@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
|
@@ -0,0 +1,32 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
import { ExperimentTargetSchema } from "../experiments.js";
|
|
3
|
+
const Id = z.string().trim().min(1).max(200);
|
|
4
|
+
const Hash = z.string().regex(/^[a-f0-9]{64}$/);
|
|
5
|
+
const Ref = z.object({ id: Id, contentHash: Hash }).strict();
|
|
6
|
+
export const TrainingHandoffOriginSchema = z
|
|
7
|
+
.object({
|
|
8
|
+
schemaVersion: z.literal("openpond.trainingHandoffOrigin.v1"),
|
|
9
|
+
teamId: Id,
|
|
10
|
+
projectId: Id.nullable(),
|
|
11
|
+
configurationId: Id,
|
|
12
|
+
source: z
|
|
13
|
+
.object({ executionId: Id, passId: Id.nullable(), manifestHash: Hash, resultHash: Hash })
|
|
14
|
+
.strict(),
|
|
15
|
+
dataset: z
|
|
16
|
+
.object({ id: Id, revision: z.number().int().positive(), contentHash: Hash })
|
|
17
|
+
.strict(),
|
|
18
|
+
target: ExperimentTargetSchema,
|
|
19
|
+
graders: z
|
|
20
|
+
.array(z
|
|
21
|
+
.object({
|
|
22
|
+
feedbackKey: Id,
|
|
23
|
+
evaluator: Ref,
|
|
24
|
+
expectedCount: z.number().int().positive(),
|
|
25
|
+
scoredCount: z.number().int().nonnegative(),
|
|
26
|
+
score: z.number().finite().nullable(),
|
|
27
|
+
})
|
|
28
|
+
.strict())
|
|
29
|
+
.max(100),
|
|
30
|
+
contentHash: Hash,
|
|
31
|
+
})
|
|
32
|
+
.strict();
|
package/dist/metrics-node.js
CHANGED
|
@@ -31,3 +31,5 @@ const executeMetric = async ({ source, module, exportName, scores, timeoutMs, si
|
|
|
31
31
|
};
|
|
32
32
|
const executeProcessMetric = ({ source, module, exportName, scores, timeoutMs, signal }) => executeJavaScriptIsolateInProcess({ source, exportName, value: scores, timeoutMs, signal, stripTypeScript: /\.[cm]?ts$/.test(module),
|
|
33
33
|
maxResultBytes: 1_024, deterministic: true, errorPrefix: "metric" });
|
|
34
|
+
/** Trusted host port for a frozen private metric; callers retain source and receipt ownership. */
|
|
35
|
+
export const executeTasksetMetricWorker = executeMetric;
|
|
@@ -0,0 +1,7 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
import { ReleaseIdSchema } from "@openpond/harness";
|
|
3
|
+
const RelativePathSchema = z.string().min(1).max(2000).refine(value => !value.includes("\\") && !value.includes(":") && !value.startsWith("/") && value.split("/").every(part => part !== "" && part !== "." && part !== ".."), "Evaluation paths must be portable and relative.");
|
|
4
|
+
export const ProfileEvaluationTargetSchema = z.discriminatedUnion("kind", [
|
|
5
|
+
z.object({ kind: z.literal("profile") }).strict(), z.object({ kind: z.literal("workflow"), workflowId: ReleaseIdSchema }).strict(),
|
|
6
|
+
z.object({ kind: z.literal("skill"), skillPath: RelativePathSchema }).strict(), z.object({ kind: z.literal("agent_action"), actionId: ReleaseIdSchema }).strict()
|
|
7
|
+
]);
|
|
@@ -1,16 +1,11 @@
|
|
|
1
1
|
import { z } from "zod";
|
|
2
|
+
import { ProfileSourceCandidateSchema } from "./profile-source-candidate.js";
|
|
2
3
|
import { ImmutableReleaseRefSchema, ReleaseHashSchema, ReleaseIdSchema, contentHash, harnessSourcePackageFiles, loadReleasedProfileWorkflowCatalogAssets, sha256, validateHarnessSourcePackage, } from "@openpond/harness";
|
|
3
4
|
import { TaskSplitSchema } from "./tasksets.js";
|
|
4
|
-
|
|
5
|
-
|
|
6
|
-
|
|
7
|
-
*
|
|
8
|
-
