openpond-sdk 0.6.7 → 0.8.0

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.
Files changed (98) hide show
  1. package/MODEL_STARTERS.md +25 -0
  2. package/TASKSETS.md +8 -0
  3. package/dist/{chunk-EJYSXE4V.js → chunk-32H2EJRA.js} +292 -456
  4. package/dist/chunk-32H2EJRA.js.map +7 -0
  5. package/dist/chunk-3HV3G6DG.js +95 -0
  6. package/dist/chunk-3HV3G6DG.js.map +7 -0
  7. package/dist/{chunk-FPOW5PU7.js → chunk-CGIMLZE5.js} +7 -7
  8. package/dist/chunk-DNBAJUVR.js +380 -0
  9. package/dist/chunk-DNBAJUVR.js.map +7 -0
  10. package/dist/{chunk-YGCHXNPI.js → chunk-DT7FRNIZ.js} +3 -94
  11. package/dist/{chunk-YGCHXNPI.js.map → chunk-DT7FRNIZ.js.map} +4 -4
  12. package/dist/{chunk-ECTMEJ5N.js → chunk-IKTZ2CD4.js} +63 -42
  13. package/dist/chunk-IKTZ2CD4.js.map +7 -0
  14. package/dist/{chunk-DNEPLQXN.js → chunk-KYCQ7LZ4.js} +5 -5
  15. package/dist/{chunk-TXQHWOFM.js → chunk-LGEF4S5I.js} +2 -2
  16. package/dist/chunk-QABCKPW6.js +171 -0
  17. package/dist/chunk-QABCKPW6.js.map +7 -0
  18. package/dist/{chunk-42YOB36J.js → chunk-WOPM32SE.js} +7 -7
  19. package/dist/{chunk-EAEYAU4R.js → chunk-Y4NGWGTA.js} +23 -2
  20. package/dist/{chunk-EAEYAU4R.js.map → chunk-Y4NGWGTA.js.map} +4 -4
  21. package/dist/dataset-workspaces.js +122 -0
  22. package/dist/dataset-workspaces.js.map +7 -0
  23. package/dist/index.js +20 -15
  24. package/dist/index.js.map +2 -2
  25. package/dist/learning.js +4 -4
  26. package/dist/model-starter-attempts.js +5 -1
  27. package/dist/model-starter-catalog.js +5 -5
  28. package/dist/model-starters.js +5 -5
  29. package/dist/model-taskset-runs.js +14 -11
  30. package/dist/model-taskset-runs.js.map +2 -2
  31. package/dist/taskset-drafts.js +10 -8
  32. package/dist/taskset-drafts.js.map +1 -1
  33. package/dist/taskset-packages.js +42 -364
  34. package/dist/taskset-packages.js.map +4 -4
  35. package/dist/training-bundle.js +10 -8
  36. package/dist/training-bundle.js.map +1 -1
  37. package/dist/training-projects.js +134 -0
  38. package/dist/training-projects.js.map +7 -0
  39. package/dist/training.js +3 -3
  40. package/dist/types/packages/cloud/src/hosted-chat.d.ts +1 -0
  41. package/dist/types/packages/cloud/src/hosted-chat.d.ts.map +1 -1
  42. package/dist/types/packages/sdk/src/dataset-workspace-client.d.ts +1631 -0
  43. package/dist/types/packages/sdk/src/dataset-workspace-client.d.ts.map +1 -0
  44. package/dist/types/packages/sdk/src/dataset-workspace-contracts.d.ts +1625 -0
  45. package/dist/types/packages/sdk/src/dataset-workspace-contracts.d.ts.map +1 -0
  46. package/dist/types/packages/sdk/src/dataset-workspaces.d.ts +3 -0
  47. package/dist/types/packages/sdk/src/dataset-workspaces.d.ts.map +1 -0
  48. package/dist/types/packages/sdk/src/evaluation-project-context.d.ts +19 -0
  49. package/dist/types/packages/sdk/src/evaluation-project-context.d.ts.map +1 -0
  50. package/dist/types/packages/sdk/src/managed-training-preparation.d.ts +2 -2
  51. package/dist/types/packages/sdk/src/model-batch-review-preparation.d.ts +3 -3
  52. package/dist/types/packages/sdk/src/model-batch-review.d.ts +7 -7
  53. package/dist/types/packages/sdk/src/model-learning-intent.d.ts +1 -1
  54. package/dist/types/packages/sdk/src/model-learning-policy.d.ts +1 -1
  55. package/dist/types/packages/sdk/src/model-projects.d.ts +122 -122
  56. package/dist/types/packages/sdk/src/model-starter-attempts.d.ts +144 -29
  57. package/dist/types/packages/sdk/src/model-starter-attempts.d.ts.map +1 -1
  58. package/dist/types/packages/sdk/src/model-starter-catalog.d.ts +23 -23
  59. package/dist/types/packages/sdk/src/model-starter-execution.d.ts +3 -3
  60. package/dist/types/packages/sdk/src/model-starters.d.ts +22 -22
  61. package/dist/types/packages/sdk/src/model-taskset-runs-contracts.d.ts +91 -16
  62. package/dist/types/packages/sdk/src/model-taskset-runs-contracts.d.ts.map +1 -1
  63. package/dist/types/packages/sdk/src/model-taskset-runs.d.ts +85 -15
  64. package/dist/types/packages/sdk/src/model-taskset-runs.d.ts.map +1 -1
  65. package/dist/types/packages/sdk/src/task-inventory.d.ts +3 -3
  66. package/dist/types/packages/sdk/src/taskset-authored-contracts.d.ts +51 -51
  67. package/dist/types/packages/sdk/src/taskset-authored-files.d.ts +30 -30
  68. package/dist/types/packages/sdk/src/taskset-draft-api-contracts.d.ts +86 -86
  69. package/dist/types/packages/sdk/src/taskset-draft-client.d.ts +183 -183
  70. package/dist/types/packages/sdk/src/taskset-draft-core.d.ts +42 -42
  71. package/dist/types/packages/sdk/src/taskset-draft-dataset-artifacts.d.ts +36 -36
  72. package/dist/types/packages/sdk/src/taskset-draft-dataset-sources.d.ts +34 -34
  73. package/dist/types/packages/sdk/src/taskset-draft-document.d.ts +43 -43
  74. package/dist/types/packages/sdk/src/taskset-draft-package-compiler.d.ts +1094 -48
  75. package/dist/types/packages/sdk/src/taskset-draft-package-compiler.d.ts.map +1 -1
  76. package/dist/types/packages/sdk/src/taskset-draft-source.d.ts +71 -71
  77. package/dist/types/packages/sdk/src/taskset-draft-workspace.d.ts +84 -84
  78. package/dist/types/packages/sdk/src/taskset-package-client.d.ts +671 -671
  79. package/dist/types/packages/sdk/src/taskset-package-contracts.d.ts +371 -371
  80. package/dist/types/packages/sdk/src/taskset-package-grader-projection.d.ts +1 -1
  81. package/dist/types/packages/sdk/src/taskset-package-learning.d.ts +5 -5
  82. package/dist/types/packages/sdk/src/taskset-package-projection.d.ts +87 -87
  83. package/dist/types/packages/sdk/src/training-evaluation-results.d.ts +1 -1
  84. package/dist/types/packages/sdk/src/training-learning-batch.d.ts +64 -64
  85. package/dist/types/packages/sdk/src/training-project-client.d.ts +401 -0
  86. package/dist/types/packages/sdk/src/training-project-client.d.ts.map +1 -0
  87. package/dist/types/packages/sdk/src/training-project-contracts.d.ts +617 -0
  88. package/dist/types/packages/sdk/src/training-project-contracts.d.ts.map +1 -0
  89. package/dist/types/packages/sdk/src/training-projects.d.ts +3 -0
  90. package/dist/types/packages/sdk/src/training-projects.d.ts.map +1 -0
  91. package/dist/types/packages/sdk/src/training.d.ts +24 -24
  92. package/package.json +10 -1
  93. package/dist/chunk-ECTMEJ5N.js.map +0 -7
  94. package/dist/chunk-EJYSXE4V.js.map +0 -7
  95. /package/dist/{chunk-FPOW5PU7.js.map → chunk-CGIMLZE5.js.map} +0 -0
  96. /package/dist/{chunk-DNEPLQXN.js.map → chunk-KYCQ7LZ4.js.map} +0 -0
  97. /package/dist/{chunk-TXQHWOFM.js.map → chunk-LGEF4S5I.js.map} +0 -0
  98. /package/dist/{chunk-42YOB36J.js.map → chunk-WOPM32SE.js.map} +0 -0
@@ -1,7 +1,7 @@
1
1
  {
2
2
  "version": 3,
3
- "sources": ["../src/taskset-authored-portable-release.ts", "../src/taskset-package-grader-projection.ts"],
4
- "sourcesContent": ["import type { Taskset } from \"./taskset-authored-contracts.js\";\nimport type {\n GraderSpec,\n TaskDataRecord,\n} from \"./taskset-draft-core.js\";\nimport {\n TaskRecordSchema,\n GraderSpecSchema as PortableGraderSpecSchema,\n TasksetReleaseContentSchema,\n TasksetReleaseSchema,\n bindTasksetExecutionReleases,\n createEnvironmentRelease,\n createVerifierSetRelease,\n EnvironmentReleaseSchema,\n VerifierSetReleaseSchema,\n verifyEnvironmentRelease,\n verifyVerifierSetRelease,\n verifyPreferenceComparisonRelease,\n type EnvironmentContract,\n type GraderSpec as PortableGraderSpec,\n type PreferenceComparisonRelease,\n type TasksetRelease,\n} from \"@openpond/evals\";\nimport type {\n ImmutableAssetRef,\n ToolDeclaration,\n} from \"@openpond/harness\";\nimport {\n CapabilityRequirementSchema,\n ImmutableAssetRefSchema,\n ToolDeclarationSchema,\n} from \"@openpond/harness\";\n\nimport { contentHash, sha256 } from \"@openpond/harness\";\nimport { TaskBatchPackageMetadataSchema } from \"@openpond/evals/learning\";\nimport { compileBoundGraders, RewardBindingSchema, RewardReleaseSchema, type RewardBinding, type RewardRelease } from \"@openpond/evals/rewards\";\nimport { portableDeterministicCheck } from \"@openpond/evals/graders\";\n\nexport type TasksetRewardExecution = { binding: RewardBinding; rewards: RewardRelease[] };\n\nexport function resolvePortableTasksetRewardExecution(taskset: Taskset, supplied?: TasksetRewardExecution): TasksetRewardExecution | null {\n const learning = taskset.metadata.learning === undefined ? null : TaskBatchPackageMetadataSchema.parse(taskset.metadata.learning);\n const embedded = taskset.metadata.rewardExecution as { binding?: unknown; rewards?: unknown } | undefined;\n const execution = supplied ?? (embedded ? {\n binding: RewardBindingSchema.parse(embedded.binding),\n rewards: RewardReleaseSchema.array().parse(embedded.rewards),\n } : learning);\n if (taskset.metadata.rewardBinding !