@openpond/evals 0.11.16 → 0.11.17
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/connected-evidence/contracts.js +72 -0
- package/dist/connected-evidence/imports.js +71 -0
- package/dist/connected-evidence/index.js +5 -0
- package/dist/connected-evidence/normalize.js +97 -0
- package/dist/connected-evidence/python-correctness.js +73 -0
- package/dist/connected-evidence/sources/claude-code.js +71 -0
- package/dist/connected-evidence/sources/codex.js +52 -0
- package/dist/connected-evidence/sources/history.js +86 -0
- package/dist/connected-evidence/sources/usage.js +15 -0
- package/dist/connected-evidence/usage-summary.js +52 -0
- package/dist/experiments.js +16 -4
- package/dist/graders.js +9 -6
- package/dist/human-review/comparison.js +161 -0
- package/dist/human-review/contracts.js +110 -0
- package/dist/human-review/execution.js +61 -0
- package/dist/human-review/handoff.js +18 -0
- package/dist/human-review/index.js +9 -0
- package/dist/human-review/local-publication.js +18 -0
- package/dist/human-review/repository.js +1 -0
- package/dist/human-review/results.js +18 -0
- package/dist/human-review/service.js +268 -0
- package/dist/human-review/task-proposals.js +39 -0
- package/dist/index.js +1 -0
- package/dist/javascript-verifier-contract.js +2 -2
- package/dist/learning/authoring.js +2 -1
- package/dist/learning/contracts.js +353 -81
- package/dist/learning/index.js +1 -0
- package/dist/learning/reward-authoring.js +5 -2
- package/dist/learning/training-handoff.js +32 -0
- package/dist/metrics-node.js +2 -0
- package/dist/profile-evaluation-target.js +7 -0
- package/dist/profile-evaluations.js +7 -10
- package/dist/profile-external-evaluation.js +35 -0
- package/dist/profile-source-candidate.js +5 -0
- package/dist/rewards.js +5 -3
- package/dist/taskset-run-contract.js +10 -0
- package/dist/tasksets.js +2 -0
- package/dist/types/builtin-benchmarks/harness-refiner.d.ts +75 -1
- package/dist/types/builtin-benchmarks/harness-refiner.d.ts.map +1 -1
- package/dist/types/conformance.d.ts +150 -2
- package/dist/types/conformance.d.ts.map +1 -1
- package/dist/types/connected-evidence/contracts.d.ts +317 -0
- package/dist/types/connected-evidence/contracts.d.ts.map +1 -0
- package/dist/types/connected-evidence/imports.d.ts +19 -0
- package/dist/types/connected-evidence/imports.d.ts.map +1 -0
- package/dist/types/connected-evidence/index.d.ts +6 -0
- package/dist/types/connected-evidence/index.d.ts.map +1 -0
- package/dist/types/connected-evidence/normalize.d.ts +25 -0
- package/dist/types/connected-evidence/normalize.d.ts.map +1 -0
- package/dist/types/connected-evidence/python-correctness.d.ts +196 -0
- package/dist/types/connected-evidence/python-correctness.d.ts.map +1 -0
- package/dist/types/connected-evidence/sources/claude-code.d.ts +74 -0
- package/dist/types/connected-evidence/sources/claude-code.d.ts.map +1 -0
- package/dist/types/connected-evidence/sources/codex.d.ts +74 -0
- package/dist/types/connected-evidence/sources/codex.d.ts.map +1 -0
- package/dist/types/connected-evidence/sources/history.d.ts +146 -0
- package/dist/types/connected-evidence/sources/history.d.ts.map +1 -0
- package/dist/types/connected-evidence/sources/usage.d.ts +3 -0
- package/dist/types/connected-evidence/sources/usage.d.ts.map +1 -0
- package/dist/types/connected-evidence/usage-summary.d.ts +162 -0
- package/dist/types/connected-evidence/usage-summary.d.ts.map +1 -0
- package/dist/types/evidence/conformance.d.ts +28 -28
- package/dist/types/evidence/contracts.d.ts +27 -27
- package/dist/types/execution-contracts.d.ts +159 -5
- package/dist/types/execution-contracts.d.ts.map +1 -1
- package/dist/types/experiments.d.ts +476 -7
- package/dist/types/experiments.d.ts.map +1 -1
- package/dist/types/graders.d.ts +21 -1
- package/dist/types/graders.d.ts.map +1 -1
- package/dist/types/human-review/comparison.d.ts +1959 -0
- package/dist/types/human-review/comparison.d.ts.map +1 -0
- package/dist/types/human-review/contracts.d.ts +2719 -0
- package/dist/types/human-review/contracts.d.ts.map +1 -0
- package/dist/types/human-review/execution.d.ts +298 -0
- package/dist/types/human-review/execution.d.ts.map +1 -0
- package/dist/types/human-review/handoff.d.ts +124 -0
- package/dist/types/human-review/handoff.d.ts.map +1 -0
