB,IAAa,OAAO,CAAC,IAAI,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AAEnH,0FAA0F;AAC1F,SAAS,gBAAgB,CAAC,IAA8B,EAAE,IAAoB;IAC5E,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC;IACvD,OAAO,KAAK,CAAC,CAAC,CAAC,IAAI,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,OAAO,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;AACpI,CAAC;AAED,SAAS,gBAAgB,CAAC,IAA8B,EAAE,IAAoB;IAC5E,MAAM,QAAQ,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,OAAO,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;IACpF,OAAO,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,GAAG,CAAC,KAAK,GAAG,CAAC,IAAI,CAAC,CAAC,MAAM,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;AAC5G,CAAC;AAED,oGAAoG;AACpG,SAAS,eAAe,CAAC,GAAkB,EAAE,SAAkB;IAC7D,IAAI,GAAG,CAAC,KAAK,IAAI,GAAG,CAAC,QAAQ,IAAI,GAAG,CAAC,QAAQ,IAAI,SAAS,KAAK,SAAS,EAAE,CAAC;QACzE,MAAM,KAAK,GAAG,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,EAAE,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC;QACnG,MAAM,KAAK,GAAG,CAAC,GAAG,GAAG,CAAC,KAAK,IAAI,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC;QAC1F,MAAM,KAAK,GAAG,EAAE,EAAE,EAAE,SAAS,EAAE,OAAO,EAAE,GAAG,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE,EAAE,EAAE,CAAC;QAChI,MAAM,MAAM,GAAG,eAAe,CAAC,KAAK,EAAE,GAAG,CAAC,QAAQ,CAAC,CAAC,MAAO,CAAC,MAAM,CAAC;QACnE,OAAO,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;IAC1E,CAAC;IACD,kFAAkF;IAClF,OAAO,GAAG,CAAC,MAAM,KAAK,OAAO,IAAI,GAAG,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC;AAClF,CAAC;AAED,SAAS,aAAa,CAAC,GAAkB,EAAE,SAAkB;IAC3D,IAAI,GAAG,CAAC,KAAK,IAAI,GAAG,CAAC,QAAQ,IAAI,SAAS,KAAK,SAAS,EAAE,CAAC;QACzD,MAAM,MAAM,GAAG,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QACpC,OAAO,MAAM,KAAK,SAAS,IAAI,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;IAClG,CAAC;IACD,OAAO,GAAG,CAAC,WAAW,CAAC;AACzB,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,OAAiC,EAAE,SAAiB;IAC1F,MAAM,WAAW,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE;QACxC,MAAM,IAAI,GAAG,eAAe,CAAC,GAAG,EAAE,SAAS,CAAC,CAAC;QAC7C,IAAI,IAAI,KAAK,SAAS;YAAE,OAAO,EAAE,CAAC;QAClC,IAAI,OAAO,GAAG,aAAa,CAAC,GAAG,EAAE,SAAS,CAAC,KAAK,IAAI,CAAC;QACrD,IAAI,GAAG,CAAC,KAAK,IAAI,GAAG,CAAC,QAAQ,IAAI,GAAG,CAAC,QAAQ,EAAE,CAAC;YAC9C,MAAM,KAAK,GAAG,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,EAAE,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC;YACnG,MAAM,KAAK,GAAG,CAAC,GAAG,GAAG,CAAC,KAAK,IAAI,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC;YAC1F,MAAM,MAAM,GAAG,eAAe,CAAC,EAAE,EAAE,EAAE,SAAS,EAAE,OAAO,EAAE,GAAG,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,EAAE;gBAC3F,KAAK,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE,EAAE,EAAE,EAAE,GAAG,CAAC,QAAQ,CAAC,CAAC,MAAO,CAAC,MAAM,CAAC;YAClF,gFAAgF;YAChF,OAAO,GAAG,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,KAAK,MAAM,CAAC;QAC5C,CAAC;QACD,OAAO,CAAC,EAAE,GAAG,GAAG,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;IACrC,CAAC,CAAC,CAAC;IACH,MAAM,MAAM,GAAG,cAAc,CAAC,WAAW,CAAC,CAAC;IAC3C,MAAM,SAAS,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,MAAM,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;IAC9G,MAAM,MAAM,GAAG,WAAW,CAAC,MAAM,IAAI,WAAW,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO;QAC7F,CAAC,CAAC,WAAW,CAAC,MAAM,IAAI,WAAW,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,aAAa,CAAC;IACvG,OAAO,EAAE,GAAG,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,QAAQ,EAAE,uBAAuB,CAAC,EAAE,GAAG,MAAM,EAAE,SAAS,EAAE,CAAC;QAChG,QAAQ,EAAE,WAAW,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,sFAAsF,CAAC,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;AACpJ,CAAC;AAED,SAAS,OAAO,CAAC,GAAkB;IACjC,OAAO,GAAG,CAAC,QAAQ,KAAK,SAAS,IAAI,GAAG,CAAC,gBAAgB,KAAK,SAAS;QACrE,CAAC,CAAC,EAAE,QAAQ,EAAE,GAAG,CAAC,QAAQ,EAAE,gBAAgB,EAAE,GAAG,CAAC,gBAAgB,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;AACrF,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,GAA6B,EAAE,OAAiC,EAAE,MAAkB,EACjH,UAAwI,EAAE;IAC1I,IAAI,CAAC,GAAG,GAAG,EAAE,GAAG,OAAO,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,OAAO,IAAI,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,GAAG,CAAC,IAAI,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;QAC1G,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IACD,MAAM,KAAK,GAAG,OAAO,CAAC,gBAAgB,IAAI,QAAQ,CAAC;IACnD,IAAI,OAAO,CAAC,gBAAgB,KAAK,SAAS,IAAI,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC,CAAC,EAAE,CAAC;QACtF,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;IACjE,CAAC;IACD,GAAG,GAAG,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;IACvC,MAAM,MAAM,GAAG,OAAO,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,uBAAuB,CAAC,OAAO,EAAE,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC;IAC7H,IAAI,IAAgC,CAAC;IACrC,IAAI,SAAS,IAAI,MAAM,EAAE,CAAC;QACxB,MAAM,CAAC,EAAE,EAAE,EAAE,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC;QAChC,IAAI,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC,IAAI,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC,EAAE,CAAC;YACnF,MAAM,IAAI,KAAK,CAAC,qCAAqC,CAAC,CAAC;QACzD,CAAC;QACD,IAAI,GAAG,MAAM,CAAC,OAAO,CAAC;IACxB,CAAC;SAAM,CAAC;QACN,MAAM,MAAM,GAAG,MAAM,CAAC,eAAe,CAAC;QACtC,IAAI,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,GAAG,IAAI,MAAM,IAAI,CAAC,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,oCAAoC,CAAC,CAAC;QACrH,iFAAiF;QACjF,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;YAClD,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,KAAK,IAAI,EAAE,EAAE,KAAK,EAAE,EAAE,CAAC;gBACjD,MAAM,SAAS,GAAmB,CAAC,KAAK,GAAG,EAAE,EAAE,CAAC,KAAK,GAAG,KAAK,CAAC,GAAG,EAAE,CAAC,CAAC;gBACrE,IAAI,gBAAgB,CAAC,GAAG,EAAE,SAAS,CAAC,IAAI,IAAI,IAAI,gBAAgB,CAAC,GAAG,EAAE,SAAS,CAAC,IAAI,MAAM,EAAE,CAAC;oBAC3F,IAAI,GAAG,SAAS,CAAC;oBAAC,MAAM;gBAC1B,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IACD,IAAI,CAAC,IAAI,IAAI,OAAO,CAAC,YAAY;QAAE,IAAI,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACjD,IAAI,CAAC,IAAI;QAAE,OAAO,EAAE,MAAM,EAAE,uBAAuB,EAAE,WAAW,EAAE,CAAC,EAAE,eAAe,EAAE,CAAC,EAAE,eAAe,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC;YAC9G,iBAAiB,EAAE,CAAC,EAAE,cAAc,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC,kDAAkD,CAAC,EAAE,GAAG,MAAM,EAAE,CAAC;IACpI,MAAM,MAAM,GAAG,IAAI,CAAC;IACpB,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,OAAO,CAAC,YAAY,IAAI,CAAC,GAAG,CAAC,OAAO,IAAI,gBAAgB,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;IACjH,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,IAAI,SAAS,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,GAAG,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IACnH,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,iBAAiB,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC;IACpG,MAAM,WAAW,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE;QACxC,MAAM,IAAI,GAAG,eAAe,CAAC,GAAG,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC;QACrD,IAAI,IAAI,KAAK,SAAS;YAAE,OAAO,EAAE,CAAC;QAClC,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC;QAC1B,MAAM,SAAS,GAAG,KAAK,KAAK,QAAQ,IAAI,CAAC,IAAI,KAAK,SAAS,IAAI,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3F,MAAM,MAAM,GAAG,OAAO,CAAC,YAAY,IAAI,CAAC,GAAG,CAAC,OAAO,IAAI,gBAAgB,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC;YACpF,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,aAAa,CAAC,GAAG,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,GAAG,CAAC;QAC7G,OAAO,CAAC,EAAE,GAAG,GAAG,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC;IACpC,CAAC,CAAC,CAAC;IACH,MAAM,iBAAiB,GAAG,IAAI,GAAG,CAAC,WAAW,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC;IAC1E,MAAM,OAAO,GAAG,WAAW,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,MAAM,KAAK,SAAS,IAAI,GAAG,CAAC,MAAM,KAAK,GAAG,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;IACtG,MAAM,eAAe,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;IAC9E,MAAM,eAAe,GAAG,QAAQ,CAAC,OAAO,EAAE,WAAW,CAAC,MAAM,CAAC,CAAC;IAC9D,MAAM,QAAQ,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,gBAAgB,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;IAC9E,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,IAAI,iBAAiB,GAAG,EAAE;QAAE,QAAQ,CAAC,IAAI,CAAC,iCAAiC,CAAC,CAAC;IAC7E,IAAI,OAAO,CAAC,YAAY,IAAI,MAAM,CAAC,aAAa;QAAE,QAAQ,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,aAAa,CAAC,QAAQ,CAAC,CAAC;IAClG,IAAI,iBAAiB,IAAI,MAAM,EAAE,CAAC;QAChC,IAAI,eAAe,GAAG,MAAM,CAAC,eAAe;YAAE,QAAQ,CAAC,IAAI,CAAC,4CAA4C,CAAC,CAAC;QAC1G,IAAI,eAAe,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,eAAe,GAAG,IAAI;YAAE,QAAQ,CAAC,IAAI,CAAC,kDAAkD,CAAC,CAAC;IAC5H,CAAC;IACD,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,QAAQ,GAAG,IAAI;QAAE,QAAQ,CAAC,IAAI,CAAC,kCAAkC,CAAC,CAAC;IAChG,MAAM,GAAG,GAAG,WAAW,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACjD,MAAM,cAAc,GAAG,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,MAAM,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;IACrE,MAAM,aAAa,GAAG,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,cAAc,GAAG,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;IACnE,MAAM,iBAAiB,GAAG,QAAQ,CAAC,cAAc,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC;IAC/D,IAAI,OAAO,CAAC,IAAI,EAAE,CAAC;QACjB,IAAI,aAAa,GAAG,IAAI;YAAE,QAAQ,CAAC,IAAI,CAAC,4BAA4B,CAAC,CAAC;QACtE,IAAI,iBAAiB,CAAC,CAAC,CAAC,GAAG,IAAI;YAAE,QAAQ,CAAC,IAAI,CAAC,+BAA+B,CAAC,CAAC;IAClF,CAAC;IACD,OAAO,EAAE,MAAM,EAAE,iBAAiB,GAAG,EAAE,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,uBAAuB,CAAC,CAAC,CAAC,YAAY,EAAE,IAAI,EAAE,MAAM;QAC7H,WAAW,EAAE,gBAAgB,CAAC,GAAG,EAAE,MAAM,CAAC,EAAE,eAAe,EAAE,eAAe,EAAE,iBAAiB;QAC/F,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,iBAAiB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7D,cAAc,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,QAAQ,EAAE,GAAG,MAAM,EAAE,CAAC;AAC5G,CAAC"}
|
|
@@ -0,0 +1,28 @@
|
|
|
1
|
+
import type { CalibrationRecord, Exclusions, PlattFit, PlattParams, QuestionCalibration } from './types.js';
|
|
2
|
+
/** Hash rank gives a balanced 50/50 group split, invariant to input row ordering. */
|
|
3
|
+
export declare function splitCalibration(records: readonly CalibrationRecord[], seed: string, exclusions?: Exclusions): {
|
|
4
|
+
fit: CalibrationRecord[];
|
|
5
|
+
heldOut: CalibrationRecord[];
|
|
6
|
+
excluded: CalibrationRecord[];
|
|
7
|
+
};
|
|
8
|
+
export declare function applyPlatt(rawPGood: number, params: PlattParams): number;
|
|
9
|
+
/** Mean binary NLL + lambda * ((a-1)^2 + b^2), lambda=0.01.
|
|
10
|
+
* This fixed regularizer is intentionally not weakened by larger instance counts.
|
|
11
|
+
*/
|
|
12
|
+
export declare function fitPlatt(records: readonly Pick<CalibrationRecord, 'rawPGood' | 'good'>[], options?: {
|
|
13
|
+
maxIterations?: number;
|
|
14
|
+
tolerance?: number;
|
|
15
|
+
}): PlattFit;
|
|
16
|
+
/** This seam accepts fit rows only; candidate selection has no held-out input. */
|
|
17
|
+
export declare function fitQuestion(fitRows: readonly CalibrationRecord[], questionId: string, type: CalibrationRecord['type']): {
|
|
18
|
+
questionSplit: CalibrationRecord[];
|
|
19
|
+
questionFit: CalibrationRecord[];
|
|
20
|
+
fitGroups: number;
|
|
21
|
+
fitSource: "type" | "question";
|
|
22
|
+
params: PlattFit;
|
|
23
|
+
};
|
|
24
|
+
export declare function calibrateQuestions(records: readonly CalibrationRecord[], options: {
|
|
25
|
+
seed: string;
|
|
26
|
+
exclusions?: Exclusions;
|
|
27
|
+
gates?: readonly string[];
|
|
28
|
+
}): QuestionCalibration[];
|
|
@@ -0,0 +1,121 @@
|
|
|
1
|
+
import { createHash } from 'node:crypto';
|
|
2
|
+
import { calibrationBars, calibrationMetrics } from './metrics.js';
|
|
3
|
+
/** Hash rank gives a balanced 50/50 group split, invariant to input row ordering. */
|
|
4
|
+
export function splitCalibration(records, seed, exclusions = {}) {
|
|
5
|
+
const lineages = new Set(exclusions.lineages);
|
|
6
|
+
const examples = new Set(exclusions.examples?.map(example => JSON.stringify([example.dataset, example.id])));
|
|
7
|
+
// Excluding an example excludes all revisions in its lineage, including revisions without example metadata.
|
|
8
|
+
for (const record of records) {
|
|
9
|
+
if (record.example && examples.has(JSON.stringify([record.example.dataset, record.example.id])))
|
|
10
|
+
lineages.add(record.group);
|
|
11
|
+
}
|
|
12
|
+
const excluded = records.filter(record => lineages.has(record.group));
|
|
13
|
+
const eligible = records.filter(record => !lineages.has(record.group));
|
|
14
|
+
const groups = [...new Set(eligible.map(record => record.group))].map(group => ({ group,
|
|
15
|
+
hash: createHash('sha256').update(JSON.stringify([seed, group])).digest('hex'),
|
|
16
|
+
})).sort((a, b) => a.hash.localeCompare(b.hash) || a.group.localeCompare(b.group));
|
|
17
|
+
const fitGroups = new Set(groups.slice(0, Math.floor(groups.length / 2)).map(item => item.group));
|
|
18
|
+
return { fit: eligible.filter(record => fitGroups.has(record.group)),
|
|
19
|
+
heldOut: eligible.filter(record => !fitGroups.has(record.group)), excluded };
|
|
20
|
+
}
|
|
21
|
+
function logit(p) {
|
|
22
|
+
if (!Number.isFinite(p) || p < 0 || p > 1)
|
|
23
|
+
throw new Error('rawPGood: expected finite probability in [0, 1]');
|
|
24
|
+
const clipped = Math.max(1e-6, Math.min(1 - 1e-6, p));
|
|
25
|
+
return Math.log(clipped / (1 - clipped));
|
|
26
|
+
}
|
|
27
|
+
function sigmoid(value) { return 1 / (1 + Math.exp(-value)); }
|
|
28
|
+
export function applyPlatt(rawPGood, params) {
|
|
29
|
+
if (!Number.isFinite(params.a) || !Number.isFinite(params.b))
|
|
30
|
+
throw new Error('params: expected finite a and b');
|
|
31
|
+
return sigmoid(params.a * logit(rawPGood) + params.b);
|
|
32
|
+
}
|
|
33
|
+
/** Mean binary NLL + lambda * ((a-1)^2 + b^2), lambda=0.01.