export const ProfileEvaluationTargetSchema = z.discriminatedUnion("kind", [
|
|
9
|
-
z.object({ kind: z.literal("profile") }).strict(),
|
|
10
|
-
z.object({ kind: z.literal("workflow"), workflowId: ReleaseIdSchema }).strict(),
|
|
11
|
-
z.object({ kind: z.literal("skill"), skillPath: RelativePathSchema }).strict(),
|
|
12
|
-
z.object({ kind: z.literal("agent_action"), actionId: ReleaseIdSchema }).strict(),
|
|
13
|
-
]);
|
|
5
|
+
import { ProfileEvaluationTargetSchema } from "./profile-evaluation-target.js";
|
|
6
|
+
import { ProfileExternalDatasetBindingSchema } from "./profile-external-evaluation.js";
|
|
7
|
+
export { ProfileEvaluationTargetSchema } from "./profile-evaluation-target.js";
|
|
8
|
+
export * from "./profile-external-evaluation.js";
|
|
14
9
|
export const ProfileEvaluationDefinitionSchema = z.object({
|
|
15
10
|
id: ReleaseIdSchema,
|
|
16
11
|
label: z.string().trim().min(1).max(240),
|
|
@@ -54,6 +49,8 @@ export const ProfileEvaluationRunSourceSchema = z.object({
|
|
|
54
49
|
definitionHash: ReleaseHashSchema,
|
|
55
50
|
target: ProfileEvaluationTargetSchema,
|
|
56
51
|
environmentHash: ReleaseHashSchema,
|
|
52
|
+
externalDatasetBinding: ProfileExternalDatasetBindingSchema.optional(),
|
|
53
|
+
sourceCandidate: ProfileSourceCandidateSchema.optional(),
|
|
57
54
|
}).strict();
|
|
58
55
|
/** Validate component and suite references against the same Profile release. */
|
|
59
56
|
export function validateProfileEvaluationCatalog(input) {
|
|
@@ -0,0 +1,35 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
import { ImmutableReleaseRefSchema, ReleaseHashSchema, ReleaseIdSchema, contentHash } from "@openpond/harness";
|
|
3
|
+
import { ProfileEvaluationTargetSchema } from "./profile-evaluation-target.js";
|
|
4
|
+
import { TaskSplitSchema } from "./tasksets.js";
|
|
5
|
+
const PopulationMemberSchema = z.object({ taskId: ReleaseIdSchema, seed: z.string().min(1).max(500), fixtureId: ReleaseIdSchema.nullable() }).strict();
|
|
6
|
+
const Segment = z.string().min(1).max(200).refine(value => !["__proto__", "prototype", "constructor"].includes(value), "Use an ordinary field name.");
|
|
7
|
+
const Slot = z.enum(["input", "output", "expectedOutput", "evaluatorContext"]);
|
|
8
|
+
export const ProfileExternalFieldMappingsSchema = z.array(z.object({ graderId: ReleaseIdSchema, fields: z.array(z.object({ destination: Slot, path: z.array(Segment).min(1).max(20), source: Slot, sourcePath: z.array(Segment).max(20) }).strict()).max(100) }).strict()).max(100);
|
|
9
|
+
const RecordedOriginSchema = z.object({ execution: ImmutableReleaseRefSchema, originalEvidence: ImmutableReleaseRefSchema, members: z.array(z.object({ caseId: ReleaseIdSchema, seed: z.string().min(1).max(500), fixtureId: ReleaseHashSchema,
|
|
10
|
+
sourceId: ReleaseIdSchema, snapshotHash: ReleaseHashSchema, boundaryRevisionHash: ReleaseHashSchema }).strict()).min(1).max(500) }).strict();
|
|
11
|
+
const MeasurementSchema = z.object({ release: ImmutableReleaseRefSchema.extend({ revision: z.union([z.number().int().positive(), z.string().min(1).max(200)]) }).strict(), feedbackKey: z.string().min(1).max(240),
|
|
12
|
+
configurationHash: ReleaseHashSchema, output: z.enum(["boolean", "score", "category"]), categories: z.array(z.string().min(1).max(120)).max(50) }).strict();
|
|
13
|
+