- package/dist/types/human-review/index.d.ts +10 -0
- package/dist/types/human-review/index.d.ts.map +1 -0
- package/dist/types/human-review/local-publication.d.ts +165 -0
- package/dist/types/human-review/local-publication.d.ts.map +1 -0
- package/dist/types/human-review/repository.d.ts +43 -0
- package/dist/types/human-review/repository.d.ts.map +1 -0
- package/dist/types/human-review/results.d.ts +118 -0
- package/dist/types/human-review/results.d.ts.map +1 -0
- package/dist/types/human-review/service.d.ts +878 -0
- package/dist/types/human-review/service.d.ts.map +1 -0
- package/dist/types/human-review/task-proposals.d.ts +70 -0
- package/dist/types/human-review/task-proposals.d.ts.map +1 -0
- package/dist/types/index.d.ts +1 -0
- package/dist/types/index.d.ts.map +1 -1
- package/dist/types/javascript-environment-contract.d.ts +1 -1
- package/dist/types/javascript-verifier-contract.d.ts +1 -1
- package/dist/types/javascript-verifier-contract.d.ts.map +1 -1
- package/dist/types/learned-preference.d.ts +2 -2
- package/dist/types/learning/acceptance-group.d.ts +13 -13
- package/dist/types/learning/acceptance-plan.d.ts +1 -1
- package/dist/types/learning/admission.d.ts +78 -1
- package/dist/types/learning/admission.d.ts.map +1 -1
- package/dist/types/learning/authoring-service.d.ts +2 -0
- package/dist/types/learning/authoring-service.d.ts.map +1 -1
- package/dist/types/learning/authoring.d.ts +20 -0
- package/dist/types/learning/authoring.d.ts.map +1 -1
- package/dist/types/learning/batch-service.d.ts +76 -2
- package/dist/types/learning/batch-service.d.ts.map +1 -1
- package/dist/types/learning/contracts.d.ts +759 -24
- package/dist/types/learning/contracts.d.ts.map +1 -1
- package/dist/types/learning/index.d.ts +1 -0
- package/dist/types/learning/index.d.ts.map +1 -1
- package/dist/types/learning/iteration-candidate-decision.d.ts +2 -2
- package/dist/types/learning/iteration-dispatch-contracts.d.ts +2 -2
- package/dist/types/learning/operations.d.ts +1118 -8
- package/dist/types/learning/operations.d.ts.map +1 -1
- package/dist/types/learning/repository.d.ts +402 -15
- 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.d.ts +74 -0
- package/dist/types/learning/reward-calibration.d.ts.map +1 -1
- package/dist/types/learning/reward-checks.d.ts +14 -2
- package/dist/types/learning/reward-checks.d.ts.map +1 -1
- package/dist/types/learning/training-handoff.d.ts +201 -0
- package/dist/types/learning/training-handoff.d.ts.map +1 -0
- package/dist/types/learning/transport.d.ts +963 -21
- package/dist/types/learning/transport.d.ts.map +1 -1
- package/dist/types/metrics-node.d.ts +3 -1
- package/dist/types/metrics-node.d.ts.map +1 -1
- package/dist/types/model-improvement-qualification.d.ts +3 -3
- package/dist/types/native-harness-experiment-attempt.d.ts +2 -2
- package/dist/types/profile-evaluation-comparisons.d.ts +230 -0
- package/dist/types/profile-evaluation-comparisons.d.ts.map +1 -1
- package/dist/types/profile-evaluation-target.d.ts +14 -0
- package/dist/types/profile-evaluation-target.d.ts.map +1 -0
- package/dist/types/profile-evaluations.d.ts +117 -14
- package/dist/types/profile-evaluations.d.ts.map +1 -1
- package/dist/types/profile-external-evaluation.d.ts +404 -0
- package/dist/types/profile-external-evaluation.d.ts.map +1 -0
- package/dist/types/profile-source-candidate.d.ts +15 -0
- package/dist/types/profile-source-candidate.d.ts.map +1 -0
- package/dist/types/review-conformance.d.ts +7 -7
- package/dist/types/rewards.d.ts +253 -0
- package/dist/types/rewards.d.ts.map +1 -1
- package/dist/types/rollouts.d.ts +3 -3
- package/dist/types/taskset-run-contract.d.ts +230 -0
- package/dist/types/taskset-run-contract.d.ts.map +1 -1
- package/dist/types/tasksets.d.ts +311 -3
- package/dist/types/tasksets.d.ts.map +1 -1
- package/dist/types/telemetry-analysis.d.ts +2 -2
- package/dist/types/telemetry-bundle.d.ts +6 -6
- package/dist/types/telemetry.d.ts +3 -3
- package/package.json +9 -1
- package/schemas/learning/v1/command-request.schema.json +1745 -409
- package/schemas/learning/v1/decision.schema.json +11 -0
- package/schemas/learning/v1/draft.schema.json +11 -0
- package/schemas/learning/v1/grade.schema.json +11 -0
- package/schemas/learning/v1/package.schema.json +532 -12
