@openpond/evals 0.12.0 → 0.12.1
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/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 +7 -0
- 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/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/contracts.d.ts +305 -298
- package/dist/types/learning/contracts.d.ts.map +1 -1
- 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.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 +304 -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 +187 -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 +2207 -2201
- 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.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 +1 -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.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 +4 -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
|
@@ -82,10 +82,10 @@ export declare const ExperimentTargetSchema: z.ZodDiscriminatedUnion<[z.ZodObjec
|
|
|
82
82
|
}, z.core.$strict>;
|
|
83
83
|
}, z.core.$strict>>;
|
|
84
84
|
split: z.ZodEnum<{
|
|
85
|
+
frozen_eval: "frozen_eval";
|
|
86
|
+
test: "test";
|
|
85
87
|
train: "train";
|
|
86
88
|
validation: "validation";
|
|
87
|
-
test: "test";
|
|
88
|
-
frozen_eval: "frozen_eval";
|
|
89
89
|
}>;
|
|
90
90
|
population: z.ZodArray<z.ZodObject<{
|
|
91
91
|
taskId: z.ZodString;
|
|
@@ -102,8 +102,8 @@ export declare const ExperimentTargetSchema: z.ZodDiscriminatedUnion<[z.ZodObjec
|
|
|
102
102
|
configurationHash: z.ZodString;
|
|
103
103
|
output: z.ZodEnum<{
|
|
104
104
|
boolean: "boolean";
|
|
105
|
-
score: "score";
|
|
106
105
|
category: "category";
|
|
106
|
+
score: "score";
|
|
107
107
|
}>;
|
|
108
108
|
categories: z.ZodArray<z.ZodString>;
|
|
109
109
|
}, z.core.$strict>>;
|
|
@@ -112,17 +112,17 @@ export declare const ExperimentTargetSchema: z.ZodDiscriminatedUnion<[z.ZodObjec
|
|
|
112
112
|
graderId: z.ZodString;
|
|
113
113
|
fields: z.ZodArray<z.ZodObject<{
|
|
114
114
|
destination: z.ZodEnum<{
|
|
115
|
-
output: "output";
|
|
116
|
-
input: "input";
|
|
117
|
-
expectedOutput: "expectedOutput";
|
|
118
115
|
evaluatorContext: "evaluatorContext";
|
|
116
|
+
expectedOutput: "expectedOutput";
|
|
117
|
+
input: "input";
|
|
118
|
+
output: "output";
|
|
119
119
|
}>;
|
|
120
120
|
path: z.ZodArray<z.ZodString>;
|
|
121
121
|
source: z.ZodEnum<{
|
|
122
|
-
output: "output";
|
|
123
|
-
input: "input";
|
|
124
|
-
expectedOutput: "expectedOutput";
|
|
125
122
|
evaluatorContext: "evaluatorContext";
|
|
123
|
+
expectedOutput: "expectedOutput";
|
|
124
|
+
input: "input";
|
|
125
|
+
output: "output";
|
|
126
126
|
}>;
|
|
127
127
|
sourcePath: z.ZodArray<z.ZodString>;
|
|
128
128
|
}, z.core.$strict>>;
|
|
@@ -181,8 +181,8 @@ export declare const ExperimentEvaluatorSchema: z.ZodObject<{
|
|
|
181
181
|
configurationHash: z.ZodOptional<z.ZodString>;
|
|
182
182
|
output: z.ZodEnum<{
|
|
183
183
|
boolean: "boolean";
|
|
184
|
-
score: "score";
|
|
185
184
|
category: "category";
|
|
185
|
+
score: "score";
|
|
186
186
|
}>;
|
|
187
187
|
categories: z.ZodArray<z.ZodString>;
|
|
188
188
|
}, z.core.$strict>;
|
|
@@ -290,10 +290,10 @@ export declare const ExperimentManifestContentSchema: z.ZodObject<{
|
|
|
290
290
|
}, z.core.$strict>;
|
|
291
291
|
}, z.core.$strict>>;
|
|
292
292
|
split: z.ZodEnum<{
|
|
293
|
+
frozen_eval: "frozen_eval";
|
|
294
|
+
test: "test";
|
|
293
295
|
train: "train";
|
|
294
296
|
validation: "validation";
|
|
295
|
-
test: "test";
|
|
296
|
-
frozen_eval: "frozen_eval";
|
|
297
297
|
}>;
|
|
298
298
|
population: z.ZodArray<z.ZodObject<{
|
|
299
299
|
taskId: z.ZodString;
|
|
@@ -310,8 +310,8 @@ export declare const ExperimentManifestContentSchema: z.ZodObject<{
|
|
|
310
310
|
configurationHash: z.ZodString;
|
|