export const ProfileExternalDatasetBindingContentSchema = z.object({ schemaVersion: z.literal("openpond.profileExternalDatasetBinding.v1"), dataset: ImmutableReleaseRefSchema.extend({ revision: z.number().int().positive() }).strict(),
|
|
14
|
+
packageHash: ReleaseHashSchema, profileId: ReleaseIdSchema, target: ProfileEvaluationTargetSchema, declaredProfileCatalogHash: ReleaseHashSchema.nullable(), protectedProfileClosureHash: ReleaseHashSchema,
|
|
15
|
+
recordedOrigin: RecordedOriginSchema.optional(), gradingSource: z.object({ pass: ImmutableReleaseRefSchema, evidence: ImmutableReleaseRefSchema }).strict().optional(), split: TaskSplitSchema, population: z.array(PopulationMemberSchema).min(1).max(10000), evaluators: z.array(MeasurementSchema).min(1).max(100), fieldMappingsHash: ReleaseHashSchema, fieldMappings: ProfileExternalFieldMappingsSchema,
|
|
16
|
+
environmentHash: ReleaseHashSchema, privateDatasetClosureHash: ReleaseHashSchema, criterion: z.object({ minimumPassRate: z.number().min(0).max(1), requireComplete: z.boolean() }).strict() }).strict().superRefine((value, ctx) => {
|
|
17
|
+
if (value.recordedOrigin && (value.recordedOrigin.members.length !== value.population.length || value.recordedOrigin.members.some((source, index) => source.caseId !== value.population[index].taskId || source.seed !== value.population[index].seed || source.fixtureId !== source.snapshotHash || value.population[index].fixtureId !== null)))
|
|
18
|
+
ctx.addIssue({ code: "custom", message: "Recorded-to-rerun mapping must preserve every exact ordered cutoff and seed while admitting new target attempts." });
|
|
19
|
+
if (contentHash(value.fieldMappings) !== value.fieldMappingsHash)
|
|
20
|
+
ctx.addIssue({ code: "custom", message: "Field mappings differ from their sealed hash." });
|
|
21
|
+
if (new Set(value.fieldMappings.map(row => row.graderId)).size !== value.fieldMappings.length)
|
|
22
|
+
ctx.addIssue({ code: "custom", message: "Field mappings must name each grader once." });
|
|
23
|
+
if (new Set(value.population.map(row => JSON.stringify([row.taskId, row.seed, row.fixtureId]))).size !== value.population.length)
|
|
24
|
+
ctx.addIssue({ code: "custom", message: "External Dataset population must be unique." });
|
|
25
|
+
if (new Set(value.evaluators.map(row => row.feedbackKey)).size !== value.evaluators.length)
|
|
26
|
+
ctx.addIssue({ code: "custom", message: "External Dataset feedback keys must be unique." });
|
|
27
|
+
for (const grader of value.evaluators)
|
|
28
|
+
if ((grader.output === "category") !== (grader.categories.length > 0))
|
|
29
|
+
ctx.addIssue({ code: "custom", message: "Only category grading declares categories." });
|
|
30
|
+
});
|
|
31
|
+
export const ProfileExternalDatasetBindingSchema = ProfileExternalDatasetBindingContentSchema.safeExtend({ contentHash: ReleaseHashSchema }).strict();
|
|
32
|
+
export function verifyProfileExternalDatasetBinding(raw) { const value = ProfileExternalDatasetBindingSchema.parse(raw), { contentHash: hash, ...body } = value; if (contentHash(body) !== hash)
|
|
33
|
+
throw new Error("External Dataset binding integrity failed."); return value; }
|
|
34
|
+
export function createProfileExternalDatasetBinding(raw) { const value = ProfileExternalDatasetBindingContentSchema.parse(raw); return verifyProfileExternalDatasetBinding({ ...value, contentHash: contentHash(value) }); }