- package/schemas/learning/v1/policy.schema.json +836 -28
- package/schemas/learning/v1/reward.schema.json +520 -0
- package/schemas/learning/v1/reward_check.schema.json +9 -0
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"transport.d.ts","sourceRoot":"","sources":["../../../../../src/learning/transport.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAQxB,mFAAmF;AACnF,wBAAgB,yBAAyB,CAAC,GAAG,EAAE,OAAO,GAAG,IAAI,CAG5D;AAED,0FAA0F;AAC1F,eAAO,MAAM,4BAA4B
|
|
1
|
+
{"version":3,"file":"transport.d.ts","sourceRoot":"","sources":["../../../../../src/learning/transport.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAQxB,mFAAmF;AACnF,wBAAgB,yBAAyB,CAAC,GAAG,EAAE,OAAO,GAAG,IAAI,CAG5D;AAED,0FAA0F;AAC1F,eAAO,MAAM,4BAA4B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAAgF,CAAC;AAC1H,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;8BAMpC,CAAC;AACH,eAAO,MAAM,kCAAkC;;;;;;;;;;;;;;;;;;;;;;;;kBAOpC,CAAC;AACZ,MAAM,MAAM,4BAA4B,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kCAAkC,CAAC,CAAC;AAC9F,eAAO,MAAM,6BAA6B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAG/B,CAAC"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { type TasksetEvaluationInput, type TasksetRunEvaluationInput, type TasksetMetricExecutionInput } from "./metrics.js";
|
|
1
|
+
import { type TasksetEvaluationInput, type TasksetRunEvaluationInput, type TasksetMetricExecutionInput, type TasksetMetricExecutor } from "./metrics.js";
|
|
2
2
|
import type { TasksetMetricResult } from "./metric-policy.js";
|
|
3
3
|
import type { EvaluationResult } from "./runs.js";
|
|
4
4
|
/** Verifies original source before stripping TS, then waits for worker termination
|
|
@@ -10,4 +10,6 @@ export declare function aggregateTasksetRunInWorker(input: TasksetRunEvaluationI
|
|
|
10
10
|
* execution. Type stripping happens in that process within its deadline. */
|
|
11
11
|
export declare function executeTasksetMetricInProcess(input: TasksetMetricExecutionInput): Promise<TasksetMetricResult>;
|
|
12
12
|
export declare function aggregateTasksetRunInProcess(input: TasksetRunEvaluationInput): Promise<TasksetMetricResult>;
|
|
13
|
+
/** Trusted host port for a frozen private metric; callers retain source and receipt ownership. */
|
|
14
|
+
export declare const executeTasksetMetricWorker: TasksetMetricExecutor;
|
|
13
15
|
//# sourceMappingURL=metrics-node.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"metrics-node.d.ts","sourceRoot":"","sources":["../../../../src/metrics-node.ts"],"names":[],"mappings":"AAEA,OAAO,EAAyF,KAAK,sBAAsB,EAAE,KAAK,yBAAyB,EAAE,KAAK,2BAA2B,
|
|
1
|
+
{"version":3,"file":"metrics-node.d.ts","sourceRoot":"","sources":["../../../../src/metrics-node.ts"],"names":[],"mappings":"AAEA,OAAO,EAAyF,KAAK,sBAAsB,EAAE,KAAK,yBAAyB,EAAE,KAAK,2BAA2B,EAAE,KAAK,qBAAqB,EAAE,MAAM,cAAc,CAAC;AAChP,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AAC9D,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,WAAW,CAAC;AAElD;wFACwF;AACxF,wBAAgB,4BAA4B,CAAC,KAAK,EAAE,2BAA2B,GAAG,OAAO,CAAC,mBAAmB,CAAC,CAE7G;AAED,wBAAgB,kCAAkC,CAAC,KAAK,EAAE,sBAAsB,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAE3G;AAED,wBAAgB,2BAA2B,CAAC,KAAK,EAAE,yBAAyB,GAAG,OAAO,CAAC,mBAAmB,CAAC,CAE1G;AAED;4EAC4E;AAC5E,wBAAgB,6BAA6B,CAAC,KAAK,EAAE,2BAA2B,GAAG,OAAO,CAAC,mBAAmB,CAAC,CAE9G;AAED,wBAAgB,4BAA4B,CAAC,KAAK,EAAE,yBAAyB,GAAG,OAAO,CAAC,mBAAmB,CAAC,CAE3G;AAgBD,kGAAkG;AAClG,eAAO,MAAM,0BAA0B,EAAE,qBAAqC,CAAC"}
|
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
import { z } from "zod";
|
|
2
2
|
export declare const ModelImprovementDecisionSchema: z.ZodEnum<{
|
|
3
|
+
preference: "preference";
|
|
3
4
|
no_training: "no_training";
|
|
4
5
|
sft: "sft";
|
|
5
|
-
preference: "preference";
|
|
6
6
|
rl: "rl";
|
|
7
7
|
}>;
|
|
8
8
|
export declare const ModelImprovementSignalSchema: z.ZodObject<{
|
|
@@ -111,9 +111,9 @@ export declare const ModelImprovementQualificationReceiptContentSchema: z.ZodObj
|
|
|
111
111
|
}, z.core.$strict>>;
|
|
112
112
|
}, z.core.$strict>;
|
|
113
113
|
decision: z.ZodEnum<{
|
|
114
|
+
preference: "preference";
|
|
114
115
|
no_training: "no_training";
|
|
115
116
|
sft: "sft";
|
|
116
|
-
preference: "preference";
|
|
117
117
|
rl: "rl";
|
|
118
118
|
}>;
|
|
119
119
|
reasons: z.ZodArray<z.ZodString>;
|
|