== undefined) {\n const reference = taskset.metadata.rewardBinding as Partial<RewardBinding> | null;\n if (!execution || !reference || reference.id !== execution.binding.id\n || reference.revision !== execution.binding.revision || reference.contentHash !== execution.binding.contentHash) {\n throw new Error(\"Portable Taskset requires its exact published Reward binding.\");\n }\n }\n if (execution) compileBoundGraders(execution.binding, execution.rewards);\n return execution ? { binding: execution.binding, rewards: execution.rewards } : null;\n}\n\nexport function materializePortableTasksetRelease(input: {\n taskset: Taskset;\n selectedTasks?: TaskDataRecord[];\n adapterId: string;\n admittedTasksetRelease?: TasksetRelease | null;\n rewardExecution?: TasksetRewardExecution;\n}): {\n environmentRelease: ReturnType<typeof createEnvironmentRelease>;\n verifierSetRelease: ReturnType<typeof createVerifierSetRelease>;\n tasksetRelease: TasksetRelease;\n} {\n const environment = portableEnvironment(input.taskset);\n const tools = portableTools(input.taskset);\n const learning = input.taskset.metadata.learning === undefined ? null : TaskBatchPackageMetadataSchema.parse(input.taskset.metadata.learning);\n const embedded = input.taskset.metadata.rewardExecution;\n const rewardExecution = resolvePortableTasksetRewardExecution(input.taskset, input.rewardExecution);\n const graders = rewardExecution ? compileBoundGraders(rewardExecution.binding, rewardExecution.rewards) : input.taskset.graders.map(portableGrader);\n const tasks = input.selectedTasks?.length\n ? input.selectedTasks\n : input.taskset.tasks;\n if (!tasks.length) throw new Error(\"A portable Taskset release requires at least one task.\");\n const pinned = input.taskset.environment.metadata.portableExecutionResources as { environment?: unknown; verifierSet?: unknown } | undefined;\n if (environment.entrypoint === \"openpond.javascript-environment.v1\" && !pinned) throw new Error(\"Tool Taskset publication requires its pinned execution resources.\");\n const environmentRelease = pinned ? EnvironmentReleaseSchema.parse(pinned.environment) : createEnvironmentRelease({\n schemaVersion: \"openpond.environmentRelease.v1\",\n id: `environment-release-${input.taskset.id}-r${input.taskset.revision}`,\n revision: input.taskset.revision,\n contract: environment,\n actionSchemaRef: null,\n observationSchemaRef: null,\n stateSchemaRef: null,\n artifactCollection: {\n maxArtifacts: 100_000,\n maxTotalBytes: 250_000_000,\n },\n adapterConformanceHashes: {\n [input.adapterId]: contentHash({ adapterId: input.adapterId, environment, tools }),\n },\n metadata: {\n sourceTasksetId: input.taskset.id,\n sourcePackageHash: input.taskset.metadata.sourcePackageHash ?? null,\n resources: input.taskset.environment.resources ?? [],\n },\n });\n const verifierSetRelease = pinned?.verifierSet !== undefined ? VerifierSetReleaseSchema.parse(pinned.verifierSet) : createVerifierSetRelease({\n schemaVersion: \"openpond.verifierSetRelease.v1\",\n id: `verifier-set-release-${input.taskset.id}-r${input.taskset.revision}`,\n revision: input.taskset.revision,\n graders,\n isolation: {\n processBoundary: \"isolated_process\",\n networkPolicy: \"none\",\n defaultTimeoutMs: Math.min(input.taskset.environment.defaultTimeoutMs, 300_000),\n },\n calibrationReceiptRefs: [],\n metadata: { sourceTasksetId: input.taskset.id },\n });\n if (!verifyEnvironmentRelease(environmentRelease) || !verifyVerifierSetRelease(verifierSetRelease) ||\n contentHash(environmentRelease.contract) !== contentHash(environment) || contentHash(verifierSetRelease.graders) !== contentHash(graders)) {\n throw new Error(\"Taskset execution resources differ from its declared environment or graders.\");\n }\n const tasksetContent = TasksetReleaseContentSchema.parse({\n schemaVersion: \"openpond.tasksetRelease.v2\",\n id: input.taskset.metadata.derivedPortableMetadata === undefined ? `taskset-release-${input.taskset.id}-r${input.taskset.revision}` : input.taskset.id,\n revision: input.taskset.revision,\n policy: {\n policyVisibleFields: input.taskset.policy.policyVisibleFields,\n privilegedFields: input.taskset.policy.privilegedFields,\n hiddenGraderRefs: input.taskset.policy.hiddenGraderRefs,\n connectedAppScopes: input.taskset.policy.connectedAppScopes,\n },\n environment,\n tools,\n capabilities: portableCapabilities(input.taskset),\n tasks: tasks.map(projectPortableTaskRecord),\n graders,\n ...(input.taskset.metrics ? { metrics: input.taskset.metrics } : {}),\n metadata: input.taskset.metadata.derivedPortableMetadata ?? {\n sourceTasksetId: input.taskset.id,\n sourceTasksetHash: input.taskset.contentHash,\n sourcePackageHash: input.taskset.metadata.sourcePackageHash ?? null,\n environmentResources: input.taskset.environment.resources ?? [],\n ordinaryAuthoring: {\n instructions: input.taskset.objective,\n graderFixtures: input.taskset.graderFixtures,\n judgeCalibrationFixtures: Object.fromEntries(input.taskset.graders\n .filter(grader => grader.kind === \"model_judge\")\n .map(grader => [grader.id, grader.calibrationFixtureRefs])),\n },\n ...(learning ? { learning } : {}),\n // A sealed batch already pins these resources in learning metadata.\n // Resolving that same binding at admission must not change its release.\n ...(rewardExecution && !learning && (input.rewardExecution || embedded) ? { rewardExecution: { binding: rewardExecution.binding, rewards: rewardExecution.rewards } } : {}),\n },\n });\n const draft = input.admittedTasksetRelease\n ? TasksetReleaseSchema.parse(input.admittedTasksetRelease)\n : TasksetReleaseSchema.parse({\n ...tasksetContent,\n contentHash: contentHash(tasksetContent),\n });\n const tasksetRelease = draft.environmentRelease\n ? assertReleaseBindings({ draft, environmentRelease, verifierSetRelease })\n : bindTasksetExecutionReleases({\n taskset: draft,\n environment: environmentRelease,\n verifierSet: verifierSetRelease,\n });\n return { environmentRelease, verifierSetRelease, tasksetRelease };\n}\n\n/**\n * Admits a published comparison policy against the exact portable Taskset\n * release it names. A comparison is deliberately not embedded in that release:\n * doing so would create a content-hash cycle between Taskset and comparison.\n */\nexport function projectPreferenceComparisonRelease(input: {\n tasksetRelease: TasksetRelease;\n preferenceComparison: PreferenceComparisonRelease;\n}): PreferenceComparisonRelease {\n if (!verifyPreferenceComparisonRelease(input.preferenceComparison)) {\n throw new Error(\"Preference comparison release failed immutable content verification.\");\n }\n const release = input.preferenceComparison;\n if (\n release.tasksetRelease.id !== input.tasksetRelease.id\n || release.tasksetRelease.contentHash !== input.tasksetRelease.contentHash\n ) {\n throw new Error(\"Preference comparison release must reference the admitted portable Taskset release.\");\n }\n return release;\n}\n\nexport function portableTasksetEnvironment(taskset: Taskset): EnvironmentContract {\n return portableEnvironment(taskset);\n}\n\nexport function portableTasksetTools(taskset: Taskset): ToolDeclaration[] {\n return portableTools(taskset);\n}\n\n/** Preserve immutable file/schema references while overlaying current edits. */\nexport function projectPortableTaskRecord(task: TaskDataRecord) {\n const admitted = task.metadata.portableTaskRecord;\n if (admitted !