@openpond/evals 0.12.0 → 0.12.2
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/dist/artifact-verification.js +2 -0
- package/dist/connected-evidence/sources/claude-branches.js +34 -1
- package/dist/connected-evidence/sources/claude-code.js +1 -1
- package/dist/learning/continuation-mode.js +9 -0
- package/dist/learning/contracts.js +2 -0
- package/dist/learning/correction-metrics.js +61 -0
- package/dist/learning/index.js +2 -0
- package/dist/native-conversations/collector-acquisition.js +17 -6
- package/dist/native-conversations/collector-source-cache.js +1 -1
- package/dist/native-conversations/collector-store.js +27 -2
- package/dist/native-conversations/collector.js +4 -3
- package/dist/native-conversations/history.js +28 -8
- package/dist/native-conversations/index.js +1 -0
- package/dist/native-conversations/selection.js +57 -0
- package/dist/native-conversations/supervisor.js +2 -0
- package/dist/native-conversations/titles.js +2 -2
- package/dist/tasksets.js +1 -0
- package/dist/types/artifact-verification.d.ts.map +1 -1
- package/dist/types/benchmarks.d.ts +22 -22
- package/dist/types/benchmarks.d.ts.map +1 -1
- package/dist/types/builtin-benchmarks/harness-refiner.d.ts +40 -39
- package/dist/types/builtin-benchmarks/harness-refiner.d.ts.map +1 -1
- package/dist/types/compatibility.d.ts.map +1 -1
- package/dist/types/conformance.d.ts +98 -96
- package/dist/types/conformance.d.ts.map +1 -1
- package/dist/types/connected-evidence/codex-command-observations.d.ts.map +1 -1
- package/dist/types/connected-evidence/collector-coverage-contracts.d.ts +6 -6
- package/dist/types/connected-evidence/collector-coverage-contracts.d.ts.map +1 -1
- package/dist/types/connected-evidence/contracts.d.ts +58 -58
- package/dist/types/connected-evidence/contracts.d.ts.map +1 -1
- package/dist/types/connected-evidence/git-file-contracts.d.ts.map +1 -1
- package/dist/types/connected-evidence/git-file-proof.d.ts.map +1 -1
- package/dist/types/connected-evidence/git-starting-state.d.ts.map +1 -1
- package/dist/types/connected-evidence/git.d.ts.map +1 -1
- package/dist/types/connected-evidence/imports.d.ts.map +1 -1
- package/dist/types/connected-evidence/index.d.ts.map +1 -1
- package/dist/types/connected-evidence/normalize.d.ts.map +1 -1
- package/dist/types/connected-evidence/python-correctness.d.ts +38 -38
- package/dist/types/connected-evidence/python-correctness.d.ts.map +1 -1
- package/dist/types/connected-evidence/sources/claude-branches.d.ts +1 -0
- package/dist/types/connected-evidence/sources/claude-branches.d.ts.map +1 -1
- package/dist/types/connected-evidence/sources/claude-code.d.ts +25 -25
- package/dist/types/connected-evidence/sources/claude-code.d.ts.map +1 -1
- package/dist/types/connected-evidence/sources/codex-authorization.d.ts.map +1 -1
- package/dist/types/connected-evidence/sources/codex-metadata.d.ts.map +1 -1
- package/dist/types/connected-evidence/sources/codex.d.ts +25 -25
- package/dist/types/connected-evidence/sources/codex.d.ts.map +1 -1
- package/dist/types/connected-evidence/sources/grok-build.d.ts +25 -25
- package/dist/types/connected-evidence/sources/grok-build.d.ts.map +1 -1
- package/dist/types/connected-evidence/sources/history.d.ts +44 -44
- package/dist/types/connected-evidence/sources/history.d.ts.map +1 -1
- package/dist/types/connected-evidence/sources/opencode.d.ts +25 -25
- package/dist/types/connected-evidence/sources/opencode.d.ts.map +1 -1
- package/dist/types/connected-evidence/sources/pi.d.ts +25 -25
- package/dist/types/connected-evidence/sources/pi.d.ts.map +1 -1
- package/dist/types/connected-evidence/sources/usage.d.ts.map +1 -1
- package/dist/types/connected-evidence/usage-summary.d.ts.map +1 -1
- package/dist/types/deterministic-graders.d.ts.map +1 -1