@@ -206,9 +206,9 @@ export declare const ModelImprovementQualificationReceiptSchema: z.ZodObject<{
|
|
|
206
206
|
}, z.core.$strict>>;
|
|
207
207
|
}, z.core.$strict>;
|
|
208
208
|
decision: z.ZodEnum<{
|
|
209
|
+
preference: "preference";
|
|
209
210
|
no_training: "no_training";
|
|
210
211
|
sft: "sft";
|
|
211
|
-
preference: "preference";
|
|
212
212
|
rl: "rl";
|
|
213
213
|
}>;
|
|
214
214
|
reasons: z.ZodArray<z.ZodString>;
|
|
@@ -33,9 +33,9 @@ export declare const NativeHarnessExperimentAttemptContentSchema: z.ZodObject<{
|
|
|
33
33
|
schemaVersion: z.ZodLiteral<"openpond.nativeHarnessExperimentAttempt.v1">;
|
|
34
34
|
taskId: z.ZodString;
|
|
35
35
|
status: z.ZodEnum<{
|
|
36
|
+
cancelled: "cancelled";
|
|
36
37
|
completed: "completed";
|
|
37
38
|
policy_failure: "policy_failure";
|
|
38
|
-
cancelled: "cancelled";
|
|
39
39
|
timed_out: "timed_out";
|
|
40
40
|
}>;
|
|
41
41
|
output: z.ZodNullable<z.ZodString>;
|
|
@@ -72,9 +72,9 @@ export declare const NativeHarnessExperimentAttemptSchema: z.ZodObject<{
|
|
|
72
72
|
schemaVersion: z.ZodLiteral<"openpond.nativeHarnessExperimentAttempt.v1">;
|
|
73
73
|
taskId: z.ZodString;
|
|
74
74
|
status: z.ZodEnum<{
|
|
75
|
+
cancelled: "cancelled";
|
|
75
76
|
completed: "completed";
|
|
76
77
|
policy_failure: "policy_failure";
|
|
77
|
-
cancelled: "cancelled";
|
|
78
78
|
timed_out: "timed_out";
|
|
79
79
|
}>;
|
|
80
80
|
output: z.ZodNullable<z.ZodString>;
|
|
@@ -41,6 +41,121 @@ export declare const ProfileEvaluationComparisonContentSchema: z.ZodObject<{
|
|
|
41
41
|
actionId: z.ZodString;
|
|
42
42
|
}, z.core.$strict>], "kind">;
|
|
43
43
|
environmentHash: z.ZodString;
|
|
44
|
+
externalDatasetBinding: z.ZodOptional<z.ZodObject<{
|
|
45
|
+
schemaVersion: z.ZodLiteral<"openpond.profileExternalDatasetBinding.v1">;
|
|
46
|
+
dataset: z.ZodObject<{
|
|
47
|
+
id: z.ZodString;
|
|
48
|
+
contentHash: z.ZodString;
|
|
49
|
+
revision: z.ZodNumber;
|
|
50
|
+
}, z.core.$strict>;
|
|
51
|
+
packageHash: z.ZodString;
|
|
52
|
+
profileId: z.ZodString;
|
|
53
|
+
target: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
54
|
+
kind: z.ZodLiteral<"profile">;
|
|
55
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
56
|
+
kind: z.ZodLiteral<"workflow">;
|
|
57
|
+
workflowId: z.ZodString;
|
|
58
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
59
|
+
kind: z.ZodLiteral<"skill">;
|
|
60
|
+
skillPath: z.ZodString;
|
|
61
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
62
|
+
kind: z.ZodLiteral<"agent_action">;
|
|
63
|
+
actionId: z.ZodString;
|
|
64
|
+
}, z.core.$strict>], "kind">;
|
|
65
|
+
declaredProfileCatalogHash: z.ZodNullable<z.ZodString>;
|
|
66
|
+
protectedProfileClosureHash: z.ZodString;
|
|
67
|
+
recordedOrigin: z.ZodOptional<z.ZodObject<{
|
|
68
|
+
execution: z.ZodObject<{
|
|
69
|
+
id: z.ZodString;
|
|
70
|
+
contentHash: z.ZodString;
|
|
71
|
+
}, z.core.$strict>;
|
|
72
|
+
originalEvidence: z.ZodObject<{
|
|
73
|
+
id: z.ZodString;
|
|
74
|
+
contentHash: z.ZodString;
|
|
75
|
+
}, z.core.$strict>;
|
|
76
|
+
members: z.ZodArray<z.ZodObject<{
|
|
77
|
+
caseId: z.ZodString;
|
|
78
|
+
seed: z.ZodString;
|
|
79
|
+
fixtureId: z.ZodString;
|
|
80
|
+
sourceId: z.ZodString;
|
|
81
|
+
snapshotHash: z.ZodString;
|
|
82
|
+
boundaryRevisionHash: z.ZodString;
|
|
83
|
+
}, z.core.$strict>>;
|
|
84
|
+
}, z.core.$strict>>;
|
|
85
|
+
gradingSource: z.ZodOptional<z.ZodObject<{
|
|
86
|
+
pass: z.ZodObject<{
|
|
87
|
+
id: z.ZodString;
|
|
88
|
+
contentHash: z.ZodString;
|
|
89
|
+
}, z.core.$strict>;
|
|
90
|
+
evidence: z.ZodObject<{
|
|
91
|
+
id: z.ZodString;
|
|
92
|
+
contentHash: z.ZodString;
|
|
93
|
+
}, z.core.$strict>;
|
|
94
|
+
}, z.core.$strict>>;
|
|
95
|
+
split: z.ZodEnum<{
|
|
96
|
+
train: "train";
|
|
97
|
+
validation: "validation";
|
|
98
|
+
test: "test";
|
|
99
|
+
frozen_eval: "frozen_eval";
|
|
100
|
+
}>;
|
|
101
|
+
population: z.ZodArray<z.ZodObject<{
|
|
102
|
+
taskId: z.ZodString;
|
|
103
|
+
seed: z.ZodString;
|
|
104
|
+
fixtureId: z.ZodNullable<z.ZodString>;
|
|
105
|
+
}, z.core.$strict>>;
|
|
106
|
+
evaluators: z.ZodArray<z.ZodObject<{
|
|
107
|
+
release: z.ZodObject<{
|
|
108
|
+
id: z.ZodString;
|
|
109
|
+
contentHash: z.ZodString;
|
|
110
|
+