311
311
|
output: z.ZodEnum<{
|
|
312
312
|
boolean: "boolean";
|
|
313
|
-
score: "score";
|
|
314
313
|
category: "category";
|
|
314
|
+
score: "score";
|
|
315
315
|
}>;
|
|
316
316
|
categories: z.ZodArray<z.ZodString>;
|
|
317
317
|
}, z.core.$strict>>;
|
|
@@ -320,17 +320,17 @@ export declare const ExperimentManifestContentSchema: z.ZodObject<{
|
|
|
320
320
|
graderId: z.ZodString;
|
|
321
321
|
fields: z.ZodArray<z.ZodObject<{
|
|
322
322
|
destination: z.ZodEnum<{
|
|
323
|
-
output: "output";
|
|
324
|
-
input: "input";
|
|
325
|
-
expectedOutput: "expectedOutput";
|
|
326
323
|
evaluatorContext: "evaluatorContext";
|
|
324
|
+
expectedOutput: "expectedOutput";
|
|
325
|
+
input: "input";
|
|
326
|
+
output: "output";
|
|
327
327
|
}>;
|
|
328
328
|
path: z.ZodArray<z.ZodString>;
|
|
329
329
|
source: z.ZodEnum<{
|
|
330
|
-
output: "output";
|
|
331
|
-
input: "input";
|
|
332
|
-
expectedOutput: "expectedOutput";
|
|
333
330
|
evaluatorContext: "evaluatorContext";
|
|
331
|
+
expectedOutput: "expectedOutput";
|
|
332
|
+
input: "input";
|
|
333
|
+
output: "output";
|
|
334
334
|
}>;
|
|
335
335
|
sourcePath: z.ZodArray<z.ZodString>;
|
|
336
336
|
}, z.core.$strict>>;
|
|
@@ -387,9 +387,9 @@ export declare const ExperimentManifestContentSchema: z.ZodObject<{
|
|
|
387
387
|
protocol: z.ZodLiteral<"openpond.evaluation-execution.v1">;
|
|
388
388
|
targetKind: z.ZodEnum<{
|
|
389
389
|
agent: "agent";
|
|
390
|
-
model: "model";
|
|
391
390
|
fixture: "fixture";
|
|
392
391
|
harness: "harness";
|
|
392
|
+
model: "model";
|
|
393
393
|
}>;
|
|
394
394
|
}, z.core.$strict>>;
|
|
395
395
|
}, z.core.$strict>>;
|
|
@@ -423,8 +423,8 @@ export declare const ExperimentManifestContentSchema: z.ZodObject<{
|
|
|
423
423
|
configurationHash: z.ZodOptional<z.ZodString>;
|
|
424
424
|
output: z.ZodEnum<{
|
|
425
425
|
boolean: "boolean";
|
|
426
|
-
score: "score";
|
|
427
426
|
category: "category";
|
|
427
|
+
score: "score";
|
|
428
428
|
}>;
|
|
429
429
|
categories: z.ZodArray<z.ZodString>;
|
|
430
430
|
}, z.core.$strict>>;
|
|
@@ -528,10 +528,10 @@ export declare const ExperimentManifestSchema: z.ZodObject<{
|
|
|
528
528
|
}, z.core.$strict>;
|
|
529
529
|
}, z.core.$strict>>;
|
|
530
530
|
split: z.ZodEnum<{
|
|
531
|
+
frozen_eval: "frozen_eval";
|
|
532
|
+
test: "test";
|
|
531
533
|
train: "train";
|
|
532
534
|
validation: "validation";
|
|
533
|
-
test: "test";
|
|
534
|
-
frozen_eval: "frozen_eval";
|
|
535
535
|
}>;
|
|
536
536
|
population: z.ZodArray<z.ZodObject<{
|
|
537
537
|
taskId: z.ZodString;
|
|
@@ -548,8 +548,8 @@ export declare const ExperimentManifestSchema: z.ZodObject<{
|
|
|
548
548
|
configurationHash: z.ZodString;
|
|
549
549
|
output: z.ZodEnum<{
|
|
550
550
|
boolean: "boolean";
|
|
551
|
-
score: "score";
|
|
552
551
|
category: "category";
|
|
552
|
+
score: "score";
|
|
553
553
|
}>;
|
|
554
554
|
categories: z.ZodArray<z.ZodString>;
|
|
555
555
|
}, z.core.$strict>>;
|
|
@@ -558,17 +558,17 @@ export declare const ExperimentManifestSchema: z.ZodObject<{
|
|
|
558
558
|
graderId: z.ZodString;
|
|
559
559
|
fields: z.ZodArray<z.ZodObject<{
|
|
560
560
|
destination: z.ZodEnum<{
|
|
561
|
-
output: "output";
|
|
562
|
-
input: "input";
|
|
563
|
-
expectedOutput: "expectedOutput";
|
|
564
561
|
evaluatorContext: "evaluatorContext";
|
|
562
|
+
expectedOutput: "expectedOutput";
|
|
563
|
+
input: "input";
|
|
564
|
+
output: "output";
|
|
565
565
|
}>;
|
|
566
566
|
path: z.ZodArray<z.ZodString>;
|
|
567
567
|
source: z.ZodEnum<{
|
|
568
|
-
output: "output";
|
|
569
|
-
input: "input";
|
|
570
|
-
expectedOutput: "expectedOutput";
|
|
571
568
|
evaluatorContext: "evaluatorContext";
|
|
569
|
+
expectedOutput: "expectedOutput";
|
|
570
|
+
input: "input";
|
|
571
|
+
output: "output";