|
|
34
|
+
* This fixed regularizer is intentionally not weakened by larger instance counts.
|
|
35
|
+
*/
|
|
36
|
+
export function fitPlatt(records, options = {}) {
|
|
37
|
+
const maxIterations = options.maxIterations ?? 100;
|
|
38
|
+
const tolerance = options.tolerance ?? 1e-9;
|
|
39
|
+
if (!Number.isInteger(maxIterations) || maxIterations < 1 || !Number.isFinite(tolerance) || tolerance <= 0) {
|
|
40
|
+
throw new Error('fit options: maxIterations must be positive integer and tolerance positive finite');
|
|
41
|
+
}
|
|
42
|
+
const rows = records.map(record => ({ x: logit(record.rawPGood), y: Number(record.good) }));
|
|
43
|
+
let a = 1, b = 0;
|
|
44
|
+
if (!rows.length)
|
|
45
|
+
return { a, b, converged: false, iterations: 0 };
|
|
46
|
+
const loss = (nextA, nextB) => rows.reduce((sum, row) => {
|
|
47
|
+
const z = nextA * row.x + nextB;
|
|
48
|
+
return sum + Math.max(z, 0) + Math.log1p(Math.exp(-Math.abs(z))) - row.y * z;
|
|
49
|
+
}, 0) / rows.length + 0.01 * ((nextA - 1) ** 2 + nextB ** 2);
|
|
50
|
+
for (let iteration = 1; iteration <= maxIterations; iteration++) {
|
|
51
|
+
let ga = 0.02 * (a - 1), gb = 0.02 * b, aa = 0.02, ab = 0, bb = 0.02;
|
|
52
|
+
for (const row of rows) {
|
|
53
|
+
const p = sigmoid(a * row.x + b), residual = (p - row.y) / rows.length;
|
|
54
|
+
const curvature = p * (1 - p) / rows.length;
|
|
55
|
+
ga += residual * row.x;
|
|
56
|
+
gb += residual;
|
|
57
|
+
aa += curvature * row.x * row.x;
|
|
58
|
+
ab += curvature * row.x;
|
|
59
|
+
bb += curvature;
|
|
60
|
+
}
|
|
61
|
+
if (Math.hypot(ga, gb) < tolerance)
|
|
62
|
+
return { a, b, converged: true, iterations: iteration };
|
|
63
|
+
const determinant = aa * bb - ab * ab;
|
|
64
|
+
const da = (bb * ga - ab * gb) / determinant, db = (aa * gb - ab * ga) / determinant;
|
|
65
|
+
let step = 1;
|
|
66
|
+
const currentLoss = loss(a, b);
|
|
67
|
+
// Summed NLL has O(n * epsilon) roundoff; do not stall a converged Newton step on that noise.
|
|
68
|
+
const roundoff = Number.EPSILON * rows.length * Math.max(1, currentLoss);
|
|
69
|
+
while (step > 1e-8 && loss(a - step * da, b - step * db) > currentLoss + roundoff)
|
|
70
|
+
step /= 2;
|
|
71
|
+
a -= step * da;
|
|
72
|
+
b -= step * db;
|
|
73
|
+
}
|
|
74
|
+
return { a, b, converged: false, iterations: maxIterations };
|
|
75
|
+
}
|
|
76
|
+
/** This seam accepts fit rows only; candidate selection has no held-out input. */
|
|
77
|
+
export function fitQuestion(fitRows, questionId, type) {
|
|
78
|
+
const questionSplit = fitRows.filter(record => record.questionId === questionId && !record.disputed);
|
|
79
|
+
const questionFit = questionSplit.filter(record => !record.missing);
|
|
80
|
+
const fitGroups = new Set(questionFit.map(record => record.group)).size;
|
|
81
|
+
const fitSource = fitGroups >= 20 ? 'question' : 'type';
|
|
82
|
+
const fit = fitSource === 'question' ? questionFit : fitRows.filter(record => record.type === type && !record.missing && !record.disputed);
|
|
83
|
+
return { questionSplit, questionFit, fitGroups, fitSource, params: fitPlatt(fit) };
|
|
84
|
+
}
|
|
85
|
+
export function calibrateQuestions(records, options) {
|
|
86
|
+
const population = splitCalibration(records, options.seed, options.exclusions);
|
|
87
|
+
const split = { fit: population.fit.filter(record => !record.disputed),
|
|
88
|
+
heldOut: population.heldOut.filter(record => !record.disputed) };
|
|
89
|
+
return [...new Set(records.map(record => record.questionId))].map(questionId => {
|
|
90
|
+
const question = records.find(record => record.questionId === questionId);
|
|
91
|
+
if (records.some(record => record.questionId === questionId && record.type !== question.type)) {
|
|
92
|
+
throw new Error(`${questionId}.type: inconsistent question types`);
|
|
93
|
+
}
|
|
94
|
+
const { questionSplit, questionFit, fitGroups, fitSource, params } = fitQuestion(split.fit, questionId, question.type);
|
|
95
|
+
const baseRate = questionSplit.length ? questionSplit.filter(record => record.good).length / questionSplit.length : 0;
|
|
96
|
+
const heldOut = split.heldOut.filter(record => record.questionId === questionId)
|
|
97
|
+
.map(record => ({ ...record, p: record.missing ? 0 : applyPlatt(record.rawPGood, params) }));
|
|
98
|
+
const gate = options.gates?.includes(questionId) ?? false;
|
|
99
|
+
const metrics = calibrationMetrics(heldOut, baseRate, `${options.seed}:${questionId}`, gate);
|
|
100
|
+
const bars = calibrationBars(metrics);
|
|
101
|
+
if (!params.converged) {
|
|
102
|
+
bars.status = 'insufficient-evidence';
|
|
103
|
+
bars.failures.push('fit.convergence');
|
|
104
|
+
}
|
|
105
|
+
if (!questionFit.length) {
|
|
106
|
+
bars.status = 'insufficient-evidence';
|
|
107
|
+
bars.failures.push('fit.questionInstances > 0');
|
|
108
|
+
}
|
|
109
|
+
const bySource = {};
|
|
110
|
+
for (const source of ['human', 'editor', 'consensus', 'single-teacher']) {
|
|
111
|
+
const subset = heldOut.filter(record => record.source === source);
|
|
112
|
+
if (subset.length)
|
|
113
|
+
bySource[source] = calibrationMetrics(subset, baseRate, `${options.seed}:${questionId}:${source}`, gate);
|
|
114
|
+
}
|
|
115
|
+
return { questionId, params: { a: params.a, b: params.b }, fitSource, fitGroups, metrics, bySource, ...bars,
|
|
116
|
+
interpretation: heldOut.some(record => record.source === 'human')
|
|
117
|
+
? 'human-adjudicated and teacher agreement' : heldOut.some(record => record.source === 'editor')
|
|
118
|
+
? 'editor-adjudicated and teacher agreement' : 'teacher agreement, not validated correctness' };
|
|
119
|
+
});
|
|
120
|
+
}
|
|
121
|
+