- package/dist/types/environment-execution-services-contract.d.ts +19 -19
- package/dist/types/environment-execution-services-contract.d.ts.map +1 -1
- package/dist/types/environment-execution-services.d.ts.map +1 -1
- package/dist/types/evidence/authoring.d.ts.map +1 -1
- package/dist/types/evidence/conformance.d.ts +98 -98
- package/dist/types/evidence/conformance.d.ts.map +1 -1
- package/dist/types/evidence/contracts.d.ts +183 -183
- package/dist/types/evidence/contracts.d.ts.map +1 -1
- package/dist/types/evidence/eligibility.d.ts +160 -160
- package/dist/types/evidence/eligibility.d.ts.map +1 -1
- package/dist/types/evidence/index.d.ts.map +1 -1
- package/dist/types/execution-contracts.d.ts +110 -110
- package/dist/types/execution-contracts.d.ts.map +1 -1
- package/dist/types/execution-receipts.d.ts.map +1 -1
- package/dist/types/experiments.d.ts +96 -96
- package/dist/types/experiments.d.ts.map +1 -1
- package/dist/types/graders.d.ts +35 -35
- package/dist/types/graders.d.ts.map +1 -1
- package/dist/types/harness-experiment-source.d.ts.map +1 -1
- package/dist/types/harness.d.ts.map +1 -1
- package/dist/types/human-review/comparison.d.ts +189 -189
- package/dist/types/human-review/comparison.d.ts.map +1 -1
- package/dist/types/human-review/contracts.d.ts +178 -178
- package/dist/types/human-review/contracts.d.ts.map +1 -1
- package/dist/types/human-review/execution.d.ts +53 -53
- package/dist/types/human-review/execution.d.ts.map +1 -1
- package/dist/types/human-review/handoff.d.ts +43 -43
- package/dist/types/human-review/handoff.d.ts.map +1 -1
- package/dist/types/human-review/index.d.ts.map +1 -1
- package/dist/types/human-review/local-publication.d.ts +4 -4
- package/dist/types/human-review/local-publication.d.ts.map +1 -1
- package/dist/types/human-review/repository.d.ts.map +1 -1
- package/dist/types/human-review/results.d.ts +8 -8
- package/dist/types/human-review/results.d.ts.map +1 -1
- package/dist/types/human-review/service.d.ts +292 -292
- package/dist/types/human-review/service.d.ts.map +1 -1
- package/dist/types/human-review/task-proposals.d.ts +2 -2
- package/dist/types/human-review/task-proposals.d.ts.map +1 -1
- package/dist/types/index.d.ts.map +1 -1
- package/dist/types/javascript-environment-attempt.d.ts.map +1 -1
- package/dist/types/javascript-environment-contract.d.ts +39 -39
- package/dist/types/javascript-environment-contract.d.ts.map +1 -1
- package/dist/types/javascript-environment-node.d.ts.map +1 -1
- package/dist/types/javascript-environment.d.ts +14 -14
- package/dist/types/javascript-environment.d.ts.map +1 -1
- package/dist/types/javascript-isolate-node.d.ts.map +1 -1
- package/dist/types/javascript-isolate-process-entry.d.ts.map +1 -1
- package/dist/types/javascript-isolate-process.d.ts.map +1 -1
- package/dist/types/javascript-isolate.d.ts.map +1 -1
- package/dist/types/javascript-verifier-contract.d.ts.map +1 -1
- package/dist/types/javascript-verifier-node.d.ts.map +1 -1
- package/dist/types/javascript-verifier-worker.d.ts.map +1 -1
- package/dist/types/javascript-verifier.d.ts.map +1 -1
- package/dist/types/learned-preference.d.ts +10 -10
- package/dist/types/learned-preference.d.ts.map +1 -1
- package/dist/types/learning/acceptance-group.d.ts +44 -44
- package/dist/types/learning/acceptance-group.d.ts.map +1 -1
- package/dist/types/learning/acceptance-plan.d.ts +8 -8
- package/dist/types/learning/acceptance-plan.d.ts.map +1 -1
- package/dist/types/learning/admission.d.ts +37 -36
- package/dist/types/learning/admission.d.ts.map +1 -1
- package/dist/types/learning/assets.d.ts +2 -2
- package/dist/types/learning/assets.d.ts.map +1 -1
- package/dist/types/learning/authoring-service.d.ts +11 -11