|
|
572
572
|
}>;
|
|
573
573
|
sourcePath: z.ZodArray<z.ZodString>;
|
|
574
574
|
}, z.core.$strict>>;
|
|
@@ -625,9 +625,9 @@ export declare const ExperimentManifestSchema: z.ZodObject<{
|
|
|
625
625
|
protocol: z.ZodLiteral<"openpond.evaluation-execution.v1">;
|
|
626
626
|
targetKind: z.ZodEnum<{
|
|
627
627
|
agent: "agent";
|
|
628
|
-
model: "model";
|
|
629
628
|
fixture: "fixture";
|
|
630
629
|
harness: "harness";
|
|
630
|
+
model: "model";
|
|
631
631
|
}>;
|
|
632
632
|
}, z.core.$strict>>;
|
|
633
633
|
}, z.core.$strict>>;
|
|
@@ -661,8 +661,8 @@ export declare const ExperimentManifestSchema: z.ZodObject<{
|
|
|
661
661
|
configurationHash: z.ZodOptional<z.ZodString>;
|
|
662
662
|
output: z.ZodEnum<{
|
|
663
663
|
boolean: "boolean";
|
|
664
|
-
score: "score";
|
|
665
664
|
category: "category";
|
|
665
|
+
score: "score";
|
|
666
666
|
}>;
|
|
667
667
|
categories: z.ZodArray<z.ZodString>;
|
|
668
668
|
}, z.core.$strict>>;
|
|
@@ -683,8 +683,8 @@ export declare const ExperimentFeedbackSchema: z.ZodObject<{
|
|
|
683
683
|
revision: z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>;
|
|
684
684
|
}, z.core.$strict>;
|
|
685
685
|
status: z.ZodEnum<{
|
|
686
|
-
pending: "pending";
|
|
687
686
|
failed: "failed";
|
|
687
|
+
pending: "pending";
|
|
688
688
|
scored: "scored";
|
|
689
689
|
unavailable: "unavailable";
|
|
690
690
|
}>;
|
|
@@ -713,8 +713,8 @@ export declare const ExperimentCaseResultSchema: z.ZodObject<{
|
|
|
713
713
|
}, z.core.$strict>;
|
|
714
714
|
status: z.ZodEnum<{
|
|
715
715
|
cancelled: "cancelled";
|
|
716
|
-
failed: "failed";
|
|
717
716
|
completed: "completed";
|
|
717
|
+
failed: "failed";
|
|
718
718
|
unavailable: "unavailable";
|
|
719
719
|
}>;
|
|
720
720
|
output: z.ZodNullable<z.ZodUnknown>;
|
|
@@ -730,8 +730,8 @@ export declare const ExperimentCaseResultSchema: z.ZodObject<{
|
|
|
730
730
|
revision: z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>;
|
|
731
731
|
}, z.core.$strict>;
|
|
732
732
|
status: z.ZodEnum<{
|
|
733
|
-
pending: "pending";
|
|
734
733
|
failed: "failed";
|
|
734
|
+
pending: "pending";
|
|
735
735
|
scored: "scored";
|
|
736
736
|
unavailable: "unavailable";
|
|
737
737
|
}>;
|
|
@@ -769,8 +769,8 @@ export declare const ExperimentResultContentSchema: z.ZodObject<{
|
|
|
769
769
|
}, z.core.$strict>;
|
|
770
770
|
status: z.ZodEnum<{
|
|
771
771
|
cancelled: "cancelled";
|
|
772
|
-
failed: "failed";
|
|
773
772
|
completed: "completed";
|
|
773
|
+
failed: "failed";
|
|
774
774
|
}>;
|
|
775
775
|
cases: z.ZodArray<z.ZodObject<{
|
|
776
776
|
identity: z.ZodObject<{
|
|
@@ -780,8 +780,8 @@ export declare const ExperimentResultContentSchema: z.ZodObject<{
|
|
|
780
780
|
}, z.core.$strict>;
|
|
781
781
|
status: z.ZodEnum<{
|
|
782
782
|
cancelled: "cancelled";
|
|
783
|
-
failed: "failed";
|
|
784
783
|
completed: "completed";
|
|
784
|
+
failed: "failed";
|
|
785
785
|
unavailable: "unavailable";
|
|
786
786
|
}>;
|
|
787
787
|
output: z.ZodNullable<z.ZodUnknown>;
|
|
@@ -797,8 +797,8 @@ export declare const ExperimentResultContentSchema: z.ZodObject<{
|
|
|
797
797
|
revision: z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>;
|
|
798
798
|
}, z.core.$strict>;
|
|
799
799
|
status: z.ZodEnum<{
|
|
800
|
-
pending: "pending";
|
|
801
800
|
failed: "failed";
|
|
801
|
+
pending: "pending";
|
|
802
802
|
scored: "scored";
|
|
803
803
|
unavailable: "unavailable";
|
|
804
804
|
}>;
|
|
@@ -838,8 +838,8 @@ export declare const ExperimentResultSchema: z.ZodObject<{
|
|
|
838
838
|
}, z.core.$strict>;
|
|
839
839
|
status: z.ZodEnum<{
|
|
840
840
|
cancelled: "cancelled";
|
|
841
|
-
failed: "failed";
|
|
842
841
|
completed: "completed";
|
|
842
|
+
failed: "failed";
|
|
843
843
|
}>;
|
|
844
844
|