//# sourceMappingURL=fitting.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"fitting.js","sourceRoot":"","sources":["../../../../src/training/graders/calibration/fitting.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,eAAe,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAGnE,qFAAqF;AACrF,MAAM,UAAU,gBAAgB,CAAC,OAAqC,EAAE,IAAY,EAAE,aAAyB,EAAE;IAG/G,MAAM,QAAQ,GAAG,IAAI,GAAG,CAAC,UAAU,CAAC,QAAQ,CAAC,CAAC;IAC9C,MAAM,QAAQ,GAAG,IAAI,GAAG,CAAC,UAAU,CAAC,QAAQ,EAAE,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,OAAO,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAC7G,4GAA4G;IAC5G,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,MAAM,CAAC,OAAO,IAAI,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,CAAC;YAAE,QAAQ,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAC9H,CAAC;IACD,MAAM,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,QAAQ,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;IACtE,MAAM,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;IACvE,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,EAAE,KAAK;QACrF,IAAI,EAAE,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;KAC/E,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC;IACnF,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC;IAClG,OAAO,EAAE,GAAG,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAClE,OAAO,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC;AACjF,CAAC;AAED,SAAS,KAAK,CAAC,CAAS;IACtB,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,iDAAiD,CAAC,CAAC;IAC9G,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IACtD,OAAO,IAAI,CAAC,GAAG,CAAC,OAAO,GAAG,CAAC,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC;AAC3C,CAAC;AAED,SAAS,OAAO,CAAC,KAAa,IAAY,OAAO,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;AAE9E,MAAM,UAAU,UAAU,CAAC,QAAgB,EAAE,MAAmB;IAC9D,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,iCAAiC,CAAC,CAAC;IACjH,OAAO,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,KAAK,CAAC,QAAQ,CAAC,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;AACxD,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,QAAQ,CAAC,OAAgE,EACvF,UAA0D,EAAE;IAC5D,MAAM,aAAa,GAAG,OAAO,CAAC,aAAa,IAAI,GAAG,CAAC;IACnD,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,IAAI,IAAI,CAAC;IAC5C,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,aAAa,CAAC,IAAI,aAAa,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,SAAS,IAAI,CAAC,EAAE,CAAC;QAC3G,MAAM,IAAI,KAAK,CAAC,mFAAmF,CAAC,CAAC;IACvG,CAAC;IACD,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC,EAAE,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC;IAC5F,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC;IACjB,IAAI,CAAC,IAAI,CAAC,MAAM;QAAE,OAAO,EAAE,CAAC,EAAE,CAAC,EAAE,SAAS,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,EAAE,CAAC;IACnE,MAAM,IAAI,GAAG,CAAC,KAAa,EAAE,KAAa,EAAU,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,EAAE;QAC9E,MAAM,CAAC,GAAG,KAAK,GAAG,GAAG,CAAC,CAAC,GAAG,KAAK,CAAC;QAChC,OAAO,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;IAC/E,CAAC,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,GAAG,IAAI,GAAG,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,IAAI,CAAC,GAAG,KAAK,IAAI,CAAC,CAAC,CAAC;IAC7D,KAAK,IAAI,SAAS,GAAG,CAAC,EAAE,SAAS,IAAI,aAAa,EAAE,SAAS,EAAE,EAAE,CAAC;QAChE,IAAI,EAAE,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,GAAG,IAAI,GAAG,CAAC,EAAE,EAAE,GAAG,IAAI,EAAE,EAAE,GAAG,CAAC,EAAE,EAAE,GAAG,IAAI,CAAC;QACrE,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;YACvB,MAAM,CAAC,GAAG,OAAO,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,EAAE,QAAQ,GAAG,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC;YACvE,MAAM,SAAS,GAAG,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC;YAC5C,EAAE,IAAI,QAAQ,GAAG,GAAG,CAAC,CAAC,CAAC;YAAC,EAAE,IAAI,QAAQ,CAAC;YACvC,EAAE,IAAI,SAAS,GAAG,GAAG,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC;YAAC,EAAE,IAAI,SAAS,GAAG,GAAG,CAAC,CAAC,CAAC;YAAC,EAAE,IAAI,SAAS,CAAC;QAC5E,CAAC;QACD,IAAI,IAAI,CAAC,KAAK,CAAC,EAAE,EAAE,EAAE,CAAC,GAAG,SAAS;YAAE,OAAO,EAAE,CAAC,EAAE,CAAC,EAAE,SAAS,EAAE,IAAI,EAAE,UAAU,EAAE,SAAS,EAAE,CAAC;QAC5F,MAAM,WAAW,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC;QACtC,MAAM,EAAE,GAAG,CAAC,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC,GAAG,WAAW,EAAE,EAAE,GAAG,CAAC,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC,GAAG,WAAW,CAAC;QACrF,IAAI,IAAI,GAAG,CAAC,CAAC;QACb,MAAM,WAAW,GAAG,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QAC/B,8FAA8F;QAC9F,MAAM,QAAQ,GAAG,MAAM,CAAC,OAAO,GAAG,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,WAAW,CAAC,CAAC;QACzE,OAAO,IAAI,GAAG,IAAI,IAAI,IAAI,CAAC,CAAC,GAAG,IAAI,GAAG,EAAE,EAAE,CAAC,GAAG,IAAI,GAAG,EAAE,CAAC,GAAG,WAAW,GAAG,QAAQ;YAAE,IAAI,IAAI,CAAC,CAAC;QAC7F,CAAC,IAAI,IAAI,GAAG,EAAE,CAAC;QAAC,CAAC,IAAI,IAAI,GAAG,EAAE,CAAC;IACjC,CAAC;IACD,OAAO,EAAE,CAAC,EAAE,CAAC,EAAE,SAAS,EAAE,KAAK,EAAE,UAAU,EAAE,aAAa,EAAE,CAAC;AAC/D,CAAC;AAED,kFAAkF;AAClF,MAAM,UAAU,WAAW,CAAC,OAAqC,EAAE,UAAkB,EAAE,IAA+B;IACpH,MAAM,aAAa,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,UAAU,KAAK,UAAU,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IACrG,MAAM,WAAW,GAAG,aAAa,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IACpE,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,WAAW,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC;IACxE,MAAM,SAAS,GAAG,SAAS,IAAI,EAAE,CAAC,CAAC,CAAC,UAAmB,CAAC,CAAC,CAAC,MAAe,CAAC;IAC1E,MAAM,GAAG,GAAG,SAAS,KAAK,UAAU,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,IAAI,KAAK,IAAI,IAAI,CAAC,MAAM,CAAC,OAAO,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IAC3I,OAAO,EAAE,aAAa,EAAE,WAAW,EAAE,SAAS,EAAE,SAAS,EAAE,MAAM,EAAE,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;AACrF,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,OAAqC,EAAE,OAEzE;IACC,MAAM,UAAU,GAAG,gBAAgB,CAAC,OAAO,EAAE,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,UAAU,CAAC,CAAC;IAC/E,MAAM,KAAK,GAAG,EAAE,GAAG,EAAE,UAAU,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC;QACpE,OAAO,EAAE,UAAU,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;IACnE,OAAO,CAAC,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,UAAU,CAAC,EAAE;QAC7E,MAAM,QAAQ,GAAG,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,UAAU,KAAK,UAAU,CAAE,CAAC;QAC3E,IAAI,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,UAAU,KAAK,UAAU,IAAI,MAAM,CAAC,IAAI,KAAK,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC;YAC9F,MAAM,IAAI,KAAK,CAAC,GAAG,UAAU,oCAAoC,CAAC,CAAC;QACrE,CAAC;QACD,MAAM,EAAE,aAAa,EAAE,WAAW,EAAE,SAAS,EAAE,SAAS,EAAE,MAAM,EAAE,GAAG,WAAW,CAAC,KAAK,CAAC,GAAG,EAAE,UAAU,EAAE,QAAQ,CAAC,IAAI,CAAC,CAAC;QACvH,MAAM,QAAQ,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC,aAAa,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,MAAM,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;QACtH,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,UAAU,KAAK,UAAU,CAAC;aAC7E,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,EAAE,GAAG,MAAM,EAAE,CAAC,EAAE,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC;QAC/F,MAAM,IAAI,GAAG,OAAO,CAAC,KAAK,EAAE,QAAQ,CAAC,UAAU,CAAC,IAAI,KAAK,CAAC;QAC1D,MAAM,OAAO,GAAG,kBAAkB,CAAC,OAAO,EAAE,QAAQ,EAAE,GAAG,OAAO,CAAC,IAAI,IAAI,UAAU,EAAE,EAAE,IAAI,CAAC,CAAC;QAC7F,MAAM,IAAI,GAAG,eAAe,CAAC,OAAO,CAAC,CAAC;QACtC,IAAI,CAAC,MAAM,CAAC,SAAS,EAAE,CAAC;YAAC,IAAI,CAAC,MAAM,GAAG,uBAAuB,CAAC;YAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