|
|
35
|
+
export const externalDatasetDefinitionId = (binding) => `external-dataset-${verifyProfileExternalDatasetBinding(binding).contentHash.slice(0, 40)}`;
|
|
@@ -0,0 +1,5 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
import { ImmutableReleaseRefSchema } from "@openpond/harness";
|
|
3
|
+
/** An identity to resolve through the actual private hosted source owner. It is
|
|
4
|
+
* never itself an authorization or a model-artifact declaration. */
|
|
5
|
+
export const ProfileSourceCandidateSchema = z.object({ candidateId: z.string().min(1).max(240), frozenRevision: z.number().int().positive(), frozenHash: z.string().regex(/^[a-f0-9]{64}$/), harnessRelease: ImmutableReleaseRefSchema, protectedClosureHash: z.string().regex(/^[a-f0-9]{64}$/) }).strict();
|
package/dist/rewards.js
CHANGED
|
@@ -101,8 +101,8 @@ export const RewardCompositionContentSchema = z.object({
|
|
|
101
101
|
taskHash: ReleaseHashSchema,
|
|
102
102
|
outputHash: ReleaseHashSchema,
|
|
103
103
|
results: z.array(BoundRewardResultSchema).max(100),
|
|
104
|
-
training: z.object({ status: z.enum(["scored", "unscorable", "not_configured"]), score: z.number().min(0).max(1).nullable(), passed: z.boolean().nullable() }).strict(),
|
|
105
|
-
evaluation: z.object({ status: z.enum(["scored", "unscorable", "not_configured"]), score: z.number().min(0).max(1).nullable(), passed: z.boolean().nullable() }).strict(),
|
|
104
|
+
training: z.object({ status: z.enum(["scored", "pending", "unscorable", "not_configured"]), score: z.number().min(0).max(1).nullable(), passed: z.boolean().nullable() }).strict(),
|
|
105
|
+
evaluation: z.object({ status: z.enum(["scored", "pending", "unscorable", "not_configured"]), score: z.number().min(0).max(1).nullable(), passed: z.boolean().nullable() }).strict(),
|
|
106
106
|
}).strict();
|
|
107
107
|
export const RewardCompositionSchema = RewardCompositionContentSchema.extend({ contentHash: ReleaseHashSchema }).strict();
|
|
108
108
|
export function createRewardRelease(input) {
|
|
@@ -229,8 +229,10 @@ export function composeBoundRewards(input) {
|
|
|
229
229
|
const selected = binding.sources.filter((source) => source.role === role).map((source) => ({ source, result: results.find((candidate) => candidate.graderId === source.graderId) }));
|
|
230
230
|
if (!selected.length)
|
|
231
231
|
return { status: "not_configured", score: null, passed: null };
|
|
232
|
-
if (selected.some(({ source, result }) => source.required && result.status !== "scored"))
|
|
232
|
+
if (selected.some(({ source, result }) => source.required && result.status !== "scored" && result.status !== "pending"))
|
|
233
233
|
return { status: "unscorable", score: null, passed: null };
|
|
234
|
+
if (selected.some(({ source, result }) => source.required && result.status === "pending"))
|
|
235
|
+
return { status: "pending", score: null, passed: null };
|
|
234
236
|
const scored = selected.filter(({ result }) => result.status === "scored");
|
|
235
237
|
const weight = scored.reduce((sum, { source }) => sum + source.weight, 0);
|
|
236
238
|
if (weight === 0)
|
|
@@ -1,3 +1,4 @@
|
|
|
1
|
+
import { verifyProfileExternalDatasetBinding, externalDatasetDefinitionId } from "./profile-external-evaluation.js";
|