== undefined) {\n const original = TaskRecordSchema.parse(admitted);\n const projected = TaskRecordSchema.parse({\n ...original,\n id: task.id,\n clusterKey: task.clusterKey,\n split: task.split,\n input: task.input,\n expectedOutput: portableExpectedOutput(task),\n policyVisibleContext: task.policyVisibleContext,\n privilegedContextRef: task.privilegedContextRef,\n tags: task.tags,\n artifactRefs: task.assets === undefined ? original.artifactRefs : task.assets.map((asset) => {\n const previous = original.artifactRefs.find(item => item.id === asset.id);\n return {\n id: asset.id,\n path: previous?.path ?? `tasks/${segment(task.id)}/${segment(asset.fileName)}`,\n contentHash: hash(asset.sha256),\n sizeBytes: asset.sizeBytes,\n mediaType: asset.mediaType,\n visibility: previous?.visibility ?? \"policy\",\n };\n }),\n requiredOutputs: task.requiredOutputs === undefined ? original.requiredOutputs : task.requiredOutputs.map((output) => {\n const previous = original.requiredOutputs?.find(item => item.path === output.path);\n return {\n path: output.path,\n mediaType: output.mediaType,\n schemaRef: previous?.schemaRef && previous.schemaRef.id === output.schemaRef ? previous.schemaRef : null,\n maxBytes: output.maxBytes ?? null,\n metadata: {\n ...output.metadata,\n ...(output.schemaRef && previous?.schemaRef?.id !== output.schemaRef\n ? { legacySchemaRef: output.schemaRef }\n : {}),\n },\n };\n }),\n });\n if (projected.requiredOutputs === undefined) delete projected.requiredOutputs;\n return projected;\n }\n return {\n id: task.id,\n clusterKey: task.clusterKey,\n split: task.split,\n input: task.input,\n expectedOutput: portableExpectedOutput(task),\n policyVisibleContext: task.policyVisibleContext,\n privilegedContextRef: task.privilegedContextRef,\n artifactRefs: (task.assets ?? []).map((item) => ({\n id: item.id,\n path: `tasks/${segment(task.id)}/${segment(item.fileName)}`,\n contentHash: hash(item.sha256),\n sizeBytes: item.sizeBytes,\n mediaType: item.mediaType,\n visibility: \"policy\" as const,\n })),\n requiredOutputs: (task.requiredOutputs ?? []).map((output) => ({\n path: output.path,\n mediaType: output.mediaType,\n schemaRef: null,\n maxBytes: output.maxBytes ?? null,\n metadata: {\n ...output.metadata,\n legacySchemaRef: output.schemaRef ?? null,\n },\n })),\n tags: task.tags,\n };\n}\n\nfunction portableExpectedOutput(\n task: TaskDataRecord,\n): Record<string, unknown> | null {\n if (!task.expectedOutput) return null;\n if (\"artifactRenderer\" in task.expectedOutput) {\n throw new Error(\n `Task ${task.id} embeds an artifact renderer. Put shared renderer configuration in Environment resources and reference it from the Scenario.`,\n );\n }\n return task.expectedOutput;\n}\n\nfunction portableEnvironment(taskset: Taskset): EnvironmentContract {\n const kind: EnvironmentContract[\"kind\"] = taskset.environment.kind === \"chat\"\n ? \"text\"\n : taskset.environment.kind === \"work\" ? \"work\"\n : taskset.environment.kind === \"program\" ? \"custom_program\" : \"agent\";\n return {\n protocolVersion: \"openpond.environment.v1\",\n kind,\n entrypoint: taskset.environment.entrypoint,\n stateful: taskset.environment.stateful,\n deterministicSeeds: taskset.environment.deterministicSeeds,\n lifecycle: [\"create\", \"reset\", \"step\", \"collect\", \"destroy\"],\n networkPolicy: taskset.environment.networkPolicy,\n defaultTimeoutMs: taskset.environment.defaultTimeoutMs,\n };\n}\n\nfunction portableTools(taskset: Taskset): ToolDeclaration[] {\n const admitted = taskset.environment.metadata.portableTools;\n if (admitted !== undefined) {\n return ToolDeclarationSchema.array().max(200).parse(admitted);\n }\n const bindings = taskset.environment.actionBindings ?? [];\n if (bindings.length) return bindings.map((binding) => ({\n name: binding.modelToolName,\n description: binding.description,\n inputSchema: binding.inputSchema,\n inputSchemaHash: hash(binding.actionSchemaHash),\n sideEffect: binding.sideEffect,\n timeoutMs: binding.timeoutMs,\n }));\n return taskset.environment.toolNames.map((name) => ({\n name,\n description: `Host-provided ${name} tool.`,\n inputSchema: {},\n inputSchemaHash: contentHash({}),\n sideEffect: \"write\" as const,\n timeoutMs: taskset.environment.defaultTimeoutMs,\n }));\n}\n\nfunction portableGrader(grader: GraderSpec): PortableGraderSpec {\n const base = {\n id: grader.id,\n version: grader.version,\n weight: grader.weight,\n hardGate: grader.hardGate,\n rewardEligible: grader.rewardEligible,\n privileged: grader.privileged,\n };\n if (grader.kind === \"model_judge\") return {\n ...base,\n kind: \"model_judge\",\n rubricRef: rubricAsset(grader),\n model: modelJudgeRef(grader),\n temperature: grader.temperature,\n calibrationStatus: grader.calibrationStatus,\n };\n if (grader.kind === \"custom_verifier\") {\n if (grader.metadata.portableVerifierRef === undefined) {\n throw new Error(`Custom verifier ${grader.id} requires its immutable executable asset before publication.`);\n }\n const imported = PortableGraderSpecSchema.safeParse(grader.metadata.portableGrader);\n // An omitted portable export means \"verify\". Preserve that encoding on an\n // unchanged import so executing it does not silently replace its release.\n const preserveDefault = imported.success && imported.data.kind === \"custom_verifier\"\n && imported.data.exportName === undefined && grader.exportName === \"verify\";\n return {\n ...base,\n kind: \"custom_verifier\",\n ...(grader.runtime === undefined ? {} : { runtime: grader.runtime }),\n verifierRef: ImmutableAssetRefSchema.parse(grader.metadata.portableVerifierRef),\n timeoutMs: grader.timeoutMs,\n ...(preserveDefault ? {} : { exportName: grader.exportName }),\n networkPolicy: \"none\",\n };\n }\n if (grader.kind === \"human\") return {\n ...base,\n kind: \"human\",\n rubricRef: rubricAsset(grader),\n reviewerRole: grader.reviewerRole,\n };\n return {\n ...base,\n ...portableDeterministicCheck(grader),\n };\n}\n\nfunction portableCapabilities(taskset: Taskset) {\n const admitted = taskset.metadata.portableCapabilities;\n if (admitted !== undefined) {\n return CapabilityRequirementSchema.array().max(200).parse(admitted);\n }\n return [\n ...(taskset.capabilities.requiresTools\n ? [{ id: \"tools\", required: true, scopes: taskset.environment.toolNames }]\n : []),\n ...(taskset.capabilities.requiresState\n ? [{ id: \"local-state\", required: true, scopes: [] }]\n : []),\n ...(taskset.capabilities.requiresPrivilegedGrading\n ? [{ id: \"private-verifier\", required: true, scopes: taskset.policy.hiddenGraderRefs }]\n : []),\n ...taskset.policy.connectedAppScopes.map((scope) => ({\n id: `connected-app-${segment(scope)}`,\n required: true,\n scopes: [scope],\n })),\n ].map((requirement) => ({\n ...requirement,\n portability: requirement.id === \"local-state\" ? \"host_adapter\" as const : \"portable\" as const,\n }));\n}\n\nfunction assertReleaseBindings(input: {\n draft: TasksetRelease;\n environmentRelease: ReturnType<typeof createEnvironmentRelease>;\n verifierSetRelease: ReturnType<typeof createVerifierSetRelease>;\n}): TasksetRelease {\n if (\n input.draft.environmentRelease?.id !== input.environmentRelease.id\n || input.draft.environmentRelease.contentHash !== input.environmentRelease.contentHash\n ) {\n throw new Error(\"Admitted Taskset Release binds a different Environment Release.\");\n }\n if (\n input.draft.verifierSetRelease?.id !== input.verifierSetRelease.id\n || input.draft.verifierSetRelease.contentHash !== input.verifierSetRelease.contentHash\n ) {\n throw new Error(\"Admitted Taskset Release binds a different Verifier Set Release.\");\n }\n if (contentHash(input.draft.environment) !== contentHash(input.environmentRelease.contract)) {\n throw new Error(\"Admitted Taskset Environment differs from its bound Environment Release.\");\n }\n if (contentHash(input.draft.graders) !== contentHash(input.verifierSetRelease.graders)) {\n throw new Error(\"Admitted Taskset graders differ from its bound Verifier Set Release.\");\n }\n return input.draft;\n}\n\nfunction modelJudgeRef(grader: Extract<GraderSpec, { kind: \"model_judge\" }>) {\n const previous = grader.metadata.portableGrader === undefined ? null : PortableGraderSpecSchema.parse(grader.metadata.portableGrader);\n const revision = previous?.kind === \"model_judge\" && previous.model?.providerId === grader.judge.providerId && previous.model.modelId === grader.judge.modelId ? previous.model.revision : null;\n return { ...grader.judge, revision };\n}\n\nfunction rubricAsset(grader: Extract<GraderSpec, { kind: \"human\" | \"model_judge\" }>): ImmutableAssetRef {\n const contentHash = sha256(grader.rubric);\n const sizeBytes = new TextEncoder().encode(grader.rubric).byteLength;\n const reference = grader.metadata.portableRubricRef;\n if (reference !