,iBAAiB,CAAC,CAAC;QAAC,CAAC;QACxG,IAAI,CAAC,WAAW,CAAC,MAAM,EAAE,CAAC;YAAC,IAAI,CAAC,MAAM,GAAG,uBAAuB,CAAC;YAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,2BAA2B,CAAC,CAAC;QAAC,CAAC;QACpH,MAAM,QAAQ,GAAoC,EAAE,CAAC;QACrD,KAAK,MAAM,MAAM,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,gBAAgB,CAAU,EAAE,CAAC;YACjF,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC;YAClE,IAAI,MAAM,CAAC,MAAM;gBAAE,QAAQ,CAAC,MAAM,CAAC,GAAG,kBAAkB,CAAC,MAAM,EAAE,QAAQ,EAAE,GAAG,OAAO,CAAC,IAAI,IAAI,UAAU,IAAI,MAAM,EAAE,EAAE,IAAI,CAAC,CAAC;QAC9H,CAAC;QACD,OAAO,EAAE,UAAU,EAAE,MAAM,EAAE,EAAE,CAAC,EAAE,MAAM,CAAC,CAAC,EAAE,CAAC,EAAE,MAAM,CAAC,CAAC,EAAE,EAAE,SAAS,EAAE,SAAS,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,IAAI;YACzG,cAAc,EAAE,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,MAAM,KAAK,OAAO,CAAC;gBAC/D,CAAC,CAAC,yCAAyC,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC;gBAC9F,CAAC,CAAC,0CAA0C,CAAC,CAAC,CAAC,8CAA8C,EAAE,CAAC;IACxG,CAAC,CAAC,CAAC;AACL,CAAC"}
|
|
@@ -0,0 +1,7 @@
|
|
|
1
|
+
export { splitCalibration, fitPlatt, applyPlatt, calibrateQuestions } from './fitting.js';
|
|
2
|
+
export { seededRandom, evidenceCounts, expectedCalibrationError, groupBootstrap, calibrationMetrics, evidenceFailures, calibrationBars, teacherAgreement, teacherAccuracyFailures, wilson90 } from './metrics.js';
|
|
3
|
+
export { inEscalationBand, planEscalation, escalationTeacherReport } from './escalation.js';
|
|
4
|
+
export type { QuestionType, GoldenSource, CalibrationRecord, Exclusions, PlattParams, PlattFit, ScoredRecord, EvidenceCounts, Interval, CalibrationMetrics, CalibrationStatus, BarResult, QuestionCalibration } from './types.js';
|
|
5
|
+
export type { TeacherAgreementRecord } from './metrics.js';
|
|
6
|
+
export type { EscalationBand, BandPolicy, CascadeRecord, BandPlan, EscalationTeacherReport, EscalationCall } from './escalation.js';
|
|
7
|
+
export { selectQuestionModel, fitNll } from './selection.js';
|
|
@@ -0,0 +1,5 @@
|
|
|
1
|
+
export { splitCalibration, fitPlatt, applyPlatt, calibrateQuestions } from './fitting.js';
|
|
2
|
+
export { seededRandom, evidenceCounts, expectedCalibrationError, groupBootstrap, calibrationMetrics, evidenceFailures, calibrationBars, teacherAgreement, teacherAccuracyFailures, wilson90 } from './metrics.js';
|
|
3
|
+
export { inEscalationBand, planEscalation, escalationTeacherReport } from './escalation.js';
|
|
4
|
+
export { selectQuestionModel, fitNll } from './selection.js';
|
|
5
|
+
//# sourceMappingURL=index.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../src/training/graders/calibration/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAE,QAAQ,EAAE,UAAU,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAC1F,OAAO,EAAE,YAAY,EAAE,cAAc,EAAE,wBAAwB,EAAE,cAAc,EAC7E,kBAAkB,EAAE,gBAAgB,EAAE,eAAe,EAAE,gBAAgB,EAAE,uBAAuB,EAAE,QAAQ,EAAE,MAAM,cAAc,CAAC;AACnI,OAAO,EAAE,gBAAgB,EAAE,cAAc,EAAE,uBAAuB,EAAE,MAAM,iBAAiB,CAAC;AAO5F,OAAO,EAAE,mBAAmB,EAAE,MAAM,EAAE,MAAM,gBAAgB,CAAC"}
|
|
@@ -0,0 +1,31 @@
|
|
|
1
|
+
import type { BarResult, CalibrationMetrics, EvidenceCounts, Interval, ScoredRecord } from './types.js';
|
|
2
|
+
/** Two-sided 90% Wilson score interval; no observations cannot establish a bound. */
|
|
3
|
+
export declare function wilson90(successes: number, total: number): Interval;
|
|
4
|
+
export declare function seededRandom(seed: string): () => number;
|
|
5
|
+
export declare function evidenceCounts(records: readonly Pick<ScoredRecord, 'group' | 'good'>[]): EvidenceCounts;
|
|
6
|
+
/** Equal-mass bins; fewer than 10 instances cannot form a valid ECE bin. */
|
|
7
|
+
export declare function expectedCalibrationError(records: readonly Pick<ScoredRecord, 'p' | 'good' | 'missing'>[]): number | null;
|
|
8
|
+
/** Every selected lineage contributes all its instances, including repeated draws. */
|
|
9
|
+
export declare function groupBootstrap(records: readonly ScoredRecord[], seed: string): {
|
|
10
|
+
eceInterval: Interval | null;
|
|
11
|
+
brierInterval: Interval;
|
|
12
|
+
};
|
|
13
|
+
export declare function calibrationMetrics(records: readonly ScoredRecord[], baseRate: number, seed: string, gate?: boolean): CalibrationMetrics;
|
|
14
|
+
export declare function evidenceFailures(counts: EvidenceCounts): string[];
|
|
15
|
+
export declare function calibrationBars(metrics: CalibrationMetrics): BarResult;
|
|
16
|
+
/** The same independent evidence/accuracy bar applies to every forced teacher route. */
|
|
17
|
+
export declare function teacherAccuracyFailures(metrics: EvidenceCounts & {
|
|
18
|
+
agreement: number;
|
|
19
|
+
}): string[];
|
|
20
|
+
export interface TeacherAgreementRecord {
|
|
21
|
+
group: string;
|
|
22
|
+
good: boolean;
|
|
23
|
+
goldenAnswer: string | number;
|
|
24
|
+
teacherAnswer?: string | number;
|
|
25
|
+
/** Labeler identities contributing to the golden answer, excluding disputed labels. */
|
|
26
|
+
goldenLabelers: readonly string[];
|
|
27
|
+
}
|
|
28
|
+
export declare function teacherAgreement(records: readonly TeacherAgreementRecord[], teacherId: string): BarResult & EvidenceCounts & {
|
|
29
|
+
agreement: number;
|
|
30
|
+
excludedSelfLabels: number;
|
|
31
|
+
};
|
|
@@ -0,0 +1,145 @@
|
|
|
1
|
+
import { createHash } from 'node:crypto';
|
|
2
|
+
/** Two-sided 90% Wilson score interval; no observations cannot establish a bound. */
|
|
3
|
+
export function wilson90(successes, total) {
|
|
4
|
+
if (!Number.isInteger(total) || total < 0 || !Number.isInteger(successes) || successes < 0 || successes > total) {
|
|
5
|
+
throw new Error('Wilson counts: expected 0 <= successes <= total integers');
|
|
6
|
+
}
|
|
7
|
+
if (!total)
|
|
8
|
+
return [0, 1];
|
|
9
|
+
const z = 1.6448536269514722, p = successes / total, denominator = 1 + z * z / total;
|
|
10
|
+
const centre = (p + z * z / (2 * total)) / denominator;
|
|
11
|
+
const radius = z * Math.sqrt(p * (1 - p) / total + z * z / (4 * total * total)) / denominator;
|
|
12
|
+
return [Math.max(0, centre - radius), Math.min(1, centre + radius)];
|
|
13
|
+
}
|
|
14
|
+
export function seededRandom(seed) {
|
|
15
|
+
let state = createHash('sha256').update(seed).digest().readUInt32LE(0);
|
|
16
|
+
return () => {
|
|
17
|
+
state = (state + 0x6D2B79F5) | 0;
|
|
18
|
+
let value = Math.imul(state ^ (state >>> 15), 1 | state);
|
|
19
|
+
value = (value + Math.imul(value ^ (value >>> 7), 61 | value)) ^ value;
|
|
20
|
+
return ((value ^ (value >>> 14)) >>> 0) / 4294967296;
|
|
21
|
+
};
|
|
22
|
+
