agent-hitch 0.2.6 → 0.2.8
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 +30 -3
- package/README.zh-CN.md +14 -2
- package/dist/bin/hitch.js +25 -3
- package/dist/bin/hitch.js.map +1 -1
- package/dist/scripts/canary-benchmark.js +54 -0
- package/dist/scripts/canary-benchmark.js.map +1 -0
- package/dist/scripts/canary-buildkit-secrets.js +166 -0
- package/dist/scripts/canary-buildkit-secrets.js.map +1 -0
- package/dist/scripts/canary-collision-domains.js +77 -0
- package/dist/scripts/canary-collision-domains.js.map +1 -0
- package/dist/scripts/canary-docker-image-observation.js +181 -0
- package/dist/scripts/canary-docker-image-observation.js.map +1 -0
- package/dist/scripts/canary-eval-scheduler-throughput.js +47 -0
- package/dist/scripts/canary-eval-scheduler-throughput.js.map +1 -0
- package/dist/scripts/canary-gpu-hardware.js +95 -0
- package/dist/scripts/canary-gpu-hardware.js.map +1 -0
- package/dist/scripts/canary-harbor-load.js +365 -0
- package/dist/scripts/canary-harbor-load.js.map +1 -0
- package/dist/scripts/canary-packaged-harness.js +75 -0
- package/dist/scripts/canary-packaged-harness.js.map +1 -0
- package/dist/scripts/canary-resource-load.js +188 -0
- package/dist/scripts/canary-resource-load.js.map +1 -0
- package/dist/scripts/check-architecture.js +10 -5
- package/dist/scripts/check-architecture.js.map +1 -1
- package/dist/scripts/docker-reaper-crash-child.js +21 -0
- package/dist/scripts/docker-reaper-crash-child.js.map +1 -0
- package/dist/src/adapters/catalog.js +3 -1
- package/dist/src/adapters/catalog.js.map +1 -1
- package/dist/src/adapters/discovery.js +1 -1
- package/dist/src/adapters/discovery.js.map +1 -1
- package/dist/src/adapters/providers/claude.js +6 -0
- package/dist/src/adapters/providers/claude.js.map +1 -1
- package/dist/src/adapters/providers/codex-auth.js +29 -0
- package/dist/src/adapters/providers/codex-auth.js.map +1 -0
- package/dist/src/adapters/providers/codex.js +29 -6
- package/dist/src/adapters/providers/codex.js.map +1 -1
- package/dist/src/adapters/providers/deepseek.js +6 -0
- package/dist/src/adapters/providers/deepseek.js.map +1 -1
- package/dist/src/adapters/providers/model-call.js +26 -0
- package/dist/src/adapters/providers/model-call.js.map +1 -0
- package/dist/src/adapters/providers/opencode.js +6 -0
- package/dist/src/adapters/providers/opencode.js.map +1 -1
- package/dist/src/adapters/providers/pi.js +6 -0
- package/dist/src/adapters/providers/pi.js.map +1 -1
- package/dist/src/artifacts/index.js +1 -0
- package/dist/src/artifacts/index.js.map +1 -1
- package/dist/src/artifacts/preparer.js +7 -2
- package/dist/src/artifacts/preparer.js.map +1 -1
- package/dist/src/backends/harbor/agent-budget.js +31 -0
- package/dist/src/backends/harbor/agent-budget.js.map +1 -0
- package/dist/src/backends/harbor/backend.js +59 -107
- package/dist/src/backends/harbor/backend.js.map +1 -1
- package/dist/src/backends/harbor/environment-config.js +92 -0
- package/dist/src/backends/harbor/environment-config.js.map +1 -0
- package/dist/src/backends/harbor/index.js +6 -1
- package/dist/src/backends/harbor/index.js.map +1 -1
- package/dist/src/backends/harbor/model-proxy-config.js +23 -0
- package/dist/src/backends/harbor/model-proxy-config.js.map +1 -0
- package/dist/src/backends/harbor/process.js +177 -0
- package/dist/src/backends/harbor/process.js.map +1 -0
- package/dist/src/backends/harbor/regrade.js +77 -0
- package/dist/src/backends/harbor/regrade.js.map +1 -0
- package/dist/src/backends/harbor/runtime-toolchain.js +30 -0
- package/dist/src/backends/harbor/runtime-toolchain.js.map +1 -0
- package/dist/src/backends/harbor/supervisor.js +74 -0
- package/dist/src/backends/harbor/supervisor.js.map +1 -0
- package/dist/src/backends/harbor/task-resources.js +206 -0
- package/dist/src/backends/harbor/task-resources.js.map +1 -0
- package/dist/src/backends/harbor/tools.js +6 -29
- package/dist/src/backends/harbor/tools.js.map +1 -1
- package/dist/src/backends/index.js +5 -1
- package/dist/src/backends/index.js.map +1 -1
- package/dist/src/benchmarks/index.js +3 -0
- package/dist/src/benchmarks/index.js.map +1 -0
- package/dist/src/benchmarks/loader.js +271 -0
- package/dist/src/benchmarks/loader.js.map +1 -0
- package/dist/src/benchmarks/metrics.js +16 -0
- package/dist/src/benchmarks/metrics.js.map +1 -0
- package/dist/src/benchmarks/toml.js +13 -0
- package/dist/src/benchmarks/toml.js.map +1 -0
- package/dist/src/benchmarks/validation.js +242 -0
- package/dist/src/benchmarks/validation.js.map +1 -0
- package/dist/src/cli/arguments.js +6 -4
- package/dist/src/cli/arguments.js.map +1 -1
- package/dist/src/cli/commands/benchmark.js +22 -0
- package/dist/src/cli/commands/benchmark.js.map +1 -0
- package/dist/src/cli/commands/capabilities.js +18 -0
- package/dist/src/cli/commands/capabilities.js.map +1 -0
- package/dist/src/cli/commands/daemon.js +135 -8
- package/dist/src/cli/commands/daemon.js.map +1 -1
- package/dist/src/cli/commands/eval.js +228 -6
- package/dist/src/cli/commands/eval.js.map +1 -1
- package/dist/src/cli/commands/images.js +40 -0
- package/dist/src/cli/commands/images.js.map +1 -0
- package/dist/src/cli/commands/run.js +42 -17
- package/dist/src/cli/commands/run.js.map +1 -1
- package/dist/src/cli/commands/runs.js +25 -3
- package/dist/src/cli/commands/runs.js.map +1 -1
- package/dist/src/cli/commands/trajectory.js +96 -5
- package/dist/src/cli/commands/trajectory.js.map +1 -1
- package/dist/src/cli/commands/verifier.js +36 -0
- package/dist/src/cli/commands/verifier.js.map +1 -0
- package/dist/src/cli/commands/worker.js +79 -0
- package/dist/src/cli/commands/worker.js.map +1 -0
- package/dist/src/cli/main.js +10 -0
- package/dist/src/cli/main.js.map +1 -1
- package/dist/src/cli/output.js +84 -4
- package/dist/src/cli/output.js.map +1 -1
- package/dist/src/control-plane/build-admission.js +69 -0
- package/dist/src/control-plane/build-admission.js.map +1 -0
- package/dist/src/control-plane/build-records.js +6 -0
- package/dist/src/control-plane/build-records.js.map +1 -0
- package/dist/src/control-plane/collisions.js +40 -0
- package/dist/src/control-plane/collisions.js.map +1 -0
- package/dist/src/control-plane/eval-control-state.js +22 -0
- package/dist/src/control-plane/eval-control-state.js.map +1 -0
- package/dist/src/control-plane/eval-control-work.js +44 -0
- package/dist/src/control-plane/eval-control-work.js.map +1 -0
- package/dist/src/control-plane/eval-image-resolution.js +52 -0
- package/dist/src/control-plane/eval-image-resolution.js.map +1 -0
- package/dist/src/control-plane/eval-image-services.js +30 -0
- package/dist/src/control-plane/eval-image-services.js.map +1 -0
- package/dist/src/control-plane/eval-records.js +252 -0
- package/dist/src/control-plane/eval-records.js.map +1 -0
- package/dist/src/control-plane/eval-recovery.js +181 -0
- package/dist/src/control-plane/eval-recovery.js.map +1 -0
- package/dist/src/control-plane/eval-scheduler.js +450 -0
- package/dist/src/control-plane/eval-scheduler.js.map +1 -0
- package/dist/src/control-plane/index.js +20 -0
- package/dist/src/control-plane/index.js.map +1 -0
- package/dist/src/control-plane/local-worker.js +49 -0
- package/dist/src/control-plane/local-worker.js.map +1 -0
- package/dist/src/control-plane/model-capture-planning.js +12 -0
- package/dist/src/control-plane/model-capture-planning.js.map +1 -0
- package/dist/src/control-plane/remote-result-transport.js +175 -0
- package/dist/src/control-plane/remote-result-transport.js.map +1 -0
- package/dist/src/control-plane/remote-work-coordinator.js +333 -0
- package/dist/src/control-plane/remote-work-coordinator.js.map +1 -0
- package/dist/src/control-plane/remote-work-inputs.js +266 -0
- package/dist/src/control-plane/remote-work-inputs.js.map +1 -0
- package/dist/src/control-plane/remote-work-item.js +62 -0
- package/dist/src/control-plane/remote-work-item.js.map +1 -0
- package/dist/src/control-plane/remote-work-recovery.js +220 -0
- package/dist/src/control-plane/remote-work-recovery.js.map +1 -0
- package/dist/src/control-plane/remote-worker-artifacts.js +135 -0
- package/dist/src/control-plane/remote-worker-artifacts.js.map +1 -0
- package/dist/src/control-plane/remote-worker-client.js +202 -0
- package/dist/src/control-plane/remote-worker-client.js.map +1 -0
- package/dist/src/control-plane/remote-worker-configuration.js +34 -0
- package/dist/src/control-plane/remote-worker-configuration.js.map +1 -0
- package/dist/src/control-plane/remote-worker-credentials.js +58 -0
- package/dist/src/control-plane/remote-worker-credentials.js.map +1 -0
- package/dist/src/control-plane/remote-worker-protocol.js +489 -0
- package/dist/src/control-plane/remote-worker-protocol.js.map +1 -0
- package/dist/src/control-plane/remote-worker-runner.js +214 -0
- package/dist/src/control-plane/remote-worker-runner.js.map +1 -0
- package/dist/src/control-plane/remote-workers.js +377 -0
- package/dist/src/control-plane/remote-workers.js.map +1 -0
- package/dist/src/control-plane/rerun-scheduler.js +448 -0
- package/dist/src/control-plane/rerun-scheduler.js.map +1 -0
- package/dist/src/control-plane/rerun-submission.js +62 -0
- package/dist/src/control-plane/rerun-submission.js.map +1 -0
- package/dist/src/control-plane/resources.js +182 -0
- package/dist/src/control-plane/resources.js.map +1 -0
- package/dist/src/control-plane/scheduler-eval-entry.js +32 -0
- package/dist/src/control-plane/scheduler-eval-entry.js.map +1 -0
- package/dist/src/control-plane/synthetic-result.js +21 -0
- package/dist/src/control-plane/synthetic-result.js.map +1 -0
- package/dist/src/control-plane/work-admission.js +23 -0
- package/dist/src/control-plane/work-admission.js.map +1 -0
- package/dist/src/control-plane/work-dispatcher.js +237 -0
- package/dist/src/control-plane/work-dispatcher.js.map +1 -0
- package/dist/src/controller-runtime/hash.js +19 -3
- package/dist/src/controller-runtime/hash.js.map +1 -1
- package/dist/src/controller-runtime/index.js +1 -1
- package/dist/src/controller-runtime/index.js.map +1 -1
- package/dist/src/controller-runtime/store.js +15 -6
- package/dist/src/controller-runtime/store.js.map +1 -1
- package/dist/src/daemon/auth.js +5 -0
- package/dist/src/daemon/auth.js.map +1 -1
- package/dist/src/daemon/client.js.map +1 -1
- package/dist/src/daemon/event-stream.js +46 -0
- package/dist/src/daemon/event-stream.js.map +1 -0
- package/dist/src/daemon/health.js +66 -0
- package/dist/src/daemon/health.js.map +1 -0
- package/dist/src/daemon/launcher.js +28 -1
- package/dist/src/daemon/launcher.js.map +1 -1
- package/dist/src/daemon/logging.js +23 -0
- package/dist/src/daemon/logging.js.map +1 -0
- package/dist/src/daemon/scheduler.js +26 -5
- package/dist/src/daemon/scheduler.js.map +1 -1
- package/dist/src/daemon/server.js +255 -98
- package/dist/src/daemon/server.js.map +1 -1
- package/dist/src/daemon/telemetry.js +163 -0
- package/dist/src/daemon/telemetry.js.map +1 -0
- package/dist/src/daemon/worker-routes.js +219 -0
- package/dist/src/daemon/worker-routes.js.map +1 -0
- package/dist/src/domain/benchmarks.js +2 -0
- package/dist/src/domain/benchmarks.js.map +1 -0
- package/dist/src/domain/bundles.js +2 -0
- package/dist/src/domain/bundles.js.map +1 -0
- package/dist/src/domain/control-plane.js +2 -0
- package/dist/src/domain/control-plane.js.map +1 -0
- package/dist/src/domain/execution-evidence.js +2 -0
- package/dist/src/domain/execution-evidence.js.map +1 -0
- package/dist/src/domain/execution-plan.js +2 -0
- package/dist/src/domain/execution-plan.js.map +1 -0
- package/dist/src/domain/images.js +2 -0
- package/dist/src/domain/images.js.map +1 -0
- package/dist/src/domain/index.js +13 -0
- package/dist/src/domain/index.js.map +1 -1
- package/dist/src/domain/interactions.js +2 -0
- package/dist/src/domain/interactions.js.map +1 -0
- package/dist/src/domain/providers.js +2 -0
- package/dist/src/domain/providers.js.map +1 -0
- package/dist/src/domain/resources.js +2 -0
- package/dist/src/domain/resources.js.map +1 -0
- package/dist/src/domain/runs.js.map +1 -1
- package/dist/src/domain/validation.js +11 -4
- package/dist/src/domain/validation.js.map +1 -1
- package/dist/src/domain/verifier-evidence-validation.js +214 -0