== undefined) {\n const asset = ImmutableAssetRefSchema.parse(reference);\n if (asset.contentHash !== contentHash || asset.sizeBytes !== sizeBytes || asset.visibility === \"policy\") throw new Error(\"Taskset rubric differs from its published private source.\");\n return asset;\n }\n return { id: `rubric-${grader.id}`, path: `graders/${segment(grader.id)}/rubric.md`, contentHash, sizeBytes, mediaType: \"text/markdown\", visibility: \"verifier\" };\n}\n\nfunction hash(value: string): string {\n return /^[a-f0-9]{64}$/.test(value) ? value : contentHash(value);\n}\n\nfunction segment(value: string): string {\n const normalized = value.trim().replace(/[^a-zA-Z0-9._-]+/g, \"-\")\n .replace(/^-+|-+$/g, \"\");\n return normalized || \"asset\";\n}\n", "import { compileBoundGraders, resolveBoundRewards, type RewardBinding, type RewardRelease } from \"@openpond/evals/rewards\";\nimport { learningRef, verifyLearningTextAsset, RewardFixtureAuthoringFieldsSchema, type LearningTextAsset } from \"@openpond/evals/learning\";\nimport type { GraderSpec } from \"./taskset-draft-core.js\";\nimport { GraderSpecSchema } from \"./taskset-draft-core.js\";\nimport { decodeTasksetPackageFile, type TasksetPackage } from \"./taskset-package-contracts.js\";\nimport { z } from \"zod\";\n\n/** Project an ordinary portable package without inventing a Reward binding. */\nexport function importedPackageGraders(value: TasksetPackage) {\n const authoring = value.taskset.metadata.ordinaryAuthoring as { judgeCalibrationFixtures?: unknown } | undefined;\n const calibrationFixtures = z.record(z.string(), z.array(z.string().min(1)).max(100_000)).parse(authoring?.judgeCalibrationFixtures ?? {});\n return value.taskset.graders.map(grader => {\n const base = { ...grader, label: grader.id, metadata: { portableGrader: grader } };\n if (grader.kind === \"custom_verifier\") return GraderSpecSchema.parse({ ...base,\n module: learningVerifierModule(grader.verifierRef.contentHash), exportName: grader.exportName ?? \"verify\",\n metadata: { ...base.metadata, portableVerifierRef: grader.verifierRef } });\n if (grader.kind === \"human\" || grader.kind === \"model_judge\") {\n const file = value.files.find(file => file.asset.id === grader.rubricRef.id);\n if (!file) throw new Error(`Imported rubric is missing: ${grader.id}.`);\n const rubric = new TextDecoder(\"utf-8\", { fatal: true }).decode(decodeTasksetPackageFile(file));\n const metadata = { ...base.metadata, portableRubricRef: grader.rubricRef };\n if (grader.kind === \"human\") return GraderSpecSchema.parse({ ...base, rubric, metadata });\n if (!grader.model) throw new Error(`Imported judge ${grader.id} requires its declared model before local preparation.`);\n return GraderSpecSchema.parse({ ...base, rubric, metadata, judge: grader.model, temperature: grader.temperature ?? 0,\n calibrationStatus: \"pending\", calibrationFixtureRefs: calibrationFixtures[grader.id] ?? [] });\n }\n return GraderSpecSchema.parse({ ...base, kind: grader.kind === \"artifact\" ? \"file\" : grader.kind,\n config: grader.kind === \"artifact\" ? { ...grader.config, ...(grader.config.refIncludes === undefined ? {} : { pathIncludes: grader.config.refIncludes }) } : grader.config });\n });\n}\n\nexport function projectLearningBatchGraders(binding: RewardBinding, rewards: RewardRelease[], assets: LearningTextAsset[] = []): GraderSpec[] {\n const resolved = resolveBoundRewards(binding, rewards);\n return compileBoundGraders(binding, rewards).map((grader) => {\n const reward = resolved.find(({ source }) => source.graderId === grader.id)!.reward;\n const base = { ...grader, label: reward.name, metadata: { rewardBinding: learningRef(binding) } };\n if (grader.kind === \"model_judge\") {\n if (!grader.model || grader.calibrationStatus !== \"passed\" || !reward.calibrationCheckRef || !reward.fixtureSetRef) throw new Error(`Reward ${grader.id} needs a retained passing calibration check before preparation.`);\n const rubric = assets.find(asset => asset.id === grader.rubricRef.id);\n const fixtures = assets.find(asset => asset.id === reward.fixtureSetRef!.id);\n if (!rubric || !fixtures) throw new Error(`Reward ${grader.id} is missing its immutable rubric or calibration fixtures.`);\n const fields = RewardFixtureAuthoringFieldsSchema.array().min(1).max(50).parse(JSON.parse(verifyLearningTextAsset(fixtures, reward.fixtureSetRef)));\n return GraderSpecSchema.parse({ ...base, rubric: verifyLearningTextAsset(rubric, grader.rubricRef), judge: grader.model,\n temperature: grader.temperature ?? 0, calibrationFixtureRefs: fields.map(fixture => fixture.id),\n metadata: { ...base.metadata, portableRubricRef: grader.rubricRef,\n calibrationEvidenceHash: reward.calibrationCheckRef.contentHash, calibrationCheckRef: reward.calibrationCheckRef },\n });\n }\n if (grader.kind === \"custom_verifier\") return GraderSpecSchema.parse({ ...base,\n module: learningVerifierModule(grader.verifierRef.contentHash), exportName: grader.exportName ?? \"verify\",\n metadata: { ...base.metadata, portableVerifierRef: grader.verifierRef },\n });\n if (grader.kind === \"human\") {\n const asset = assets.find((asset) => asset.id === grader.rubricRef.id);\n if (!asset) throw new Error(`Reward ${grader.id} is missing its immutable rubric.`);\n return GraderSpecSchema.parse({ ...base, rubric: verifyLearningTextAsset(asset, grader.rubricRef) });\n }\n return GraderSpecSchema.parse({ ...base, kind: grader.kind === \"artifact\" ? \"file\" : grader.kind });\n });\n}\n\nexport function learningVerifierModule(hash: string) { return `graders/reward-${hash}.js`; }\n"],
5
- "mappings": ";;;;;;;;;;;;;;;;;AAKA;AAAA,EACE;AAAA,EACA,oBAAoB;AAAA,EACpB;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OAKK;AAYP,SAAS,sCAAsC;AAC/C,SAAS,qBAAqB,qBAAqB,2BAAmE;AACtH,SAAS,kCAAkC;AAIpC,SAAS,sCAAsC,SAAkB,UAAkE;AACxI,QAAM,WAAW,QAAQ,SAAS,aAAa,SAAY,OAAO,+BAA+B,MAAM,QAAQ,SAAS,QAAQ;AAChI,QAAM,WAAW,QAAQ,SAAS;AAClC,QAAM,YAAY,aAAa,WAAW;AAAA,IACxC,SAAS,oBAAoB,MAAM,SAAS,OAAO;AAAA,IACnD,SAAS,oBAAoB,MAAM,EAAE,MAAM,SAAS,OAAO;AAAA,EAC7D,IAAI;AACJ,MAAI,QAAQ,SAAS,kBAAkB,QAAW;AAChD,UAAM,YAAY,QAAQ,SAAS;AACnC,QAAI,CAAC,aAAa,CAAC,aAAa,UAAU,OAAO,UAAU,QAAQ,MAC9D,UAAU,aAAa,UAAU,QAAQ,YAAY,UAAU,gBAAgB,UAAU,QAAQ,aAAa;AACjH,YAAM,IAAI,MAAM,+DAA+D;AAAA,IACjF;AAAA,EACF;AACA,MAAI,UAAW,qBAAoB,UAAU,SAAS,UAAU,OAAO;AACvE,SAAO,YAAY,EAAE,SAAS,UAAU,SAAS,SAAS,UAAU,QAAQ,IAAI;AAClF;AAEO,SAAS,kCAAkC,OAUhD;AACA,QAAM,cAAc,oBAAoB,MAAM,OAAO;AACrD,QAAM,QAAQ,cAAc,MAAM,OAAO;AACzC,QAAM,WAAW,MAAM,QAAQ,SAAS,aAAa,SAAY,OAAO,+BAA+B,MAAM,MAAM,QAAQ,SAAS,QAAQ;AAC5I,QAAM,WAAW,MAAM,QAAQ,SAAS;AACxC,QAAM,kBAAkB,sCAAsC,MAAM,SAAS,MAAM,eAAe;AAClG,QAAM,UAAU,kBAAkB,oBAAoB,gBAAgB,SAAS,gBAAgB,OAAO,IAAI,MAAM,QAAQ,QAAQ,IAAI,cAAc;AAClJ,QAAM,QAAQ,MAAM,eAAe,SAC/B,MAAM,gBACN,MAAM,QAAQ;AAClB,MAAI,CAAC,MAAM,OAAQ,OAAM,IAAI,MAAM,wDAAwD;AAC3F,QAAM,SAAS,MAAM,QAAQ,YAAY,SAAS;AAClD,MAAI,YAAY,eAAe,wCAAwC,CAAC,OAAQ,OAAM,IAAI,MAAM,mEAAmE;AACnK,QAAM,qBAAqB,SAAS,yBAAyB,MAAM,OAAO,WAAW,IAAI,yBAAyB;AAAA,IAChH,eAAe;AAAA,IACf,IAAI,uBAAuB,MAAM,QAAQ,EAAE,KAAK,MAAM,QAAQ,QAAQ;AAAA,IACtE,UAAU,MAAM,QAAQ;AAAA,IACxB,UAAU;AAAA,IACV,iBAAiB;AAAA,IACjB,sBAAsB;AAAA,IACtB,gBAAgB;AAAA,IAChB,oBAAoB;AAAA,MAClB,cAAc;AAAA,MACd,eAAe;AAAA,IACjB;AAAA,IACA,0BAA0B;AAAA,MACxB,CAAC,MAAM,SAAS,GAAG,YAAY,EAAE,WAAW,MAAM,WAAW,aAAa,MAAM,CAAC;AAAA,IACnF;AAAA,IACA,UAAU;AAAA,MACR,iBAAiB,MAAM,QAAQ;AAAA,MAC/B,mBAAmB,MAAM,QAAQ,SAAS,qBAAqB;AAAA,MAC/D,WAAW,MAAM,QAAQ,YAAY,aAAa,CAAC;AAAA,IACrD;AAAA,EACF,CAAC;AACD,QAAM,qBAAqB,QAAQ,gBAAgB,SAAY,yBAAyB,MAAM,OAAO,WAAW,IAAI,yBAAyB;AAAA,IAC3I,eAAe;AAAA,IACf,IAAI,wBAAwB,MAAM,QAAQ,EAAE,KAAK,MAAM,QAAQ,QAAQ;AAAA,IACvE,UAAU,MAAM,QAAQ;AAAA,IACxB;AAAA,IACA,WAAW;AAAA,MACT,iBAAiB;AAAA,MACjB,eAAe;AAAA,MACf,kBAAkB,KAAK,IAAI,MAAM,QAAQ,YAAY,kBAAkB,GAAO;AAAA,IAChF;AAAA,IACA,wBAAwB,CAAC;AAAA,IACzB,UAAU,EAAE,iBAAiB,MAAM,QAAQ,GAAG;AAAA,EAChD,CAAC;AACD,MAAI,CAAC,yBAAyB,kBAAkB,KAAK,CAAC,yBAAyB,kBAAkB,KAC7F,YAAY,mBAAmB,QAAQ,MAAM,YAAY,WAAW,KAAK,YAAY,mBAAmB,OAAO,MAAM,YAAY,OAAO,GAAG;AAC7I,UAAM,IAAI,MAAM,8EAA8E;AAAA,EAChG;AACA,QAAM,iBAAiB,4BAA4B,MAAM;AAAA,IACvD,eAAe;AAAA,IACf,IAAI,MAAM,QAAQ,SAAS,4BAA4B,SAAY,mBAAmB,MAAM,QAAQ,EAAE,KAAK,MAAM,QAAQ,QAAQ,KAAK,MAAM,QAAQ;AAAA,IACpJ,UAAU,MAAM,QAAQ;AAAA,IACxB,QAAQ;AAAA,MACN,qBAAqB,MAAM,