cases: z.ZodArray<z.ZodObject<{
|
|
845
845
|
identity: z.ZodObject<{
|
|
@@ -849,8 +849,8 @@ export declare const ExperimentResultSchema: z.ZodObject<{
|
|
|
849
849
|
}, z.core.$strict>;
|
|
850
850
|
status: z.ZodEnum<{
|
|
851
851
|
cancelled: "cancelled";
|
|
852
|
-
failed: "failed";
|
|
853
852
|
completed: "completed";
|
|
853
|
+
failed: "failed";
|
|
854
854
|
unavailable: "unavailable";
|
|
855
855
|
}>;
|
|
856
856
|
output: z.ZodNullable<z.ZodUnknown>;
|
|
@@ -866,8 +866,8 @@ export declare const ExperimentResultSchema: z.ZodObject<{
|
|
|
866
866
|
revision: z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>;
|
|
867
867
|
}, z.core.$strict>;
|
|
868
868
|
status: z.ZodEnum<{
|
|
869
|
-
pending: "pending";
|
|
870
869
|
failed: "failed";
|
|
870
|
+
pending: "pending";
|
|
871
871
|
scored: "scored";
|
|
872
872
|
unavailable: "unavailable";
|
|
873
873
|
}>;
|
|
@@ -1004,7 +1004,35 @@ export declare function verifyExperimentEvidence(input: {
|
|
|
1004
1004
|
};
|
|
1005
1005
|
declaredProfileCatalogHash: string | null;
|
|
1006
1006
|
protectedProfileClosureHash: string;
|
|
1007
|
-
|
|
1007
|
+
recordedOrigin?: {
|
|
1008
|
+
execution: {
|
|
1009
|
+
id: string;
|
|
1010
|
+
contentHash: string;
|
|
1011
|
+
};
|
|
1012
|
+
originalEvidence: {
|
|
1013
|
+
id: string;
|
|
1014
|
+
contentHash: string;
|
|
1015
|
+
};
|
|
1016
|
+
members: {
|
|
1017
|
+
caseId: string;
|
|
1018
|
+
seed: string;
|
|
1019
|
+
fixtureId: string;
|
|
1020
|
+
sourceId: string;
|
|
1021
|
+
snapshotHash: string;
|
|
1022
|
+
boundaryRevisionHash: string;
|
|
1023
|
+
}[];
|
|
1024
|
+
} | undefined;
|
|
1025
|
+
gradingSource?: {
|
|
1026
|
+
pass: {
|
|
1027
|
+
id: string;
|
|
1028
|
+
contentHash: string;
|
|
1029
|
+
};
|
|
1030
|
+
evidence: {
|
|
1031
|
+
id: string;
|
|
1032
|
+
contentHash: string;
|
|
1033
|
+
};
|
|
1034
|
+
} | undefined;
|
|
1035
|
+
split: "frozen_eval" | "test" | "train" | "validation";
|
|
1008
1036
|
population: {
|
|
1009
1037
|
taskId: string;
|
|
1010
1038
|
seed: string;
|
|
@@ -1018,16 +1046,16 @@ export declare function verifyExperimentEvidence(input: {
|
|
|
1018
1046
|
};
|
|
1019
1047
|
feedbackKey: string;
|
|
1020
1048
|
configurationHash: string;
|
|
1021
|
-
output: "boolean" | "
|
|
1049
|
+
output: "boolean" | "category" | "score";
|
|
1022
1050
|
categories: string[];
|
|
1023
1051
|
}[];
|
|
1024
1052
|
fieldMappingsHash: string;
|
|
1025
1053
|
fieldMappings: {
|
|
1026
1054
|
graderId: string;
|
|
1027
1055
|
fields: {
|
|
1028
|
-
destination: "
|
|
1056
|
+
destination: "evaluatorContext" | "expectedOutput" | "input" | "output";
|
|
1029
1057
|
path: string[];
|
|
1030
|
-
source: "
|
|
1058
|
+
source: "evaluatorContext" | "expectedOutput" | "input" | "output";
|
|
1031
1059
|
sourcePath: string[];
|
|
1032
1060
|
}[];
|
|
1033
1061
|
}[];
|
|
@@ -1038,34 +1066,6 @@ export declare function verifyExperimentEvidence(input: {
|
|
|
1038
1066
|
requireComplete: boolean;
|
|
1039
1067
|
};
|
|
1040
1068
|
contentHash: string;
|
|
1041
|
-
recordedOrigin?: {
|
|
1042
|
-
execution: {
|
|
1043
|
-
id: string;
|
|
1044
|
-
contentHash: string;
|
|
1045
|
-
};
|
|
1046
|
-
originalEvidence: {
|
|
1047
|
-
id: string;
|
|
1048
|
-
contentHash: string;
|
|
1049
|
-
};
|
|
1050
|
-
members: {
|
|
1051
|
-
caseId: string;
|
|
1052
|
-
seed: string;
|
|
1053
|
-
fixtureId: string;
|
|
1054
|
-
sourceId: string;
|
|
1055
|
-
snapshotHash: string;
|
|
1056
|
-
boundaryRevisionHash: string;
|
|
1057
|
-
}[];
|
|
1058
|
-
} | undefined;
|
|
1059
|
-
gradingSource?: {
|
|
1060
|
-
pass: {
|
|
1061
|
-
id: string;
|
|
1062
|
-
contentHash: string;
|
|
1063
|
-
};
|
|
1064
|
-
evidence: {
|
|
1065
|
-
id: string;
|
|
1066
|
-
contentHash: string;
|
|
1067
|
-
};
|
|
1068
|
-
} | undefined;