|
1
2
|
import { z } from "zod";
|
|
2
3
|
import { assertContentHash, contentHash, ImmutableReleaseRefSchema, MetadataSchema, ModelRefSchema, ReleaseHashSchema, ReleaseIdSchema, ReleaseTimestampSchema } from "@openpond/harness";
|
|
3
4
|
import { RunLimitsSchema, RuntimeTargetBindingSchema, verifyAttemptReceipt } from "./runs.js";
|
|
@@ -100,6 +101,15 @@ export function assertProfileEvaluationRunAdmission(manifest, taskset, catalogIn
|
|
|
100
101
|
if (definition.taskIds.some((id) => tasks.get(id)?.split !== definition.split)) {
|
|
101
102
|
throw new Error("Profile evaluation case is absent from its declared Taskset split.");
|
|
102
103
|
}
|
|
104
|
+
if (source.externalDatasetBinding) {
|
|
105
|
+
const binding = verifyProfileExternalDatasetBinding(source.externalDatasetBinding);
|
|
106
|
+
if (source.definitionId !== externalDatasetDefinitionId(binding) || binding.dataset.id !== taskset.id || binding.dataset.revision !== taskset.revision || binding.dataset.contentHash !== taskset.contentHash
|
|
107
|
+
|| binding.packageHash !== manifest.packageHash || binding.profileId !== source.profileId || contentHash(binding.target) !== contentHash(source.target) || binding.split !== definition.split
|
|
108
|
+
|| binding.environmentHash !== contentHash(taskset.environment) || binding.population.length !== manifest.population.length
|
|
109
|
+
|| binding.population.some((member, index) => contentHash(member) !== contentHash({ taskId: manifest.population[index].taskId, seed: manifest.population[index].seed, fixtureId: manifest.population[index].fixtureId })))
|
|
110
|
+
throw new Error("Profile external Dataset execution differs from its exact admitted recipe.");
|
|
111
|
+
return;
|
|
112
|
+
}
|
|
103
113
|
const expected = new Set(definition.taskIds.flatMap((taskId) => definition.seeds.map((seed) => JSON.stringify([taskId, seed]))));
|
|
104
114
|
const actual = new Set(manifest.population.map(({ taskId, seed }) => JSON.stringify([taskId, seed])));
|
|
105
115
|
if (manifest.population.some((member) => member.fixtureId !== null)
|
package/dist/tasksets.js
CHANGED
|
@@ -1,6 +1,7 @@
|
|
|
1
1
|
import { z } from "zod";
|
|
2
2
|
import { CapabilityRequirementSchema, ImmutableAssetRefSchema, MetadataSchema, ReleaseHashSchema, ReleaseIdSchema, ToolDeclarationSchema, assertContentHash, contentHash, } from "@openpond/harness";
|
|
3
3
|
import { TasksetMetricPolicySchema } from "./metric-policy.js";
|
|
4
|
+
import { HumanFormSchema } from "./human-review/contracts.js";
|
|
4
5
|
export const TaskSplitSchema = z.enum(["train", "validation", "test", "frozen_eval"]);
|
|
5
6
|
export const RequiredOutputContractSchema = z.object({
|
|
6
7
|
path: z.string().trim().min(1).max(2_000).refine(safeRelativePath),
|
|
@@ -58,6 +59,7 @@ export const HumanGraderSpecSchema = GraderBaseSchema.extend({
|
|
|
58
59
|
kind: z.literal("human"),
|
|
59
60
|
rubricRef: ImmutableAssetRefSchema,
|
|
60
61
|
reviewerRole: z.string().trim().min(1).max(500),
|
|
62
|
+
form: HumanFormSchema.optional(),
|
|
61
63
|
}).strict();
|
|
62
64
|
export const GraderSpecSchema = z.union([
|
|
63
65
|
DeterministicGraderSpecSchema,
|
|
@@ -14,7 +14,7 @@ export declare const harnessRefinerBenchmarkRelease: {
|
|
|
14
14
|
entrypoint: string;
|
|
15
15
|
stateful: boolean;
|
|
16
16
|