revision: z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>;
|
|
111
|
+
}, z.core.$strict>;
|
|
112
|
+
feedbackKey: z.ZodString;
|
|
113
|
+
configurationHash: z.ZodString;
|
|
114
|
+
output: z.ZodEnum<{
|
|
115
|
+
boolean: "boolean";
|
|
116
|
+
score: "score";
|
|
117
|
+
category: "category";
|
|
118
|
+
}>;
|
|
119
|
+
categories: z.ZodArray<z.ZodString>;
|
|
120
|
+
}, z.core.$strict>>;
|
|
121
|
+
fieldMappingsHash: z.ZodString;
|
|
122
|
+
fieldMappings: z.ZodArray<z.ZodObject<{
|
|
123
|
+
graderId: z.ZodString;
|
|
124
|
+
fields: z.ZodArray<z.ZodObject<{
|
|
125
|
+
destination: z.ZodEnum<{
|
|
126
|
+
output: "output";
|
|
127
|
+
input: "input";
|
|
128
|
+
expectedOutput: "expectedOutput";
|
|
129
|
+
evaluatorContext: "evaluatorContext";
|
|
130
|
+
}>;
|
|
131
|
+
path: z.ZodArray<z.ZodString>;
|
|
132
|
+
source: z.ZodEnum<{
|
|
133
|
+
output: "output";
|
|
134
|
+
input: "input";
|
|
135
|
+
expectedOutput: "expectedOutput";
|
|
136
|
+
evaluatorContext: "evaluatorContext";
|
|
137
|
+
}>;
|
|
138
|
+
sourcePath: z.ZodArray<z.ZodString>;
|
|
139
|
+
}, z.core.$strict>>;
|
|
140
|
+
}, z.core.$strict>>;
|
|
141
|
+
environmentHash: z.ZodString;
|
|
142
|
+
privateDatasetClosureHash: z.ZodString;
|
|
143
|
+
criterion: z.ZodObject<{
|
|
144
|
+
minimumPassRate: z.ZodNumber;
|
|
145
|
+
requireComplete: z.ZodBoolean;
|
|
146
|
+
}, z.core.$strict>;
|
|
147
|
+
contentHash: z.ZodString;
|
|
148
|
+
}, z.core.$strict>>;
|
|
149
|
+
sourceCandidate: z.ZodOptional<z.ZodObject<{
|
|
150
|
+
candidateId: z.ZodString;
|
|
151
|
+
frozenRevision: z.ZodNumber;
|
|
152
|
+
frozenHash: z.ZodString;
|
|
153
|
+
harnessRelease: z.ZodObject<{
|
|
154
|
+
id: z.ZodString;
|
|
155
|
+
contentHash: z.ZodString;
|
|
156
|
+
}, z.core.$strict>;
|
|
157
|
+
protectedClosureHash: z.ZodString;
|
|
158
|
+
}, z.core.$strict>>;
|
|
44
159
|
}, z.core.$strict>;
|
|
45
160
|
policy: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
46
161
|
kind: z.ZodLiteral<"model">;
|
|
@@ -100,6 +215,121 @@ export declare const ProfileEvaluationComparisonSchema: z.ZodObject<{
|
|
|
100
215
|
actionId: z.ZodString;
|
|
101
216
|
}, z.core.$strict>], "kind">;
|
|
102
217
|
environmentHash: z.ZodString;
|
|
218
|
+
externalDatasetBinding: z.ZodOptional<z.ZodObject<{
|
|
219
|
+
schemaVersion: z.ZodLiteral<"openpond.profileExternalDatasetBinding.v1">;
|
|
220
|
+
dataset: z.ZodObject<{
|
|
221
|
+
id: z.ZodString;
|
|
222
|
+
contentHash: z.ZodString;
|
|
223
|
+
revision: z.ZodNumber;
|
|
224
|
+
}, z.core.$strict>;
|
|
225
|
+
packageHash: z.ZodString;
|
|
226
|
+
profileId: z.ZodString;
|
|
227
|
+
target: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
228
|
+
kind: z.ZodLiteral<"profile">;
|
|
229
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
230
|
+
kind: z.ZodLiteral<"workflow">;
|
|
231
|
+
workflowId: z.ZodString;
|
|
232
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
233
|
+
kind: z.ZodLiteral<"skill">;
|
|
234
|
+
skillPath: z.ZodString;
|
|
235
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
236
|
+
kind: z.ZodLiteral<"agent_action">;
|
|
237
|
+
actionId: z.ZodString;
|
|
238
|
+
}, z.core.$strict>], "kind">;
|
|
239
|
+
declaredProfileCatalogHash: z.ZodNullable<z.ZodString>;
|
|
240
|
+
protectedProfileClosureHash: z.ZodString;
|
|
241
|
+
recordedOrigin: z.ZodOptional<z.ZodObject<{
|
|
242
|
+
execution: z.ZodObject<{
|
|
243
|
+
id: z.ZodString;
|
|
244
|
+
contentHash: z.ZodString;
|
|
245
|
+
}, z.core.$strict>;
|
|
246
|
+
originalEvidence: z.ZodObject<{
|
|
247
|
+
id: z.ZodString;
|
|
248
|
+
contentHash: z.ZodString;
|
|
249
|
+
}, z.core.$strict>;
|
|
250
|
+
members: z.ZodArray<z.ZodObject<{
|
|
251
|
+
caseId: z.ZodString;
|
|
252
|
+
seed: z.ZodString;
|
|
253
|
+
fixtureId: z.ZodString;
|
|
254
|
+
sourceId: z.ZodString;
|
|
255
|
+
snapshotHash: z.ZodString;
|
|
256
|
+
boundaryRevisionHash: z.ZodString;
|
|
257
|
+
}, z.core.$strict>>;
|
|
258
|
+
}, z.core.$strict>>;
|
|
259
|
+
gradingSource: z.ZodOptional<z.ZodObject<{
|
|
260
|
+
pass: z.ZodObject<{
|
|
261
|
+
id: z.ZodString;
|
|
262
|
+
contentHash: z.ZodString;
|
|
263
|
+
}, z.core.$strict>;
|
|
264
|
+
evidence: z.ZodObject<{
|
|
265
|
+
id: z.ZodString;
|
|
266
|
+
contentHash: z.ZodString;
|
|
267
|
+
}, z.core.$strict>;