QAAQ,OAAO;AAAA,MAC1C,kBAAkB,MAAM,QAAQ,OAAO;AAAA,MACvC,kBAAkB,MAAM,QAAQ,OAAO;AAAA,MACvC,oBAAoB,MAAM,QAAQ,OAAO;AAAA,IAC3C;AAAA,IACA;AAAA,IACA;AAAA,IACA,cAAc,qBAAqB,MAAM,OAAO;AAAA,IAChD,OAAO,MAAM,IAAI,yBAAyB;AAAA,IAC1C;AAAA,IACA,GAAI,MAAM,QAAQ,UAAU,EAAE,SAAS,MAAM,QAAQ,QAAQ,IAAI,CAAC;AAAA,IAClE,UAAU,MAAM,QAAQ,SAAS,2BAA2B;AAAA,MAC1D,iBAAiB,MAAM,QAAQ;AAAA,MAC/B,mBAAmB,MAAM,QAAQ;AAAA,MACjC,mBAAmB,MAAM,QAAQ,SAAS,qBAAqB;AAAA,MAC/D,sBAAsB,MAAM,QAAQ,YAAY,aAAa,CAAC;AAAA,MAC9D,mBAAmB;AAAA,QACjB,cAAc,MAAM,QAAQ;AAAA,QAC5B,gBAAgB,MAAM,QAAQ;AAAA,QAC9B,0BAA0B,OAAO,YAAY,MAAM,QAAQ,QACxD,OAAO,YAAU,OAAO,SAAS,aAAa,EAC9C,IAAI,YAAU,CAAC,OAAO,IAAI,OAAO,sBAAsB,CAAC,CAAC;AAAA,MAC9D;AAAA,MACA,GAAI,WAAW,EAAE,SAAS,IAAI,CAAC;AAAA;AAAA;AAAA,MAG/B,GAAI,mBAAmB,CAAC,aAAa,MAAM,mBAAmB,YAAY,EAAE,iBAAiB,EAAE,SAAS,gBAAgB,SAAS,SAAS,gBAAgB,QAAQ,EAAE,IAAI,CAAC;AAAA,IAC3K;AAAA,EACF,CAAC;AACD,QAAM,QAAQ,MAAM,yBAChB,qBAAqB,MAAM,MAAM,sBAAsB,IACvD,qBAAqB,MAAM;AAAA,IAC3B,GAAG;AAAA,IACH,aAAa,YAAY,cAAc;AAAA,EACzC,CAAC;AACH,QAAM,iBAAiB,MAAM,qBACzB,sBAAsB,EAAE,OAAO,oBAAoB,mBAAmB,CAAC,IACvE,6BAA6B;AAAA,IAC7B,SAAS;AAAA,IACT,aAAa;AAAA,IACb,aAAa;AAAA,EACf,CAAC;AACH,SAAO,EAAE,oBAAoB,oBAAoB,eAAe;AAClE;AAOO,SAAS,mCAAmC,OAGnB;AAC9B,MAAI,CAAC,kCAAkC,MAAM,oBAAoB,GAAG;AAClE,UAAM,IAAI,MAAM,sEAAsE;AAAA,EACxF;AACA,QAAM,UAAU,MAAM;AACtB,MACE,QAAQ,eAAe,OAAO,MAAM,eAAe,MAChD,QAAQ,eAAe,gBAAgB,MAAM,eAAe,aAC/D;AACA,UAAM,IAAI,MAAM,qFAAqF;AAAA,EACvG;AACA,SAAO;AACT;AAEO,SAAS,2BAA2B,SAAuC;AAChF,SAAO,oBAAoB,OAAO;AACpC;AAEO,SAAS,qBAAqB,SAAqC;AACxE,SAAO,cAAc,OAAO;AAC9B;AAGO,SAAS,0BAA0B,MAAsB;AAC9D,QAAM,WAAW,KAAK,SAAS;AAC/B,MAAI,aAAa,QAAW;AAC1B,UAAM,WAAW,iBAAiB,MAAM,QAAQ;AAChD,UAAM,YAAY,iBAAiB,MAAM;AAAA,MACvC,GAAG;AAAA,MACH,IAAI,KAAK;AAAA,MACT,YAAY,KAAK;AAAA,MACjB,OAAO,KAAK;AAAA,MACZ,OAAO,KAAK;AAAA,MACZ,gBAAgB,uBAAuB,IAAI;AAAA,MAC3C,sBAAsB,KAAK;AAAA,MAC3B,sBAAsB,KAAK;AAAA,MAC3B,MAAM,KAAK;AAAA,MACX,cAAc,KAAK,WAAW,SAAY,SAAS,eAAe,KAAK,OAAO,IAAI,CAAC,UAAU;AAC3F,cAAM,WAAW,SAAS,aAAa,KAAK,UAAQ,KAAK,OAAO,MAAM,EAAE;AACxE,eAAO;AAAA,UACL,IAAI,MAAM;AAAA,UACV,MAAM,UAAU,QAAQ,SAAS,QAAQ,KAAK,EAAE,CAAC,IAAI,QAAQ,MAAM,QAAQ,CAAC;AAAA,UAC5E,aAAa,KAAK,MAAM,MAAM;AAAA,UAC9B,WAAW,MAAM;AAAA,UACjB,WAAW,MAAM;AAAA,UACjB,YAAY,UAAU,cAAc;AAAA,QACtC;AAAA,MACF,CAAC;AAAA,MACD,iBAAiB,KAAK,oBAAoB,SAAY,SAAS,kBAAkB,KAAK,gBAAgB,IAAI,CAAC,WAAW;AACpH,cAAM,WAAW,SAAS,iBAAiB,KAAK,UAAQ,KAAK,SAAS,OAAO,IAAI;AACjF,eAAO;AAAA,UACL,MAAM,OAAO;AAAA,UACb,WAAW,OAAO;AAAA,UAClB,WAAW,UAAU,aAAa,SAAS,UAAU,OAAO,OAAO,YAAY,SAAS,YAAY;AAAA,UACpG,UAAU,OAAO,YAAY;AAAA,UAC7B,UAAU;AAAA,YACR,GAAG,OAAO;AAAA,YACV,GAAI,OAAO,aAAa,UAAU,WAAW,OAAO,OAAO,YACvD,EAAE,iBAAiB,OAAO,UAAU,IACpC,CAAC;AAAA,UACP;AAAA,QACF;AAAA,MACF,CAAC;AAAA,IACH,CAAC;AACD,QAAI,UAAU,oBAAoB,OAAW,QAAO,UAAU;AAC9D,WAAO;AAAA,EACT;AACA,SAAO;AAAA,IACL,IAAI,KAAK;AAAA,IACT,YAAY,KAAK;AAAA,IACjB,OAAO,KAAK;AAAA,IACZ,OAAO,KAAK;AAAA,IACZ,gBAAgB,uBAAuB,IAAI;AAAA,IAC3C,sBAAsB,KAAK;AAAA,IAC3B,sBAAsB,KAAK;AAAA,IAC3B,eAAe,KAAK,UAAU,CAAC,GAAG,IAAI,CAAC,UAAU;AAAA,MAC/C,IAAI,KAAK;AAAA,MACT,MAAM,SAAS,QAAQ,KAAK,EAAE,CAAC,IAAI,QAAQ,KAAK,QAAQ,CAAC;AAAA,MACzD,aAAa,KAAK,KAAK,MAAM;AAAA,MAC7B,WAAW,KAAK;AAAA,MAChB,WAAW,KAAK;AAAA,MAChB,YAAY;AAAA,IACd,EAAE;AAAA,IACF,kBAAkB,KAAK,mBAAmB,CAAC,GAAG,IAAI,CAAC,YAAY;AAAA,MAC7D,MAAM,OAAO;AAAA,MACb,WAAW,OAAO;AAAA,MAClB,WAAW;AAAA,MACX,UAAU,OAAO,YAAY;AAAA,MAC7B,UAAU;AAAA,QACR,GAAG,OAAO;AAAA,QACV,iBAAiB,OAAO,aAAa;AAAA,MACvC;AAAA,IACF,EAAE;AAAA,IACF,MAAM,KAAK;AAAA,EACb;AACF;AAEA,SAAS,uBACP,MACgC;AAChC,MAAI,CAAC,KAAK,eAAgB,QAAO;AACjC,MAAI,sBAAsB,KAAK,gBAAgB;AAC7C,UAAM,IAAI;AAAA,MACR,QAAQ,KAAK,EAAE;AAAA,IACjB;AAAA,EACF;AACA,SAAO,KAAK;AACd;AAEA,SAAS,oBAAoB,SAAuC;AAClE,QAAM,OAAoC,QAAQ,YAAY,SAAS,SACnE,SACA,QAAQ,YAAY,SAAS,SAAS,SACpC,QAAQ,YAAY,SAAS,YAAY,mBAAmB;AAClE,SAAO;AAAA,IACL,iBAAiB;AAAA,IACjB;AAAA,IACA,YAAY,QAAQ,YAAY;AAAA,IAChC,UAAU,QAAQ,YAAY;AAAA,IAC9B,oBAAoB,QAAQ,YAAY;AAAA,IACxC,WAAW,CAAC,UAAU,SAAS,QAAQ,WAAW,SAAS;AAAA,IAC3D,eAAe,QAAQ,YAAY;AAAA,IACnC,kBAAkB,QAAQ,YAAY;AAAA,EACxC;AACF;AAEA,SAAS,cAAc,SAAqC;AAC1D,QAAM,WAAW,QAAQ,YAAY,SAAS;AAC9C,MAAI,aAAa,QAAW;AAC1B,WAAO,sBAAsB,MAAM,EAAE,IAAI,GAAG,EAAE,MAAM,QAAQ;AAAA,EAC9D;AACA,QAAM,WAAW,QAAQ,YAAY,kBAAkB,CAAC;AACxD,MAAI,SAAS,OAAQ,QAAO,SAAS,IAAI,CAAC,aAAa;AAAA,IACrD,MAAM,QAAQ;AAAA,IACd,aAAa,QAAQ;AAAA,IACrB,aAAa,QAAQ;AAAA,IACrB,iBAAiB,KAAK,QAAQ,gBAAgB;AAAA,IAC9C,YAAY,QAAQ;AAAA,IACpB,WAAW,QAAQ;AAAA,EACrB,EAAE;AACF,SAAO,QAAQ,YAAY,UAAU,IAAI,CAAC,UAAU;AAAA,IAClD;AAAA,IACA,aAAa,iBAAiB,IAAI;AAAA,IAClC,aAAa,CAAC;AAAA,IACd,iBAAiB,YAAY,CAAC,CAAC;AAAA,IAC/B,YAAY;AAAA,IACZ,WAAW,QAAQ,YAAY;AAAA,EACjC,EAAE;AACJ;AAEA,SAAS,eAAe,QAAwC;AAC9D,QAAM,OAAO;AAAA,IACX,IAAI,OAAO;AAAA,IACX,SAAS,OAAO;AAAA,IAChB,QAAQ,OAAO;AAAA,IACf,UAAU,OAAO;AAAA,IACjB,gBAAgB,OAAO;AAAA,IACvB,YAAY,OAAO;AAAA,EACrB;AACA,MAAI,OAAO,SAAS,cAAe,QAAO;AAAA,IACxC,GAAG;AAAA,IACH,MAAM;AAAA,IACN,WAAW,YAAY,MAAM;AAAA,IAC7B,OAAO,cAAc,MAAM;AAAA,IAC3B,aAAa,OAAO;AAAA,IACpB,mBAAmB,OAAO;AAAA,EAC5B;AACA,MAAI,OAAO,SAAS,mBAAmB;AACrC,QAAI,OAAO,SAAS,wBAAwB,QAAW;AACrD,YAAM,IAAI,MAAM,mBAAmB,OAAO,EAAE,8DAA8D;AAAA,IAC5G;AACA,UAAM,WAAW,yBAAyB,UAAU,OAAO,SAAS,cAAc;AAGlF,UAAM,kBAAkB,SAAS,WAAW,SAAS,KAAK,SAAS,qBAC9D,SAAS,KAAK,eAAe,UAAa,OAAO,eAAe;AACrE,WAAO;AAAA,MACL,GAAG;AAAA,MACH,MAAM;AAAA,MACN,GAAI,OAAO,YAAY,SAAY,CAAC,IAAI,EAAE,SAAS,OAAO,QAAQ;AAAA,MAClE,aAAa,wBAAwB,MAAM,OAAO,SAAS,mBAAmB;AAAA,MAC9E,WAAW,OAAO;AAAA,MAClB,GAAI,kBAAkB,CAAC,IAAI,EAAE,YAAY,OAAO,WAAW;AAAA,MAC3D,eAAe;AAAA,IACjB;AAAA,EACF;AACA,MAAI,OAAO,SAAS,QAAS,QAAO;AAAA,IAClC,GAAG;AAAA,IACH,MAAM;AAAA,IACN,WAAW,YAAY,MAAM;AAAA,IAC7B,cAAc,OAAO;AAAA,EACvB;AACA,SAAO;AAAA,IACL,GAAG;AAAA,IACH,GAAG,2BAA2B,MAAM;AAAA,EACtC;AACF;AAEA,SAAS,qBAAqB,SAAkB;AAC9C,QAAM,WAAW,QAAQ,SAAS;AAClC,MAAI,aAAa,QAAW;AAC1B,WAAO,4BAA4B,MAAM,EAAE,IAAI,GAAG,EAAE,MAAM,QAAQ;AAAA,EACpE;AACA,SAAO;AAAA,IACL,GAAI,QAAQ,aAAa,gBACrB,CAAC,EAAE,IAAI,SAAS,UAAU,MAAM,QAAQ,QAAQ,YAAY,UAAU,CAAC,IACvE,CAAC;AAAA,IACL,GAAI,QAAQ,aAAa,gBACrB,CAAC,EAAE,IAAI,eAAe,UAAU,MAAM,QAAQ,CAAC,EAAE,CAAC,IAClD,CAAC;AAAA,IACL,GAAI,QAAQ,aAAa,4BACrB,CAAC,EAAE,IAAI,oBAAoB,UAAU,MAAM,QAAQ,QAAQ,OAAO,iBAAiB,CAAC,IACpF,CAAC;AAAA,IACL,GAAG,QAAQ,OAAO,mBAAmB,IAAI,CAAC,WAAW;AAAA,MACnD,IAAI,iBAAiB,QAAQ,KAAK,CAAC;AAAA,MACnC,UAAU;AAAA,MACV,QAAQ,CAAC,KAAK;AAAA,IAChB,EAAE;AAAA,EACJ,EAAE,IAAI,CAAC,iBAAiB;AAAA,IACtB,GAAG;AAAA,IACH,aAAa,YAAY,OAAO,gBAAgB,iBAA0B;AAAA,EAC5E,EAAE;AACJ;AAEA,SAAS,sBAAsB,OAIZ;AACjB,MACE,MAAM,MAAM,oBAAoB,OAAO,MAAM,mBAAmB,MAC7D,MAAM,MAAM,mBAAmB,gBAAgB,MAAM,mBAAmB,aAC3E;AACA,UAAM,IAAI,MAAM,iEAAiE;AAAA,EACnF;AACA,MACE,MAAM,MAAM,oBAAoB,OAAO,MAAM,mBAAmB,MAC7D,MAAM,MAAM,mBAAmB,gBAAgB,MAAM,mBAAmB,aAC3E;AACA,UAAM,IAAI,MAAM,kEAAkE;AAAA,EACpF;AACA,MAAI,YAAY,MAAM,MAAM,WAAW,MAAM,YAAY,MAAM,mBAAmB,QAAQ,GAAG;AAC3F,UAAM,IAAI,MAAM,0EAA0E;AAAA,EAC5F;AACA,MAAI,YAAY,MAAM,MAAM,OAAO,MAAM,YAAY,MAAM,mBAAmB,OAAO,GAAG;AACtF,UAAM,IAAI,MAAM,sEAAsE;AAAA,EACxF;AACA,SAAO,MAAM;AACf;AAEA,SAAS,cAAc,QAAsD;AAC3E,QAAM,WAAW,OAAO,SAAS,mBAAmB,SAAY,OAAO,yBAAyB,MAAM,OAAO,SAAS,cAAc;AACpI,QAAM,WAAW,UAAU,SAAS,iBAAiB,SAAS,OAAO,eAAe,OAAO,MAAM,cAAc,SAAS,MAAM,YAAY,OAAO,MAAM,UAAU,SAAS,MAAM,WAAW;AAC3L,SAAO,EAAE,GAAG,OAAO,OAAO,SAAS;AACrC;AAEA,SAAS,YAAY,QAAmF;AACtG,QAAMA,eAAc,OAAO,OAAO,MAAM;AACxC,QAAM,YAAY,IAAI,YAAY,EAAE,OAAO,OAAO,MAAM,EAAE;AAC1D,QAAM,YAAY,OAAO,SAAS;AAClC,MAAI,cAAc,QAAW;AAC3B,UAAM,QAAQ,wBAAwB,MAAM,SAAS;AACrD,QAAI,MAAM,gBAAgBA,gBAAe,MAAM,cAAc,aAAa,MAAM,eAAe,SAAU,OAAM,IAAI,MAAM,2DAA2D;AACpL,WAAO;AAAA,EACT;AACA,SAAO,EAAE,IAAI,UAAU,OAAO,EAAE,IAAI,MAAM,WAAW,QAAQ,OAAO,EAAE,CAAC,cAAc,aAAAA,cAAa,WAAW,WAAW,iBAAiB,YAAY,WAAW;AAClK;AAEA,SAAS,KAAK,OAAuB;AACnC,SAAO,iBAAiB,KAAK,KAAK,IAAI,QAAQ,YAAY,KAAK;AACjE;AAEA,SAAS,QAAQ,OAAuB;AACtC,QAAM,aAAa,MAAM,KAAK,EAAE,QAAQ,qBAAqB,GAAG,EAC7D,QAAQ,YAAY,EAAE;AACzB,SAAO,cAAc;AACvB;;;ACrcA,SAAS,uBAAAC,sBAAqB,2BAAmE;AACjG,SAAS,aAAa,yBAAyB,0CAAkE;AAIjH,SAAS,SAAS;AAGX,SAAS,uBAAuB,OAAuB;AAC5D,QAAM,YAAY,MAAM,QAAQ,SAAS;AACzC,QAAM,sBAAsB,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,MAAM,EAAE,OAAO,EAAE,IAAI,CAAC,CAAC,EAAE,IAAI,GAAO,CAAC,EAAE,MAAM,WAAW,4BAA4B,CAAC,CAAC;AACzI,SAAO,MAAM,QAAQ,QAAQ,IAAI,YAAU;AACzC,UAAM,OAAO,EAAE,GAAG,QAAQ,OAAO,OAAO,IAAI,UAAU,EAAE,gBAAgB,OAAO,EAAE;AACjF,QAAI,OAAO,SAAS,kBAAmB,QAAO,iBAAiB,MAAM;AAAA,MAAE,GAAG;AAAA,MACxE,QAAQ,uBAAuB,OAAO,YAAY,WAAW;AAAA,MAAG,YAAY,OAAO,cAAc;AAAA,MACjG,UAAU,EAAE,GAAG,KAAK,UAAU,qBAAqB,OAAO,YAAY;AAAA,IAAE,CAAC;AAC3E,QAAI,OAAO,SAAS,WAAW,OAAO,SAAS,eAAe;AAC5D,YAAM,OAAO,MAAM,MAAM,KAAK,CAAAC,UAAQA,MAAK,MAAM,OAAO,OAAO,UAAU,EAAE;AAC3E,UAAI,CAAC,KAAM,OAAM,IAAI,MAAM,+BAA+B,OAAO,EAAE,GAAG;AACtE,YAAM,SAAS,IAAI,YAAY,SAAS,