- package/dist/src/domain/verifier-evidence-validation.js.map +1 -0
- package/dist/src/domain/verifier-evidence.js +3 -0
- package/dist/src/domain/verifier-evidence.js.map +1 -0
- package/dist/src/domain/verifier-score-contract.js +232 -0
- package/dist/src/domain/verifier-score-contract.js.map +1 -0
- package/dist/src/domain/workers.js +2 -0
- package/dist/src/domain/workers.js.map +1 -0
- package/dist/src/evals/benchmark-adapter-manifest.js +219 -0
- package/dist/src/evals/benchmark-adapter-manifest.js.map +1 -0
- package/dist/src/evals/benchmark-candidate.js +36 -0
- package/dist/src/evals/benchmark-candidate.js.map +1 -0
- package/dist/src/evals/benchmark-run.js +166 -0
- package/dist/src/evals/benchmark-run.js.map +1 -0
- package/dist/src/evals/collect-only-rerun.js +183 -0
- package/dist/src/evals/collect-only-rerun.js.map +1 -0
- package/dist/src/evals/directory.js +30 -0
- package/dist/src/evals/directory.js.map +1 -0
- package/dist/src/evals/docker-engine-stats.js +89 -0
- package/dist/src/evals/docker-engine-stats.js.map +1 -0
- package/dist/src/evals/docker-ownership.js +51 -0
- package/dist/src/evals/docker-ownership.js.map +1 -0
- package/dist/src/evals/docker-reaper.js +144 -0
- package/dist/src/evals/docker-reaper.js.map +1 -0
- package/dist/src/evals/docker-resource-observer.js +207 -0
- package/dist/src/evals/docker-resource-observer.js.map +1 -0
- package/dist/src/evals/duration-estimator.js +126 -0
- package/dist/src/evals/duration-estimator.js.map +1 -0
- package/dist/src/evals/environment-image-planning.js +164 -0
- package/dist/src/evals/environment-image-planning.js.map +1 -0
- package/dist/src/evals/eval-artifact-planning.js +56 -0
- package/dist/src/evals/eval-artifact-planning.js.map +1 -0
- package/dist/src/evals/eval-finalization.js +10 -0
- package/dist/src/evals/eval-finalization.js.map +1 -0
- package/dist/src/evals/eval-lifecycle-events.js +29 -0
- package/dist/src/evals/eval-lifecycle-events.js.map +1 -0
- package/dist/src/evals/eval-logical-plan.js +51 -0
- package/dist/src/evals/eval-logical-plan.js.map +1 -0
- package/dist/src/evals/eval-result-builder.js +81 -0
- package/dist/src/evals/eval-result-builder.js.map +1 -0
- package/dist/src/evals/events.js +53 -5
- package/dist/src/evals/events.js.map +1 -1
- package/dist/src/evals/evolution-baseline.js +191 -0
- package/dist/src/evals/evolution-baseline.js.map +1 -0
- package/dist/src/evals/execution-evidence.js +116 -0
- package/dist/src/evals/execution-evidence.js.map +1 -0
- package/dist/src/evals/execution-leases.js +342 -0
- package/dist/src/evals/execution-leases.js.map +1 -0
- package/dist/src/evals/execution-plan-artifacts.js +72 -0
- package/dist/src/evals/execution-plan-artifacts.js.map +1 -0
- package/dist/src/evals/execution-plan-images.js +69 -0
- package/dist/src/evals/execution-plan-images.js.map +1 -0
- package/dist/src/evals/execution-plan-resources.js +174 -0
- package/dist/src/evals/execution-plan-resources.js.map +1 -0
- package/dist/src/evals/execution-plan.js +451 -0
- package/dist/src/evals/execution-plan.js.map +1 -0
- package/dist/src/evals/failure-classifier.js +61 -0
- package/dist/src/evals/failure-classifier.js.map +1 -0
- package/dist/src/evals/harbor-artifact-builder-image.js +99 -0
- package/dist/src/evals/harbor-artifact-builder-image.js.map +1 -0
- package/dist/src/evals/harbor-artifact-builder.js +358 -0
- package/dist/src/evals/harbor-artifact-builder.js.map +1 -0
- package/dist/src/evals/harbor-bridge-error.js +58 -3
- package/dist/src/evals/harbor-bridge-error.js.map +1 -1
- package/dist/src/evals/harbor-node-runtime.js +146 -0
- package/dist/src/evals/harbor-node-runtime.js.map +1 -0
- package/dist/src/evals/harbor-phase-timings.js +60 -0
- package/dist/src/evals/harbor-phase-timings.js.map +1 -0
- package/dist/src/evals/index.js +40 -2
- package/dist/src/evals/index.js.map +1 -1
- package/dist/src/evals/infrastructure-retry.js +208 -84
- package/dist/src/evals/infrastructure-retry.js.map +1 -1
- package/dist/src/evals/interaction-capture-import.js +49 -0
- package/dist/src/evals/interaction-capture-import.js.map +1 -0
- package/dist/src/evals/local-docker-provider.js +372 -0
- package/dist/src/evals/local-docker-provider.js.map +1 -0
- package/dist/src/evals/local-eval-planning.js +80 -0
- package/dist/src/evals/local-eval-planning.js.map +1 -0
- package/dist/src/evals/model-capture-plan.js +22 -0
- package/dist/src/evals/model-capture-plan.js.map +1 -0
- package/dist/src/evals/model-capture-runtime.js +114 -0
- package/dist/src/evals/model-capture-runtime.js.map +1 -0
- package/dist/src/evals/model-proxy-runtime-state.js +53 -0
- package/dist/src/evals/model-proxy-runtime-state.js.map +1 -0
- package/dist/src/evals/native-phase-evidence.js +440 -0
- package/dist/src/evals/native-phase-evidence.js.map +1 -0
- package/dist/src/evals/physical-retry-work.js +31 -0
- package/dist/src/evals/physical-retry-work.js.map +1 -0
- package/dist/src/evals/planned-execution-support.js +66 -0
- package/dist/src/evals/planned-execution-support.js.map +1 -0
- package/dist/src/evals/planned-execution.js +469 -0
- package/dist/src/evals/planned-execution.js.map +1 -0
- package/dist/src/evals/planned-progress-publisher.js +62 -0
- package/dist/src/evals/planned-progress-publisher.js.map +1 -0
- package/dist/src/evals/planned-remote-execution.js +40 -0
- package/dist/src/evals/planned-remote-execution.js.map +1 -0
- package/dist/src/evals/planned-retry-execution.js +174 -0
- package/dist/src/evals/planned-retry-execution.js.map +1 -0
- package/dist/src/evals/planned-retry-lifecycle.js +188 -0
- package/dist/src/evals/planned-retry-lifecycle.js.map +1 -0
- package/dist/src/evals/preparation-rerun.js +225 -0
- package/dist/src/evals/preparation-rerun.js.map +1 -0
- package/dist/src/evals/prepared-harness.js +76 -0
- package/dist/src/evals/prepared-harness.js.map +1 -0
- package/dist/src/evals/progress.js +52 -12
- package/dist/src/evals/progress.js.map +1 -1
- package/dist/src/evals/records.js +20 -3
- package/dist/src/evals/records.js.map +1 -1
- package/dist/src/evals/recovery.js +242 -0
- package/dist/src/evals/recovery.js.map +1 -0
- package/dist/src/evals/regrade-evidence.js +77 -0
- package/dist/src/evals/regrade-evidence.js.map +1 -0
- package/dist/src/evals/remote-infrastructure-retry.js +165 -0
- package/dist/src/evals/remote-infrastructure-retry.js.map +1 -0
- package/dist/src/evals/request.js +21 -3
- package/dist/src/evals/request.js.map +1 -1
- package/dist/src/evals/rerun-inputs.js +108 -0
- package/dist/src/evals/rerun-inputs.js.map +1 -0
- package/dist/src/evals/rerun-slots.js +4 -3
- package/dist/src/evals/rerun-slots.js.map +1 -1
- package/dist/src/evals/rerun-types.js +45 -0
- package/dist/src/evals/rerun-types.js.map +1 -0
- package/dist/src/evals/rerun.js +133 -88
- package/dist/src/evals/rerun.js.map +1 -1
- package/dist/src/evals/result-helpers.js +78 -0
- package/dist/src/evals/result-helpers.js.map +1 -0
- package/dist/src/evals/resume-state.js +36 -0
- package/dist/src/evals/resume-state.js.map +1 -0
- package/dist/src/evals/retry-backoff.js +20 -0
- package/dist/src/evals/retry-backoff.js.map +1 -0
- package/dist/src/evals/retry-state.js +195 -0
- package/dist/src/evals/retry-state.js.map +1 -0
- package/dist/src/evals/scheduler-metrics.js +95 -0
- package/dist/src/evals/scheduler-metrics.js.map +1 -0
- package/dist/src/evals/scheduler-trace-replay.js +107 -0
- package/dist/src/evals/scheduler-trace-replay.js.map +1 -0
- package/dist/src/evals/service-types.js +2 -0
- package/dist/src/evals/service-types.js.map +1 -0
- package/dist/src/evals/service.js +331 -347
- package/dist/src/evals/service.js.map +1 -1
- package/dist/src/evals/task-resources.js +209 -0
- package/dist/src/evals/task-resources.js.map +1 -0
- package/dist/src/evals/trial-environment-evidence.js +87 -0
- package/dist/src/evals/trial-environment-evidence.js.map +1 -0
- package/dist/src/evals/trial-execution-evidence.js +13 -0
- package/dist/src/evals/trial-execution-evidence.js.map +1 -0
- package/dist/src/evals/trial-import-identity.js +31 -0
- package/dist/src/evals/trial-import-identity.js.map +1 -0
- package/dist/src/evals/trial-import.js +183 -188
- package/dist/src/evals/trial-import.js.map +1 -1
- package/dist/src/evals/trial-publication-recovery.js +144 -0
- package/dist/src/evals/trial-publication-recovery.js.map +1 -0
- package/dist/src/evals/trial-publication.js +25 -0
- package/dist/src/evals/trial-publication.js.map +1 -0
- package/dist/src/evals/trial-reference-validation.js +49 -0
- package/dist/src/evals/trial-reference-validation.js.map +1 -0
- package/dist/src/evals/verifier-artifacts.js +317 -0
- package/dist/src/evals/verifier-artifacts.js.map +1 -0
- package/dist/src/evals/verifier-eligibility.js +25 -0
- package/dist/src/evals/verifier-eligibility.js.map +1 -0
- package/dist/src/evals/verifier-only-rerun.js +210 -0
- package/dist/src/evals/verifier-only-rerun.js.map +1 -0
- package/dist/src/evals/verifier-runtime.js +61 -0
- package/dist/src/evals/verifier-runtime.js.map +1 -0
- package/dist/src/evals/verifier-score-artifacts.js +119 -0
- package/dist/src/evals/verifier-score-artifacts.js.map +1 -0
- package/dist/src/evals/work-item-artifacts.js +23 -0
- package/dist/src/evals/work-item-artifacts.js.map +1 -0
- package/dist/src/foundation/config.js +17 -0
- package/dist/src/foundation/config.js.map +1 -1
- package/dist/src/foundation/contained-file.js +85 -0
- package/dist/src/foundation/contained-file.js.map +1 -0
- package/dist/src/foundation/credential-redaction.js +162 -0
- package/dist/src/foundation/credential-redaction.js.map +1 -0
- package/dist/src/foundation/environment-image-references.js +59 -0
- package/dist/src/foundation/environment-image-references.js.map +1 -0
- package/dist/src/foundation/executable.js +45 -5
- package/dist/src/foundation/executable.js.map +1 -1
- package/dist/src/foundation/fs.js +2 -1
- package/dist/src/foundation/fs.js.map +1 -1
- package/dist/src/foundation/index.js +6 -2
- package/dist/src/foundation/index.js.map +1 -1
- package/dist/src/foundation/process-identity.js +69 -0
- package/dist/src/foundation/process-identity.js.map +1 -0
- package/dist/src/images/context.js +47 -0
- package/dist/src/images/context.js.map +1 -0
- package/dist/src/images/docker-buildkit.js +137 -0
- package/dist/src/images/docker-buildkit.js.map +1 -0
- package/dist/src/images/dockerfile.js +67 -0
- package/dist/src/images/dockerfile.js.map +1 -0
- package/dist/src/images/gc.js +261 -0
- package/dist/src/images/gc.js.map +1 -0
- package/dist/src/images/index.js +10 -0
- package/dist/src/images/index.js.map +1 -0
- package/dist/src/images/manifest.js +109 -0
- package/dist/src/images/manifest.js.map +1 -0
- package/dist/src/images/ownership.js +5 -0
- package/dist/src/images/ownership.js.map +1 -0
- package/dist/src/images/records.js +32 -0
- package/dist/src/images/records.js.map +1 -0
- package/dist/src/images/registry.js +154 -0
- package/dist/src/images/registry.js.map +1 -0
- package/dist/src/images/service.js +224 -0
- package/dist/src/images/service.js.map +1 -0
- package/dist/src/model-access/capture.js +255 -0
- package/dist/src/model-access/capture.js.map +1 -0
- package/dist/src/model-access/index.js +5 -0
- package/dist/src/model-access/index.js.map +1 -0
- package/dist/src/model-access/policy.js +49 -0
- package/dist/src/model-access/policy.js.map +1 -0
- package/dist/src/model-access/proxy.js +336 -0
- package/dist/src/model-access/proxy.js.map +1 -0
- package/dist/src/model-access/records.js +167 -0
- package/dist/src/model-access/records.js.map +1 -0
- package/dist/src/runs/adapter-process.js +12 -0
- package/dist/src/runs/adapter-process.js.map +1 -0
- package/dist/src/runs/bundle.js +450 -0
- package/dist/src/runs/bundle.js.map +1 -0
- package/dist/src/runs/events.js +12 -1
- package/dist/src/runs/events.js.map +1 -1
- package/dist/src/runs/executor.js +46 -45
- package/dist/src/runs/executor.js.map +1 -1
- package/dist/src/runs/finalizer.js +17 -0
- package/dist/src/runs/finalizer.js.map +1 -1
- package/dist/src/runs/index.js +8 -1
- package/dist/src/runs/index.js.map +1 -1
- package/dist/src/runs/manifest.js +5 -5
- package/dist/src/runs/manifest.js.map +1 -1