|
|
268
|
+
}, z.core.$strict>>;
|
|
269
|
+
split: z.ZodEnum<{
|
|
270
|
+
train: "train";
|
|
271
|
+
validation: "validation";
|
|
272
|
+
test: "test";
|
|
273
|
+
frozen_eval: "frozen_eval";
|
|
274
|
+
}>;
|
|
275
|
+
population: z.ZodArray<z.ZodObject<{
|
|
276
|
+
taskId: z.ZodString;
|
|
277
|
+
seed: z.ZodString;
|
|
278
|
+
fixtureId: z.ZodNullable<z.ZodString>;
|
|
279
|
+
}, z.core.$strict>>;
|
|
280
|
+
evaluators: z.ZodArray<z.ZodObject<{
|
|
281
|
+
release: z.ZodObject<{
|
|
282
|
+
id: z.ZodString;
|
|
283
|
+
contentHash: z.ZodString;
|
|
284
|
+
revision: z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>;
|
|
285
|
+
}, z.core.$strict>;
|
|
286
|
+
feedbackKey: z.ZodString;
|
|
287
|
+
configurationHash: z.ZodString;
|
|
288
|
+
output: z.ZodEnum<{
|
|
289
|
+
boolean: "boolean";
|
|
290
|
+
score: "score";
|
|
291
|
+
category: "category";
|
|
292
|
+
}>;
|
|
293
|
+
categories: z.ZodArray<z.ZodString>;
|
|
294
|
+
}, z.core.$strict>>;
|
|
295
|
+
fieldMappingsHash: z.ZodString;
|
|
296
|
+
fieldMappings: z.ZodArray<z.ZodObject<{
|
|
297
|
+
graderId: z.ZodString;
|
|
298
|
+
fields: z.ZodArray<z.ZodObject<{
|
|
299
|
+
destination: z.ZodEnum<{
|
|
300
|
+
output: "output";
|
|
301
|
+
input: "input";
|
|
302
|
+
expectedOutput: "expectedOutput";
|
|
303
|
+
evaluatorContext: "evaluatorContext";
|
|
304
|
+
}>;
|
|
305
|
+
path: z.ZodArray<z.ZodString>;
|
|
306
|
+
source: z.ZodEnum<{
|
|
307
|
+
output: "output";
|
|
308
|
+
input: "input";
|
|
309
|
+
expectedOutput: "expectedOutput";
|
|
310
|
+
evaluatorContext: "evaluatorContext";
|
|
311
|
+
}>;
|
|
312
|
+
sourcePath: z.ZodArray<z.ZodString>;
|
|
313
|
+
}, z.core.$strict>>;
|
|
314
|
+
}, z.core.$strict>>;
|
|
315
|
+
environmentHash: z.ZodString;
|
|
316
|
+
privateDatasetClosureHash: z.ZodString;
|
|
317
|
+
criterion: z.ZodObject<{
|
|
318
|
+
minimumPassRate: z.ZodNumber;
|
|
319
|
+
requireComplete: z.ZodBoolean;
|
|
320
|
+
}, z.core.$strict>;
|
|
321
|
+
contentHash: z.ZodString;
|
|
322
|
+
}, z.core.$strict>>;
|
|
323
|
+
sourceCandidate: z.ZodOptional<z.ZodObject<{
|
|
324
|
+
candidateId: z.ZodString;
|
|
325
|
+
frozenRevision: z.ZodNumber;
|
|
326
|
+
frozenHash: z.ZodString;
|
|
327
|
+
harnessRelease: z.ZodObject<{
|
|
328
|
+
id: z.ZodString;
|
|
329
|
+
contentHash: z.ZodString;
|
|
330
|
+
}, z.core.$strict>;
|
|
331
|
+
protectedClosureHash: z.ZodString;
|
|
332
|
+
}, z.core.$strict>>;
|
|
103
333
|
}, z.core.$strict>;
|
|
104
334
|
policy: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
105
335
|
kind: z.ZodLiteral<"model">;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"profile-evaluation-comparisons.d.ts","sourceRoot":"","sources":["../../../../src/profile-evaluation-comparisons.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAMxB,OAAO,EAA6B,KAAK,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AAEzF,OAAO,EAA4B,KAAK,kBAAkB,EAAE,MAAM,2BAA2B,CAAC;AAE9F,eAAO,MAAM,wCAAwC
|
|
1
|
+
{"version":3,"file":"profile-evaluation-comparisons.d.ts","sourceRoot":"","sources":["../../../../src/profile-evaluation-comparisons.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAMxB,OAAO,EAA6B,KAAK,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AAEzF,OAAO,EAA4B,KAAK,kBAAkB,EAAE,MAAM,2BAA2B,CAAC;AAE9F,eAAO,MAAM,wCAAwC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAgB1C,CAAC;AACZ,eAAO,MAAM,iCAAiC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAEnC,CAAC;AACZ,MAAM,MAAM,2BAA2B,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,iCAAiC,CAAC,CAAC;AAE5F;;+DAE+D;AAC/D,wBAAgB,iCAAiC,CAAC,KAAK,EAAE;IACvD,EAAE,EAAE,MAAM,CAAC;IACX,OAAO,EAAE,KAAK,CAAC;QAAE,QAAQ,EAAE,kBAAkB,CAAC;QAAC,MAAM,EAAE,mBAAmB,CAAA;KAAE,CAAC,CAAC;IAC9E,SAAS,EAAE,MAAM,CAAC;CACnB,GAAG,2BAA2B,CAuD9B"}
|
|
@@ -0,0 +1,14 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
export declare const ProfileEvaluationTargetSchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
3
|
+
kind: z.ZodLiteral<"profile">;
|
|
4
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
5
|
+
kind: z.ZodLiteral<"workflow">;
|
|
6
|
+
workflowId: z.ZodString;
|
|
7
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
8
|
+
kind: z.ZodLiteral<"skill">;
|
|
9
|
+
skillPath: z.ZodString;
|
|
10
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