- package/dist/types/learning/authoring-service.d.ts.map +1 -1
- package/dist/types/learning/authoring.d.ts +38 -38
- package/dist/types/learning/authoring.d.ts.map +1 -1
- package/dist/types/learning/batch-service.d.ts +75 -72
- package/dist/types/learning/batch-service.d.ts.map +1 -1
- package/dist/types/learning/budgeted-reward-fixtures.d.ts.map +1 -1
- package/dist/types/learning/continuation-mode.d.ts +18 -0
- package/dist/types/learning/continuation-mode.d.ts.map +1 -0
- package/dist/types/learning/contracts.d.ts +341 -298
- package/dist/types/learning/contracts.d.ts.map +1 -1
- package/dist/types/learning/correction-metrics.d.ts +274 -0
- package/dist/types/learning/correction-metrics.d.ts.map +1 -0
- package/dist/types/learning/errors.d.ts.map +1 -1
- package/dist/types/learning/grade-judge-budget.d.ts.map +1 -1
- package/dist/types/learning/grade-worker.d.ts.map +1 -1
- package/dist/types/learning/index.d.ts +2 -0
- package/dist/types/learning/index.d.ts.map +1 -1
- package/dist/types/learning/intake-contracts.d.ts +10 -10
- package/dist/types/learning/intake-contracts.d.ts.map +1 -1
- package/dist/types/learning/intake-csv.d.ts.map +1 -1
- package/dist/types/learning/intake-history.d.ts.map +1 -1
- package/dist/types/learning/intake-service.d.ts.map +1 -1
- package/dist/types/learning/intake-text.d.ts.map +1 -1
- package/dist/types/learning/intake.d.ts.map +1 -1
- package/dist/types/learning/iteration-budget.d.ts.map +1 -1
- package/dist/types/learning/iteration-candidate-decision.d.ts +16 -16
- package/dist/types/learning/iteration-candidate-decision.d.ts.map +1 -1
- package/dist/types/learning/iteration-dispatch-commands.d.ts.map +1 -1
- package/dist/types/learning/iteration-dispatch-contracts.d.ts +4 -4
- package/dist/types/learning/iteration-dispatch-contracts.d.ts.map +1 -1
- package/dist/types/learning/iteration-dispatch-worker.d.ts.map +1 -1
- package/dist/types/learning/iteration-eligibility.d.ts +1 -1
- package/dist/types/learning/iteration-eligibility.d.ts.map +1 -1
- package/dist/types/learning/iteration-reservation-cancellation.d.ts.map +1 -1
- package/dist/types/learning/iteration-reservation-contracts.d.ts.map +1 -1
- package/dist/types/learning/iteration-reservation-service.d.ts.map +1 -1
- package/dist/types/learning/iteration-training-parent.d.ts.map +1 -1
- package/dist/types/learning/judge-budget.d.ts +1 -1
- package/dist/types/learning/judge-budget.d.ts.map +1 -1
- package/dist/types/learning/model-judge.d.ts.map +1 -1
- package/dist/types/learning/nightly-schedule.d.ts.map +1 -1
- package/dist/types/learning/operations.d.ts +352 -298
- package/dist/types/learning/operations.d.ts.map +1 -1
- package/dist/types/learning/policy-inspection.d.ts.map +1 -1
- package/dist/types/learning/rating-reward-fixture.d.ts.map +1 -1
- package/dist/types/learning/repository.d.ts +199 -184
- package/dist/types/learning/repository.d.ts.map +1 -1
- package/dist/types/learning/reward-authoring.d.ts.map +1 -1
- package/dist/types/learning/reward-calibration-closure.d.ts +248 -248
- package/dist/types/learning/reward-calibration-closure.d.ts.map +1 -1
- package/dist/types/learning/reward-calibration.d.ts +62 -62
- package/dist/types/learning/reward-calibration.d.ts.map +1 -1
- package/dist/types/learning/reward-check-budget.d.ts.map +1 -1
- package/dist/types/learning/reward-check-service.d.ts.map +1 -1
- package/dist/types/learning/reward-check-worker.d.ts.map +1 -1
- package/dist/types/learning/reward-checks.d.ts +30 -30
- package/dist/types/learning/reward-checks.d.ts.map +1 -1
- package/dist/types/learning/reward-fixture-executor.d.ts.map +1 -1
- package/dist/types/learning/schedule-contracts.d.ts +3 -3
- package/dist/types/learning/schedule-contracts.d.ts.map +1 -1