}
|
|
23
|
+
export function evidenceCounts(records) {
|
|
24
|
+
const good = records.filter(record => record.good).length;
|
|
25
|
+
return { groups: new Set(records.map(record => record.group)).size, instances: records.length,
|
|
26
|
+
good, bad: records.length - good };
|
|
27
|
+
}
|
|
28
|
+
/** Equal-mass bins; fewer than 10 instances cannot form a valid ECE bin. */
|
|
29
|
+
export function expectedCalibrationError(records) {
|
|
30
|
+
const observed = records.filter(row => !row.missing);
|
|
31
|
+
const missing = records.length - observed.length;
|
|
32
|
+
if (observed.length < 10)
|
|
33
|
+
return null;
|
|
34
|
+
const sorted = [...observed].sort((a, b) => a.p - b.p);
|
|
35
|
+
const bins = Math.min(10, Math.floor(sorted.length / 10));
|
|
36
|
+
let total = missing;
|
|
37
|
+
for (let index = 0; index < bins; index++) {
|
|
38
|
+
const bin = sorted.slice(Math.floor(index * sorted.length / bins), Math.floor((index + 1) * sorted.length / bins));
|
|
39
|
+
total += Math.abs(bin.reduce((sum, row) => sum + row.p - Number(row.good), 0));
|
|
40
|
+
}
|
|
41
|
+
return total / records.length;
|
|
42
|
+
}
|
|
43
|
+
function brier(records) {
|
|
44
|
+
return records.length ? records.reduce((sum, row) => sum + (row.missing ? 1 : (row.p - Number(row.good)) ** 2), 0) / records.length : 1;
|
|
45
|
+
}
|
|
46
|
+
function percentile(values, fraction) {
|
|
47
|
+
values.sort((a, b) => a - b);
|
|
48
|
+
const position = (values.length - 1) * fraction;
|
|
49
|
+
const low = Math.floor(position), high = Math.ceil(position);
|
|
50
|
+
return values[low] + (values[high] - values[low]) * (position - low);
|
|
51
|
+
}
|
|
52
|
+
/** Every selected lineage contributes all its instances, including repeated draws. */
|
|
53
|
+
export function groupBootstrap(records, seed) {
|
|
54
|
+
if (!records.length)
|
|
55
|
+
return { eceInterval: null, brierInterval: [1, 1] };
|
|
56
|
+
const byGroup = new Map();
|
|
57
|
+
for (const record of records) {
|
|
58
|
+
const group = byGroup.get(record.group) ?? [];
|
|
59
|
+
group.push(record);
|
|
60
|
+
byGroup.set(record.group, group);
|
|
61
|
+
}
|
|
62
|
+
const groups = [...byGroup.entries()].sort(([a], [b]) => a.localeCompare(b)).map(([, rows]) => rows);
|
|
63
|
+
const random = seededRandom(seed), eces = [], briers = [];
|
|
64
|
+
for (let iteration = 0; iteration < 1000; iteration++) {
|
|
65
|
+
const sample = [];
|
|
66
|
+
for (let draw = 0; draw < groups.length; draw++)
|
|
67
|
+
sample.push(...groups[Math.floor(random() * groups.length)]);
|
|
68
|
+
const ece = expectedCalibrationError(sample);
|
|
69
|
+
if (ece !== null)
|
|
70
|
+
eces.push(ece);
|
|
71
|
+
briers.push(brier(sample));
|
|
72
|
+
}
|
|
73
|
+
return { eceInterval: expectedCalibrationError(records) === null || eces.length !== 1000 ? null
|
|
74
|
+
: [percentile(eces, 0.05), percentile(eces, 0.95)],
|
|
75
|
+
brierInterval: [percentile(briers, 0.05), percentile(briers, 0.95)] };
|
|
76
|
+
}
|
|
77
|
+
export function calibrationMetrics(records, baseRate, seed, gate = false) {
|
|
78
|
+
if (!Number.isFinite(baseRate) || baseRate < 0 || baseRate > 1)
|
|
79
|
+
throw new Error('baseRate: expected probability');
|
|
80
|
+
if (records.some(record => !record.missing && (!Number.isFinite(record.p) || record.p < 0 || record.p > 1))) {
|
|
81
|
+
throw new Error('p: expected finite probability');
|
|
82
|
+
}
|
|
83
|
+
const counts = evidenceCounts(records);
|
|
84
|
+
return { ...counts, baseRate,
|
|
85
|
+
accuracy: records.length ? records.filter(row => !row.missing && (row.p >= 0.5) === row.good).length / records.length : 0,
|
|
86
|
+
brier: brier(records),
|
|
87
|
+
baseRateBrier: brier(records.map(row => ({ ...row, p: baseRate, missing: false }))),
|
|
88
|
+
ece: expectedCalibrationError(records), ...groupBootstrap(records, seed),
|
|
89
|
+
...(gate ? { falsePassRate: counts.bad ? records.filter(row => !row.missing && row.p >= 0.5 && !row.good).length / counts.bad : 0,
|
|
90
|
+
falsePassInterval: wilson90(records.filter(row => !row.missing && row.p >= 0.5 && !row.good).length, counts.bad) } : {}),
|
|
91
|
+
};
|
|
92
|
+
}
|
|
93
|
+
export function evidenceFailures(counts) {
|
|
94
|
+
const failures = [];
|
|
95
|
+
if (counts.groups < 15)
|
|
96
|
+
failures.push('evidence.groups >= 15');
|
|
97
|
+
if (counts.instances < 30)
|
|
98
|
+
failures.push('evidence.instances >= 30');
|
|
99
|
+
if (counts.good < 5)
|
|
100
|
+
failures.push('evidence.good >= 5');
|
|
101
|
+
if (counts.bad < 5)
|
|
102
|
+
failures.push('evidence.bad >= 5');
|
|
103
|
+
return failures;
|
|
104
|
+
}
|
|
105
|
+
export function calibrationBars(metrics) {
|
|
106
|
+
const evidence = evidenceFailures(metrics);
|
|
107
|
+
if (metrics.ece === null || metrics.eceInterval === null)
|
|
108
|
+
evidence.push('evidence.eceBins >= 1 (including bootstrap resamples)');
|
|
109
|
+
if (evidence.length)
|
|
110
|
+
return { status: 'insufficient-evidence', failures: evidence };
|
|
111
|
+
const failures = [];
|
|
112
|
+
if (!(metrics.ece <= 0.10))
|
|
113
|
+
failures.push('calibration.ece <= 0.10');
|
|
114
|
+
if (!(metrics.eceInterval[1] <= 0.15))
|
|
115
|
+
failures.push('calibration.ece90Upper <= 0.15');
|
|
116
|
+
if (!(metrics.brier <= 0.20))
|
|
117
|
+
failures.push('calibration.brier <= 0.20');
|
|
118
|
+
if (!(metrics.brier < metrics.baseRateBrier))
|
|
119
|
+
failures.push('calibration.brier < baseRateBrier');
|
|
120
|
+
if (metrics.falsePassRate !== undefined && !(metrics.falsePassRate <= 0.10))
|
|
121
|
+
failures.push('gate.falsePassRate <= 0.10');
|
|
122
|
+
if (metrics.falsePassRate !== undefined && !(metrics.falsePassInterval?.[1] !== undefined && metrics.falsePassInterval[1] <= 0.20)) {
|
|
123
|
+
failures.push('gate.falsePass90Upper <= 0.20');
|
|
124
|
+
}
|
|
125
|
+
return { status: failures.length ? 'uncalibrated' : 'calibrated', failures };
|
|
126
|
+
}
|
|
127
|
+
/** The same independent evidence/accuracy bar applies to every forced teacher route. */
|
|
128
|
+
export function teacherAccuracyFailures(metrics) {
|
|
129
|
+
const failures = evidenceFailures(metrics).map(failure => `teacher.${failure}`);
|
|
130
|
+
if (metrics.agreement < 0.85)
|
|
131
|
+
failures.push('teacher.agreement >= 0.85');
|
|
132
|
+
return failures;
|
|
133
|
+
}
|
|
134
|
+
export function teacherAgreement(records, teacherId) {
|
|
135
|
+
// A consensus containing this teacher must be rederived from other labelers by the caller.