|
|
1069
1069
|
} | undefined;
|
|
1070
1070
|
sourceCandidate?: {
|
|
1071
1071
|
candidateId: string;
|
|
@@ -1103,31 +1103,13 @@ export declare function verifyExperimentEvidence(input: {
|
|
|
1103
1103
|
kind: "fixture";
|
|
1104
1104
|
configurationHash: string;
|
|
1105
1105
|
};
|
|
1106
|
-
evaluators: {
|
|
1107
|
-
release: {
|
|
1108
|
-
id: string;
|
|
1109
|
-
contentHash: string;
|
|
1110
|
-
revision: string | number;
|
|
1111
|
-
};
|
|
1112
|
-
feedbackKey: string;
|
|
1113
|
-
output: "boolean" | "score" | "category";
|
|
1114
|
-
categories: string[];
|
|
1115
|
-
configurationHash?: string | undefined;
|
|
1116
|
-
}[];
|
|
1117
|
-
population: {
|
|
1118
|
-
caseId: string;
|
|
1119
|
-
seed: string;
|
|
1120
|
-
fixtureId: string | null;
|
|
1121
|
-
}[];
|
|
1122
|
-
createdAt: string;
|
|
1123
|
-
contentHash: string;
|
|
1124
1106
|
execution?: {
|
|
1125
1107
|
packageHash: string;
|
|
1126
1108
|
runtimeTargetHash: string;
|
|
1127
1109
|
metricPolicyHash: string;
|
|
1128
1110
|
compatibility?: {
|
|
1129
1111
|
protocol: "openpond.evaluation-execution.v1";
|
|
1130
|
-
targetKind: "agent" | "
|
|
1112
|
+
targetKind: "agent" | "fixture" | "harness" | "model";
|
|
1131
1113
|
} | undefined;
|
|
1132
1114
|
} | undefined;
|
|
1133
1115
|
lineage?: {
|
|
@@ -1150,6 +1132,24 @@ export declare function verifyExperimentEvidence(input: {
|
|
|
1150
1132
|
rulesHash: string;
|
|
1151
1133
|
} | undefined;
|
|
1152
1134
|
} | undefined;
|
|
1135
|
+
evaluators: {
|
|
1136
|
+
release: {
|
|
1137
|
+
id: string;
|
|
1138
|
+
contentHash: string;
|
|
1139
|
+
revision: string | number;
|
|
1140
|
+
};
|
|
1141
|
+
feedbackKey: string;
|
|
1142
|
+
configurationHash?: string | undefined;
|
|
1143
|
+
output: "boolean" | "category" | "score";
|
|
1144
|
+
categories: string[];
|
|
1145
|
+
}[];
|
|
1146
|
+
population: {
|
|
1147
|
+
caseId: string;
|
|
1148
|
+
seed: string;
|
|
1149
|
+
fixtureId: string | null;
|
|
1150
|
+
}[];
|
|
1151
|
+
createdAt: string;
|
|
1152
|
+
contentHash: string;
|
|
1153
1153
|
};
|
|
1154
1154
|
result: {
|
|
1155
1155
|
schemaVersion: "openpond.experimentResult.v1";
|
|
@@ -1157,14 +1157,14 @@ export declare function verifyExperimentEvidence(input: {
|
|
|
1157
1157
|
id: string;
|
|
1158
1158
|
contentHash: string;
|
|
1159
1159
|
};
|
|
1160
|
-
status: "cancelled" | "
|
|
1160
|
+
status: "cancelled" | "completed" | "failed";
|
|
1161
1161
|
cases: {
|
|
1162
1162
|
identity: {
|
|
1163
1163
|
caseId: string;
|
|
1164
1164
|
seed: string;
|
|
1165
1165
|
fixtureId: string | null;
|
|
1166
1166
|
};
|
|
1167
|
-
status: "cancelled" | "
|
|
1167
|
+
status: "cancelled" | "completed" | "failed" | "unavailable";
|
|
1168
1168
|
output: unknown;
|
|
1169
1169
|
error: {
|
|
1170
1170
|
code: string;
|
|
@@ -1177,8 +1177,9 @@ export declare function verifyExperimentEvidence(input: {
|
|
|
1177
1177
|
contentHash: string;
|
|
1178
1178
|
revision: string | number;
|
|
1179
1179
|
};
|
|
1180
|
-
status: "
|
|
1180
|
+
status: "failed" | "pending" | "scored" | "unavailable";
|
|
1181
1181
|
value: string | number | boolean | null;
|
|
1182
|
+
passed?: boolean | null | undefined;
|
|
1182
1183
|
reasoning: string | null;
|
|
1183
1184
|
evidenceRefs: {
|
|
1184
1185
|
id: string;
|
|
@@ -1186,7 +1187,6 @@ export declare function verifyExperimentEvidence(input: {
|
|
|
1186
1187
|
mediaType: string | null;
|
|
1187
1188
|
sizeBytes: number | null;
|
|
1188
1189
|
}[];
|
|
1189
|
-
passed?: boolean | null | undefined;
|
|
1190
1190
|
}[];
|
|
1191
1191
|
usage: {
|
|
1192
1192
|
inputTokens: number | null;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"experiments.d.ts","sourceRoot":"","sources":["
|
|
1
|
+