- package/dist/src/runs/phase-bundle.js +86 -0
- package/dist/src/runs/phase-bundle.js.map +1 -0
- package/dist/src/runs/phase-cancellation.js +107 -0
- package/dist/src/runs/phase-cancellation.js.map +1 -0
- package/dist/src/runs/phase-group.js +113 -0
- package/dist/src/runs/phase-group.js.map +1 -0
- package/dist/src/runs/query.js +1 -1
- package/dist/src/runs/query.js.map +1 -1
- package/dist/src/runs/records.js +3 -0
- package/dist/src/runs/records.js.map +1 -1
- package/dist/src/runs/request.js +10 -3
- package/dist/src/runs/request.js.map +1 -1
- package/dist/src/runs/training-candidate.js +161 -0
- package/dist/src/runs/training-candidate.js.map +1 -0
- package/dist/src/runs/verifier-evidence-redaction.js +61 -0
- package/dist/src/runs/verifier-evidence-redaction.js.map +1 -0
- package/dist/src/runs/verifier-evidence.js +459 -0
- package/dist/src/runs/verifier-evidence.js.map +1 -0
- package/dist/src/runs/verifier-structured-evidence.js +83 -0
- package/dist/src/runs/verifier-structured-evidence.js.map +1 -0
- package/dist/src/trajectories/analysis.js +288 -0
- package/dist/src/trajectories/analysis.js.map +1 -0
- package/dist/src/trajectories/chunk-projection.js +148 -0
- package/dist/src/trajectories/chunk-projection.js.map +1 -0
- package/dist/src/trajectories/content-projection.js +389 -0
- package/dist/src/trajectories/content-projection.js.map +1 -0
- package/dist/src/trajectories/dsh-chunk-contract.js +176 -0
- package/dist/src/trajectories/dsh-chunk-contract.js.map +1 -0
- package/dist/src/trajectories/dsh-contract.js +284 -0
- package/dist/src/trajectories/dsh-contract.js.map +1 -0
- package/dist/src/trajectories/events-chunk-drill.js +95 -0
- package/dist/src/trajectories/events-chunk-drill.js.map +1 -0
- package/dist/src/trajectories/events-page.js +367 -0
- package/dist/src/trajectories/events-page.js.map +1 -0
- package/dist/src/trajectories/index.js +3 -0
- package/dist/src/trajectories/index.js.map +1 -1
- package/dist/src/trajectories/provider-capture.js +15 -24
- package/dist/src/trajectories/provider-capture.js.map +1 -1
- package/dist/src/trajectories/providers/deepseek.js +3 -3
- package/dist/src/trajectories/providers/deepseek.js.map +1 -1
- package/dist/src/trajectories/request-attempt.js +47 -0
- package/dist/src/trajectories/request-attempt.js.map +1 -0
- package/dist/src/trajectories/stream-reader.js +311 -0
- package/dist/src/trajectories/stream-reader.js.map +1 -0
- package/dist/src/trajectories/surface-fold.js +126 -0
- package/dist/src/trajectories/surface-fold.js.map +1 -0
- package/dist/src/workers/index.js +3 -0
- package/dist/src/workers/index.js.map +1 -0
- package/dist/src/workers/remote-harbor-work-spec.js +97 -0
- package/dist/src/workers/remote-harbor-work-spec.js.map +1 -0
- package/dist/src/workers/remote-harbor-worker.js +223 -0
- package/dist/src/workers/remote-harbor-worker.js.map +1 -0
- package/docs/schemas/benchmark-hook-request-v1.schema.json +65 -0
- package/docs/schemas/benchmark-hook-response-v1.schema.json +75 -0
- package/docs/schemas/benchmark-hook-v1.schema.json +33 -0
- package/docs/schemas/benchmark-lock-v1.schema.json +351 -0
- package/docs/schemas/benchmark-metric-v1.schema.json +37 -0
- package/docs/schemas/benchmark-package-v1.schema.json +167 -0
- package/docs/schemas/benchmark-phase-group.schema.json +41 -0
- package/docs/schemas/benchmark-profile-v1.schema.json +141 -0
- package/docs/schemas/benchmark-task-v1.schema.json +384 -0
- package/docs/schemas/benchmark-tool-result-v1.schema.json +30 -0
- package/docs/schemas/environment-build-index.schema.json +13 -0
- package/docs/schemas/environment-build-record.schema.json +25 -0
- package/docs/schemas/environment-image-manifest.schema.json +62 -0
- package/docs/schemas/error.schema.json +22 -0
- package/docs/schemas/eval-control.schema.json +36 -0
- package/docs/schemas/eval-execution-plan.schema.json +216 -0
- package/docs/schemas/eval-progress.schema.json +1 -15
- package/docs/schemas/eval-rerun-result.schema.json +58 -0
- package/docs/schemas/eval-rerun-state.schema.json +61 -0
- package/docs/schemas/eval-rerun-submission.schema.json +60 -0
- package/docs/schemas/eval-result.schema.json +32 -13
- package/docs/schemas/eval-submission-input.schema.json +14 -0
- package/docs/schemas/eval-submission.schema.json +68 -0
- package/docs/schemas/eval-trial-publication.schema.json +15 -0
- package/docs/schemas/eval-trial-reference.schema.json +129 -0
- package/docs/schemas/execution-evidence.schema.json +96 -0
- package/docs/schemas/execution-lease.schema.json +50 -0
- package/docs/schemas/execution-provider-status.schema.json +65 -0
- package/docs/schemas/execution-worker.schema.json +53 -0
- package/docs/schemas/harbor-process-exit.schema.json +14 -0
- package/docs/schemas/interaction-capture-ref.schema.json +45 -0
- package/docs/schemas/local-provider-execution.schema.json +33 -0
- package/docs/schemas/model-interaction.schema.json +41 -0
- package/docs/schemas/regrade-assessment-reference.schema.json +11 -0
- package/docs/schemas/remote-result-envelope.schema.json +32 -0
- package/docs/schemas/remote-tree-envelope.schema.json +25 -0
- package/docs/schemas/remote-work-offer.schema.json +58 -0
- package/docs/schemas/remote-work-receipt.schema.json +50 -0
- package/docs/schemas/remote-worker-artifact.schema.json +17 -0
- package/docs/schemas/remote-worker-event.schema.json +18 -0
- package/docs/schemas/remote-worker-heartbeat.schema.json +37 -0
- package/docs/schemas/remote-worker-registration.schema.json +52 -0
- package/docs/schemas/result-bundle-index.schema.json +133 -0
- package/docs/schemas/run-context.schema.json +15 -0
- package/docs/schemas/run-environment-images.schema.json +39 -0
- package/docs/schemas/run-manifest.schema.json +4 -0
- package/docs/schemas/run-request.schema.json +4 -0
- package/docs/schemas/training-data-candidate.schema.json +46 -0
- package/docs/schemas/trajectory-analysis.schema.json +258 -0
- package/docs/schemas/trajectory-events-page.schema.json +40 -0
- package/docs/schemas/verifier-assessment.schema.json +77 -0
- package/docs/schemas/verifier-evidence.schema.json +276 -0
- package/integrations/harbor/hitch_benchmark.py +252 -0
- package/integrations/harbor/hitch_candidate_recycle.py +262 -0
- package/integrations/harbor/hitch_harbor_agent.py +698 -568
- package/integrations/harbor/hitch_harbor_environment.py +469 -0
- package/integrations/harbor/hitch_harbor_task_resources.py +275 -0
- package/integrations/harbor/hitch_harbor_verifier.py +83 -2
- package/integrations/harbor/hitch_phase_supervisor.py +452 -0
- package/integrations/harbor/hitch_tool_client.mjs +91 -0
- package/integrations/model-call/cli.js +57 -0
- package/package.json +30 -6
|
@@ -0,0 +1,144 @@
|
|
|
1
|
+
import { readdir, readFile } from "node:fs/promises";
|
|
2
|
+
import path from "node:path";
|
|
3
|
+
import { HitchError, readJSON, sha256Bytes, statePaths } from "../foundation/index.js";
|
|
4
|
+
import { verifyResultBundleIndex } from "../runs/index.js";
|
|
5
|
+
import { evalTrialKey, mergeEvalProgressTrial, replaceInvalidEvalProgressTrial, writeEvalProgress } from "./progress.js";
|
|
6
|
+
import { validateEvalTrialReferences } from "./trial-import.js";
|
|
7
|
+
import { EVAL_TRIAL_PUBLICATION_REF, parseEvalTrialPublication } from "./trial-publication.js";
|
|
8
|
+
const RUN_ID = /^run_[a-f0-9]{32}$/;
|
|
9
|
+
export async function recoverPromotedEvalTrialPublications(input) {
|
|
10
|
+
const evalId = input.plan.eval_id;
|
|
11
|
+
const publications = await readEvalTrialPublications(input.root, evalId);
|
|
12
|
+
const bySlot = new Map();
|
|
13
|
+
for (const publication of publications) {
|
|
14
|
+
assertPublicationBelongsToPlan(publication, input.plan);
|
|
15
|
+
if (!publication.trial.run_group)
|
|
16
|
+
await verifyResultBundleIndex(path.join(statePaths(input.root).runs, publication.trial.run_id));
|
|
17
|
+
await validateEvalTrialReferences(input.root, evalId, [publication.trial], {
|
|
18
|
+
benchmarkId: input.plan.benchmark.id,
|
|
19
|
+
benchmarkRevision: input.plan.benchmark.revision,
|
|
20
|
+
});
|
|
21
|
+
const key = evalTrialKey(publication.trial);
|
|
22
|
+
const entries = bySlot.get(key) ?? [];
|
|
23
|
+
entries.push(publication);
|
|
24
|
+
bySlot.set(key, entries);
|
|
25
|
+
}
|
|
26
|
+
let progress = input.progress;
|
|
27
|
+
const recovered = [];
|
|
28
|
+
for (const key of [...bySlot.keys()].sort()) {
|
|
29
|
+
const entries = bySlot.get(key);
|
|
30
|
+
const settlements = entries.filter((entry) => entry.mode === "settle");
|
|
31
|
+
const repairs = entries.filter((entry) => entry.mode === "replace-invalid" && entry.trial.observation_status === "valid");
|
|
32
|
+
if (settlements.length > 1)
|
|
33
|
+
throw conflict(`multiple promoted authoritative runs exist for ${formatTrial(settlements[0].trial)}`);
|
|
34
|
+
if (repairs.length > 1)
|
|
35
|
+
throw conflict(`multiple promoted repair runs exist for ${formatTrial(repairs[0].trial)}`);
|
|
36
|
+
const selected = repairs[0] ?? settlements[0];
|
|
37
|
+
if (!selected)
|
|
38
|
+
continue;
|
|
39
|
+
const current = progress.trials.find((trial) => evalTrialKey(trial) === key);
|
|
40
|
+
if (current && JSON.stringify(current) === JSON.stringify(selected.trial))
|
|
41
|
+
continue;
|
|
42
|
+
if (current && selected.mode === "settle") {
|
|
43
|
+
throw conflict(`promoted run conflicts with durable progress for ${formatTrial(selected.trial)}`);
|
|
44
|
+
}
|
|
45
|
+
if (current?.observation_status === "valid") {
|
|
46
|
+
throw conflict(`promoted repair conflicts with an existing valid run for ${formatTrial(selected.trial)}`);
|
|
47
|
+
}
|
|
48
|
+
progress = selected.mode === "replace-invalid"
|
|
49
|
+
? replaceInvalidEvalProgressTrial(progress, selected.trial)
|
|
50
|
+
: mergeEvalProgressTrial(progress, selected.trial);
|
|
51
|
+
recovered.push({ mode: selected.mode, trial: selected.trial });
|
|
52
|
+
}
|
|
53
|
+
if (recovered.length > 0)
|
|
54
|
+
await writeEvalProgress(input.evalDirectory, progress);
|
|
55
|
+
for (const item of recovered)
|
|
56
|
+
input.sink?.emit({
|
|
57
|
+
type: "eval.trial.publication-recovered", mode: item.mode, trial_id: item.trial.trial_id,
|
|
58
|
+
task_id: item.trial.task_id, attempt: item.trial.attempt, run_id: item.trial.run_id,
|
|
59
|
+
...(item.trial.run_group ? { run_group_id: item.trial.run_group.run_group_id } : {}),
|
|
60
|
+
observation_status: item.trial.observation_status, generation: progress.generation,
|
|
61
|
+
});
|
|
62
|
+
return { progress, recovered };
|
|
63
|
+
}
|
|
64
|
+
async function readEvalTrialPublications(root, evalId) {
|
|
65
|
+
let entries;
|
|
66
|
+
try {
|
|
67
|
+
entries = await readdir(statePaths(root).runs, { withFileTypes: true });
|
|
68
|
+
}
|
|
69
|
+
catch (error) {
|
|
70
|
+
if (error.code === "ENOENT")
|
|
71
|
+
entries = [];
|
|
72
|
+
else
|
|
73
|
+
throw error;
|
|
74
|
+
}
|
|
75
|
+
const result = [];
|
|
76
|
+
for (const entry of entries.sort((left, right) => left.name.localeCompare(right.name))) {
|
|
77
|
+
if (!entry.isDirectory() || !RUN_ID.test(entry.name))
|
|
78
|
+
continue;
|
|
79
|
+
const value = await readJSON(path.join(statePaths(root).runs, entry.name, EVAL_TRIAL_PUBLICATION_REF), null);
|
|
80
|
+
if (value === null || !belongsToEval(value, evalId))
|
|
81
|
+
continue;
|
|
82
|
+
result.push(parseEvalTrialPublication(value, entry.name));
|
|
83
|
+
}
|
|
84
|
+
const assessments = path.join(statePaths(root).evals, evalId, "assessments");
|
|
85
|
+
let assessmentEntries;
|
|
86
|
+
try {
|
|
87
|
+
assessmentEntries = await readdir(assessments, { withFileTypes: true });
|
|
88
|
+
}
|
|
89
|
+
catch (error) {
|
|
90
|
+
if (error.code === "ENOENT")
|
|
91
|
+
return result;
|
|
92
|
+
throw error;
|
|
93
|
+
}
|
|
94
|
+
for (const entry of assessmentEntries) {
|
|
95
|
+
if (!entry.isDirectory() || !/^assessment_[a-f0-9]{32}$/.test(entry.name))
|
|
96
|
+
continue;
|
|
97
|
+
let value = await readJSON(path.join(assessments, entry.name, EVAL_TRIAL_PUBLICATION_REF), null);
|
|
98
|
+
if (value === null) {
|
|
99
|
+
// assessment.json is the immutable commit. Recover a crash between its
|
|
100
|
+
// publication and writing the convenience publication receipt.