|
|
136
|
+
const independent = records.filter(row => row.goldenLabelers.length > 0 && !row.goldenLabelers.includes(teacherId));
|
|
137
|
+
const counts = evidenceCounts(independent);
|
|
138
|
+
const agreement = independent.length
|
|
139
|
+
? independent.filter(row => row.teacherAnswer !== undefined && row.teacherAnswer === row.goldenAnswer).length / independent.length : 0;
|
|
140
|
+
const failures = teacherAccuracyFailures({ ...counts, agreement });
|
|
141
|
+
const insufficient = evidenceFailures(counts).length > 0;
|
|
142
|
+
return { ...counts, agreement, excludedSelfLabels: records.length - independent.length, failures,
|
|
143
|
+
status: insufficient ? 'insufficient-evidence' : failures.length ? 'uncalibrated' : 'calibrated' };
|
|
144
|
+
}
|
|
145
|
+
//# sourceMappingURL=metrics.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"metrics.js","sourceRoot":"","sources":["../../../../src/training/graders/calibration/metrics.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAGzC,qFAAqF;AACrF,MAAM,UAAU,QAAQ,CAAC,SAAiB,EAAE,KAAa;IACvD,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,SAAS,CAAC,IAAI,SAAS,GAAG,CAAC,IAAI,SAAS,GAAG,KAAK,EAAE,CAAC;QAChH,MAAM,IAAI,KAAK,CAAC,0DAA0D,CAAC,CAAC;IAC9E,CAAC;IACD,IAAI,CAAC,KAAK;QAAE,OAAO,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAC1B,MAAM,CAAC,GAAG,kBAAkB,EAAE,CAAC,GAAG,SAAS,GAAG,KAAK,EAAE,WAAW,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,KAAK,CAAC;IACrF,MAAM,MAAM,GAAG,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,GAAG,KAAK,CAAC,CAAC,GAAG,WAAW,CAAC;IACvD,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,KAAK,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,GAAG,KAAK,GAAG,KAAK,CAAC,CAAC,GAAG,WAAW,CAAC;IAC9F,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,GAAG,MAAM,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC;AACtE,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,IAAY;IACvC,IAAI,KAAK,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;IACvE,OAAO,GAAG,EAAE;QACV,KAAK,GAAG,CAAC,KAAK,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC;QACjC,IAAI,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,KAAK,KAAK,EAAE,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,CAAC;QACzD,KAAK,GAAG,CAAC,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,KAAK,KAAK,CAAC,CAAC,EAAE,EAAE,GAAG,KAAK,CAAC,CAAC,GAAG,KAAK,CAAC;QACvE,OAAO,CAAC,CAAC,KAAK,GAAG,CAAC,KAAK,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,GAAG,UAAU,CAAC;IACvD,CAAC,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,OAAwD;IACrF,MAAM,IAAI,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;IAC1D,OAAO,EAAE,MAAM,EAAE,IAAI,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,SAAS,EAAE,OAAO,CAAC,MAAM;QAC3F,IAAI,EAAE,GAAG,EAAE,OAAO,CAAC,MAAM,GAAG,IAAI,EAAE,CAAC;AACvC,CAAC;AAED,4EAA4E;AAC5E,MAAM,UAAU,wBAAwB,CAAC,OAAgE;IACvG,MAAM,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IACrD,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,GAAG,QAAQ,CAAC,MAAM,CAAC;IACjD,IAAI,QAAQ,CAAC,MAAM,GAAG,EAAE;QAAE,OAAO,IAAI,CAAC;IACtC,MAAM,MAAM,GAAG,CAAC,GAAG,QAAQ,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IACvD,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,GAAG,EAAE,CAAC,CAAC,CAAC;IAC1D,IAAI,KAAK,GAAG,OAAO,CAAC;IACpB,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;QAC1C,MAAM,GAAG,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,GAAG,MAAM,CAAC,MAAM,GAAG,IAAI,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC,CAAC;QACnH,KAAK,IAAI,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,EAAE,CAAC,GAAG,GAAG,GAAG,CAAC,CAAC,GAAG,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IACjF,CAAC;IACD,OAAO,KAAK,GAAG,OAAO,CAAC,MAAM,CAAC;AAChC,CAAC;AAED,SAAS,KAAK,CAAC,OAAgE;IAC7E,OAAO,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;AAC1I,CAAC;AAED,SAAS,UAAU,CAAC,MAAgB,EAAE,QAAgB;IACpD,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC7B,MAAM,QAAQ,GAAG,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,GAAG,QAAQ,CAAC;IAChD,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAC7D,OAAO,MAAM,CAAC,GAAG,CAAE,GAAG,CAAC,MAAM,CAAC,IAAI,CAAE,GAAG,MAAM,CAAC,GAAG,CAAE,CAAC,GAAG,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC;AAC1E,CAAC;AAED,sFAAsF;AACtF,MAAM,UAAU,cAAc,CAAC,OAAgC,EAAE,IAAY;IAG3E,IAAI,CAAC,OAAO,CAAC,MAAM;QAAE,OAAO,EAAE,WAAW,EAAE,IAAI,EAAE,aAAa,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC;IACzE,MAAM,OAAO,GAAG,IAAI,GAAG,EAA0B,CAAC;IAClD,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,MAAM,KAAK,GAAG,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC;QAC9C,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAAC,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;IACvD,CAAC;IACD,MAAM,MAAM,GAAG,CAAC,GAAG,OAAO,CAAC,OAAO,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC;IACrG,MAAM,MAAM,GAAG,YAAY,CAAC,IAAI,CAAC,EAAE,IAAI,GAAa,EAAE,EAAE,MAAM,GAAa,EAAE,CAAC;IAC9E,KAAK,IAAI,SAAS,GAAG,CAAC,EAAE,SAAS,GAAG,IAAI,EAAE,SAAS,EAAE,EAAE,CAAC;QACtD,MAAM,MAAM,GAAmB,EAAE,CAAC;QAClC,KAAK,IAAI,IAAI,GAAG,CAAC,EAAE,IAAI,GAAG,MAAM,CAAC,MAAM,EAAE,IAAI,EAAE;YAAE,MAAM,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,GAAG,MAAM,CAAC,MAAM,CAAC,CAAE,CAAC,CAAC;QAC/G,MAAM,GAAG,GAAG,wBAAwB,CAAC,MAAM,CAAC,CAAC;QAC7C,IAAI,GAAG,KAAK,IAAI;YAAE,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACjC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC;IAC7B,CAAC;IACD,OAAO,EAAE,WAAW,EAAE,wBAAwB,CAAC,OAAO,CAAC,KAAK,IAAI,IAAI,IAAI,CAAC,MAAM,KAAK,IAAI,CAAC,CAAC,CAAC,IAAI;YAC7F,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,EAAE,IAAI,CAAC,EAAE,UAAU,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;QAClD,aAAa,EAAE,CAAC,UAAU,CAAC,MAAM,EAAE,IAAI,CAAC,EAAE,UAAU,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC,EAAE,CAAC;AAC1E,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,OAAgC,EAAE,QAAgB,EAAE,IAAY,EACjG,IAAI,GAAG,KAAK;IACZ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,QAAQ,GAAG,CAAC,IAAI,QAAQ,GAAG,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IAClH,IAAI,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,MAAM,CAAC,OAAO,IAAI,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,MAAM,CAAC,CAAC,GAAG,CAAC,IAAI,MAAM,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;QAC5G,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IACD,MAAM,MAAM,GAAG,cAAc,CAAC,OAAO,CAAC,CAAC;IACvC,OAAO,EAAE,GAAG,MAAM,EAAE,QAAQ;QAC1B,QAAQ,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,GAAG,CAAC,KAAK,GAAG,CAAC,IAAI,CAAC,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QACzH,KAAK,EAAE,KAAK,CAAC,OAAO,CAAC;QACrB,aAAa,EAAE,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,EAAE,GAAG,GAAG,EAAE,CAAC,EAAE,QAAQ,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC;QACnF,GAAG,EAAE,wBAAwB,CAAC,OAAO,CAAC,EAAE,GAAG,cAAc,CAAC,OAAO,EAAE,IAAI,CAAC;QACxE,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,OAAO,IAAI,GAAG,CAAC,CAAC,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,MAAM,