- package/dist/types/learning/schedule-service.d.ts +2 -2
- package/dist/types/learning/schedule-service.d.ts.map +1 -1
- package/dist/types/learning/schedule-worker.d.ts.map +1 -1
- package/dist/types/learning/service.d.ts +3 -3
- package/dist/types/learning/service.d.ts.map +1 -1
- package/dist/types/learning/task-queue-inspection.d.ts.map +1 -1
- package/dist/types/learning/training-handoff.d.ts +9 -9
- package/dist/types/learning/training-handoff.d.ts.map +1 -1
- package/dist/types/learning/transport.d.ts +2270 -2228
- package/dist/types/learning/transport.d.ts.map +1 -1
- package/dist/types/metric-policy.d.ts +6 -6
- package/dist/types/metric-policy.d.ts.map +1 -1
- package/dist/types/metrics-node.d.ts.map +1 -1
- package/dist/types/metrics.d.ts.map +1 -1
- package/dist/types/model-improvement-qualification.d.ts +17 -17
- package/dist/types/model-improvement-qualification.d.ts.map +1 -1
- package/dist/types/native-conversations/appimage-runtime.d.ts.map +1 -1
- package/dist/types/native-conversations/codex-window-inventory.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-acquisition.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-branches.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-calendar-gate.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-contracts.d.ts +3 -0
- package/dist/types/native-conversations/collector-contracts.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-coverage.d.ts +2 -2
- package/dist/types/native-conversations/collector-coverage.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-cron.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-destinations.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-errors.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-progress.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-schedule.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-service-files.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-snapshots.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-source-cache.d.ts +2 -2
- package/dist/types/native-conversations/collector-source-cache.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-store.d.ts.map +1 -1
- package/dist/types/native-conversations/collector-sync-requests.d.ts.map +1 -1
- package/dist/types/native-conversations/collector.d.ts.map +1 -1
- package/dist/types/native-conversations/contracts.d.ts +2 -0
- package/dist/types/native-conversations/contracts.d.ts.map +1 -1
- package/dist/types/native-conversations/database.d.ts.map +1 -1
- package/dist/types/native-conversations/discovery.d.ts.map +1 -1
- package/dist/types/native-conversations/git-assets.d.ts.map +1 -1
- package/dist/types/native-conversations/history.d.ts +2 -0
- package/dist/types/native-conversations/history.d.ts.map +1 -1
- package/dist/types/native-conversations/index.d.ts +1 -0
- package/dist/types/native-conversations/index.d.ts.map +1 -1
- package/dist/types/native-conversations/location.d.ts.map +1 -1
- package/dist/types/native-conversations/openclaw-cold-archive.d.ts.map +1 -1
- package/dist/types/native-conversations/openclaw-database.d.ts.map +1 -1
- package/dist/types/native-conversations/readiness.d.ts.map +1 -1
- package/dist/types/native-conversations/selection.d.ts +32 -0
- package/dist/types/native-conversations/selection.d.ts.map +1 -0
- package/dist/types/native-conversations/source-locations.d.ts.map +1 -1
- package/dist/types/native-conversations/supervisor.d.ts.map +1 -1
- package/dist/types/native-conversations/titles.d.ts +1 -1
- package/dist/types/native-conversations/titles.d.ts.map +1 -1
- package/dist/types/native-harness-experiment-attempt.d.ts.map +1 -1
- package/dist/types/preferences.d.ts +22 -22
- package/dist/types/preferences.d.ts.map +1 -1