EAAE,OAAO,KAAK,CAAC,EAAE,OAAO,yBAAyB,IAAI,CAAC;AAC9F,YAAM,WAAW,EAAE,GAAG,KAAK,UAAU,mBAAmB,OAAO,UAAU;AACzE,UAAI,OAAO,SAAS,QAAS,QAAO,iBAAiB,MAAM,EAAE,GAAG,MAAM,QAAQ,SAAS,CAAC;AACxF,UAAI,CAAC,OAAO,MAAO,OAAM,IAAI,MAAM,kBAAkB,OAAO,EAAE,wDAAwD;AACtH,aAAO,iBAAiB,MAAM;AAAA,QAAE,GAAG;AAAA,QAAM;AAAA,QAAQ;AAAA,QAAU,OAAO,OAAO;AAAA,QAAO,aAAa,OAAO,eAAe;AAAA,QACjH,mBAAmB;AAAA,QAAW,wBAAwB,oBAAoB,OAAO,EAAE,KAAK,CAAC;AAAA,MAAE,CAAC;AAAA,IAChG;AACA,WAAO,iBAAiB,MAAM;AAAA,MAAE,GAAG;AAAA,MAAM,MAAM,OAAO,SAAS,aAAa,SAAS,OAAO;AAAA,MAC1F,QAAQ,OAAO,SAAS,aAAa,EAAE,GAAG,OAAO,QAAQ,GAAI,OAAO,OAAO,gBAAgB,SAAY,CAAC,IAAI,EAAE,cAAc,OAAO,OAAO,YAAY,EAAG,IAAI,OAAO;AAAA,IAAO,CAAC;AAAA,EAChL,CAAC;AACH;AAEO,SAAS,4BAA4B,SAAwB,SAA0B,SAA8B,CAAC,GAAiB;AAC5I,QAAM,WAAW,oBAAoB,SAAS,OAAO;AACrD,SAAOC,qBAAoB,SAAS,OAAO,EAAE,IAAI,CAAC,WAAW;AAC3D,UAAM,SAAS,SAAS,KAAK,CAAC,EAAE,OAAO,MAAM,OAAO,aAAa,OAAO,EAAE,EAAG;AAC7E,UAAM,OAAO,EAAE,GAAG,QAAQ,OAAO,OAAO,MAAM,UAAU,EAAE,eAAe,YAAY,OAAO,EAAE,EAAE;AAChG,QAAI,OAAO,SAAS,eAAe;AACjC,UAAI,CAAC,OAAO,SAAS,OAAO,sBAAsB,YAAY,CAAC,OAAO,uBAAuB,CAAC,OAAO,cAAe,OAAM,IAAI,MAAM,UAAU,OAAO,EAAE,iEAAiE;AACxN,YAAM,SAAS,OAAO,KAAK,WAAS,MAAM,OAAO,OAAO,UAAU,EAAE;AACpE,YAAM,WAAW,OAAO,KAAK,WAAS,MAAM,OAAO,OAAO,cAAe,EAAE;AAC3E,UAAI,CAAC,UAAU,CAAC,SAAU,OAAM,IAAI,MAAM,UAAU,OAAO,EAAE,2DAA2D;AACxH,YAAM,SAAS,mCAAmC,MAAM,EAAE,IAAI,CAAC,EAAE,IAAI,EAAE,EAAE,MAAM,KAAK,MAAM,wBAAwB,UAAU,OAAO,aAAa,CAAC,CAAC;AAClJ,aAAO,iBAAiB,MAAM;AAAA,QAAE,GAAG;AAAA,QAAM,QAAQ,wBAAwB,QAAQ,OAAO,SAAS;AAAA,QAAG,OAAO,OAAO;AAAA,QAChH,aAAa,OAAO,eAAe;AAAA,QAAG,wBAAwB,OAAO,IAAI,aAAW,QAAQ,EAAE;AAAA,QAC9F,UAAU;AAAA,UAAE,GAAG,KAAK;AAAA,UAAU,mBAAmB,OAAO;AAAA,UACtD,yBAAyB,OAAO,oBAAoB;AAAA,UAAa,qBAAqB,OAAO;AAAA,QAAoB;AAAA,MACrH,CAAC;AAAA,IACH;AACA,QAAI,OAAO,SAAS,kBAAmB,QAAO,iBAAiB,MAAM;AAAA,MAAE,GAAG;AAAA,MACxE,QAAQ,uBAAuB,OAAO,YAAY,WAAW;AAAA,MAAG,YAAY,OAAO,cAAc;AAAA,MACjG,UAAU,EAAE,GAAG,KAAK,UAAU,qBAAqB,OAAO,YAAY;AAAA,IACxE,CAAC;AACD,QAAI,OAAO,SAAS,SAAS;AAC3B,YAAM,QAAQ,OAAO,KAAK,CAACC,WAAUA,OAAM,OAAO,OAAO,UAAU,EAAE;AACrE,UAAI,CAAC,MAAO,OAAM,IAAI,MAAM,UAAU,OAAO,EAAE,mCAAmC;AAClF,aAAO,iBAAiB,MAAM,EAAE,GAAG,MAAM,QAAQ,wBAAwB,OAAO,OAAO,SAAS,EAAE,CAAC;AAAA,IACrG;AACA,WAAO,iBAAiB,MAAM,EAAE,GAAG,MAAM,MAAM,OAAO,SAAS,aAAa,SAAS,OAAO,KAAK,CAAC;AAAA,EACpG,CAAC;AACH;AAEO,SAAS,uBAAuBC,OAAc;AAAE,SAAO,kBAAkBA,KAAI;AAAO;",
6
- "names": ["contentHash", "compileBoundGraders", "file", "compileBoundGraders", "asset", "hash"]
3
+ "sources": ["../src/taskset-authored-portable-release.ts"],
4
+ "sourcesContent": ["import type { Taskset } from \"./taskset-authored-contracts.js\";\nimport type {\n GraderSpec,\n TaskDataRecord,\n} from \"./taskset-draft-core.js\";\nimport {\n TaskRecordSchema,\n GraderSpecSchema as PortableGraderSpecSchema,\n TasksetReleaseContentSchema,\n TasksetReleaseSchema,\n bindTasksetExecutionReleases,\n createEnvironmentRelease,\n createVerifierSetRelease,\n EnvironmentReleaseSchema,\n VerifierSetReleaseSchema,\n verifyEnvironmentRelease,\n verifyVerifierSetRelease,\n verifyPreferenceComparisonRelease,\n type EnvironmentContract,\n type GraderSpec as PortableGraderSpec,\n type PreferenceComparisonRelease,\n type TasksetRelease,\n} from \"@openpond/evals\";\nimport type {\n ImmutableAssetRef,\n ToolDeclaration,\n} from \"@openpond/harness\";\nimport {\n CapabilityRequirementSchema,\n ImmutableAssetRefSchema,\n ToolDeclarationSchema,\n} from \"@openpond/harness\";\n\nimport { contentHash, sha256 } from \"@openpond/harness\";\nimport { TaskBatchPackageMetadataSchema } from \"@openpond/evals/learning\";\nimport { compileBoundGraders, RewardBindingSchema, RewardReleaseSchema, type RewardBinding, type RewardRelease } from \"@openpond/evals/rewards\";\nimport { portableDeterministicCheck } from \"@openpond/evals/graders\";\n\nexport type TasksetRewardExecution = { binding: RewardBinding; rewards: RewardRelease[] };\n\nexport function resolvePortableTasksetRewardExecution(taskset: Taskset, supplied?: TasksetRewardExecution): TasksetRewardExecution | null {\n const learning = taskset.metadata.learning === undefined ? null : TaskBatchPackageMetadataSchema.parse(taskset.metadata.learning);\n const embedded = taskset.metadata.rewardExecution as { binding?: unknown; rewards?: unknown } | undefined;\n const execution = supplied ?? (embedded ? {\n binding: RewardBindingSchema.parse(embedded.binding),\n rewards: RewardReleaseSchema.array().parse(embedded.rewards),\n } : learning);\n if (taskset.metadata.rewardBinding !== undefined) {\n const reference = taskset.metadata.rewardBinding as Partial<RewardBinding> | null;\n if (!execution || !reference || reference.id !== execution.binding.id\n || reference.revision !== execution.binding.revision || reference.contentHash !== execution.binding.contentHash) {\n throw new Error(\"Portable Taskset requires its exact published Reward binding.\");\n }\n }\n if (execution) compileBoundGraders(execution.binding, execution.rewards);\n return execution ? { binding: execution.binding, rewards: execution.rewards } : null;\n}\n\nexport function materializePortableTasksetRelease(input: {\n taskset: Taskset;\n selectedTasks?: TaskDataRecord[];\n adapterId: string;\n admittedTasksetRelease?: TasksetRelease | null;\n rewardExecution?: TasksetRewardExecution;\n}): {\n environmentRelease: ReturnType<typeof createEnvironmentRelease>;\n verifierSetRelease: ReturnType<typeof createVerifierSetRelease>;\n tasksetRelease: TasksetRelease;\n} {\n const environment = portableEnvironment(input.taskset);\n const tools = portableTools(input.taskset);\n const learning = input.taskset.metadata.learning === undefined ? null : TaskBatchPackageMetadataSchema.parse(input.taskset.metadata.learning);\n const embedded = input.taskset.metadata.rewardExecution;\n const rewardExecution = resolvePortableTasksetRewardExecution(input.taskset, input.rewardExecution);\n const graders = rewardExecution ? compileBoundGraders(rewardExecution.binding, rewardExecution.rewards) : input.taskset.graders.map(portableGrader);\n const tasks = input.selectedTasks?.length\n ? input.selectedTasks\n : input.taskset.tasks;\n if (!tasks.length) throw new Error(\"A portable Taskset release requires at least one task.\");\n const pinned = input.taskset.environment.metadata.portableExecutionResources as { environment?: unknown; verifierSet?: unknown } | undefined;\n if (environment.entrypoint === \"openpond.javascript-environment.v1\" && !pinned) throw new Error(\"Tool Taskset publication requires its pinned execution resources.\");\n const environmentRelease = pinned ? EnvironmentReleaseSchema.parse(pinned.environment) : createEnvironmentRelease({\n schemaVersion: \"openpond.environmentRelease.v1\",\n id: `environment-release-${input.taskset.id}-r${input.taskset.revision}`,\n revision: input.taskset.revision,\n contract: environment,\n actionSchemaRef: null,\n observationSchemaRef: null,\n stateSchemaRef: null,\n artifactCollection: {\n maxArtifacts: 100_000,\n maxTotalBytes: 250_000_000,\n },\n adapterConformanceHashes: {\n [input.adapterId]: contentHash({ adapterId: input.adapterId, environment, tools }),\n },\n metadata: {\n sourceTasksetId: input.taskset.id,\n sourcePackageHash: input.taskset.metadata.sourcePackageHash ?? null,\n resources: input.taskset.environment.resources ?? [],\n },\n });\n const verifierSetRelease = pinned?.verifierSet !== undefined ? VerifierSetReleaseSchema.parse(pinned.verifierSet) : createVerifierSetRelease({\n schemaVersion: \"openpond.verifierSetRelease.v1\",\n id: `verifier-set-release-${input.taskset.id}-r${input.taskset.revision}`,\n revision: input.taskset.revision,\n graders,\n isolation: {\n processBoundary: \"isolated_process\",\n networkPolicy: \"none\",\n defaultTimeoutMs: Math.min(input.taskset.environment.defaultTimeoutMs, 300_000),\n },\n calibrationReceiptRefs: [],\n metadata: { sourceTasksetId: input.taskset.id },\n });\n if (!verifyEnvironmentRelease(environmentRelease) || !verifyVerifierSetRelease(verifierSetRelease) ||\n contentHash(environmentRelease.contract) !== contentHash(environment) || contentHash(verifierSetRelease.graders) !== contentHash(graders)) {\n throw new Error(\"Taskset execution resources differ from its declared environment or graders.