|
|
101
|
+
const manifestPath = path.join(assessments, entry.name, "assessment.json");
|
|
102
|
+
const record = await readJSON(manifestPath, null);
|
|
103
|
+
if (record?.kind === "native-phase-assessment" && record.eval_id === evalId) {
|
|
104
|
+
const observation = record.observation;
|
|
105
|
+
value = { schema_version: "1", eval_id: evalId, mode: record.publication_mode, created_at: record.created_at,
|
|
106
|
+
trial: { trial_id: record.trial_id, task_id: record.task_id, attempt: record.attempt, run_group: record.run_group,
|
|
107
|
+
assessment: { id: entry.name, digest: sha256Bytes(await readFile(manifestPath)) }, observation_status: observation.status,
|
|
108
|
+
...(observation.reward !== undefined ? { reward: observation.reward } : {}),
|
|
109
|
+
...(record.scores !== undefined ? { scores: record.scores } : {}),
|
|
110
|
+
...(observation.invalid_reason ? { invalid_reason: observation.invalid_reason } : {}),
|
|
111
|
+
...(observation.verifier_result_ref ? { verifier_result_ref: observation.verifier_result_ref } : {}) } };
|
|
112
|
+
}
|
|
113
|
+
}
|
|
114
|
+
if (value === null || !belongsToEval(value, evalId))
|
|
115
|
+
continue;
|
|
116
|
+
const publication = parseEvalTrialPublication(value);
|
|
117
|
+
if (!publication.trial.run_group || publication.trial.assessment.id !== entry.name)
|
|
118
|
+
throw conflict("native phase publication assessment mismatch");
|
|
119
|
+
result.push(publication);
|
|
120
|
+
}
|
|
121
|
+
return result;
|
|
122
|
+
}
|
|
123
|
+
function belongsToEval(value, evalId) {
|
|
124
|
+
return Boolean(value && typeof value === "object" && !Array.isArray(value)
|
|
125
|
+
&& value.eval_id === evalId);
|
|
126
|
+
}
|
|
127
|
+
function assertPublicationBelongsToPlan(publication, plan) {
|
|
128
|
+
if (publication.eval_id !== plan.eval_id)
|
|
129
|
+
throw conflict("promoted run belongs to another eval plan");
|
|
130
|
+
const trial = publication.trial;
|
|
131
|
+
const planned = plan.membership === "known"
|
|
132
|
+
? plan.slots.some((slot) => slot.task_id === trial.task_id && slot.attempt === trial.attempt)
|
|
133
|
+
: plan.work_items.some((work) => work.logical_attempt === trial.attempt
|
|
134
|
+
&& (work.task_ids.length === 0 || work.task_ids.includes(trial.task_id)));
|
|
135
|
+
if (!planned)
|
|
136
|
+
throw conflict(`promoted run is outside the frozen plan: ${formatTrial(trial)}`);
|
|
137
|
+
}
|
|
138
|
+
function formatTrial(trial) {
|
|
139
|
+
return `${trial.task_id} attempt ${trial.attempt}`;
|
|
140
|
+
}
|
|
141
|
+
function conflict(message) {
|
|
142
|
+
return new HitchError(message, { code: "eval_trial_publication_conflict", exitCode: 12 });
|
|
143
|
+
}
|
|
144
|
+
//# sourceMappingURL=trial-publication-recovery.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"trial-publication-recovery.js","sourceRoot":"","sources":["../../../src/evals/trial-publication-recovery.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAErD,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,WAAW,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AACvF,OAAO,EAAE,uBAAuB,EAAE,MAAM,kBAAkB,CAAC;AAC3D,OAAO,EAAE,YAAY,EAAE,sBAAsB,EAAE,+BAA+B,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAEzH,OAAO,EAAE,2BAA2B,EAAE,MAAM,mBAAmB,CAAC;AAChE,OAAO,EAAE,0BAA0B,EAAE,yBAAyB,EAAE,MAAM,wBAAwB,CAAC;AAG/F,MAAM,MAAM,GAAG,oBAAoB,CAAC;AAOpC,MAAM,CAAC,KAAK,UAAU,oCAAoC,CAAC,KAM1D;IACC,MAAM,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC;IAClC,MAAM,YAAY,GAAG,MAAM,yBAAyB,CAAC,KAAK,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IACzE,MAAM,MAAM,GAAG,IAAI,GAAG,EAAoC,CAAC;IAC3D,KAAK,MAAM,WAAW,IAAI,YAAY,EAAE,CAAC;QACvC,8BAA8B,CAAC,WAAW,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;QACxD,IAAI,CAAC,WAAW,CAAC,KAAK,CAAC,SAAS;YAAE,MAAM,uBAAuB,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,WAAW,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC;QAClI,MAAM,2BAA2B,CAAC,KAAK,CAAC,IAAI,EAAE,MAAM,EAAE,CAAC,WAAW,CAAC,KAAK,CAAC,EAAE;YACzE,WAAW,EAAE,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE;YACpC,iBAAiB,EAAE,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ;SACjD,CAAC,CAAC;QACH,MAAM,GAAG,GAAG,YAAY,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC;QAC5C,MAAM,OAAO,GAAG,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC;QACtC,OAAO,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QAC1B,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;IAC3B,CAAC;IAED,IAAI,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC;IAC9B,MAAM,SAAS,GAAoC,EAAE,CAAC;IACtD,KAAK,MAAM,GAAG,IAAI,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC;QAC5C,MAAM,OAAO,GAAG,MAAM,CAAC,GAAG,CAAC,GAAG,CAA6B,CAAC;QAC5D,MAAM,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC;QACvE,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,iBAAiB,IAAI,KAAK,CAAC,KAAK,CAAC,kBAAkB,KAAK,OAAO,CAAC,CAAC;QAC1H,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC;YAAE,MAAM,QAAQ,CAAC,kDAAkD,WAAW,CAAC,WAAW,CAAC,CAAC,CAAE,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACnI,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC;YAAE,MAAM,QAAQ,CAAC,2CAA2C,WAAW,CAAC,OAAO,CAAC,CAAC,CAAE,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACpH,MAAM,QAAQ,GAAG,OAAO,CAAC,CAAC,CAAC,IAAI,WAAW,CAAC,CAAC,CAAC,CAAC;QAC9C,IAAI,CAAC,QAAQ;YAAE,SAAS;QACxB,MAAM,OAAO,GAAG,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,YAAY,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC;QAC7E,IAAI,OAAO,IAAI,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,KAAK,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,KAAK,CAAC;YAAE,SAAS;QACpF,IAAI,OAAO,IAAI,QAAQ,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YAC1C,MAAM,QAAQ,CAAC,oDAAoD,WAAW,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACpG,CAAC;QACD,IAAI,OAAO,EAAE,kBAAkB,KAAK,OAAO,EAAE,CAAC;YAC5C,MAAM,QAAQ,CAAC,4DAA4D,WAAW,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAC5G,CAAC;QACD,QAAQ,GAAG,QAAQ,CAAC,IAAI,KAAK,iBAAiB;YAC5C,CAAC,CAAC,+BAA+B,CAAC,QAAQ,EAAE,QAAQ,CAAC,KAAK,CAAC;YAC3D,CAAC,CAAC,sBAAsB,CAAC,QAAQ,EAAE,QAAQ,CAAC,KAAK,CAAC,CAAC;QACrD,SAAS,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,CAAC,IAAI,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC;IACjE,CAAC;IACD,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC;QAAE,MAAM,iBAAiB,CAAC,KAAK,CAAC,aAAa,EAAE,QAAQ,CAAC,CAAC;IACjF,KAAK,MAAM,IAAI,IAAI,SAAS;QAAE,KAAK,CAAC,IAAI,EAAE,IAAI,CAAC;YAC7C,IAAI,EAAE,kCAAkC,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,QAAQ,EAAE,IAAI,CAAC,KAAK,CAAC,QAAQ;YACxF,OAAO,EAAE,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,OAAO,EAAE,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,MAAM,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM;YACnF,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACpF,kBAAkB,EAAE,IAAI,CAAC,KAAK,CAAC,kBAAkB,EAAE,UAAU,EAAE,QAAQ,CAAC,UAAU;SACnF,CAAC,CAAC;IACH,OAAO,EAAE,QAAQ,EAAE,SAAS,EAAE,CAAC;AACjC,CAAC;AAED,KAAK,UAAU,yBAAyB,CAAC,IAAY,EAAE,MAAc;IACnE,IAAI,OAAiB,CAAC;IACtB,IAAI,CAAC;QAAC,OAAO,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;IAAC,CAAC;IAChF,OAAO,KAAK,EAAE,CAAC;QACb,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,GAAG,EAAE,CAAC;;YAChE,MAAM,KAAK,CAAC;IACnB,CAAC;IACD,MAAM,MAAM,GAA6B,EAAE,CAAC;IAC5C,KAAK,MAAM,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;QACvF,IAAI,CAAC,KAAK,CAAC,WAAW,EAAE,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC;YAAE,SAAS;QAC/D,MAAM,KAAK,GAAG,MAAM,QAAQ,CAAiB,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,0BAA0B,CAAC,EAAE,IAAI,CAAC,CAAC;QAC7H,IAAI,KAAK,KAAK,IAAI,IAAI,CAAC,aAAa,CAAC,KAAK,EAAE,MAAM,CAAC;YAAE,SAAS;QAC9D,MAAM,CAAC,IAAI,CAAC,yBAAyB,CAAC,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;IAC5D,CAAC;IACD,MAAM,WAAW,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,aAAa,CAAC,CAAC;IAC7E,IAAI,iBAAiB,CAAC;IACtB,IAAI,CAAC;QAAC,iBAAiB,GAAG,MAAM,OAAO,CAAC,WAAW,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;IAAC,CAAC;IAChF,OAAO,KAAK,EAAE,CAAC;QAAC,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,MAAM,CAAC;QAAC,MAAM,KAAK,CAAC;IAAC,CAAC;IACrG,KAAK,MAAM,KAAK,IAAI,iBAAiB,EAAE,CAAC;QACtC,IAAI,CAAC,KAAK,CAAC,WAAW,EAAE,IAAI,CAAC,2BAA2B,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC;YAAE,SAAS;QACpF,IAAI,KAAK,GAAG,MAAM,QAAQ,CAAiB,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,KAAK,CAAC,IAAI,EAAE,0BAA0B,CAAC,EAAE,IAAI,CAAC,CAAC;QACjH,IAAI,KAAK,KAAK,IAAI,EAAE,CAAC;YACnB,uEAAuE;YACvE,+DAA+D;YAC/D,MAAM,YAAY,GAAG,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,KAAK,CAAC,IAAI,EAAE,iBAAiB,CAAC,CAAC;YAC3E,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAiC,YAAY,EAAE,IAAI,CAAC,CAAC;YAClF,IAAI,MAAM,EAAE,IAAI,KAAK,yBAAyB,IAAI,MAAM,CAAC,OAAO,KAAK,MAAM,EAAE,CAAC;gBAC5E,MAAM,WAAW,GAAG,MAAM,CAAC,WAAsC,CAAC;gBAClE,KAAK,GAAG,EAAE,cAAc,EAAE,GAAG,EAAE,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,gBAAgB,EAAE,UAAU,EAAE,MAAM,CAAC,UAAU;oBAC1G,KAAK,EAAE,EAAE,QAAQ,EAAE,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,SAAS,EAAE,MAAM,CAAC,SAAS;wBAC/G,UAAU,EAAE,EAAE,EAAE,EAAE,KAAK,CAAC,IAAI,EAAE,MAAM,EAAE,WAAW,CAAC,MAAM,QAAQ,CAAC,YAAY,CAAC,CAAC,EAAE,EAAE,kBAAkB,EAAE,WAAW,CAAC,MAAM;wBACzH,GAAG,CAAC,WAAW,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,WAAW,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;wBAC3E,GAAG,CAAC,MAAM,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;wBACjE,GAAG,CAAC,WAAW,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,cAAc,EAAE,WAAW,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;wBACrF,GAAG,CAAC,WAAW,CAAC,mBAAmB,CAAC,CAAC,CAAC,EAAE,mBAAmB,EAAE,WAAW,CAAC,mBAAmB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;YAC/G,CAAC;QACH,CAAC;QACD,IAAI,KAAK,KAAK,IAAI,IAAI,CAAC,aAAa,CAAC,KAAK,EAAE,MAAM,CAAC;YAAE,SAAS;QAC9D,MAAM,WAAW,GAAG,yBAAyB,CAAC,KAAK,CAAC,CAAC;QACrD,IAAI,CAAC,WAAW,CAAC,KAAK,CAAC,SAAS,IAAI,WAAW,CAAC,KAAK,CAAC,UAAU,CAAC,EAAE,KAAK,KAAK,CAAC,IAAI;YAAE,MAAM,QAAQ,CAAC,8CAA8C,CAAC,CAAC;QACnJ,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;IAC3B,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,SAAS,aAAa,CAAC,KAAc,EAAE,MAAc;IACnD,OAAO,OAAO,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;WACpE,KAAiC,CAAC,OAAO,KAAK,MAAM,CAAC,CAAC;AAC9D,CAAC;AAED,SAAS,8BAA8B,CAAC,WAAmC,EAAE,IAAyB;IACpG,IAAI,WAAW,CAAC,OAAO,KAAK,IAAI,CAAC,OAAO;QAAE,MAAM,QAAQ,CAAC,2CAA2C,CAAC,CAAC;IACtG,MAAM,KAAK,GAAG,WAAW,CAAC,KAAK,CAAC;IAChC,MAAM,OAAO,GAAG,IAAI,CAAC,UAAU,KAAK,OAAO;QACzC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,KAAK,KAAK,CAAC,OAAO,IAAI,IAAI,CAAC,OAAO,KAAK,KAAK,CAAC,OAAO,CAAC;QAC7F,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,eAAe,KAAK,KAAK,CAAC,OAAO;eAClE,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,IAAI,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;IAC9E,IAAI,CAAC,OAAO;QAAE,MAAM,QAAQ,CAAC,4CAA4C,WAAW,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;AACjG,CAAC;AAED,SAAS,WAAW,CAAC,KAAkD;IACrE,OAAO,GAAG,KAAK,CAAC,OAAO,YAAY,KAAK,CAAC,OAAO,EAAE,CAAC;AACrD,CAAC;AAED,SAAS,QAAQ,CAAC,OAAe;IAC/B,OAAO,IAAI,UAAU,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,iCAAiC,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC;AAC5F,CAAC"}