- package/dist/types/prepared-process-isolation.d.ts.map +1 -1
- package/dist/types/profile-evaluation-comparisons.d.ts +18 -18
- package/dist/types/profile-evaluation-comparisons.d.ts.map +1 -1
- package/dist/types/profile-evaluation-execution.d.ts.map +1 -1
- package/dist/types/profile-evaluation-reports.d.ts +2 -2
- package/dist/types/profile-evaluation-reports.d.ts.map +1 -1
- package/dist/types/profile-evaluation-suites.d.ts.map +1 -1
- package/dist/types/profile-evaluation-target.d.ts.map +1 -1
- package/dist/types/profile-evaluations.d.ts +15 -15
- package/dist/types/profile-evaluations.d.ts.map +1 -1
- package/dist/types/profile-external-evaluation.d.ts +84 -84
- package/dist/types/profile-external-evaluation.d.ts.map +1 -1
- package/dist/types/profile-source-candidate.d.ts.map +1 -1
- package/dist/types/review-conformance.d.ts +53 -53
- package/dist/types/review-conformance.d.ts.map +1 -1
- package/dist/types/rewards.d.ts +104 -104
- package/dist/types/rewards.d.ts.map +1 -1
- package/dist/types/rollouts.d.ts +16 -16
- package/dist/types/rollouts.d.ts.map +1 -1
- package/dist/types/runs.d.ts +22 -22
- package/dist/types/runs.d.ts.map +1 -1
- package/dist/types/sql-execution-contract.d.ts +3 -3
- package/dist/types/sql-execution-contract.d.ts.map +1 -1
- package/dist/types/sql-execution-engine.d.ts.map +1 -1
- package/dist/types/sql-execution-node.d.ts.map +1 -1
- package/dist/types/sql-execution-process-entry.d.ts.map +1 -1
- package/dist/types/task-schema-validation.d.ts.map +1 -1
- package/dist/types/task-schema.d.ts.map +1 -1
- package/dist/types/task-value-validator.d.ts.map +1 -1
- package/dist/types/taskset-run-contract.d.ts +26 -26
- package/dist/types/taskset-run-contract.d.ts.map +1 -1
- package/dist/types/tasksets.d.ts +73 -69
- package/dist/types/tasksets.d.ts.map +1 -1
- package/dist/types/telemetry/index.d.ts.map +1 -1
- package/dist/types/telemetry-analysis.d.ts +6 -6
- package/dist/types/telemetry-analysis.d.ts.map +1 -1
- package/dist/types/telemetry-bundle.d.ts +42 -42
- package/dist/types/telemetry-bundle.d.ts.map +1 -1
- package/dist/types/telemetry-catalog.d.ts +67 -67
- package/dist/types/telemetry-catalog.d.ts.map +1 -1
- package/dist/types/telemetry.d.ts +44 -44
- package/dist/types/telemetry.d.ts.map +1 -1
- package/package.json +1 -1
- package/schemas/learning/v1/command-request.schema.json +44 -0
- package/schemas/learning/v1/definition.schema.json +4 -0
- package/schemas/learning/v1/evidence.schema.json +4 -0
- package/schemas/learning/v1/example-submission.schema.json +4 -0
- package/schemas/learning/v1/package.schema.json +4 -0
- package/schemas/learning/v1/policy.schema.json +40 -0
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"benchmarks.d.ts","sourceRoot":"","sources":["
|
|
1
|
+
{"version":3,"file":"benchmarks.d.ts","sourceRoot":"","sources":["../../src/benchmarks.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAYxB,OAAO,EAAE,KAAK,cAAc,EAAE,KAAK,gBAAgB,EAAE,MAAM,WAAW,CAAC;AAGvE,OAAO,EACL,6BAA6B,EAC7B,8BAA8B,GAC/B,MAAM,yCAAyC,CAAC;AAEjD,eAAO,MAAM,qBAAqB;;;;;EAKhC,CAAC;AACH,eAAO,MAAM,uBAAuB;;;EAAoC,CAAC;AAEzE,eAAO,MAAM,uBAAuB;;;;;;;;;;;;;;kBASzB,CAAC;AAEZ,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAe3B,CAAC;AAEZ,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;;;;;;;kBAS3B,CAAC;AAQZ,eAAO,MAAM,gCAAgC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAkBlC,CAAC;AACZ,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAE3B,CAAC;AAEZ,eAAO,MAAM,gCAAgC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAelC,CAAC;AACZ,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAE3B,CAAC;AAEZ,wBAAgB,yBAAyB,CACvC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,GAC/C,mBAAmB,CAErB;AAED,wBAAgB,yBAAyB,CAAC,KAAK,EAAE;IAC/C,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,iBAAiB,CAAC;IACzB,UAAU,EAAE,gBAAgB,CAAC;IAC7B,QAAQ,EAAE,cAAc,EAAE,CAAC;IAC3B,eAAe,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,QAAQ,EAAE,iBAAiB,CAAC;IAC5B,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC,GAAG,mBAAmB,CAuCtB;AAED,wBAAgB,oBAAoB,CAAC,KAAK,EAAE;IAC1C,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,mBAAmB,CAAC;IAC9B,SAAS,EAAE,mBAAmB,CAAC;IAC/B,aAAa,EAAE,eAAe,CAAC;IAC/B,WAAW,EAAE,MAAM,GAAG,gBAAgB,GAAG,UAAU,CAAC;IACpD,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC,GAAG,mBAAmB,CAuDtB;AAkDD,MAAM,MAAM,eAAe,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,qBAAqB,CAAC,CAAC;AACpE,MAAM,MAAM,iBAAiB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAC;AACxE,MAAM,MAAM,iBAAiB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAC;AACxE,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC"}
|
|
@@ -10,14 +10,18 @@ export declare const harnessRefinerBenchmarkRelease: {
|
|
|
10
10
|
};
|
|
11
11
|
environment: {
|
|
12
12
|
protocolVersion: "openpond.environment.v1";
|
|
13
|
-
kind: "
|
|
13
|
+
kind: "agent" | "custom_program" | "text" | "work";
|
|
14
14
|
entrypoint: string;
|
|
15
15
|
stateful: boolean;
|
|
16
16
|
deterministicSeeds: boolean;
|
|
17
|
-
lifecycle: ("
|
|
18
|
-
networkPolicy: "
|
|
17
|
+
lifecycle: ("collect" | "create" | "destroy" | "reset" | "step")[];
|
|
18
|
+
networkPolicy: "declared_read_only" | "declared_scoped" | "none";
|
|
19
19
|
defaultTimeoutMs: number;
|
|
20
20
|
};
|
|
21
|
+
environmentRelease?: {
|
|
22
|
+
id: string;
|
|
23
|
+
contentHash: string;
|
|
24
|
+
} | undefined;
|
|
21
25
|
tools: {
|
|
22
26
|
name: string;
|
|
23
27
|
description: string;
|
|
@@ -30,12 +34,12 @@ export declare const harnessRefinerBenchmarkRelease: {
|
|
|
30
34
|
id: string;
|
|
31
35
|
required: boolean;
|
|
32
36
|
scopes: string[];
|
|
33
|
-
portability: "
|
|
37
|
+
portability: "host_adapter" | "hosted_only" | "local_only" | "portable";
|
|
34
38
|
}[];
|
|
35
39
|
tasks: {
|
|
36
40
|
id: string;
|
|
37
41
|
clusterKey: string;
|
|
38
|
-
split: "
|
|
42
|
+
split: "frozen_eval" | "test" | "train" | "validation";
|
|
39
43
|
input: Record<string, unknown>;
|
|
40
44
|
expectedOutput: Record<string, unknown> | null;
|
|
41
45
|
policyVisibleContext: Record<string, unknown>;
|
|
@@ -46,9 +50,8 @@ export declare const harnessRefinerBenchmarkRelease: {
|
|
|
46
50
|
contentHash: string;
|
|
47
51
|
sizeBytes: number;
|
|
48
52
|
mediaType: string;
|
|
49
|
-
visibility: "
|
|
53
|
+
visibility: "host_private" | "policy" | "verifier";
|
|
50
54
|
}[];
|
|
51
|
-
tags: string[];
|
|
52
55
|
requiredOutputs?: {
|
|
53
56
|
path: string;
|
|
54
57
|
mediaType: string;
|
|
@@ -58,24 +61,27 @@ export declare const harnessRefinerBenchmarkRelease: {
|
|
|
58
61
|
contentHash: string;
|
|
59
62
|
sizeBytes: number;
|
|
60
63
|
mediaType: string;
|
|
61
|
-
visibility: "
|
|
64
|
+
visibility: "host_private" | "policy" | "verifier";
|
|
62
65
|
} | null;
|
|
66
|
+
submissionMode?: "structured_json" | undefined;
|
|
63
67
|
maxBytes: number | null;
|
|
64
68
|
metadata: Record<string, unknown>;
|
|
65
69
|
}[] | undefined;
|
|
70
|
+
tags: string[];
|
|
66
71
|
}[];
|
|
67
72
|
graders: ({
|
|
68
73
|
id: string;
|
|
74
|
+
taskIds?: string[] | undefined;
|
|
69
75
|
version: string;
|
|
70
76
|
weight: number;
|
|
71
77
|
hardGate: boolean;
|
|
72
78
|
rewardEligible: boolean;
|
|
73
79
|
privileged: boolean;
|
|
74
|
-
kind: "
|
|
80
|
+
kind: "artifact" | "content" | "runtime_event" | "schema" | "state";
|
|
75
81
|
config: Record<string, unknown>;
|
|
76
|
-
taskIds?: string[] | undefined;
|
|
77
82
|
} | {
|
|
78
83
|
id: string;
|
|
84
|
+