\");\n }\n const tasksetContent = TasksetReleaseContentSchema.parse({\n schemaVersion: \"openpond.tasksetRelease.v2\",\n id: input.taskset.metadata.derivedPortableMetadata === undefined ? `taskset-release-${input.taskset.id}-r${input.taskset.revision}` : input.taskset.id,\n revision: input.taskset.revision,\n policy: {\n policyVisibleFields: input.taskset.policy.policyVisibleFields,\n privilegedFields: input.taskset.policy.privilegedFields,\n hiddenGraderRefs: input.taskset.policy.hiddenGraderRefs,\n connectedAppScopes: input.taskset.policy.connectedAppScopes,\n },\n environment,\n tools,\n capabilities: portableCapabilities(input.taskset),\n tasks: tasks.map(projectPortableTaskRecord),\n graders,\n ...(input.taskset.metrics ? { metrics: input.taskset.metrics } : {}),\n metadata: input.taskset.metadata.derivedPortableMetadata ?? {\n sourceTasksetId: input.taskset.id,\n sourceTasksetHash: input.taskset.contentHash,\n sourcePackageHash: input.taskset.metadata.sourcePackageHash ?? null,\n environmentResources: input.taskset.environment.resources ?? [],\n ordinaryAuthoring: {\n instructions: input.taskset.objective,\n graderFixtures: input.taskset.graderFixtures,\n judgeCalibrationFixtures: Object.fromEntries(input.taskset.graders\n .filter(grader => grader.kind === \"model_judge\")\n .map(grader => [grader.id, grader.calibrationFixtureRefs])),\n },\n ...(learning ? { learning } : {}),\n // A sealed batch already pins these resources in learning metadata.\n // Resolving that same binding at admission must not change its release.\n ...(rewardExecution && !learning && (input.rewardExecution || embedded) ? { rewardExecution: { binding: rewardExecution.binding, rewards: rewardExecution.rewards } } : {}),\n },\n });\n const draft = input.admittedTasksetRelease\n ? TasksetReleaseSchema.parse(input.admittedTasksetRelease)\n : TasksetReleaseSchema.parse({\n ...tasksetContent,\n contentHash: contentHash(tasksetContent),\n });\n const tasksetRelease = draft.environmentRelease\n ? assertReleaseBindings({ draft, environmentRelease, verifierSetRelease })\n : bindTasksetExecutionReleases({\n taskset: draft,\n environment: environmentRelease,\n verifierSet: verifierSetRelease,\n });\n return { environmentRelease, verifierSetRelease, tasksetRelease };\n}\n\n/**\n * Admits a published comparison policy against the exact portable Taskset\n * release it names. A comparison is deliberately not embedded in that release:\n * doing so would create a content-hash cycle between Taskset and comparison.\n */\nexport function projectPreferenceComparisonRelease(input: {\n tasksetRelease: TasksetRelease;\n preferenceComparison: PreferenceComparisonRelease;\n}): PreferenceComparisonRelease {\n if (!verifyPreferenceComparisonRelease(input.preferenceComparison)) {\n throw new Error(\"Preference comparison release failed immutable content verification.\");\n }\n const release = input.preferenceComparison;\n if (\n release.tasksetRelease.id !== input.tasksetRelease.id\n || release.tasksetRelease.contentHash !== input.tasksetRelease.contentHash\n ) {\n throw new Error(\"Preference comparison release must reference the admitted portable Taskset release.\");\n }\n return release;\n}\n\nexport function portableTasksetEnvironment(taskset: Taskset): EnvironmentContract {\n return portableEnvironment(taskset);\n}\n\nexport function portableTasksetTools(taskset: Taskset): ToolDeclaration[] {\n return portableTools(taskset);\n}\n\n/** Preserve immutable file/schema references while overlaying current edits. */\nexport function projectPortableTaskRecord(task: TaskDataRecord) {\n const admitted = task.metadata.portableTaskRecord;\n if (admitted !== undefined) {\n const original = TaskRecordSchema.parse(admitted);\n const projected = TaskRecordSchema.parse({\n ...original,\n id: task.id,\n clusterKey: task.clusterKey,\n split: task.split,\n input: task.input,\n expectedOutput: portableExpectedOutput(task),\n policyVisibleContext: task.policyVisibleContext,\n privilegedContextRef: task.privilegedContextRef,\n tags: task.tags,\n artifactRefs: task.assets === undefined ? original.artifactRefs : task.assets.map((asset) => {\n const previous = original.artifactRefs.find(item => item.id === asset.id);\n return {\n id: asset.id,\n path: previous?.path ?? `tasks/${segment(task.id)}/${segment(asset.fileName)}`,\n contentHash: hash(asset.sha256),\n sizeBytes: asset.sizeBytes,\n mediaType: asset.mediaType,\n visibility: previous?.visibility ?? \"policy\",\n };\n }),\n requiredOutputs: task.requiredOutputs === undefined ? original.requiredOutputs : task.requiredOutputs.map((output) => {\n const previous = original.requiredOutputs?.find(item => item.path === output.path);\n return {\n path: output.path,\n mediaType: output.mediaType,\n schemaRef: previous?.schemaRef && previous.schemaRef.id === output.schemaRef ? previous.schemaRef : null,\n maxBytes: output.maxBytes ?? null,\n metadata: {\n ...output.metadata,\n ...(output.schemaRef && previous?.schemaRef?.id !== output.schemaRef\n ? { legacySchemaRef: output.schemaRef }\n : {}),\n },\n };\n }),\n });\n if (projected.requiredOutputs === undefined) delete projected.requiredOutputs;\n return projected;\n }\n return {\n id: task.id,\n clusterKey: task.clusterKey,\n split: task.split,\n input: task.input,\n expectedOutput: portableExpectedOutput(task),\n policyVisibleContext: task.policyVisibleContext,\n privilegedContextRef: task.privilegedContextRef,\n artifactRefs: (task.assets ?? []).map((item) => ({\n id: item.id,\n path: `tasks/${segment(task.id)}/${segment(item.fileName)}`,\n contentHash: hash(item.sha256),\n sizeBytes: item.sizeBytes,\n mediaType: item.mediaType,\n visibility: \"policy\" as const,\n })),\n requiredOutputs: (task.requiredOutputs ?? []).map((output) => ({\n path: output.path,\n mediaType: output.mediaType,\n schemaRef: null,\n maxBytes: output.maxBytes ?? null,\n metadata: {\n ...output.metadata,\n legacySchemaRef: output.schemaRef ?? null,\n },\n })),\n tags: task.tags,\n };\n}\n\nfunction portableExpectedOutput(\n task: TaskDataRecord,\n): Record<string, unknown> | null {\n if (!task.expectedOutput) return null;\n if (\"artifactRenderer\" in task.expectedOutput) {\n throw new Error(\n `Task ${task.id} embeds an artifact renderer. Put shared renderer configuration in Environment resources and reference it from the Scenario.`,\n );\n }\n return task.expectedOutput;\n}\n\nfunction portableEnvironment(taskset: Taskset): EnvironmentContract {\n const kind: EnvironmentContract[\"kind\"] = taskset.environment.kind === \"chat\"\n ? \"text\"\n : taskset.environment.kind === \"work\" ? \"work\"\n : taskset.environment.kind === \"program\" ? \"custom_program\" : \"agent\";\n return {\n protocolVersion: \"openpond.environment.v1\",\n kind,\n entrypoint: taskset.environment.entrypoint,\n stateful: taskset.environment.stateful,\n deterministicSeeds: taskset.environment.deterministicSeeds,\n lifecycle: [\"create\", \"reset\", \"step\", \"collect\", \"destroy\"],\n networkPolicy: taskset.environment.networkPolicy,\n defaultTimeoutMs: taskset.environment.defaultTimeoutMs,\n };\n}\n\nfunction portableTools(taskset: Taskset): ToolDeclaration[] {\n const admitted = taskset.environment.metadata.portableTools;\n if (admitted !== undefined) {\n return ToolDeclarationSchema.array().max(200).parse(admitted);\n }\n const bindings = taskset.environment.actionBindings ?? [];\n if (bindings.length) return bindings.map((binding) => ({\n name: binding.modelToolName,\n description: binding.description,\n inputSchema: binding.inputSchema,\n inputSchemaHash: hash(binding.actionSchemaHash),\n sideEffect: binding.sideEffect,\n timeoutMs: binding.timeoutMs,\n }));\n return taskset.environment.toolNames.map((name) => ({\n name,\n description: `Host-provided ${name} tool.`,\n inputSchema: {},\n inputSchemaHash: contentHash({}),\n sideEffect: \"write\" as const,\n timeoutMs: taskset.environment.defaultTimeoutMs,\n }));\n}\n\nfunction portableGrader(grader: GraderSpec): PortableGraderSpec {\n const base = {\n id: grader.id,\n version: grader.version,\n weight: grader.weight,\n hardGate: grader.hardGate,\n rewardEligible: grader.rewardEligible,\n privileged: grader.privileged,\n };\n if (grader.kind === \"model_judge\") return {\n ...base,\n kind: \"model_judge\",\n rubricRef: rubricAsset(grader),\n model: modelJudgeRef(grader),\n temperature: grader.temperature,\n calibrationStatus: grader.calibrationStatus,\n };\n if (grader.kind === \"custom_verifier\") {\n if (grader.metadata.portableVerifierRef === undefined) {\n throw new Error(`Custom verifier ${grader.id} requires its immutable executable asset before publication.