deterministicSeeds: boolean;
|
|
17
|
-
lifecycle: ("
|
|
17
|
+
lifecycle: ("step" | "create" | "reset" | "collect" | "destroy")[];
|
|
18
18
|
networkPolicy: "none" | "declared_read_only" | "declared_scoped";
|
|
19
19
|
defaultTimeoutMs: number;
|
|
20
20
|
};
|
|
@@ -133,6 +133,80 @@ export declare const harnessRefinerBenchmarkRelease: {
|
|
|
133
133
|
visibility: "policy" | "verifier" | "host_private";
|
|
134
134
|
};
|
|
135
135
|
reviewerRole: string;
|
|
136
|
+
form?: {
|
|
137
|
+
schemaVersion: "openpond.humanForm.v1";
|
|
138
|
+
mode: "individual" | "pairwise";
|
|
139
|
+
instructions: string;
|
|
140
|
+
criteria: ({
|
|
141
|
+
id: string;
|
|
142
|
+
label: string;
|
|
143
|
+
instructions: string;
|
|
144
|
+
required: boolean;
|
|
145
|
+
allowAbstain: boolean;
|
|
146
|
+
kind: "score";
|
|
147
|
+
minimum: number;
|
|
148
|
+
maximum: number;
|
|
149
|
+
step: number;
|
|
150
|
+
requireRationale?: boolean | undefined;
|
|
151
|
+
anchors?: {
|
|
152
|
+
value: string | number | boolean;
|
|
153
|
+
description: string;
|
|
154
|
+
}[] | undefined;
|
|
155
|
+
} | {
|
|
156
|
+
id: string;
|
|
157
|
+
label: string;
|
|
158
|
+
instructions: string;
|
|
159
|
+
required: boolean;
|
|
160
|
+
allowAbstain: boolean;
|
|
161
|
+
kind: "category";
|
|
162
|
+
options: string[];
|
|
163
|
+
requireRationale?: boolean | undefined;
|
|
164
|
+
anchors?: {
|
|
165
|
+
value: string | number | boolean;
|
|
166
|
+
description: string;
|
|
167
|
+
}[] | undefined;
|
|
168
|
+
} | {
|
|
169
|
+
id: string;
|
|
170
|
+
label: string;
|
|
171
|
+
instructions: string;
|
|
172
|
+
required: boolean;
|
|
173
|
+
allowAbstain: boolean;
|
|
174
|
+
kind: "boolean";
|
|
175
|
+
requireRationale?: boolean | undefined;
|
|
176
|
+
anchors?: {
|
|
177
|
+
value: string | number | boolean;
|
|
178
|
+
description: string;
|
|
179
|
+
}[] | undefined;
|
|
180
|
+
} | {
|
|
181
|
+
id: string;
|
|
182
|
+
label: string;
|
|
183
|
+
instructions: string;
|
|
184
|
+
required: boolean;
|
|
185
|
+
allowAbstain: boolean;
|
|
186
|
+
kind: "text";
|
|
187
|
+
maximumLength: number;
|
|
188
|
+
requireRationale?: boolean | undefined;
|
|
189
|
+
anchors?: {
|
|
190
|
+
value: string | number | boolean;
|
|
191
|
+
description: string;
|
|
192
|
+
}[] | undefined;
|
|
193
|
+
minimumLength?: number | undefined;
|
|
194
|
+
} | {
|
|
195
|
+
id: string;
|
|
196
|
+
label: string;
|
|
197
|
+
instructions: string;
|
|
198
|
+
required: boolean;
|
|
199
|
+
allowAbstain: boolean;
|
|
200
|
+
kind: "preference";
|
|
201
|
+
allowTie: boolean;
|
|
202
|
+
requireRationale?: boolean | undefined;
|
|
203
|
+
anchors?: {
|
|
204
|
+
value: string | number | boolean;
|
|
205
|
+
description: string;
|
|
206
|
+
}[] | undefined;
|
|
207
|
+
requireStrength?: boolean | undefined;
|
|
208
|
+
})[];
|
|
209
|
+
} | undefined;
|
|
136
210
|
})[];
|
|
137
211
|
metadata: Record<string, unknown>;
|
|
138
212
|
contentHash: string;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"harness-refiner.d.ts","sourceRoot":"","sources":["../../../../../src/builtin-benchmarks/harness-refiner.ts"],"names":[],"mappings":"AAGA,eAAO,MAAM,8BAA8B
|
|
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"}
|
|
@@ -124,7 +124,7 @@ export declare const genericToolConformance: {
|
|
|
124
124
|