11
|
+
kind: z.ZodLiteral<"agent_action">;
|
|
12
|
+
actionId: z.ZodString;
|
|
13
|
+
}, z.core.$strict>], "kind">;
|
|
14
|
+
//# sourceMappingURL=profile-evaluation-target.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"profile-evaluation-target.d.ts","sourceRoot":"","sources":["../../../../src/profile-evaluation-target.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAGxB,eAAO,MAAM,6BAA6B;;;;;;;;;;;4BAEgH,CAAC"}
|
|
@@ -1,19 +1,7 @@
|
|
|
1
1
|
import { z } from "zod";
|
|
2
2
|
import { type AgentSnapshot, type HarnessRelease } from "@openpond/harness";
|
|
3
|
-
|
|
4
|
-
*
|
|
5
|
-
export declare const ProfileEvaluationTargetSchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
6
|
-
kind: z.ZodLiteral<"profile">;
|
|
7
|
-
}, z.core.$strict>, z.ZodObject<{
|
|
8
|
-
kind: z.ZodLiteral<"workflow">;
|
|
9
|
-
workflowId: z.ZodString;
|
|
10
|
-
}, z.core.$strict>, z.ZodObject<{
|
|
11
|
-
kind: z.ZodLiteral<"skill">;
|
|
12
|
-
skillPath: z.ZodString;
|
|
13
|
-
}, z.core.$strict>, z.ZodObject<{
|
|
14
|
-
kind: z.ZodLiteral<"agent_action">;
|
|
15
|
-
actionId: z.ZodString;
|
|
16
|
-
}, z.core.$strict>], "kind">;
|
|
3
|
+
export { ProfileEvaluationTargetSchema } from "./profile-evaluation-target.js";
|
|
4
|
+
export * from "./profile-external-evaluation.js";
|
|
17
5
|
export declare const ProfileEvaluationDefinitionSchema: z.ZodObject<{
|
|
18
6
|
id: z.ZodString;
|
|
19
7
|
label: z.ZodString;
|
|
@@ -128,6 +116,121 @@ export declare const ProfileEvaluationRunSourceSchema: z.ZodObject<{
|
|
|
128
116
|
actionId: z.ZodString;
|
|
129
117
|
}, z.core.$strict>], "kind">;
|
|
130
118
|
environmentHash: z.ZodString;
|
|
119
|
+
externalDatasetBinding: z.ZodOptional<z.ZodObject<{
|
|
120
|
+
schemaVersion: z.ZodLiteral<"openpond.profileExternalDatasetBinding.v1">;
|
|
121
|
+
dataset: z.ZodObject<{
|
|
122
|
+
id: z.ZodString;
|
|
123
|
+
contentHash: z.ZodString;
|
|
124
|
+
revision: z.ZodNumber;
|
|
125
|
+
}, z.core.$strict>;
|
|
126
|
+
packageHash: z.ZodString;
|
|
127
|
+
profileId: z.ZodString;
|
|
128
|
+
target: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
129
|
+
kind: z.ZodLiteral<"profile">;
|
|
130
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
131
|
+
kind: z.ZodLiteral<"workflow">;
|
|
132
|
+
workflowId: z.ZodString;
|
|
133
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
134
|
+
kind: z.ZodLiteral<"skill">;
|
|
135
|
+
skillPath: z.ZodString;
|
|
136
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
137
|
+
kind: z.ZodLiteral<"agent_action">;
|
|
138
|
+
actionId: z.ZodString;
|
|
139
|
+
}, z.core.$strict>], "kind">;
|
|
140
|
+
declaredProfileCatalogHash: z.ZodNullable<z.ZodString>;
|
|
141
|
+
protectedProfileClosureHash: z.ZodString;
|
|
142
|
+
recordedOrigin: z.ZodOptional<z.ZodObject<{
|
|
143
|
+
execution: z.ZodObject<{
|
|
144
|
+
id: z.ZodString;
|
|
145
|
+
contentHash: z.ZodString;
|
|
146
|
+
}, z.core.$strict>;
|
|
147
|
+
originalEvidence: z.ZodObject<{
|
|
148
|
+
id: z.ZodString;
|
|
149
|
+
contentHash: z.ZodString;
|
|
150
|
+
}, z.core.$strict>;
|
|
151
|
+
members: z.ZodArray<z.ZodObject<{
|
|
152
|
+
caseId: z.ZodString;
|
|
153
|
+
seed: z.ZodString;
|
|
154
|
+
fixtureId: z.ZodString;
|
|
155
|
+
sourceId: z.ZodString;
|
|
156
|
+
snapshotHash: z.ZodString;
|
|
157
|
+
boundaryRevisionHash: z.ZodString;
|
|
158
|
+
}, z.core.$strict>>;
|
|
159
|
+
}, z.core.$strict>>;
|
|
160
|
+
gradingSource: z.ZodOptional<z.ZodObject<{
|
|
161
|
+
pass: z.ZodObject<{
|
|
162
|
+
id: z.ZodString;
|
|
163
|
+
contentHash: z.ZodString;
|
|
164
|
+
}, z.core.$strict>;
|
|
165
|
+
evidence: z.ZodObject<{
|
|
166
|
+
id: z.ZodString;
|
|
167
|
+
contentHash: z.ZodString;
|
|
168
|
+
}, z.core.$strict>;
|
|
169
|
+
}, z.core.$strict>>;
|
|
170
|
+
split: z.ZodEnum<{
|
|
171
|
+
train: "train";
|
|
172
|
+
validation: "validation";
|
|
173
|
+
test: "test";
|
|
174
|
+
frozen_eval: "frozen_eval";
|
|
175
|
+
}>;
|
|
176
|
+
population: z.ZodArray<z.ZodObject<{
|
|
177
|
+
taskId: z.ZodString;
|
|
178
|
+
seed: z.ZodString;
|
|
179
|
+
fixtureId: z.ZodNullable<z.ZodString>;
|
|
180
|
+
}, z.core.$strict>>;
|
|
181
|
+
evaluators: z.ZodArray<z.ZodObject<{
|
|
182
|
+
release: z.ZodObject<{
|
|
183
|
+
id: z.ZodString;
|
|