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;YAC/H,iBAAiB,EAAE,QAAQ,CAAC,OAAO,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,OAAO,IAAI,GAAG,CAAC,CAAC,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC3H,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,gBAAgB,CAAC,MAAsB;IACrD,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,IAAI,MAAM,CAAC,MAAM,GAAG,EAAE;QAAE,QAAQ,CAAC,IAAI,CAAC,uBAAuB,CAAC,CAAC;IAC/D,IAAI,MAAM,CAAC,SAAS,GAAG,EAAE;QAAE,QAAQ,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;IACrE,IAAI,MAAM,CAAC,IAAI,GAAG,CAAC;QAAE,QAAQ,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;IACzD,IAAI,MAAM,CAAC,GAAG,GAAG,CAAC;QAAE,QAAQ,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC;IACvD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,OAA2B;IACzD,MAAM,QAAQ,GAAG,gBAAgB,CAAC,OAAO,CAAC,CAAC;IAC3C,IAAI,OAAO,CAAC,GAAG,KAAK,IAAI,IAAI,OAAO,CAAC,WAAW,KAAK,IAAI;QAAE,QAAQ,CAAC,IAAI,CAAC,uDAAuD,CAAC,CAAC;IACjI,IAAI,QAAQ,CAAC,MAAM;QAAE,OAAO,EAAE,MAAM,EAAE,uBAAuB,EAAE,QAAQ,EAAE,QAAQ,EAAE,CAAC;IACpF,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,IAAI,CAAC,CAAC,OAAO,CAAC,GAAI,IAAI,IAAI,CAAC;QAAE,QAAQ,CAAC,IAAI,CAAC,yBAAyB,CAAC,CAAC;IACtE,IAAI,CAAC,CAAC,OAAO,CAAC,WAAY,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC;QAAE,QAAQ,CAAC,IAAI,CAAC,gCAAgC,CAAC,CAAC;IACxF,IAAI,CAAC,CAAC,OAAO,CAAC,KAAK,IAAI,IAAI,CAAC;QAAE,QAAQ,CAAC,IAAI,CAAC,2BAA2B,CAAC,CAAC;IACzE,IAAI,CAAC,CAAC,OAAO,CAAC,KAAK,GAAG,OAAO,CAAC,aAAa,CAAC;QAAE,QAAQ,CAAC,IAAI,CAAC,mCAAmC,CAAC,CAAC;IACjG,IAAI,OAAO,CAAC,aAAa,KAAK,SAAS,IAAI,CAAC,CAAC,OAAO,CAAC,aAAa,IAAI,IAAI,CAAC;QAAE,QAAQ,CAAC,IAAI,CAAC,4BAA4B,CAAC,CAAC;IACzH,IAAI,OAAO,CAAC,aAAa,KAAK,SAAS,IAAI,CAAC,CAAC,OAAO,CAAC,iBAAiB,EAAE,CAAC,CAAC,CAAC,KAAK,SAAS,IAAI,OAAO,CAAC,iBAAiB,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,EAAE,CAAC;QACnI,QAAQ,CAAC,IAAI,CAAC,+BAA+B,CAAC,CAAC;IACjD,CAAC;IACD,OAAO,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,CAAC,YAAY,EAAE,QAAQ,EAAE,CAAC;AAC/E,CAAC;AAED,wFAAwF;AACxF,MAAM,UAAU,uBAAuB,CAAC,OAA+C;IACrF,MAAM,QAAQ,GAAG,gBAAgB,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC,WAAW,OAAO,EAAE,CAAC,CAAC;IAChF,IAAI,OAAO,CAAC,SAAS,GAAG,IAAI;QAAE,QAAQ,CAAC,IAAI,CAAC,2BAA2B,CAAC,CAAC;IACzE,OAAO,QAAQ,CAAC;AAClB,CAAC;AAWD,MAAM,UAAU,gBAAgB,CAAC,OAA0C,EAAE,SAAiB;IAE5F,2FAA2F;IAC3F,MAAM,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,cAAc,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,cAAc,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC,CAAC;IACpH,MAAM,MAAM,GAAG,cAAc,CAAC,WAAW,CAAC,CAAC;IAC3C,MAAM,SAAS,GAAG,WAAW,CAAC,MAAM;QAClC,CAAC,CAAC,WAAW,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,aAAa,KAAK,SAAS,IAAI,GAAG,CAAC,aAAa,KAAK,GAAG,CAAC,YAAY,CAAC,CAAC,MAAM,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;IACzI,MAAM,QAAQ,GAAG,uBAAuB,CAAC,EAAE,GAAG,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IACnE,MAAM,YAAY,GAAG,gBAAgB,CAAC,MAAM,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC;IACzD,OAAO,EAAE,GAAG,MAAM,EAAE,SAAS,EAAE,kBAAkB,EAAE,OAAO,CAAC,MAAM,GAAG,WAAW,CAAC,MAAM,EAAE,QAAQ;QAC9F,MAAM,EAAE,YAAY,CAAC,CAAC,CAAC,uBAAuB,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,CAAC,YAAY,EAAE,CAAC;AACvG,CAAC"}
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
import type { CalibrationRecord, PlattParams } from './types.js';
|
|
2
|
+
export interface ModelFitRows {
|
|
3
|
+
id: string;
|
|
4
|
+
teacher: boolean;
|
|
5
|
+
fit: readonly CalibrationRecord[];
|
|
6
|
+
}
|
|
7
|
+
/** Calibrated mean binary NLL, with the entire golden fit denominator retained.
|
|
8
|
+
* A missing answer has maximum binary error, as in held-out metrics: truth
|
|
9
|
+
* probability zero, clipped to 1e-6 for finite NLL. Never impute p(good)=0.
|
|
10
|
+
*/
|
|
11
|
+
export declare function fitNll(rows: readonly CalibrationRecord[], params: PlattParams): number;
|
|
12
|
+
/** Strict improvement preserves authored order on ties, including no fit evidence. */
|
|
13
|
+
export declare function selectQuestionModel(questionId: string, type: CalibrationRecord['type'], candidates: readonly ModelFitRows[]): string;
|
|
@@ -0,0 +1,30 @@
|
|
|
1
|
+
import { applyPlatt, fitQuestion } from './fitting.js';
|
|
2
|
+
/** Calibrated mean binary NLL, with the entire golden fit denominator retained.
|
|
3
|
+
* A missing answer has maximum binary error, as in held-out metrics: truth
|
|
4
|
+
* probability zero, clipped to 1e-6 for finite NLL. Never impute p(good)=0.
|
|
5
|
+
*/
|
|
6
|
+
export function fitNll(rows, params) {
|
|
7
|
+
if (!rows.length)
|
|
8
|
+
return Infinity;
|
|
9
|
+
return rows.reduce((sum, row) => {
|
|
10
|
+
const p = row.missing ? 0 : applyPlatt(row.rawPGood, params);
|
|
11
|
+
const truth = row.missing ? 0 : row.good ? p : 1 - p;
|
|
12
|
+
return sum - Math.log(Math.max(1e-6, Math.min(1 - 1e-6, truth)));
|
|
13
|
+
}, 0) / rows.length;
|
|
14
|
+
}
|
|
15
|
+
/** Strict improvement preserves authored order on ties, including no fit evidence. */
|
|
16
|
+
export function selectQuestionModel(questionId, type, candidates) {
|
|
17
|
+
if (!candidates.length)
|
|
18
|
+
throw new Error('Model selection requires candidates');
|
|
19
|
+
let selected = candidates[0].id, lowest = Infinity;
|
|
20
|
+
for (const candidate of candidates) {
|
|
21
|
+
const fit = fitQuestion(candidate.fit, questionId, type);
|
|
22
|
+
const loss = fitNll(fit.questionSplit, candidate.teacher ? { a: 1, b: 0 } : fit.params);
|
|
23
|
+
if (loss < lowest) {
|
|
24
|
+
selected = candidate.id;
|
|
25
|
+
lowest = loss;
|
|
26
|
+
}
|
|
27
|
+
}
|
|
28
|
+
return selected;
|
|
29
|
+
}
|
|
30
|
+
//# sourceMappingURL=selection.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"selection.js","sourceRoot":"","sources":["../../../../src/training/graders/calibration/selection.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AASvD;;;GAGG;AACH,MAAM,UAAU,MAAM,CAAC,IAAkC,EAAE,MAAmB;IAC5E,IAAI,CAAC,IAAI,CAAC,MAAM;QAAE,OAAO,QAAQ,CAAC;IAClC,OAAO,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,EAAE;QAC9B,MAAM,CAAC,GAAG,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;QAC7D,MAAM,KAAK,GAAG,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QACrD,OAAO,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC;IACnE,CAAC,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC;AACtB,CAAC;AAED,sFAAsF;AACtF,MAAM,UAAU,mBAAmB,CAAC,UAAkB,EAAE,IAA+B,EAAE,UAAmC;IAC1H,IAAI,CAAC,UAAU,CAAC,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,qCAAqC,CAAC,CAAC;IAC/E,IAAI,QAAQ,GAAG,UAAU,CAAC,CAAC,CAAE,CAAC,EAAE,EAAE,MAAM,GAAG,QAAQ,CAAC;IACpD,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACnC,MAAM,GAAG,GAAG,WAAW,CAAC,SAAS,CAAC,GAAG,EAAE,UAAU,EAAE,IAAI,CAAC,CAAC;QACzD,MAAM,IAAI,GAAG,MAAM,CAAC,GAAG,CAAC,aAAa,EAAE,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QACxF,IAAI,IAAI,GAAG,MAAM,EAAE,CAAC;YAAC,QAAQ,GAAG,SAAS,CAAC,EAAE,CAAC;YAAC,MAAM,GAAG,IAAI,CAAC;QAAC,CAAC;IAChE,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC"}
|