entrypoint: string;
|
|
125
125
|
stateful: boolean;
|
|
126
126
|
deterministicSeeds: boolean;
|
|
127
|
-
lifecycle: ("
|
|
127
|
+
lifecycle: ("step" | "create" | "reset" | "collect" | "destroy")[];
|
|
128
128
|
networkPolicy: "none" | "declared_read_only" | "declared_scoped";
|
|
129
129
|
defaultTimeoutMs: number;
|
|
130
130
|
};
|
|
@@ -243,6 +243,80 @@ export declare const genericToolConformance: {
|
|
|
243
243
|
visibility: "policy" | "verifier" | "host_private";
|
|
244
244
|
};
|
|
245
245
|
reviewerRole: string;
|
|
246
|
+
form?: {
|
|
247
|
+
schemaVersion: "openpond.humanForm.v1";
|
|
248
|
+
mode: "individual" | "pairwise";
|
|
249
|
+
instructions: string;
|
|
250
|
+
criteria: ({
|
|
251
|
+
id: string;
|
|
252
|
+
label: string;
|
|
253
|
+
instructions: string;
|
|
254
|
+
required: boolean;
|
|
255
|
+
allowAbstain: boolean;
|
|
256
|
+
kind: "score";
|
|
257
|
+
minimum: number;
|
|
258
|
+
maximum: number;
|
|
259
|
+
step: number;
|
|
260
|
+
requireRationale?: boolean | undefined;
|
|
261
|
+
anchors?: {
|
|
262
|
+
value: string | number | boolean;
|
|
263
|
+
description: string;
|
|
264
|
+
}[] | undefined;
|
|
265
|
+
} | {
|
|
266
|
+
id: string;
|
|
267
|
+
label: string;
|
|
268
|
+
instructions: string;
|
|
269
|
+
required: boolean;
|
|
270
|
+
allowAbstain: boolean;
|
|
271
|
+
kind: "category";
|
|
272
|
+
options: string[];
|
|
273
|
+
requireRationale?: boolean | undefined;
|
|
274
|
+
anchors?: {
|
|
275
|
+
value: string | number | boolean;
|
|
276
|
+
description: string;
|
|
277
|
+
}[] | undefined;
|
|
278
|
+
} | {
|
|
279
|
+
id: string;
|
|
280
|
+
label: string;
|
|
281
|
+
instructions: string;
|
|
282
|
+
required: boolean;
|
|
283
|
+
allowAbstain: boolean;
|
|
284
|
+
kind: "boolean";
|
|
285
|
+
requireRationale?: boolean | undefined;
|
|
286
|
+
anchors?: {
|
|
287
|
+
value: string | number | boolean;
|
|
288
|
+
description: string;
|
|
289
|
+
}[] | undefined;
|
|
290
|
+
} | {
|
|
291
|
+
id: string;
|
|
292
|
+
label: string;
|
|
293
|
+
instructions: string;
|
|
294
|
+
required: boolean;
|
|
295
|
+
allowAbstain: boolean;
|
|
296
|
+
kind: "text";
|
|
297
|
+
maximumLength: number;
|
|
298
|
+
requireRationale?: boolean | undefined;
|
|
299
|
+
anchors?: {
|
|
300
|
+
value: string | number | boolean;
|
|
301
|
+
description: string;
|
|
302
|
+
}[] | undefined;
|
|
303
|
+
minimumLength?: number | undefined;
|
|
304
|
+
} | {
|
|
305
|
+
id: string;
|
|
306
|
+
label: string;
|
|
307
|
+
instructions: string;
|
|
308
|
+
required: boolean;
|
|
309
|
+
allowAbstain: boolean;
|
|
310
|
+
kind: "preference";
|
|
311
|
+
allowTie: boolean;
|
|
312
|
+
requireRationale?: boolean | undefined;
|
|
313
|
+
anchors?: {
|
|
314
|
+
value: string | number | boolean;
|
|
315
|
+
description: string;
|
|
316
|
+
}[] | undefined;
|
|
317
|
+
requireStrength?: boolean | undefined;
|
|
318
|
+
})[];
|
|
319
|
+
} | undefined;
|
|
246
320
|
})[];
|
|
247
321
|
metadata: Record<string, unknown>;
|
|
248
322
|
contentHash: string;
|
|
@@ -435,7 +509,7 @@ export declare const marketingPortfolioConformance: {
|
|
|
435
509
|
entrypoint: string;
|
|
436
510
|
stateful: boolean;
|
|
437
511
|
deterministicSeeds: boolean;
|
|
438
|
-
lifecycle: ("
|
|
512
|
+
lifecycle: ("step" | "create" | "reset" | "collect" | "destroy")[];