{"version":3,"file":"experiments.d.ts","sourceRoot":"","sources":["../../src/experiments.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAaxB,OAAO,EACL,uCAAuC,EACvC,sCAAsC,EACtC,gCAAgC,EAChC,gCAAgC,EAChC,sCAAsC,EACtC,qCAAqC,EACrC,KAAK,iCAAiC,GACvC,MAAM,gCAAgC,CAAC;AACxC,OAAO,EACL,qCAAqC,EACrC,2CAA2C,EAC3C,oCAAoC,EACpC,oCAAoC,EACpC,oCAAoC,EACpC,KAAK,+BAA+B,EACpC,KAAK,8BAA8B,GACpC,MAAM,wCAAwC,CAAC;AAIhD,eAAO,MAAM,sBAAsB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;4BAiBjC,CAAC;AAEH,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;kBAepC,CAAC;AAEH,eAAO,MAAM,4BAA4B;;;;kBAI9B,CAAC;AACZ,MAAM,MAAM,sBAAsB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,4BAA4B,CAAC,CAAC;AAClF,wBAAgB,iBAAiB,CAAC,QAAQ,EAAE,sBAAsB,GAAG,MAAM,CAG1E;AAED,wBAAgB,qBAAqB,CAAC,gBAAgB,EAAE;IAAC,EAAE,EAAC,MAAM,CAAC;IAAA,WAAW,EAAC,MAAM,CAAA;CAAC,EAAE,aAAa,EAAC,MAAM,EAAE,SAAS,EAAC,MAAM,GAAG,MAAM,CAEtI;AAED,eAAO,MAAM,+BAA+B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBA8C1C,CAAC;AACH,eAAO,MAAM,wBAAwB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAAsF,CAAC;AAC5H,MAAM,MAAM,kBAAkB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,wBAAwB,CAAC,CAAC;AAE1E,eAAO,MAAM,wBAAwB;;;;;;;;;;;;;;;;;;;;;;kBAQ1B,CAAC;AAEZ,eAAO,MAAM,qBAAqB;;;;;;kBAUhC,CAAC;AAEH,eAAO,MAAM,0BAA0B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAU5B,CAAC;AAEZ,eAAO,MAAM,6BAA6B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAM/B,CAAC;AACZ,eAAO,MAAM,sBAAsB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAAoF,CAAC;AACxH,MAAM,MAAM,gBAAgB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,sBAAsB,CAAC,CAAC;AAEtE,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,+BAA+B,CAAC,GAAG,kBAAkB,CAGnH;AAED,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,6BAA6B,CAAC,EAAE,QAAQ,EAAE,kBAAkB,GAAG,gBAAgB,CAgC3I;AAED,MAAM,MAAM,oBAAoB,GAAG;IACjC,UAAU,EAAE,OAAO,CAAC;IACpB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,KAAK,CAAC;QAAE,QAAQ,EAAE,sBAAsB,CAAC;QAAC,QAAQ,EAAE,oBAAoB,GAAG,IAAI,CAAC;QAAC,SAAS,EAAE,oBAAoB,GAAG,IAAI,CAAC;QAC7H,QAAQ,EAAE,4BAA4B,EAAE,CAAA;KAAE,CAAC,CAAC;IAC9C,OAAO,EAAE,0BAA0B,EAAE,CAAC;CACvC,CAAC;AACF,MAAM,MAAM,oBAAoB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,0BAA0B,CAAC,CAAC;AAC9E,MAAM,MAAM,4BAA4B,GAAG;IACzC,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,OAAO,CAAC;IAClB,MAAM,EAAE,0BAA0B,GAAG,cAAc,GAAG,oBAAoB,GAAG,sBAAsB,GAAG,sBAAsB,GAAG,IAAI,CAAC;IACpI,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,cAAc,EAAE,OAAO,GAAG,IAAI,CAAC;IAC/B,eAAe,EAAE,OAAO,GAAG,IAAI,CAAC;CACjC,CAAC;AACF,MAAM,MAAM,0BAA0B,GAAG;IACvC,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,EAAE,MAAM,CAAC;IACtB,aAAa,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,gBAAgB,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,iBAAiB,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;CAC9B,CAAC;AAEF,wBAAgB,wBAAwB,CAAC,KAAK,EAAE;IAAE,QAAQ,EAAE,OAAO,CAAC;IAAC,MAAM,EAAE,OAAO,CAAA;CAAE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAOrF;AAED,iFAAiF;AACjF,wBAAgB,kBAAkB,CAChC,QAAQ,EAAE;IAAE,QAAQ,EAAE,kBAAkB,CAAC;IAAC,MAAM,EAAE,gBAAgB,CAAA;CAAE,EACpE,SAAS,EAAE;IAAE,QAAQ,EAAE,kBAAkB,CAAC;IAAC,MAAM,EAAE,gBAAgB,CAAA;CAAE,GACpE,oBAAoB,CA+EtB"}
|
package/dist/types/graders.d.ts
CHANGED
|
@@ -5,12 +5,12 @@ export declare const ModelJudgeTransportPolicySchema: z.ZodObject<{
|
|
|
5
5
|
schemaVersion: z.ZodLiteral<"openpond.modelJudgeTransportPolicy.v1">;
|
|
6
6
|
responseFormat: z.ZodLiteral<"judgment_json_schema">;
|
|
7
7
|
reasoningEffort: z.ZodEnum<{
|
|
8
|
-
|
|
8
|
+
high: "high";
|
|
9
9
|
low: "low";
|
|
10
|
+
max: "max";
|
|
10
11
|
medium: "medium";
|
|
11
|
-
|
|
12
|
+
off: "off";
|
|
12
13
|
xhigh: "xhigh";
|
|
13
|
-
max: "max";
|
|
14
14
|
}>;
|
|
15
15
|
maximumOutputTokens: z.ZodNumber;
|
|
16
16
|
}, z.core.$strict>;
|
|
@@ -30,12 +30,12 @@ export declare const ModelJudgeReceiptSchema: z.ZodObject<{