|
|
@@ -0,0 +1,25 @@
|
|
|
1
|
+
import path from "node:path";
|
|
2
|
+
import { atomicWriteJSON } from "../foundation/index.js";
|
|
3
|
+
import { parseEvalTrialRef } from "./progress.js";
|
|
4
|
+
const EVAL_ID = /^eval_[a-f0-9]{32}$/;
|
|
5
|
+
export const EVAL_TRIAL_PUBLICATION_REF = path.join("eval", "publication.json");
|
|
6
|
+
export async function writeEvalTrialPublication(runDirectory, evalId, mode, trial) {
|
|
7
|
+
await atomicWriteJSON(path.join(runDirectory, EVAL_TRIAL_PUBLICATION_REF), {
|
|
8
|
+
schema_version: "1", eval_id: evalId, mode, trial, created_at: new Date().toISOString(),
|
|
9
|
+
});
|
|
10
|
+
}
|
|
11
|
+
export function parseEvalTrialPublication(value, expectedRunId) {
|
|
12
|
+
if (!value || typeof value !== "object" || Array.isArray(value))
|
|
13
|
+
throw new TypeError("eval trial publication must be an object");
|
|
14
|
+
const record = value;
|
|
15
|
+
if (record.schema_version !== "1" || typeof record.eval_id !== "string" || !EVAL_ID.test(record.eval_id)
|
|
16
|
+
|| (record.mode !== "settle" && record.mode !== "replace-invalid")
|
|
17
|
+
|| typeof record.created_at !== "string" || !Number.isFinite(Date.parse(record.created_at))) {
|
|
18
|
+
throw new TypeError("eval trial publication is invalid");
|
|
19
|
+
}
|
|
20
|
+
const trial = parseEvalTrialRef(record.trial, "eval trial publication trial");
|
|
21
|
+
if (expectedRunId !== undefined && trial.run_id !== expectedRunId)
|
|
22
|
+
throw new TypeError("eval trial publication run identity is invalid");
|
|
23
|
+
return { schema_version: "1", eval_id: record.eval_id, mode: record.mode, trial, created_at: record.created_at };
|
|
24
|
+
}
|
|
25
|
+
//# sourceMappingURL=trial-publication.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"trial-publication.js","sourceRoot":"","sources":["../../../src/evals/trial-publication.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AACzD,OAAO,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAElD,MAAM,OAAO,GAAG,qBAAqB,CAAC;AAEtC,MAAM,CAAC,MAAM,0BAA0B,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,kBAAkB,CAAC,CAAC;AAWhF,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAC7C,YAAoB,EACpB,MAAc,EACd,IAA8B,EAC9B,KAAqB;IAErB,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,0BAA0B,CAAC,EAAE;QACzE,cAAc,EAAE,GAAG,EAAE,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,UAAU,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;KACxF,CAAC,CAAC;AACL,CAAC;AAED,MAAM,UAAU,yBAAyB,CAAC,KAAc,EAAE,aAAsB;IAC9E,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,0CAA0C,CAAC,CAAC;IACjI,MAAM,MAAM,GAAG,KAAgC,CAAC;IAChD,IAAI,MAAM,CAAC,cAAc,KAAK,GAAG,IAAI,OAAO,MAAM,CAAC,OAAO,KAAK,QAAQ,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC;WACnG,CAAC,MAAM,CAAC,IAAI,KAAK,QAAQ,IAAI,MAAM,CAAC,IAAI,KAAK,iBAAiB,CAAC;WAC/D,OAAO,MAAM,CAAC,UAAU,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,EAAE,CAAC;QAC9F,MAAM,IAAI,SAAS,CAAC,mCAAmC,CAAC,CAAC;IAC3D,CAAC;IACD,MAAM,KAAK,GAAG,iBAAiB,CAAC,MAAM,CAAC,KAAK,EAAE,8BAA8B,CAAC,CAAC;IAC9E,IAAI,aAAa,KAAK,SAAS,IAAI,KAAK,CAAC,MAAM,KAAK,aAAa;QAAE,MAAM,IAAI,SAAS,CAAC,gDAAgD,CAAC,CAAC;IACzI,OAAO,EAAE,cAAc,EAAE,GAAG,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,CAAC,UAAU,EAAE,CAAC;AACnH,CAAC"}
|
|
@@ -0,0 +1,49 @@
|
|
|
1
|
+
import path from "node:path";
|
|
2
|
+
import { isDeepStrictEqual } from "node:util";
|
|
3
|
+
import { statePaths } from "../foundation/index.js";
|
|
4
|
+
import { benchmarkVerifierIdentity, loadRunRecord } from "../runs/index.js";
|
|
5
|
+
import { parseVerifierScores } from "../domain/index.js";
|
|
6
|
+
import { readJSON } from "../foundation/index.js";
|
|
7
|
+
import { readRegradeObservation } from "./regrade-evidence.js";
|
|
8
|
+
import { readNativePhaseObservation } from "./native-phase-evidence.js";
|
|
9
|
+
export async function validateEvalTrialReferences(root, evalId, trials, expected) {
|
|
10
|
+
for (const trial of trials) {
|
|
11
|
+
let observation;
|
|
12
|
+
if (trial.run_group)
|
|
13
|
+
observation = await readNativePhaseObservation(root, evalId, trial, expected);
|
|
14
|
+
else {
|
|
15
|
+
const loaded = await loadRunRecord(path.join(statePaths(root).runs, trial.run_id), { verifyTrajectory: false });
|
|
16
|
+
const record = loaded.record;
|
|
17
|
+
if (record.context.kind !== "benchmark_task")
|
|
18
|
+
throw new Error(`eval trial ${trial.trial_id} references a non-benchmark run`);
|
|
19
|
+
if (expected && (record.context.benchmark_id !== expected.benchmarkId
|
|
20
|
+
|| record.context.benchmark_revision !== expected.benchmarkRevision
|
|
21
|
+
|| record.context.verifier_identity !== benchmarkVerifierIdentity(expected.benchmarkId, expected.benchmarkRevision)))
|
|
22
|
+
throw new Error(`eval trial ${trial.trial_id} benchmark identity mismatch`);
|
|
23
|
+
if (record.parent?.eval_id !== evalId || record.parent.trial_id !== trial.trial_id || record.parent.attempt !== trial.attempt) {
|
|
24
|
+
throw new Error(`eval trial ${trial.trial_id} parent mismatch`);
|
|
25
|
+
}
|
|
26
|
+
if (record.context.task_id !== trial.task_id)
|
|
27
|
+
throw new Error(`eval trial ${trial.trial_id} task mismatch`);
|
|
28
|
+
observation = trial.assessment ? await readRegradeObservation(root, evalId, trial) : record.observation;
|
|
29
|
+
}
|
|
30
|
+
if (observation?.status !== trial.observation_status)
|
|
31
|
+
throw new Error(`eval trial ${trial.trial_id} observation status mismatch`);
|
|
32
|
+
if (observation?.reward !== trial.reward)
|
|
33
|
+
throw new Error(`eval trial ${trial.trial_id} reward mismatch`);
|
|
34
|
+
if (observation?.verifier_result_ref !== trial.verifier_result_ref)
|
|
35
|
+
throw new Error(`eval trial ${trial.trial_id} verifier ref mismatch`);
|
|
36
|
+
if (observation?.invalid_reason !== trial.invalid_reason)
|
|
37
|
+
throw new Error(`eval trial ${trial.trial_id} invalid reason mismatch`);
|
|
38
|
+
if (trial.scores !== undefined && !trial.run_group) {
|
|
39
|
+
const directory = trial.assessment
|
|
40
|
+
? path.join(statePaths(root).evals, evalId, "assessments", trial.assessment.id)
|
|
41
|
+
: path.join(statePaths(root).runs, trial.run_id);
|
|
42
|
+
const result = await readJSON(path.join(directory, observation?.verifier_result_ref ?? "verifier/result.json"));
|
|
43
|
+
if (!isDeepStrictEqual(parseVerifierScores(result), trial.scores)) {
|
|
44
|
+
throw new Error(`eval trial ${trial.trial_id} score channels mismatch`);
|
|
45
|
+
}
|
|
46
|
+
}
|
|
47
|
+
}
|
|
48
|
+
}
|
|
49
|
+
//# sourceMappingURL=trial-reference-validation.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"trial-reference-validation.js","sourceRoot":"","sources":["../../../src/evals/trial-reference-validation.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,EAAE,iBAAiB,EAAE,MAAM,WAAW,CAAC;AAE9C,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AACpD,OAAO,EAAE,yBAAyB,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AAC5E,OAAO,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AACzD,OAAO,EAAE,QAAQ,EAAE,MAAM,wBAAwB,CAAC;AAClD,OAAO,EAAE,sBAAsB,EAAE,MAAM,uBAAuB,CAAC;AAC/D,OAAO,EAAE,0BAA0B,EAAE,MAAM,4BAA4B,CAAC;AAExE,MAAM,CAAC,KAAK,UAAU,2BAA2B,CAC/C,IAAY,EACZ,MAAc,EACd,MAAwB,EACxB,QAA6D;IAE7D,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;QAC3B,IAAI,WAAyC,CAAC;QAC9C,IAAI,KAAK,CAAC,SAAS;YAAE,WAAW,GAAG,MAAM,0BAA0B,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,QAAQ,CAAC,CAAC;aAC9F,CAAC;YACJ,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,EAAE,gBAAgB,EAAE,KAAK,EAAE,CAAC,CAAC;YAChH,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC;YAC7B,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,KAAK,gBAAgB;gBAAE,MAAM,IAAI,KAAK,CAAC,cAAc,KAAK,CAAC,QAAQ,iCAAiC,CAAC,CAAC;YAC7H,IAAI,QAAQ,IAAI,CACd,MAAM,CAAC,OAAO,CAAC,YAAY,KAAK,QAAQ,CAAC,WAAW;mBACjD,MAAM,CAAC,OAAO,CAAC,kBAAkB,KAAK,QAAQ,CAAC,iBAAiB;mBAChE,MAAM,CAAC,OAAO,CAAC,iBAAiB,KAAK,yBAAyB,CAAC,QAAQ,CAAC,WAAW,EAAE,QAAQ,CAAC,iBAAiB,CAAC,CACpH;gBAAE,MAAM,IAAI,KAAK,CAAC,cAAc,KAAK,CAAC,QAAQ,8BAA8B,CAAC,CAAC;YAC/E,IAAI,MAAM,CAAC,MAAM,EAAE,OAAO,KAAK,MAAM,IAAI,MAAM,CAAC,MAAM,CAAC,QAAQ,KAAK,KAAK,CAAC,QAAQ,IAAI,MAAM,CAAC,MAAM,CAAC,OAAO,KAAK,KAAK,CAAC,OAAO,EAAE,CAAC;gBAC9H,MAAM,IAAI,KAAK,CAAC,cAAc,KAAK,CAAC,QAAQ,kBAAkB,CAAC,CAAC;YAClE,CAAC;YACD,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,KAAK,KAAK,CAAC,OAAO;gBAAE,MAAM,IAAI,KAAK,CAAC,cAAc,KAAK,CAAC,QAAQ,gBAAgB,CAAC,CAAC;YAC5G,WAAW,GAAG,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,MAAM,sBAAsB,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,WAAW,CAAC;QAC1G,CAAC;QACD,IAAI,WAAW,EAAE,MAAM,KAAK,KAAK,CAAC,kBAAkB;YAAE,MAAM,IAAI,KAAK,CAAC,cAAc,KAAK,CAAC,QAAQ,8BAA8B,CAAC,CAAC;QAClI,IAAI,WAAW,EAAE,MAAM,KAAK,KAAK,CAAC,MAAM;YAAE,MAAM,IAAI,KAAK,CAAC,cAAc,KAAK,CAAC,QAAQ,kBAAkB,CAAC,CAAC;QAC1G,IAAI,WAAW,EAAE,mBAAmB,KAAK,KAAK,CAAC,mBAAmB;YAAE,MAAM,IAAI,KAAK,CAAC,cAAc,KAAK,CAAC,QAAQ,wBAAwB,CAAC,CAAC;QAC1I,IAAI,WAAW,EAAE,cAAc,KAAK,KAAK,CAAC,cAAc;YAAE,MAAM,IAAI,KAAK,CAAC,cAAc,KAAK,CAAC,QAAQ,0BAA0B,CAAC,CAAC;QAClI,IAAI,KAAK,CAAC,MAAM,KAAK,SAAS,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC;YACnD,MAAM,SAAS,GAAG,KAAK,CAAC,UAAU;gBAChC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,aAAa,EAAE,KAAK,CAAC,UAAU,CAAC,EAAE,CAAC;gBAC/E,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;YACnD,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,WAAW,EAAE,mBAAmB,IAAI,sBAAsB,CAAC,CAAC,CAAC;YAChH,IAAI,CAAC,iBAAiB,CAAC,mBAAmB,CAAC,MAAM,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC;gBAClE,MAAM,IAAI,KAAK,CAAC,cAAc,KAAK,CAAC,QAAQ,0BAA0B,CAAC,CAAC;YAC1E,CAAC;QACH,CAAC;IACH,CAAC;AACH,CAAC"}