taskIds?: string[] | undefined;
|
|
79
85
|
version: string;
|
|
80
86
|
weight: number;
|
|
81
87
|
hardGate: boolean;
|
|
@@ -88,10 +94,9 @@ export declare const harnessRefinerBenchmarkRelease: {
|
|
|
88
94
|
contentHash: string;
|
|
89
95
|
sizeBytes: number;
|
|
90
96
|
mediaType: string;
|
|
91
|
-
visibility: "
|
|
97
|
+
visibility: "host_private" | "policy" | "verifier";
|
|
92
98
|
};
|
|
93
|
-
calibrationStatus: "
|
|
94
|
-
taskIds?: string[] | undefined;
|
|
99
|
+
calibrationStatus: "failed" | "passed" | "pending";
|
|
95
100
|
model?: {
|
|
96
101
|
providerId: string;
|
|
97
102
|
modelId: string;
|
|
@@ -100,27 +105,28 @@ export declare const harnessRefinerBenchmarkRelease: {
|
|
|
100
105
|
temperature?: number | undefined;
|
|
101
106
|
} | {
|
|
102
107
|
id: string;
|
|
108
|
+
taskIds?: string[] | undefined;
|
|
103
109
|
version: string;
|
|
104
110
|
weight: number;
|
|
105
111
|
hardGate: boolean;
|
|
106
112
|
rewardEligible: boolean;
|
|
107
113
|
privileged: boolean;
|
|
108
114
|
kind: "custom_verifier";
|
|
115
|
+
runtime?: "isolated_javascript" | "sandbox_process" | undefined;
|
|
109
116
|
verifierRef: {
|
|
110
117
|
id: string;
|
|
111
118
|
path: string;
|
|
112
119
|
contentHash: string;
|
|
113
120
|
sizeBytes: number;
|
|
114
121
|
mediaType: string;
|
|
115
|
-
visibility: "
|
|
122
|
+
visibility: "host_private" | "policy" | "verifier";
|
|
116
123
|
};
|
|
124
|
+
exportName?: string | undefined;
|
|
117
125
|
timeoutMs: number;
|
|
118
126
|
networkPolicy: "none";
|
|
119
|
-
taskIds?: string[] | undefined;
|
|
120
|
-
runtime?: "isolated_javascript" | "sandbox_process" | undefined;
|
|
121
|
-
exportName?: string | undefined;
|
|
122
127
|
} | {
|
|
123
128
|
id: string;
|
|
129
|
+
taskIds?: string[] | undefined;
|
|
124
130
|
version: string;
|
|
125
131
|
weight: number;
|
|
126
132
|
hardGate: boolean;
|
|
@@ -133,10 +139,9 @@ export declare const harnessRefinerBenchmarkRelease: {
|
|
|
133
139
|
contentHash: string;
|
|
134
140
|
sizeBytes: number;
|
|
135
141
|
mediaType: string;
|
|
136
|
-
visibility: "
|
|
142
|
+
visibility: "host_private" | "policy" | "verifier";
|
|
137
143
|
};
|
|
138
144
|
reviewerRole: string;
|
|
139
|
-
taskIds?: string[] | undefined;
|
|
140
145
|
form?: {
|
|
141
146
|
schemaVersion: "openpond.humanForm.v1";
|
|
142
147
|
mode: "individual" | "pairwise";
|
|
@@ -147,77 +152,71 @@ export declare const harnessRefinerBenchmarkRelease: {
|
|
|
147
152
|
instructions: string;
|
|
148
153
|
required: boolean;
|
|
149
154
|
allowAbstain: boolean;
|
|
150
|
-
kind: "score";
|
|
151
|
-
minimum: number;
|
|
152
|
-
maximum: number;
|
|
153
|
-
step: number;
|
|
154
155
|
requireRationale?: boolean | undefined;
|
|
155
156
|
anchors?: {
|
|
156
157
|
value: string | number | boolean;
|
|
157
158
|
description: string;
|
|
158
159
|
}[] | undefined;
|
|
160
|
+
kind: "score";
|
|
161
|
+
minimum: number;
|
|
162
|
+
maximum: number;
|
|
163
|
+
step: number;
|
|
159
164
|
} | {
|
|
160
165
|
id: string;
|
|
161
166
|
label: string;
|
|
162
167
|
instructions: string;
|
|
163
168
|
required: boolean;
|
|
164
169
|
allowAbstain: boolean;
|
|
165
|
-
kind: "category";
|
|
166
|
-
options: string[];
|
|
167
170
|
requireRationale?: boolean | undefined;
|
|
168
171
|
anchors?: {
|
|
169
172
|
value: string | number | boolean;
|
|
170
173
|
description: string;
|
|
171
174
|
}[] | undefined;
|
|
175
|
+
kind: "category";
|
|
176
|
+
options: string[];
|
|
172
177
|
} | {
|
|
173
178
|
id: string;
|
|
174
179
|
label: string;
|
|
175
180
|
instructions: string;
|
|
176
181
|
required: boolean;
|
|
177
182
|
allowAbstain: boolean;
|
|
178
|
-
kind: "boolean";
|
|
179
183
|
requireRationale?: boolean | undefined;