184
|
+
contentHash: z.ZodString;
|
|
185
|
+
revision: z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>;
|
|
186
|
+
}, z.core.$strict>;
|
|
187
|
+
feedbackKey: z.ZodString;
|
|
188
|
+
configurationHash: z.ZodString;
|
|
189
|
+
output: z.ZodEnum<{
|
|
190
|
+
boolean: "boolean";
|
|
191
|
+
score: "score";
|
|
192
|
+
category: "category";
|
|
193
|
+
}>;
|
|
194
|
+
categories: z.ZodArray<z.ZodString>;
|
|
195
|
+
}, z.core.$strict>>;
|
|
196
|
+
fieldMappingsHash: z.ZodString;
|
|
197
|
+
fieldMappings: z.ZodArray<z.ZodObject<{
|
|
198
|
+
graderId: z.ZodString;
|
|
199
|
+
fields: z.ZodArray<z.ZodObject<{
|
|
200
|
+
destination: z.ZodEnum<{
|
|
201
|
+
output: "output";
|
|
202
|
+
input: "input";
|
|
203
|
+
expectedOutput: "expectedOutput";
|
|
204
|
+
evaluatorContext: "evaluatorContext";
|
|
205
|
+
}>;
|
|
206
|
+
path: z.ZodArray<z.ZodString>;
|
|
207
|
+
source: z.ZodEnum<{
|
|
208
|
+
output: "output";
|
|
209
|
+
input: "input";
|
|
210
|
+
expectedOutput: "expectedOutput";
|
|
211
|
+
evaluatorContext: "evaluatorContext";
|
|
212
|
+
}>;
|
|
213
|
+
sourcePath: z.ZodArray<z.ZodString>;
|
|
214
|
+
}, z.core.$strict>>;
|
|
215
|
+
}, z.core.$strict>>;
|
|
216
|
+
environmentHash: z.ZodString;
|
|
217
|
+
privateDatasetClosureHash: z.ZodString;
|
|
218
|
+
criterion: z.ZodObject<{
|
|
219
|
+
minimumPassRate: z.ZodNumber;
|
|
220
|
+
requireComplete: z.ZodBoolean;
|
|
221
|
+
}, z.core.$strict>;
|
|
222
|
+
contentHash: z.ZodString;
|
|
223
|
+
}, z.core.$strict>>;
|
|
224
|
+
sourceCandidate: z.ZodOptional<z.ZodObject<{
|
|
225
|
+
candidateId: z.ZodString;
|
|
226
|
+
frozenRevision: z.ZodNumber;
|
|
227
|
+
frozenHash: z.ZodString;
|
|
228
|
+
harnessRelease: z.ZodObject<{
|
|
229
|
+
id: z.ZodString;
|
|
230
|
+
contentHash: z.ZodString;
|
|
231
|
+
}, z.core.$strict>;
|
|
232
|
+
protectedClosureHash: z.ZodString;
|
|
233
|
+
}, z.core.$strict>>;
|
|
131
234
|
}, z.core.$strict>;
|
|
132
235
|
export type ProfileEvaluationRunSource = z.infer<typeof ProfileEvaluationRunSourceSchema>;
|
|
133
236
|
/** Validate component and suite references against the same Profile release. */
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"profile-evaluations.d.ts","sourceRoot":"","sources":["../../../../src/profile-evaluations.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;
|
|
1
|
+
{"version":3,"file":"profile-evaluations.d.ts","sourceRoot":"","sources":["../../../../src/profile-evaluations.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAGxB,OAAO,EAGyB,KAAK,aAAa,EAAE,KAAK,cAAc,EACtE,MAAM,mBAAmB,CAAC;AAK3B,OAAO,EAAE,6BAA6B,EAAE,MAAM,gCAAgC,CAAC;AAC/E,cAAc,kCAAkC,CAAC;AAEjD,eAAO,MAAM,iCAAiC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAoB5C,CAAC;AAEH,eAAO,MAAM,4BAA4B;;;;;;;;kBAK9B,CAAC;AAEZ,eAAO,MAAM,8BAA8B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAIhC,CAAC;AAEZ,MAAM,MAAM,wBAAwB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,8BAA8B,CAAC,CAAC;AACtF,MAAM,MAAM,2BAA2B,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,iCAAiC,CAAC,CAAC;AAE5F;gFACgF;AAChF,eAAO,MAAM,gCAAgC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAWlC,CAAC;AACZ,MAAM,MAAM,0BAA0B,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,gCAAgC,CAAC,CAAC;AAE1F,gFAAgF;AAChF,wBAAgB,gCAAgC,CAAC,KAAK,EAAE;IACtD,OAAO,EAAE,OAAO,CAAC;IACjB,WAAW,EAAE,WAAW,CAAC,MAAM,CAAC,CAAC;IACjC,UAAU,EAAE,WAAW,CAAC,MAAM,CAAC,CAAC;IAChC,SAAS,EAAE,WAAW,CAAC,MAAM,CAAC,CAAC;CAChC,GAAG;IAAE,OAAO,EAAE,wBAAwB,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,CA+B7D;AAED,wBAAgB,iCAAiC,CAAC,KAAK,EAAE;IACvD,OAAO,EAAE,wBAAwB,CAAC;IAClC,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;IAClB,cAAc,EAAE,MAAM,CAAC;IACvB,cAAc,EAAE;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC;IACpD,eAAe,EAAE,MAAM,CAAC;CACzB,GAAG,0BAA0B,CAc7B;AAED;;mCAEmC;AACnC,wBAAgB,oCAAoC,CAAC,KAAK,EAAE,OAAO,GAAG;IACpE,OAAO,EAAE,wBAAwB,CAAC;IAClC,WAAW,EAAE,MAAM,CAAC;IACpB,cAAc,EAAE;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC;CACrD,CAaA;AAED,iFAAiF;AACjF,wBAAgB,0CAA0C,CAAC,KAAK,EAAE;IAChE,aAAa,EAAE,aAAa,CAAC;IAC7B,cAAc,EAAE,cAAc,CAAC;IAC/B,YAAY,CAAC,EAAE,UAAU,CAAC;IAC1B,aAAa,CAAC,EAAE,UAAU,CAAC;IAC3B,WAAW,CAAC,EAAE,UAAU,CAAC;CAC1B,GAAG;IAAE,OAAO,EAAE,wBAAwB,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,CAqB7D"}
|