|
|
|
30
30
|
schemaVersion: z.ZodLiteral<"openpond.modelJudgeTransportPolicy.v1">;
|
|
31
31
|
responseFormat: z.ZodLiteral<"judgment_json_schema">;
|
|
32
32
|
reasoningEffort: z.ZodEnum<{
|
|
33
|
-
|
|
33
|
+
high: "high";
|
|
34
34
|
low: "low";
|
|
35
|
+
max: "max";
|
|
35
36
|
medium: "medium";
|
|
36
|
-
|
|
37
|
+
off: "off";
|
|
37
38
|
xhigh: "xhigh";
|
|
38
|
-
max: "max";
|
|
39
39
|
}>;
|
|
40
40
|
maximumOutputTokens: z.ZodNumber;
|
|
41
41
|
}, z.core.$strict>>;
|
|
@@ -48,8 +48,8 @@ export declare const GraderEvidenceContentSchema: z.ZodObject<{
|
|
|
48
48
|
graderId: z.ZodString;
|
|
49
49
|
graderVersion: z.ZodString;
|
|
50
50
|
status: z.ZodOptional<z.ZodEnum<{
|
|
51
|
-
pending: "pending";
|
|
52
51
|
failed: "failed";
|
|
52
|
+
pending: "pending";
|
|
53
53
|
scored: "scored";
|
|
54
54
|
unavailable: "unavailable";
|
|
55
55
|
}>>;
|
|
@@ -57,12 +57,12 @@ export declare const GraderEvidenceContentSchema: z.ZodObject<{
|
|
|
57
57
|
passed: z.ZodBoolean;
|
|
58
58
|
rewardEligible: z.ZodBoolean;
|
|
59
59
|
failureClass: z.ZodNullable<z.ZodEnum<{
|
|
60
|
-
|
|
61
|
-
grader_failure: "grader_failure";
|
|
60
|
+
cancelled: "cancelled";
|
|
62
61
|
environment_failure: "environment_failure";
|
|
62
|
+
grader_failure: "grader_failure";
|
|
63
63
|
infrastructure_failure: "infrastructure_failure";
|
|
64
|
+
policy_failure: "policy_failure";
|
|
64
65
|
timeout: "timeout";
|
|
65
|
-
cancelled: "cancelled";
|
|
66
66
|
}>>;
|
|
67
67
|
feedback: z.ZodArray<z.ZodString>;
|
|
68
68
|
visibleEvidenceRefs: z.ZodArray<z.ZodString>;
|
|
@@ -82,12 +82,12 @@ export declare const GraderEvidenceContentSchema: z.ZodObject<{
|
|
|
82
82
|
schemaVersion: z.ZodLiteral<"openpond.modelJudgeTransportPolicy.v1">;
|
|
83
83
|
responseFormat: z.ZodLiteral<"judgment_json_schema">;
|
|
84
84
|
reasoningEffort: z.ZodEnum<{
|
|
85
|
-
|
|
85
|
+
high: "high";
|
|
86
86
|
low: "low";
|
|
87
|
+
max: "max";
|
|
87
88
|
medium: "medium";
|
|
88
|
-
|
|
89
|
+
off: "off";
|
|
89
90
|
xhigh: "xhigh";
|
|
90
|
-
max: "max";
|
|
91
91
|
}>;
|
|
92
92
|
maximumOutputTokens: z.ZodNumber;
|
|
93
93
|
}, z.core.$strict>>;
|
|
@@ -101,8 +101,8 @@ export declare const GraderEvidenceSchema: z.ZodObject<{
|
|
|
101
101
|
graderId: z.ZodString;
|
|
102
102
|
graderVersion: z.ZodString;
|
|
103
103
|
status: z.ZodOptional<z.ZodEnum<{
|
|
104
|
-
pending: "pending";
|
|
105
104
|
failed: "failed";
|
|
105
|
+
pending: "pending";
|
|
106
106
|
scored: "scored";
|
|
107
107
|
unavailable: "unavailable";
|
|
108
108
|
}>>;
|
|
@@ -110,12 +110,12 @@ export declare const GraderEvidenceSchema: z.ZodObject<{
|
|
|
110
110
|
passed: z.ZodBoolean;
|
|
111
111
|
rewardEligible: z.ZodBoolean;
|
|
112
112
|
failureClass: z.ZodNullable<z.ZodEnum<{
|
|
113
|
-
|
|
114
|
-
grader_failure: "grader_failure";
|
|
113
|
+
cancelled: "cancelled";
|
|
115
114
|
environment_failure: "environment_failure";
|
|
115
|
+
grader_failure: "grader_failure";
|
|
116
116
|
infrastructure_failure: "infrastructure_failure";
|
|
117
|
+
policy_failure: "policy_failure";
|
|
117
118
|
timeout: "timeout";
|
|
118
|
-
cancelled: "cancelled";
|
|
119
119
|
}>>;
|
|
120
120
|
feedback: z.ZodArray<z.ZodString>;
|
|
121
121
|
visibleEvidenceRefs: z.ZodArray<z.ZodString>;
|
|
@@ -135,12 +135,12 @@ export declare const GraderEvidenceSchema: z.ZodObject<{
|
|
|
135
135
|
schemaVersion: z.ZodLiteral<"openpond.modelJudgeTransportPolicy.v1">;
|
|
136
136
|
responseFormat: z.ZodLiteral<"judgment_json_schema">;
|
|
137
137
|
reasoningEffort: z.ZodEnum<{
|
|
138
|
-
|
|
138
|
+
high: "high";
|
|
139
139
|
low: "low";
|
|
140
|
+
max: "max";
|
|
140
141
|
medium: "medium";
|
|
141
|
-
|
|
142
|
+
off: "off";
|
|
142
143
|
xhigh: "xhigh";
|
|
143