|
|
439
513
|
networkPolicy: "none" | "declared_read_only" | "declared_scoped";
|
|
440
514
|
defaultTimeoutMs: number;
|
|
441
515
|
};
|
|
@@ -554,6 +628,80 @@ export declare const marketingPortfolioConformance: {
|
|
|
554
628
|
visibility: "policy" | "verifier" | "host_private";
|
|
555
629
|
};
|
|
556
630
|
reviewerRole: string;
|
|
631
|
+
form?: {
|
|
632
|
+
schemaVersion: "openpond.humanForm.v1";
|
|
633
|
+
mode: "individual" | "pairwise";
|
|
634
|
+
instructions: string;
|
|
635
|
+
criteria: ({
|
|
636
|
+
id: string;
|
|
637
|
+
label: string;
|
|
638
|
+
instructions: string;
|
|
639
|
+
required: boolean;
|
|
640
|
+
allowAbstain: boolean;
|
|
641
|
+
kind: "score";
|
|
642
|
+
minimum: number;
|
|
643
|
+
maximum: number;
|
|
644
|
+
step: number;
|
|
645
|
+
requireRationale?: boolean | undefined;
|
|
646
|
+
anchors?: {
|
|
647
|
+
value: string | number | boolean;
|
|
648
|
+
description: string;
|
|
649
|
+
}[] | undefined;
|
|
650
|
+
} | {
|
|
651
|
+
id: string;
|
|
652
|
+
label: string;
|
|
653
|
+
instructions: string;
|
|
654
|
+
required: boolean;
|
|
655
|
+
allowAbstain: boolean;
|
|
656
|
+
kind: "category";
|
|
657
|
+
options: string[];
|
|
658
|
+
requireRationale?: boolean | undefined;
|
|
659
|
+
anchors?: {
|
|
660
|
+
value: string | number | boolean;
|
|
661
|
+
description: string;
|
|
662
|
+
}[] | undefined;
|
|
663
|
+
} | {
|
|
664
|
+
id: string;
|
|
665
|
+
label: string;
|
|
666
|
+
instructions: string;
|
|
667
|
+
required: boolean;
|
|
668
|
+
allowAbstain: boolean;
|
|
669
|
+
kind: "boolean";
|
|
670
|
+
requireRationale?: boolean | undefined;
|
|
671
|
+
anchors?: {
|
|
672
|
+
value: string | number | boolean;
|
|
673
|
+
description: string;
|
|
674
|
+
}[] | undefined;
|
|
675
|
+
} | {
|
|
676
|
+
id: string;
|
|
677
|
+
label: string;
|
|
678
|
+
instructions: string;
|
|
679
|
+
required: boolean;
|
|
680
|
+
allowAbstain: boolean;
|
|
681
|
+
kind: "text";
|
|
682
|
+
maximumLength: number;
|
|
683
|
+
requireRationale?: boolean | undefined;
|
|
684
|
+
anchors?: {
|
|
685
|
+
value: string | number | boolean;
|
|
686
|
+
description: string;
|
|
687
|
+
}[] | undefined;
|
|
688
|
+
minimumLength?: number | undefined;
|
|
689
|
+
} | {
|
|
690
|
+
id: string;
|
|
691
|
+
label: string;
|
|
692
|
+
instructions: string;
|
|
693
|
+
required: boolean;
|
|
694
|
+
allowAbstain: boolean;
|
|
695
|
+
kind: "preference";
|
|
696
|
+
allowTie: boolean;
|
|
697
|
+
requireRationale?: boolean | undefined;
|
|
698
|
+
anchors?: {
|
|
699
|
+
value: string | number | boolean;
|
|
700
|
+
description: string;
|
|
701
|
+
}[] | undefined;
|
|
702
|
+
requireStrength?: boolean | undefined;
|
|
703
|
+
})[];
|
|
704
|
+
} | undefined;
|
|
557
705
|
})[];
|
|
558
706
|
metadata: Record<string, unknown>;
|
|
559
707
|
contentHash: string;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"conformance.d.ts","sourceRoot":"","sources":["../../../../src/conformance.ts"],"names":[],"mappings":"AAaA,eAAO,MAAM,sBAAsB
|
|
1
|
+
{"version":3,"file":"conformance.d.ts","sourceRoot":"","sources":["../../../../src/conformance.ts"],"names":[],"mappings":"AAaA,eAAO,MAAM,sBAAsB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAEjC,CAAC;AAEH,eAAO,MAAM,6BAA6B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAGxC,CAAC"}
|