|
|
180
184
|
anchors?: {
|
|
181
185
|
value: string | number | boolean;
|
|
182
186
|
description: string;
|
|
183
187
|
}[] | undefined;
|
|
188
|
+
kind: "boolean";
|
|
184
189
|
} | {
|
|
185
190
|
id: string;
|
|
186
191
|
label: string;
|
|
187
192
|
instructions: string;
|
|
188
193
|
required: boolean;
|
|
189
194
|
allowAbstain: boolean;
|
|
190
|
-
kind: "text";
|
|
191
|
-
maximumLength: number;
|
|
192
195
|
requireRationale?: boolean | undefined;
|
|
193
196
|
anchors?: {
|
|
194
197
|
value: string | number | boolean;
|
|
195
198
|
description: string;
|
|
196
199
|
}[] | undefined;
|
|
200
|
+
kind: "text";
|
|
197
201
|
minimumLength?: number | undefined;
|
|
202
|
+
maximumLength: number;
|
|
198
203
|
} | {
|
|
199
204
|
id: string;
|
|
200
205
|
label: string;
|
|
201
206
|
instructions: string;
|
|
202
207
|
required: boolean;
|
|
203
208
|
allowAbstain: boolean;
|
|
204
|
-
kind: "preference";
|
|
205
|
-
allowTie: boolean;
|
|
206
209
|
requireRationale?: boolean | undefined;
|
|
207
210
|
anchors?: {
|
|
208
211
|
value: string | number | boolean;
|
|
209
212
|
description: string;
|
|
210
213
|
}[] | undefined;
|
|
214
|
+
kind: "preference";
|
|
211
215
|
requireStrength?: boolean | undefined;
|
|
216
|
+
allowTie: boolean;
|
|
212
217
|
})[];
|
|
213
218
|
} | undefined;
|
|
214
219
|
})[];
|
|
215
|
-
metadata: Record<string, unknown>;
|
|
216
|
-
contentHash: string;
|
|
217
|
-
environmentRelease?: {
|
|
218
|
-
id: string;
|
|
219
|
-
contentHash: string;
|
|
220
|
-
} | undefined;
|
|
221
220
|
verifierSetRelease?: {
|
|
222
221
|
id: string;
|
|
223
222
|
contentHash: string;
|
|
@@ -225,8 +224,9 @@ export declare const harnessRefinerBenchmarkRelease: {
|
|
|
225
224
|
metrics?: {
|
|
226
225
|
schemaVersion: "openpond.tasksetMetricPolicy.v1";
|
|
227
226
|
primaryMetric: string;
|
|
228
|
-
aggregation: "
|
|
229
|
-
missingReward: "
|
|
227
|
+
aggregation: "custom" | "mean_score" | "pass_rate" | "weighted_mean";
|
|
228
|
+
missingReward: "exclude" | "zero";
|
|
229
|
+
taskWeights?: Record<string, number> | undefined;
|
|
230
230
|
customAggregator: {
|
|
231
231
|
module: string;
|
|
232
232
|
exportName: string;
|
|
@@ -234,8 +234,9 @@ export declare const harnessRefinerBenchmarkRelease: {
|
|
|
234
234
|
timeoutMs: number;
|
|
235
235
|
networkPolicy: "none";
|
|
236
236
|
} | null;
|
|
237
|
-
taskWeights?: Record<string, number> | undefined;
|
|
238
237
|
} | undefined;
|
|
238
|
+
metadata: Record<string, unknown>;
|
|
239
|
+
contentHash: string;
|
|
239
240
|
};
|
|
240
241
|
export declare const harnessRefinerBenchmarkAssets: Readonly<Record<string, string>>;
|
|
241
242
|
//# sourceMappingURL=harness-refiner.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"harness-refiner.d.ts","sourceRoot":"","sources":["
|
|
1
|
+
{"version":3,"file":"harness-refiner.d.ts","sourceRoot":"","sources":["../../../src/builtin-benchmarks/harness-refiner.ts"],"names":[],"mappings":"AAGA,eAAO,MAAM,8BAA8B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAqqD1C,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CASzE,CAAC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"compatibility.d.ts","sourceRoot":"","sources":["
|
|
1
|
+
{"version":3,"file":"compatibility.d.ts","sourceRoot":"","sources":["../../src/compatibility.ts"],"names":[],"mappings":"AACA,OAAO,EAEL,KAAK,cAAc,EACpB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EAEL,KAAK,2BAA2B,EACjC,MAAM,WAAW,CAAC;AACnB,OAAO,EAEL,KAAK,cAAc,EACpB,MAAM,eAAe,CAAC;AAEvB,wBAAgB,yCAAyC,CAAC,KAAK,EAAE;IAC/D,EAAE,EAAE,MAAM,CAAC;IACX,kBAAkB,EAAE,cAAc,CAAC;IACnC,uBAAuB,EAAE,cAAc,CAAC;IACxC,cAAc,EAAE,cAAc,CAAC;IAC/B,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC,GAAG,2BAA2B,CA4C9B"}
|