|
-
max: "max";
|
|
144
144
|
}>;
|
|
145
145
|
maximumOutputTokens: z.ZodNumber;
|
|
146
146
|
}, z.core.$strict>>;
|
|
@@ -192,18 +192,18 @@ declare const GradeSummarySchema: z.ZodObject<{
|
|
|
192
192
|
passed: z.ZodBoolean;
|
|
193
193
|
rewardEligible: z.ZodBoolean;
|
|
194
194
|
failureClass: z.ZodNullable<z.ZodEnum<{
|
|
195
|
-
|
|
196
|
-
grader_failure: "grader_failure";
|
|
195
|
+
cancelled: "cancelled";
|
|
197
196
|
environment_failure: "environment_failure";
|
|
197
|
+
grader_failure: "grader_failure";
|
|
198
198
|
infrastructure_failure: "infrastructure_failure";
|
|
199
|
+
policy_failure: "policy_failure";
|
|
199
200
|
timeout: "timeout";
|
|
200
|
-
cancelled: "cancelled";
|
|
201
201
|
}>>;
|
|
202
202
|
gradingStatus: z.ZodEnum<{
|
|
203
|
+
not_configured: "not_configured";
|
|
203
204
|
pending: "pending";
|
|
204
205
|
scored: "scored";
|
|
205
206
|
unscorable: "unscorable";
|
|
206
|
-
not_configured: "not_configured";
|
|
207
207
|
}>;
|
|
208
208
|
}, z.core.$strict>;
|
|
209
209
|
/** Shared ordinary grading policy. Results must match the declared population;
|
|
@@ -218,18 +218,18 @@ export declare const TaskGradeSchema: z.ZodObject<{
|
|
|
218
218
|
passed: z.ZodBoolean;
|
|
219
219
|
rewardEligible: z.ZodBoolean;
|
|
220
220
|
failureClass: z.ZodNullable<z.ZodEnum<{
|
|
221
|
-
|
|
222
|
-
grader_failure: "grader_failure";
|
|
221
|
+
cancelled: "cancelled";
|
|
223
222
|
environment_failure: "environment_failure";
|
|
223
|
+
grader_failure: "grader_failure";
|
|
224
224
|
infrastructure_failure: "infrastructure_failure";
|
|
225
|
+
policy_failure: "policy_failure";
|
|
225
226
|
timeout: "timeout";
|
|
226
|
-
cancelled: "cancelled";
|
|
227
227
|
}>>;
|
|
228
228
|
gradingStatus: z.ZodEnum<{
|
|
229
|
+
not_configured: "not_configured";
|
|
229
230
|
pending: "pending";
|
|
230
231
|
scored: "scored";
|
|
231
232
|
unscorable: "unscorable";
|
|
232
|
-
not_configured: "not_configured";
|
|
233
233
|
}>;
|
|
234
234
|
schemaVersion: z.ZodLiteral<"openpond.taskGrade.v1">;
|
|
235
235
|
taskHash: z.ZodString;
|
|
@@ -240,8 +240,8 @@ export declare const TaskGradeSchema: z.ZodObject<{
|
|
|
240
240
|
graderId: z.ZodString;
|
|
241
241
|
graderVersion: z.ZodString;
|
|
242
242
|
status: z.ZodOptional<z.ZodEnum<{
|
|
243
|
-
pending: "pending";
|
|
244
243
|
failed: "failed";
|
|
244
|
+
pending: "pending";
|
|
245
245
|
scored: "scored";
|
|
246
246
|
unavailable: "unavailable";
|
|
247
247
|
}>>;
|
|
@@ -249,12 +249,12 @@ export declare const TaskGradeSchema: z.ZodObject<{
|
|
|
249
249
|
passed: z.ZodBoolean;
|
|
250
250
|
rewardEligible: z.ZodBoolean;
|
|
251
251
|
failureClass: z.ZodNullable<z.ZodEnum<{
|
|
252
|
-
|
|
253
|
-
grader_failure: "grader_failure";
|
|
252
|
+
cancelled: "cancelled";
|
|
254
253
|
environment_failure: "environment_failure";
|
|
254
|
+
grader_failure: "grader_failure";
|
|
255
255
|
infrastructure_failure: "infrastructure_failure";
|
|
256
|
+
policy_failure: "policy_failure";
|
|
256
257
|
timeout: "timeout";
|
|
257
|
-
cancelled: "cancelled";
|
|
258
258
|
}>>;
|
|
259
259
|
feedback: z.ZodArray<z.ZodString>;
|
|
260
260
|
visibleEvidenceRefs: z.ZodArray<z.ZodString>;
|
|
@@ -274,12 +274,12 @@ export declare const TaskGradeSchema: z.ZodObject<{
|
|
|
274
274
|
schemaVersion: z.ZodLiteral<"openpond.modelJudgeTransportPolicy.v1">;
|
|
275
275
|
responseFormat: z.ZodLiteral<"judgment_json_schema">;
|
|
276
276
|
reasoningEffort: z.ZodEnum<{
|
|
277
|
-
|
|
277
|
+
high: "high";
|
|
278
278
|
low: "low";
|
|
279
|
+
max: "max";
|
|
279
280
|
medium: "medium";
|
|
280
|
-
|
|
281
|
+
off: "off";
|
|
281
282
|
xhigh: "xhigh";
|
|
282
|
-
max: "max";
|
|
283
283
|
}>;
|
|
284
284
|
maximumOutputTokens: z.ZodNumber;
|
|
285
285
|
}, z.core.$strict>>;
|