|
|
@@ -0,0 +1,317 @@
|
|
|
1
|
+
import { createHash } from "node:crypto";
|
|
2
|
+
import { realpath } from "node:fs/promises";
|
|
3
|
+
import { StringDecoder } from "node:string_decoder";
|
|
4
|
+
import path from "node:path";
|
|
5
|
+
import { PROVIDER_ENVIRONMENT_NAMES, atomicWriteJSON, credentialValuesFromEnv, openContainedRegularFile, redactCredentialText, writePrivateFile } from "../foundation/index.js";
|
|
6
|
+
import { captureVerifierScoreEvidence } from "./verifier-score-artifacts.js";
|
|
7
|
+
export const DEFAULT_VERIFIER_DIAGNOSTIC_MAX_BYTES = 64 * 1024;
|
|
8
|
+
export const MAX_VERIFIER_DIAGNOSTIC_MAX_BYTES = 16 * 1024 * 1024;
|
|
9
|
+
export const VERIFIER_DIAGNOSTICS_INDEX_REF = "verifier/diagnostics.json";
|
|
10
|
+
export const VERIFIER_ARTIFACT_NAMES = [
|
|
11
|
+
"ctrf.json", "test-stdout.txt", "test-stderr.txt", "stdout.txt", "stderr.txt",
|
|
12
|
+
];
|
|
13
|
+
const TRUNCATION_MARKER = Buffer.from("\n[... verifier artifact truncated ...]\n", "utf8");
|
|
14
|
+
const MAX_LOG_LINE_BYTES = 1024 * 1024;
|
|
15
|
+
export async function persistTrialVerifierDiagnostics(input) {
|
|
16
|
+
const env = input.env ?? process.env;
|
|
17
|
+
const credentialValues = credentialValuesFromEnv([...PROVIDER_ENVIRONMENT_NAMES, ...(input.passEnv ?? [])], env);
|
|
18
|
+
throwIfAborted(input.signal);
|
|
19
|
+
await copyVerifierRetryHistory(input.trialDirectory, input.runDirectory, credentialValues, input.signal);
|
|
20
|
+
await copyCandidateIneligibleDiagnostic(input.trialDirectory, input.runDirectory, input.signal);
|
|
21
|
+
const scores = await captureVerifierScoreEvidence({
|
|
22
|
+
trialDirectory: input.trialDirectory,
|
|
23
|
+
runDirectory: input.runDirectory,
|
|
24
|
+
verifierResult: input.verifierResult ?? null,
|
|
25
|
+
credentialValues,
|
|
26
|
+
...(input.dataset === undefined ? {} : { dataset: input.dataset }),
|
|
27
|
+
...(input.benchmarkRevision === undefined ? {} : { benchmarkRevision: input.benchmarkRevision }),
|
|
28
|
+
...(input.signal ? { signal: input.signal } : {}),
|
|
29
|
+
});
|
|
30
|
+
await captureVerifierDiagnostics(input.trialDirectory, input.runDirectory, {
|
|
31
|
+
...(input.maxArtifactBytes === undefined ? {} : { maxArtifactBytes: input.maxArtifactBytes }),
|
|
32
|
+
credentialValues,
|
|
33
|
+
...(input.signal ? { signal: input.signal } : {}),
|
|
34
|
+
});
|
|
35
|
+
return scores;
|
|
36
|
+
}
|
|
37
|
+
async function copyCandidateIneligibleDiagnostic(trialDirectory, runDirectory, signal) {
|
|
38
|
+
const root = await realpath(trialDirectory);
|
|
39
|
+
throwIfAborted(signal);
|
|
40
|
+
const safe = await safeArtifact(root, "candidate-ineligible.json", 16 * 1024);
|
|
41
|
+
if (!safe)
|
|
42
|
+
return;
|
|
43
|
+
let bytes;
|
|
44
|
+
try {
|
|
45
|
+
bytes = await safe.handle.readFile();
|
|
46
|
+
await safe.assertUnchanged();
|
|
47
|
+
}
|
|
48
|
+
finally {
|
|
49
|
+
await safe.handle.close();
|
|
50
|
+
}
|
|
51
|
+
const value = JSON.parse(bytes.toString("utf8"));
|
|
52
|
+
if (value.schema_version !== "1" || value.code !== "candidate_evidence_unavailable" || value.verifier_executed !== false)
|
|
53
|
+
return;
|
|
54
|
+
await atomicWriteJSON(path.join(runDirectory, "verifier", "candidate-ineligible.json"), value);
|
|
55
|
+
}
|
|
56
|
+
/** Persist bounded, redacted verifier artifacts in the immutable run bundle. */
|
|
57
|
+
export async function captureVerifierDiagnostics(trialDirectory, runDirectory, options = {}) {
|
|
58
|
+
const maxArtifactBytes = options.maxArtifactBytes ?? DEFAULT_VERIFIER_DIAGNOSTIC_MAX_BYTES;
|
|
59
|
+
if (!Number.isSafeInteger(maxArtifactBytes) || maxArtifactBytes <= TRUNCATION_MARKER.length + 2
|
|
60
|
+
|| maxArtifactBytes > MAX_VERIFIER_DIAGNOSTIC_MAX_BYTES) {
|
|
61
|
+
throw new TypeError("verifier diagnostic artifact limit is invalid");
|
|
62
|
+
}
|
|
63
|
+
throwIfAborted(options.signal);
|
|
64
|
+
const root = await realpath(trialDirectory);
|
|
65
|
+
const artifacts = [];
|
|
66
|
+
const redactions = new Map();
|
|
67
|
+
for (const name of VERIFIER_ARTIFACT_NAMES) {
|
|
68
|
+
throwIfAborted(options.signal);
|
|
69
|
+
const safe = await safeArtifact(root, name);
|
|
70
|
+
if (!safe)
|
|
71
|
+
continue;
|
|
72
|
+
let captured;
|
|
73
|
+
try {
|
|
74
|
+
captured = await captureText(safe.handle, maxArtifactBytes, options.credentialValues ?? [], options.signal);
|
|
75
|
+
await safe.assertUnchanged();
|
|
76
|
+
if (captured.sourceBytes !== safe.size)
|
|
77
|
+
throw new TypeError(`verifier artifact changed while being captured: ${name}`);
|
|
78
|
+
}
|
|
79
|
+
finally {
|
|
80
|
+
await safe.handle.close();
|
|
81
|
+
}
|
|
82
|
+
mergeCounts(redactions, captured.redactions);
|
|
83
|
+
const ref = `verifier/${name}`;
|
|
84
|
+
const destination = path.join(runDirectory, ...ref.split("/"));
|
|
85
|
+
await writePrivateFile(destination, captured.stored);
|
|
86
|
+
const stored = captured.stored;
|
|
87
|
+
artifacts.push({
|
|
88
|
+
name,
|
|
89
|
+
ref,
|
|
90
|
+
media_type: name === "ctrf.json" ? "application/json" : "text/plain",
|
|
91
|
+
source_bytes: captured.sourceBytes,
|
|
92
|
+
source_sha256: captured.sourceSha256,
|
|
93
|
+
bytes: captured.contentBytes,
|
|
94
|
+
sha256: captured.contentSha256,
|
|
95
|
+
stored_bytes: stored.length,
|
|
96
|
+
stored_sha256: sha256(stored),
|
|
97
|
+
truncated: captured.truncated,
|
|
98
|
+
});
|
|
99
|
+
}
|
|
100
|
+
if (artifacts.length === 0)
|
|
101
|
+
return null;
|
|
102
|
+
const index = {
|
|
103
|
+
schema_version: "1",
|
|
104
|
+
kind: "verifier-diagnostics",
|
|
105
|
+
artifacts,
|
|
106
|
+
redactions: canonicalCounts(redactions),
|
|
107
|
+
};
|
|
108
|
+
await atomicWriteJSON(path.join(runDirectory, ...VERIFIER_DIAGNOSTICS_INDEX_REF.split("/")), index);
|
|
109
|
+
return index;
|
|
110
|
+
}
|
|
111
|
+
async function safeArtifact(root, name, maximumBytes) {
|
|
112
|
+
try {
|
|
113
|
+
return await openContainedRegularFile(root, `verifier/${name}`, maximumBytes);
|
|
114
|
+
}
|
|
115
|
+
catch (error) {
|
|
116
|
+
if (error.code === "ENOENT")
|
|
117
|
+
return null;
|
|
118
|
+
throw new TypeError(`unsafe verifier artifact: ${name}`, { cause: error });
|
|
119
|
+
}
|
|
120
|
+
}
|
|
121
|
+
async function captureText(handle, maxBytes, credentialValues, signal) {
|
|
122
|
+
const sourceHash = createHash("sha256");
|
|
123
|
+
const contentHash = createHash("sha256");
|
|
124
|
+
const decoder = new StringDecoder("utf8");
|
|
125
|
+
const bounded = new BoundedBuffer(maxBytes);
|
|
126
|
+
const redactions = new Map();
|
|
127
|
+
let sourceBytes = 0;
|
|
128
|
+
let pending = "";
|
|
129
|
+
let discarding = false;
|
|
130
|
+
const emit = (value) => {
|
|
131
|
+
const redacted = redactCredentialText(value, credentialValues);
|
|
132
|
+
mergeCounts(redactions, redacted.redactions);
|
|
133
|
+
const bytes = Buffer.from(redacted.text, "utf8");
|
|
134
|
+
contentHash.update(bytes);
|
|
135
|
+
bounded.append(bytes);
|
|
136
|
+
};
|
|
137
|
+
const drain = () => {
|
|
138
|
+
for (;;) {
|
|
139
|
+
const newline = pending.indexOf("\n");
|
|
140
|
+
if (discarding) {
|
|
141
|
+
if (newline < 0) {
|
|
142
|
+
pending = "";
|
|
143
|
+
return;
|
|
144
|
+
}
|
|
145
|
+
pending = pending.slice(newline + 1);
|
|
146
|
+
discarding = false;
|
|
147
|
+
continue;
|
|
148
|
+
}
|
|
149
|
+
if (newline >= 0) {
|
|
150
|
+
emit(pending.slice(0, newline + 1));
|
|
151
|
+
pending = pending.slice(newline + 1);
|
|
152
|
+
continue;
|
|
153
|
+
}
|
|
154
|
+
if (Buffer.byteLength(pending, "utf8") > MAX_LOG_LINE_BYTES) {
|
|
155
|
+
emit("[REDACTED OVERSIZED VERIFIER LOG LINE]\n");
|
|
156
|
+
increment(redactions, "oversized-verifier-log-line-v1");
|
|
157
|
+
bounded.markOmitted();
|
|
158
|
+
pending = "";
|
|
159
|
+
discarding = true;
|
|
160
|
+
}
|
|
161
|
+
return;
|
|
162
|
+
}
|
|
163
|
+
};
|
|
164
|
+
const buffer = Buffer.allocUnsafe(64 * 1024);
|
|
165
|
+
let position = 0;
|
|
166
|
+
for (;;) {
|
|
167
|
+
throwIfAborted(signal);
|
|
168
|
+
const { bytesRead } = await handle.read(buffer, 0, buffer.length, position);
|
|
169
|
+
if (bytesRead === 0)
|
|
170
|
+
break;
|
|
171
|
+
position += bytesRead;
|
|
172
|
+
const bytes = buffer.subarray(0, bytesRead);
|
|
173
|
+
sourceHash.update(bytes);
|
|
174
|
+
sourceBytes += bytes.length;
|
|
175
|
+
pending += decoder.write(bytes);
|
|
176
|
+
drain();
|
|
177
|
+
}
|
|
178
|
+
pending += decoder.end();
|
|
179
|
+
drain();
|
|
180
|
+
if (!discarding && pending)
|
|
181
|
+
emit(pending);
|
|
182
|
+
throwIfAborted(signal);
|
|
183
|
+
return {
|
|
184
|
+
sourceSha256: `sha256:${sourceHash.digest("hex")}`,
|
|
185
|
+
sourceBytes,
|
|
186
|
+
contentSha256: `sha256:${contentHash.digest("hex")}`,
|
|
187
|
+
contentBytes: bounded.totalBytes,
|
|
188
|
+
stored: bounded.value(),
|
|
189
|
+
truncated: bounded.truncated,
|
|
190
|
+
redactions,
|
|
191
|
+
};
|
|
192
|
+
}
|
|
193
|
+
class BoundedBuffer {
|
|
194
|
+
limit;
|
|
195
|
+
payloadLimit;
|
|
196
|
+
headLimit;
|
|
197
|
+
tailLimit;
|
|
198
|
+
totalBytes = 0;
|
|
199
|
+
clipped = false;
|
|
200
|
+
omitted = false;
|
|
201
|
+
complete = [];
|
|
202
|
+
head = Buffer.alloc(0);
|
|
203
|
+
tail = Buffer.alloc(0);
|
|
204
|
+
constructor(limit) {
|
|
205
|
+
this.limit = limit;
|
|
206
|
+
this.payloadLimit = limit - TRUNCATION_MARKER.length;
|
|
207
|
+
this.headLimit = Math.floor(this.payloadLimit / 2);
|
|
208
|
+
this.tailLimit = this.payloadLimit - this.headLimit;
|
|
209
|
+
}
|
|
210
|
+
append(value) {
|
|
211
|
+
if (value.length === 0)
|
|
212
|
+
return;
|
|
213
|
+
this.totalBytes += value.length;
|
|
214
|
+
if (!this.clipped) {
|
|
215
|
+
this.complete.push(value);
|
|
216
|
+
const combined = Buffer.concat(this.complete);
|
|
217
|
+
if (combined.length <= this.limit)
|
|
218
|
+
return;
|
|
219
|
+
this.clipped = true;
|
|
220
|
+
this.complete = [];
|
|
221
|
+
this.head = combined.subarray(0, this.headLimit);
|
|
222
|
+
this.tail = combined.subarray(Math.max(0, combined.length - this.tailLimit));
|
|
223
|
+
return;
|
|
224
|
+
}
|
|
225
|
+
this.tail = rollingTail(this.tail, value, this.tailLimit);
|
|
226
|
+
}
|
|
227
|
+
value() {
|
|
228
|
+
return this.clipped
|
|
229
|
+
? Buffer.concat([
|
|
230
|
+
validUtf8Prefix(this.head),
|
|
231
|
+
TRUNCATION_MARKER,
|
|
232
|
+
validUtf8Suffix(this.tail),
|
|
233
|
+
])
|
|
234
|
+
: Buffer.concat(this.complete);
|
|
235
|
+
}
|
|
236
|
+
markOmitted() {
|
|
237
|
+
this.omitted = true;
|
|
238
|
+
}
|
|
239
|
+
get truncated() {
|
|
240
|
+
return this.clipped || this.omitted;
|
|
241
|
+
}
|
|
242
|
+
}
|
|
243
|
+
function validUtf8Prefix(value) {
|
|
244
|
+
let end = value.length;
|
|
245
|
+
while (end > 0 && (value[end - 1] & 0xc0) === 0x80)
|
|
246
|
+
end -= 1;
|
|
247
|
+
const lead = end - 1;
|
|
248
|
+
return lead >= 0 && utf8SequenceLength(value[lead]) > value.length - lead ? value.subarray(0, lead) : value;
|
|
249
|
+
}
|
|
250
|
+
function validUtf8Suffix(value) {
|
|
251
|
+
let start = 0;
|
|
252
|
+
while (start < value.length && (value[start] & 0xc0) === 0x80)
|
|
253
|
+
start += 1;
|
|
254
|
+
return value.subarray(start);
|
|
255
|
+
}
|
|
256
|
+
function utf8SequenceLength(byte) {
|
|
257
|
+
if ((byte & 0x80) === 0)
|
|
258
|
+
return 1;
|
|
259
|
+
if ((byte & 0xe0) === 0xc0)
|
|
260
|
+
return 2;
|
|
261
|
+
if ((byte & 0xf0) === 0xe0)
|
|
262
|
+
return 3;
|
|
263
|
+
if ((byte & 0xf8) === 0xf0)
|
|
264
|
+
return 4;
|
|
265
|
+
return 1;
|
|
266
|
+
}
|
|
267
|
+
function rollingTail(current, next, limit) {
|
|
268
|
+
if (next.length >= limit)
|
|
269
|
+
return next.subarray(next.length - limit);
|
|
270
|
+
const keep = Math.max(0, limit - next.length);
|
|
271
|
+
return Buffer.concat([current.subarray(Math.max(0, current.length - keep)), next]);
|
|
272
|
+
}
|
|
273
|
+
function canonicalCounts(counts) {
|
|
274
|
+
return [...counts.entries()]
|
|
275
|
+
.filter(([, count]) => count > 0)
|
|
276
|
+
.sort(([left], [right]) => left.localeCompare(right))
|
|
277
|
+
.map(([rule_id, count]) => ({ rule_id, count }));
|
|
278
|
+
}
|
|
279
|
+
function mergeCounts(target, source) {
|
|
280
|
+
for (const [rule, count] of source)
|
|
281
|
+
increment(target, rule, count);
|
|
282
|
+
}
|
|
283
|
+
function increment(counts, rule, count = 1) {
|
|
284
|
+
counts.set(rule, (counts.get(rule) ?? 0) + count);
|
|
285
|
+
}
|
|
286
|
+
function sha256(value) {
|
|
287
|
+
return `sha256:${createHash("sha256").update(value).digest("hex")}`;
|
|
288
|
+
}
|
|
289
|
+
function throwIfAborted(signal) {
|
|
290
|
+
if (signal?.aborted)
|
|
291
|
+
throw signal.reason ?? new DOMException("The operation was aborted", "AbortError");
|
|
292
|
+
}
|
|
293
|
+
async function copyVerifierRetryHistory(trialDirectory, runDirectory, credentialValues, signal) {
|
|
294
|
+
const root = await realpath(trialDirectory);
|
|
295
|
+
throwIfAborted(signal);
|
|
296
|
+
const safe = await safeArtifact(root, "infrastructure-retry-history.json", 1024 * 1024);
|
|
297
|
+
if (!safe)
|
|
298
|
+
return;
|
|
299
|
+
let bytes;
|
|
300
|
+
try {
|
|
301
|
+
bytes = await safe.handle.readFile();
|
|
302
|
+
await safe.assertUnchanged();
|
|
303
|
+
}
|
|
304
|
+
finally {
|
|
305
|
+
await safe.handle.close();
|
|
306
|
+
}
|
|
307
|
+
throwIfAborted(signal);
|
|
308
|
+
const history = JSON.parse(bytes.toString("utf8"));
|
|
309
|
+
if (history?.schema_version !== "1"
|
|
310
|
+
|| history.code !== "verifier_infrastructure_retry_history"
|
|
311
|
+
|| history.candidate_rerun !== false
|
|
312
|
+
|| !Array.isArray(history.attempts))
|
|
313
|
+
return;
|
|
314
|
+
const redacted = redactCredentialText(JSON.stringify(history), credentialValues).text;
|
|
315
|
+
await atomicWriteJSON(path.join(runDirectory, "verifier", "infrastructure-retry-history.json"), JSON.parse(redacted));
|
|
316
|
+
}
|
|
317
|
+
//# sourceMappingURL=verifier-artifacts.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"verifier-artifacts.js","sourceRoot":"","sources":["../../../src/evals/verifier-artifacts.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,QAAQ,EAAmB,MAAM,kBAAkB,CAAC;AAC7D,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,0BAA0B,EAAE,eAAe,EAAE,uBAAuB,EAAE,wBAAwB,EAAE,oBAAoB,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAEhL,OAAO,EAAE,4BAA4B,EAAE,MAAM,+BAA+B,CAAC;AAG7E,MAAM,CAAC,MAAM,qCAAqC,GAAG,EAAE,GAAG,IAAI,CAAC;AAC/D,MAAM,CAAC,MAAM,iCAAiC,GAAG,EAAE,GAAG,IAAI,GAAG,IAAI,CAAC;AAClE,MAAM,CAAC,MAAM,8BAA8B,GAAG,2BAA2B,CAAC;AAC1E,MAAM,CAAC,MAAM,uBAAuB,GAAiD;IACnF,WAAW,EAAE,iBAAiB,EAAE,iBAAiB,EAAE,YAAY,EAAE,YAAY;CAC9E,CAAC;AAEF,MAAM,iBAAiB,GAAG,MAAM,CAAC,IAAI,CAAC,2CAA2C,EAAE,MAAM,CAAC,CAAC;AAC3F,MAAM,kBAAkB,GAAG,IAAI,GAAG,IAAI,CAAC;AA4BvC,MAAM,CAAC,KAAK,UAAU,+BAA+B,CAAC,KAUrD;IACC,MAAM,GAAG,GAAG,KAAK,CAAC,GAAG,IAAI,OAAO,CAAC,GAAG,CAAC;IACrC,MAAM,gBAAgB,GAAG,uBAAuB,CAC9C,CAAC,GAAG,0BAA0B,EAAE,GAAG,CAAC,KAAK,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,EACzD,GAAG,CACJ,CAAC;IACF,cAAc,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IAC7B,MAAM,wBAAwB,CAAC,KAAK,CAAC,cAAc,EAAE,KAAK,CAAC,YAAY,EAAE,gBAAgB,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IACzG,MAAM,iCAAiC,CAAC,KAAK,CAAC,cAAc,EAAE,KAAK,CAAC,YAAY,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IAChG,MAAM,MAAM,GAAG,MAAM,4BAA4B,CAAC;QAChD,cAAc,EAAE,KAAK,CAAC,cAAc;QACpC,YAAY,EAAE,KAAK,CAAC,YAAY;QAChC,cAAc,EAAE,KAAK,CAAC,cAAc,IAAI,IAAI;QAC5C,gBAAgB;QAChB,GAAG,CAAC,KAAK,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC;QAClE,GAAG,CAAC,KAAK,CAAC,iBAAiB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,KAAK,CAAC,iBAAiB,EAAE,CAAC;QAChG,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAClD,CAAC,CAAC;IACH,MAAM,0BAA0B,CAAC,KAAK,CAAC,cAAc,EAAE,KAAK,CAAC,YAAY,EAAE;QACzE,GAAG,CAAC,KAAK,CAAC,gBAAgB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC7F,gBAAgB;QAChB,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAClD,CAAC,CAAC;IACH,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,KAAK,UAAU,iCAAiC,CAAC,cAAsB,EAAE,YAAoB,EAAE,MAAoB;IACjH,MAAM,IAAI,GAAG,MAAM,QAAQ,CAAC,cAAc,CAAC,CAAC;IAC5C,cAAc,CAAC,MAAM,CAAC,CAAC;IACvB,MAAM,IAAI,GAAG,MAAM,YAAY,CAAC,IAAI,EAAE,2BAA2B,EAAE,EAAE,GAAG,IAAI,CAAC,CAAC;IAC9E,IAAI,CAAC,IAAI;QAAE,OAAO;IAClB,IAAI,KAAa,CAAC;IAClB,IAAI,CAAC;QACH,KAAK,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,QAAQ,EAAE,CAAC;QACrC,MAAM,IAAI,CAAC,eAAe,EAAE,CAAC;IAC/B,CAAC;YAAS,CAAC;QACT,MAAM,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC;IAC5B,CAAC;IACD,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAA4B,CAAC;IAC5E,IAAI,KAAK,CAAC,cAAc,KAAK,GAAG,IAAI,KAAK,CAAC,IAAI,KAAK,gCAAgC,IAAI,KAAK,CAAC,iBAAiB,KAAK,KAAK;QAAE,OAAO;IACjI,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,UAAU,EAAE,2BAA2B,CAAC,EAAE,KAAK,CAAC,CAAC;AACjG,CAAC;AAED,gFAAgF;AAChF,MAAM,CAAC,KAAK,UAAU,0BAA0B,CAC9C,cAAsB,EACtB,YAAoB,EACpB,UAA6C,EAAE;IAE/C,MAAM,gBAAgB,GAAG,OAAO,CAAC,gBAAgB,IAAI,qCAAqC,CAAC;IAC3F,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,gBAAgB,CAAC,IAAI,gBAAgB,IAAI,iBAAiB,CAAC,MAAM,GAAG,CAAC;WAC1F,gBAAgB,GAAG,iCAAiC,EAAE,CAAC;QAC1D,MAAM,IAAI,SAAS,CAAC,+CAA+C,CAAC,CAAC;IACvE,CAAC;IACD,cAAc,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;IAC/B,MAAM,IAAI,GAAG,MAAM,QAAQ,CAAC,cAAc,CAAC,CAAC;IAC5C,MAAM,SAAS,GAAkC,EAAE,CAAC;IACpD,MAAM,UAAU,GAAG,IAAI,GAAG,EAAkB,CAAC;IAC7C,KAAK,MAAM,IAAI,IAAI,uBAAuB,EAAE,CAAC;QAC3C,cAAc,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC/B,MAAM,IAAI,GAAG,MAAM,YAAY,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;QAC5C,IAAI,CAAC,IAAI;YAAE,SAAS;QACpB,IAAI,QAAiD,CAAC;QACtD,IAAI,CAAC;YACH,QAAQ,GAAG,MAAM,WAAW,CAAC,IAAI,CAAC,MAAM,EAAE,gBAAgB,EAAE,OAAO,CAAC,gBAAgB,IAAI,EAAE,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;YAC5G,MAAM,IAAI,CAAC,eAAe,EAAE,CAAC;YAC7B,IAAI,QAAQ,CAAC,WAAW,KAAK,IAAI,CAAC,IAAI;gBAAE,MAAM,IAAI,SAAS,CAAC,mDAAmD,IAAI,EAAE,CAAC,CAAC;QACzH,CAAC;gBAAS,CAAC;YACT,MAAM,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC;QAC5B,CAAC;QACD,WAAW,CAAC,UAAU,EAAE,QAAQ,CAAC,UAAU,CAAC,CAAC;QAC7C,MAAM,GAAG,GAAG,YAAY,IAAI,EAAE,CAAC;QAC/B,MAAM,WAAW,GAAG,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,GAAG,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC;QAC/D,MAAM,gBAAgB,CAAC,WAAW,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;QACrD,MAAM,MAAM,GAAG,QAAQ,CAAC,MAAM,CAAC;QAC/B,SAAS,CAAC,IAAI,CAAC;YACb,IAAI;YACJ,GAAG;YACH,UAAU,EAAE,IAAI,KAAK,WAAW,CAAC,CAAC,CAAC,kBAAkB,CAAC,CAAC,CAAC,YAAY;YACpE,YAAY,EAAE,QAAQ,CAAC,WAAW;YAClC,aAAa,EAAE,QAAQ,CAAC,YAAY;YACpC,KAAK,EAAE,QAAQ,CAAC,YAAY;YAC5B,MAAM,EAAE,QAAQ,CAAC,aAAa;YAC9B,YAAY,EAAE,MAAM,CAAC,MAAM;YAC3B,aAAa,EAAE,MAAM,CAAC,MAAM,CAAC;YAC7B,SAAS,EAAE,QAAQ,CAAC,SAAS;SAC9B,CAAC,CAAC;IACL,CAAC;IACD,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IACxC,MAAM,KAAK,GAA+B;QACxC,cAAc,EAAE,GAAG;QACnB,IAAI,EAAE,sBAAsB;QAC5B,SAAS;QACT,UAAU,EAAE,eAAe,CAAC,UAAU,CAAC;KACxC,CAAC;IACF,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,GAAG,8BAA8B,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC;IACpG,OAAO,KAAK,CAAC;AACf,CAAC;AAED,KAAK,UAAU,YAAY,CAAC,IAAY,EAAE,IAAY,EAAE,YAAqB;IAC3E,IAAI,CAAC;QACH,OAAO,MAAM,wBAAwB,CAAC,IAAI,EAAE,YAAY,IAAI,EAAE,EAAE,YAAY,CAAC,CAAC;IAChF,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,IAAI,CAAC;QACpE,MAAM,IAAI,SAAS,CAAC,6BAA6B,IAAI,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,CAAC;IAC7E,CAAC;AACH,CAAC;AAED,KAAK,UAAU,WAAW,CACxB,MAAkB,EAClB,QAAgB,EAChB,gBAAmC,EACnC,MAAgC;IAUhC,MAAM,UAAU,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC;IACxC,MAAM,WAAW,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC;IACzC,MAAM,OAAO,GAAG,IAAI,aAAa,CAAC,MAAM,CAAC,CAAC;IAC1C,MAAM,OAAO,GAAG,IAAI,aAAa,CAAC,QAAQ,CAAC,CAAC;IAC5C,MAAM,UAAU,GAAG,IAAI,GAAG,EAAkB,CAAC;IAC7C,IAAI,WAAW,GAAG,CAAC,CAAC;IACpB,IAAI,OAAO,GAAG,EAAE,CAAC;IACjB,IAAI,UAAU,GAAG,KAAK,CAAC;IACvB,MAAM,IAAI,GAAG,CAAC,KAAa,EAAQ,EAAE;QACnC,MAAM,QAAQ,GAAG,oBAAoB,CAAC,KAAK,EAAE,gBAAgB,CAAC,CAAC;QAC/D,WAAW,CAAC,UAAU,EAAE,QAAQ,CAAC,UAAU,CAAC,CAAC;QAC7C,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACjD,WAAW,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAC1B,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACxB,CAAC,CAAC;IACF,MAAM,KAAK,GAAG,GAAS,EAAE;QACvB,SAAS,CAAC;YACR,MAAM,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;YACtC,IAAI,UAAU,EAAE,CAAC;gBACf,IAAI,OAAO,GAAG,CAAC,EAAE,CAAC;oBAAC,OAAO,GAAG,EAAE,CAAC;oBAAC,OAAO;gBAAC,CAAC;gBAC1C,OAAO,GAAG,OAAO,CAAC,KAAK,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC;gBACrC,UAAU,GAAG,KAAK,CAAC;gBACnB,SAAS;YACX,CAAC;YACD,IAAI,OAAO,IAAI,CAAC,EAAE,CAAC;gBACjB,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC;gBACpC,OAAO,GAAG,OAAO,CAAC,KAAK,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC;gBACrC,SAAS;YACX,CAAC;YACD,IAAI,MAAM,CAAC,UAAU,CAAC,OAAO,EAAE,MAAM,CAAC,GAAG,kBAAkB,EAAE,CAAC;gBAC5D,IAAI,CAAC,0CAA0C,CAAC,CAAC;gBACjD,SAAS,CAAC,UAAU,EAAE,gCAAgC,CAAC,CAAC;gBACxD,OAAO,CAAC,WAAW,EAAE,CAAC;gBACtB,OAAO,GAAG,EAAE,CAAC;gBACb,UAAU,GAAG,IAAI,CAAC;YACpB,CAAC;YACD,OAAO;QACT,CAAC;IACH,CAAC,CAAC;IAEF,MAAM,MAAM,GAAG,MAAM,CAAC,WAAW,CAAC,EAAE,GAAG,IAAI,CAAC,CAAC;IAC7C,IAAI,QAAQ,GAAG,CAAC,CAAC;IACjB,SAAS,CAAC;QACR,cAAc,CAAC,MAAM,CAAC,CAAC;QACvB,MAAM,EAAE,SAAS,EAAE,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,QAAQ,CAAC,CAAC;QAC5E,IAAI,SAAS,KAAK,CAAC;YAAE,MAAM;QAC3B,QAAQ,IAAI,SAAS,CAAC;QACtB,MAAM,KAAK,GAAG,MAAM,CAAC,QAAQ,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC;QAC5C,UAAU,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACzB,WAAW,IAAI,KAAK,CAAC,MAAM,CAAC;QAC5B,OAAO,IAAI,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;QAChC,KAAK,EAAE,CAAC;IACV,CAAC;IACD,OAAO,IAAI,OAAO,CAAC,GAAG,EAAE,CAAC;IACzB,KAAK,EAAE,CAAC;IACR,IAAI,CAAC,UAAU,IAAI,OAAO;QAAE,IAAI,CAAC,OAAO,CAAC,CAAC;IAC1C,cAAc,CAAC,MAAM,CAAC,CAAC;IACvB,OAAO;QACL,YAAY,EAAE,UAAU,UAAU,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE;QAClD,WAAW;QACX,aAAa,EAAE,UAAU,WAAW,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE;QACpD,YAAY,EAAE,OAAO,CAAC,UAAU;QAChC,MAAM,EAAE,OAAO,CAAC,KAAK,EAAE;QACvB,SAAS,EAAE,OAAO,CAAC,SAAS;QAC5B,UAAU;KACX,CAAC;AACJ,CAAC;AAED,MAAM,aAAa;IAWI;IAVZ,YAAY,CAAS;IACrB,SAAS,CAAS;IAClB,SAAS,CAAS;IAC3B,UAAU,GAAG,CAAC,CAAC;IACP,OAAO,GAAG,KAAK,CAAC;IAChB,OAAO,GAAG,KAAK,CAAC;IAChB,QAAQ,GAAa,EAAE,CAAC;IACxB,IAAI,GAAW,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IAC/B,IAAI,GAAW,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IAEvC,YAAqB,KAAa;QAAb,UAAK,GAAL,KAAK,CAAQ;QAChC,IAAI,CAAC,YAAY,GAAG,KAAK,GAAG,iBAAiB,CAAC,MAAM,CAAC;QACrD,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,YAAY,GAAG,CAAC,CAAC,CAAC;QACnD,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,YAAY,GAAG,IAAI,CAAC,SAAS,CAAC;IACtD,CAAC;IAED,MAAM,CAAC,KAAa;QAClB,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO;QAC/B,IAAI,CAAC,UAAU,IAAI,KAAK,CAAC,MAAM,CAAC;QAChC,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC;YAClB,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;YAC1B,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YAC9C,IAAI,QAAQ,CAAC,MAAM,IAAI,IAAI,CAAC,KAAK;gBAAE,OAAO;YAC1C,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC;YACpB,IAAI,CAAC,QAAQ,GAAG,EAAE,CAAC;YACnB,IAAI,CAAC,IAAI,GAAG,QAAQ,CAAC,QAAQ,CAAC,CAAC,EAAE,IAAI,CAAC,SAAS,CAAC,CAAC;YACjD,IAAI,CAAC,IAAI,GAAG,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,QAAQ,CAAC,MAAM,GAAG,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC;YAC7E,OAAO;QACT,CAAC;QACD,IAAI,CAAC,IAAI,GAAG,WAAW,CAAC,IAAI,CAAC,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,SAAS,CAAC,CAAC;IAC5D,CAAC;IAED,KAAK;QACH,OAAO,IAAI,CAAC,OAAO;YACjB,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC;gBACd,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC;gBAC1B,iBAAiB;gBACjB,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC;aAC3B,CAAC;YACF,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACnC,CAAC;IAED,WAAW;QACT,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC;IACtB,CAAC;IAED,IAAI,SAAS;QACX,OAAO,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,OAAO,CAAC;IACtC,CAAC;CACF;AAED,SAAS,eAAe,CAAC,KAAa;IACpC,IAAI,GAAG,GAAG,KAAK,CAAC,MAAM,CAAC;IACvB,OAAO,GAAG,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,GAAG,CAAC,CAAE,GAAG,IAAI,CAAC,KAAK,IAAI;QAAE,GAAG,IAAI,CAAC,CAAC;IAC9D,MAAM,IAAI,GAAG,GAAG,GAAG,CAAC,CAAC;IACrB,OAAO,IAAI,IAAI,CAAC,IAAI,kBAAkB,CAAC,KAAK,CAAC,IAAI,CAAE,CAAC,GAAG,KAAK,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC;AAC/G,CAAC;AAED,SAAS,eAAe,CAAC,KAAa;IACpC,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,OAAO,KAAK,GAAG,KAAK,CAAC,MAAM,IAAI,CAAC,KAAK,CAAC,KAAK,CAAE,GAAG,IAAI,CAAC,KAAK,IAAI;QAAE,KAAK,IAAI,CAAC,CAAC;IAC3E,OAAO,KAAK,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;AAC/B,CAAC;AAED,SAAS,kBAAkB,CAAC,IAAY;IACtC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAClC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,KAAK,IAAI;QAAE,OAAO,CAAC,CAAC;IACrC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,KAAK,IAAI;QAAE,OAAO,CAAC,CAAC;IACrC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,KAAK,IAAI;QAAE,OAAO,CAAC,CAAC;IACrC,OAAO,CAAC,CAAC;AACX,CAAC;AAED,SAAS,WAAW,CAAC,OAAe,EAAE,IAAY,EAAE,KAAa;IAC/D,IAAI,IAAI,CAAC,MAAM,IAAI,KAAK;QAAE,OAAO,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC,CAAC;IACpE,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC;IAC9C,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;AACrF,CAAC;AAED,SAAS,eAAe,CAAC,MAA2B;IAClD,OAAO,CAAC,GAAG,MAAM,CAAC,OAAO,EAAE,CAAC;SACzB,MAAM,CAAC,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,KAAK,GAAG,CAAC,CAAC;SAChC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC;SACpD,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC;AACrD,CAAC;AAED,SAAS,WAAW,CAAC,MAA2B,EAAE,MAA2B;IAC3E,KAAK,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,MAAM;QAAE,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,KAAK,CAAC,CAAC;AACrE,CAAC;AAED,SAAS,SAAS,CAAC,MAA2B,EAAE,IAAY,EAAE,KAAK,GAAG,CAAC;IACrE,MAAM,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,KAAK,CAAC,CAAC;AACpD,CAAC;AAED,SAAS,MAAM,CAAC,KAAa;IAC3B,OAAO,UAAU,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC;AACtE,CAAC;AAED,SAAS,cAAc,CAAC,MAAoB;IAC1C,IAAI,MAAM,EAAE,OAAO;QAAE,MAAM,MAAM,CAAC,MAAM,IAAI,IAAI,YAAY,CAAC,2BAA2B,EAAE,YAAY,CAAC,CAAC;AAC1G,CAAC;AAED,KAAK,UAAU,wBAAwB,CACrC,cAAsB,EACtB,YAAoB,EACpB,gBAAmC,EACnC,MAAoB;IAEpB,MAAM,IAAI,GAAG,MAAM,QAAQ,CAAC,cAAc,CAAC,CAAC;IAC5C,cAAc,CAAC,MAAM,CAAC,CAAC;IACvB,MAAM,IAAI,GAAG,MAAM,YAAY,CAAC,IAAI,EAAE,mCAAmC,EAAE,IAAI,GAAG,IAAI,CAAC,CAAC;IACxF,IAAI,CAAC,IAAI;QAAE,OAAO;IAClB,IAAI,KAAa,CAAC;IAClB,IAAI,CAAC;QACH,KAAK,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,QAAQ,EAAE,CAAC;QACrC,MAAM,IAAI,CAAC,eAAe,EAAE,CAAC;IAC/B,CAAC;YAAS,CAAC;QACT,MAAM,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC;IAC5B,CAAC;IACD,cAAc,CAAC,MAAM,CAAC,CAAC;IACvB,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAA4B,CAAC;IAC9E,IAAI,OAAO,EAAE,cAAc,KAAK,GAAG;WAC9B,OAAO,CAAC,IAAI,KAAK,uCAAuC;WACxD,OAAO,CAAC,eAAe,KAAK,KAAK;WACjC,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,QAAQ,CAAC;QAAE,OAAO;IAC9C,MAAM,QAAQ,GAAG,oBAAoB,CAAC,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,EAAE,gBAAgB,CAAC,CAAC,IAAI,CAAC;IACtF,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,UAAU,EAAE,mCAAmC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC;AACxH,CAAC"}
|
|
@@ -0,0 +1,25 @@
|
|
|
1
|
+
import { lstat, readFile } from "node:fs/promises";
|
|
2
|
+
import path from "node:path";
|
|
3
|
+
export async function readCandidateIneligibleDiagnostic(trialDirectory) {
|
|
4
|
+
const source = path.join(trialDirectory, "verifier", "candidate-ineligible.json");
|
|
5
|
+
try {
|
|
6
|
+
const info = await lstat(source);
|
|
7
|
+
if (!info.isFile() || info.size <= 0 || info.size > 16 * 1024)
|
|
8
|
+
return null;
|
|
9
|
+
const value = JSON.parse(await readFile(source, "utf8"));
|
|
10
|
+
if (!value || typeof value !== "object" || Array.isArray(value))
|
|
11
|
+
return null;
|
|
12
|
+
const record = value;
|
|
13
|
+
if (Object.keys(record).some((key) => !["schema_version", "code", "run_id", "candidate_bundle", "reason_code", "verifier_executed"].includes(key))
|
|
14
|
+
|| record.schema_version !== "1" || record.code !== "candidate_evidence_unavailable"
|
|
15
|
+
|| typeof record.run_id !== "string" || !/^run_[a-f0-9]{32}$/.test(record.run_id)
|
|
16
|
+
|| record.candidate_bundle !== "missing" && record.candidate_bundle !== "invalid"
|
|
17
|
+
|| typeof record.reason_code !== "string" || !record.reason_code || record.verifier_executed !== false)
|
|
18
|
+
return null;
|
|
19
|
+
return record;
|
|
20
|
+
}
|
|
21
|
+
catch {
|
|
22
|
+
return null;
|
|
23
|
+
}
|
|
24
|
+
}
|
|
25
|
+
//# sourceMappingURL=verifier-eligibility.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"verifier-eligibility.js","sourceRoot":"","sources":["../../../src/evals/verifier-eligibility.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AACnD,OAAO,IAAI,MAAM,WAAW,CAAC;AAW7B,MAAM,CAAC,KAAK,UAAU,iCAAiC,CAAC,cAAsB;IAC5E,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,UAAU,EAAE,2BAA2B,CAAC,CAAC;IAClF,IAAI,CAAC;QACH,MAAM,IAAI,GAAG,MAAM,KAAK,CAAC,MAAM,CAAC,CAAC;QACjC,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,IAAI,IAAI,CAAC,IAAI,IAAI,CAAC,IAAI,IAAI,CAAC,IAAI,GAAG,EAAE,GAAG,IAAI;YAAE,OAAO,IAAI,CAAC;QAC3E,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC,CAAY,CAAC;QACpE,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;YAAE,OAAO,IAAI,CAAC;QAC7E,MAAM,MAAM,GAAG,KAAgC,CAAC;QAChD,IAAI,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC,CAAC,gBAAgB,EAAE,MAAM,EAAE,QAAQ,EAAE,kBAAkB,EAAE,aAAa,EAAE,mBAAmB,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;eAC7I,MAAM,CAAC,cAAc,KAAK,GAAG,IAAI,MAAM,CAAC,IAAI,KAAK,gCAAgC;eACjF,OAAO,MAAM,CAAC,MAAM,KAAK,QAAQ,IAAI,CAAC,oBAAoB,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC;eAC9E,MAAM,CAAC,gBAAgB,KAAK,SAAS,IAAI,MAAM,CAAC,gBAAgB,KAAK,SAAS;eAC9E,OAAO,MAAM,CAAC,WAAW,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,WAAW,IAAI,MAAM,CAAC,iBAAiB,KAAK,KAAK;YAAE,OAAO,IAAI,CAAC;QACtH,OAAO,MAAoD,CAAC;IAC9D,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC"}
|