`);\n }\n const imported = PortableGraderSpecSchema.safeParse(grader.metadata.portableGrader);\n // An omitted portable export means \"verify\". Preserve that encoding on an\n // unchanged import so executing it does not silently replace its release.\n const preserveDefault = imported.success && imported.data.kind === \"custom_verifier\"\n && imported.data.exportName === undefined && grader.exportName === \"verify\";\n return {\n ...base,\n kind: \"custom_verifier\",\n ...(grader.runtime === undefined ? {} : { runtime: grader.runtime }),\n verifierRef: ImmutableAssetRefSchema.parse(grader.metadata.portableVerifierRef),\n timeoutMs: grader.timeoutMs,\n ...(preserveDefault ? {} : { exportName: grader.exportName }),\n networkPolicy: \"none\",\n };\n }\n if (grader.kind === \"human\") return {\n ...base,\n kind: \"human\",\n rubricRef: rubricAsset(grader),\n reviewerRole: grader.reviewerRole,\n };\n return {\n ...base,\n ...portableDeterministicCheck(grader),\n };\n}\n\nfunction portableCapabilities(taskset: Taskset) {\n const admitted = taskset.metadata.portableCapabilities;\n if (admitted !== undefined) {\n return CapabilityRequirementSchema.array().max(200).parse(admitted);\n }\n return [\n ...(taskset.capabilities.requiresTools\n ? [{ id: \"tools\", required: true, scopes: taskset.environment.toolNames }]\n : []),\n ...(taskset.capabilities.requiresState\n ? [{ id: \"local-state\", required: true, scopes: [] }]\n : []),\n ...(taskset.capabilities.requiresPrivilegedGrading\n ? [{ id: \"private-verifier\", required: true, scopes: taskset.policy.hiddenGraderRefs }]\n : []),\n ...taskset.policy.connectedAppScopes.map((scope) => ({\n id: `connected-app-${segment(scope)}`,\n required: true,\n scopes: [scope],\n })),\n ].map((requirement) => ({\n ...requirement,\n portability: requirement.id === \"local-state\" ? \"host_adapter\" as const : \"portable\" as const,\n }));\n}\n\nfunction assertReleaseBindings(input: {\n draft: TasksetRelease;\n environmentRelease: ReturnType<typeof createEnvironmentRelease>;\n verifierSetRelease: ReturnType<typeof createVerifierSetRelease>;\n}): TasksetRelease {\n if (\n input.draft.environmentRelease?.id !== input.environmentRelease.id\n || input.draft.environmentRelease.contentHash !== input.environmentRelease.contentHash\n ) {\n throw new Error(\"Admitted Taskset Release binds a different Environment Release.\");\n }\n if (\n input.draft.verifierSetRelease?.id !== input.verifierSetRelease.id\n || input.draft.verifierSetRelease.contentHash !== input.verifierSetRelease.contentHash\n ) {\n throw new Error(\"Admitted Taskset Release binds a different Verifier Set Release.\");\n }\n if (contentHash(input.draft.environment) !== contentHash(input.environmentRelease.contract)) {\n throw new Error(\"Admitted Taskset Environment differs from its bound Environment Release.\");\n }\n if (contentHash(input.draft.graders) !== contentHash(input.verifierSetRelease.graders)) {\n throw new Error(\"Admitted Taskset graders differ from its bound Verifier Set Release.\");\n }\n return input.draft;\n}\n\nfunction modelJudgeRef(grader: Extract<GraderSpec, { kind: \"model_judge\" }>) {\n const previous = grader.metadata.portableGrader === undefined ? null : PortableGraderSpecSchema.parse(grader.metadata.portableGrader);\n const revision = previous?.kind === \"model_judge\" && previous.model?.providerId === grader.judge.providerId && previous.model.modelId === grader.judge.modelId ? previous.model.revision : null;\n return { ...grader.judge, revision };\n}\n\nfunction rubricAsset(grader: Extract<GraderSpec, { kind: \"human\" | \"model_judge\" }>): ImmutableAssetRef {\n const contentHash = sha256(grader.rubric);\n const sizeBytes = new TextEncoder().encode(grader.rubric).byteLength;\n const reference = grader.metadata.portableRubricRef;\n if (reference !== undefined) {\n const asset = ImmutableAssetRefSchema.parse(reference);\n if (asset.contentHash !== contentHash || asset.sizeBytes !== sizeBytes || asset.visibility === \"policy\") throw new Error(\"Taskset rubric differs from its published private source.\");\n return asset;\n }\n return { id: `rubric-${grader.id}`, path: `graders/${segment(grader.id)}/rubric.md`, contentHash, sizeBytes, mediaType: \"text/markdown\", visibility: \"verifier\" };\n}\n\nfunction hash(value: string): string {\n return /^[a-f0-9]{64}$/.test(value) ? value : contentHash(value);\n}\n\nfunction segment(value: string): string {\n const normalized = value.trim().replace(/[^a-zA-Z0-9._-]+/g, \"-\")\n .replace(/^-+|-+$/g, \"\");\n return normalized || \"asset\";\n}\n"],
5
+ "mappings": ";;;;;;;;;;;AAKA;AAAA,EACE;AAAA,EACA,oBAAoB;AAAA,EACpB;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OAKK;AAYP,SAAS,sCAAsC;AAC/C,SAAS,qBAAqB,qBAAqB,2BAAmE;AACtH,SAAS,kCAAkC;AAIpC,SAAS,sCAAsC,SAAkB,UAAkE;AACxI,QAAM,WAAW,QAAQ,SAAS,aAAa,SAAY,OAAO,+BAA+B,MAAM,QAAQ,SAAS,QAAQ;AAChI,QAAM,WAAW,QAAQ,SAAS;AAClC,QAAM,YAAY,aAAa,WAAW;AAAA,IACxC,SAAS,oBAAoB,MAAM,SAAS,OAAO;AAAA,IACnD,SAAS,oBAAoB,MAAM,EAAE,MAAM,SAAS,OAAO;AAAA,EAC7D,IAAI;AACJ,MAAI,QAAQ,SAAS,kBAAkB,QAAW;AAChD,UAAM,YAAY,QAAQ,SAAS;AACnC,QAAI,CAAC,aAAa,CAAC,aAAa,UAAU,OAAO,UAAU,QAAQ,MAC9D,UAAU,aAAa,UAAU,QAAQ,YAAY,UAAU,gBAAgB,UAAU,QAAQ,aAAa;AACjH,YAAM,IAAI,MAAM,+DAA+D;AAAA,IACjF;AAAA,EACF;AACA,MAAI,UAAW,qBAAoB,UAAU,SAAS,UAAU,OAAO;AACvE,SAAO,YAAY,EAAE,SAAS,UAAU,SAAS,SAAS,UAAU,QAAQ,IAAI;AAClF;AAEO,SAAS,kCAAkC,OAUhD;AACA,QAAM,cAAc,oBAAoB,MAAM,OAAO;AACrD,QAAM,QAAQ,cAAc,MAAM,OAAO;AACzC,QAAM,WAAW,MAAM,QAAQ,SAAS,aAAa,SAAY,OAAO,+BAA+B,MAAM,MAAM,QAAQ,SAAS,QAAQ;AAC5I,QAAM,WAAW,MAAM,QAAQ,SAAS;AACxC,QAAM,kBAAkB,sCAAsC,MAAM,SAAS,MAAM,eAAe;AAClG,QAAM,UAAU,kBAAkB,oBAAoB,gBAAgB,SAAS,gBAAgB,OAAO,IAAI,MAAM,QAAQ,QAAQ,IAAI,cAAc;AAClJ,QAAM,QAAQ,MAAM,eAAe,SAC/B,MAAM,gBACN,MAAM,QAAQ;AAClB,MAAI,CAAC,MAAM,OAAQ,OAAM,IAAI,MAAM,wDAAwD;AAC3F,QAAM,SAAS,MAAM,QAAQ,YAAY,SAAS;AAClD,MAAI,YAAY,eAAe,wCAAwC,CAAC,OAAQ,OAAM,IAAI,MAAM,mEAAmE;AACnK,QAAM,qBAAqB,SAAS,yBAAyB,MAAM,OAAO,WAAW,IAAI,yBAAyB;AAAA,IAChH,eAAe;AAAA,IACf,IAAI,uBAAuB,MAAM,QAAQ,EAAE,KAAK,MAAM,QAAQ,QAAQ;AAAA,IACtE,UAAU,MAAM,QAAQ;AAAA,IACxB,UAAU;AAAA,IACV,iBAAiB;AAAA,IACjB,sBAAsB;AAAA,IACtB,gBAAgB;AAAA,IAChB,oBAAoB;AAAA,MAClB,cAAc;AAAA,MACd,eAAe;AAAA,IACjB;AAAA,IACA,0BAA0B;AAAA,MACxB,CAAC,MAAM,SAAS,GAAG,YAAY,EAAE,WAAW,MAAM,WAAW,aAAa,MAAM,CAAC;AAAA,IACnF;AAAA,IACA,UAAU;AAAA,MACR,iBAAiB,MAAM,QAAQ;AAAA,MAC/B,mBAAmB,MAAM,QAAQ,SAAS,qBAAqB;AAAA,MAC/D,WAAW,MAAM,QAAQ,YAAY,aAAa,CAAC;AAAA,IACrD;AAAA,EACF,CAAC;AACD,QAAM,qBAAqB,QAAQ,gBAAgB,SAAY,yBAAyB,MAAM,OAAO,WAAW,IAAI,yBAAyB;AAAA,IAC3I,eAAe;AAAA,IACf,IAAI,wBAAwB,MAAM,QAAQ,EAAE,KAAK,MAAM,QAAQ,QAAQ;AAAA,IACvE,UAAU,MAAM,QAAQ;AAAA,IACxB;AAAA,IACA,WAAW;AAAA,MACT,iBAAiB;AAAA,MACjB,eAAe;AAAA,MACf,kBAAkB,KAAK,IAAI,MAAM,QAAQ,YAAY,kBAAkB,GAAO;AAAA,IAChF;AAAA,IACA,wBAAwB,CAAC;AAAA,IACzB,UAAU,EAAE,iBAAiB,MAAM,QAAQ,GAAG;AAAA,EAChD,CAAC;AACD,MAAI,CAAC,yBAAyB,kBAAkB,KAAK,CAAC,yBAAyB,kBAAkB,KAC7F,YAAY,mBAAmB,QAAQ,MAAM,YAAY,WAAW,KAAK,YAAY,mBAAmB,OAAO,MAAM,YAAY,OAAO,GAAG;AAC7I,UAAM,IAAI,MAAM,8EAA8E;AAAA,EAChG;AACA,QAAM,iBAAiB,4BAA4B,MAAM;AAAA,IACvD,eAAe;AAAA,IACf,IAAI,MAAM,QAAQ,SAAS,4BAA4B,SAAY,mBAAmB,MAAM,QAAQ,EAAE,KAAK,MAAM,QAAQ,QAAQ,KAAK,MAAM,QAAQ;AAAA,IACpJ,UAAU,MAAM,QAAQ;AAAA,IACxB,QAAQ;AAAA,MACN,qBAAqB,MAAM,QAAQ,OAAO;AAAA,MAC1C,kBAAkB,MAAM,QAAQ,OAAO;AAAA,MACvC,kBAAkB,MAAM,QAAQ,OAAO;AAAA,MACvC,oBAAoB,MAAM,QAAQ,OAAO;AAAA,IAC3C;AAAA,IACA;AAAA,IACA;AAAA,IACA,cAAc,qBAAqB,MAAM,OAAO;AAAA,IAChD,OAAO,MAAM,IAAI,yBAAyB;AAAA,IAC1C;AAAA,IACA,GAAI,MAAM,QAAQ,UAAU,EAAE,SAAS,MAAM,QAAQ,QAAQ,IAAI,CAAC;AAAA,IAClE,UAAU,MAAM,QAAQ,SAAS,2BAA2B;AAAA,MAC1D,iBAAiB,MAAM,QAAQ;AAAA,MAC/B,mBAAmB,MAAM,QAAQ;AAAA,MACjC,mBAAmB,MAAM,QAAQ,SAAS,qBAAqB;AAAA,MAC/D,sBAAsB,MAAM,QAAQ,YAAY,aAAa,CAAC;AAAA,MAC9D,mBAAmB;AAAA,QACjB,cAAc,MAAM,QAAQ;AAAA,QAC5B,gBAAgB,MAAM,QAAQ;AAAA,QAC9B,0BAA0B,OAAO,YAAY,MAAM,QAAQ,QACxD,OAAO,YAAU,OAAO,SAAS,aAAa,EAC9C,IAAI,YAAU,CAAC,OAAO,IAAI,OAAO,sBAAsB,CAAC,CAAC;AAAA,MAC9D;AAAA,MACA,GAAI,WAAW,EAAE,SAAS,IAAI,CAAC;AAAA;AAAA;AAAA,MAG/B,GAAI,mBAAmB,CAAC,aAAa,MAAM,mBAAmB,YAAY,EAAE,iBAAiB,EAAE,SAAS,gBAAgB,SAAS,SAAS,gBAAgB,QAAQ,EAAE,IAAI,CAAC;AAAA,IAC3K;AAAA,EACF,CAAC;AACD,QAAM,QAAQ,MAAM,yBAChB,qBAAqB,MAAM,MAAM,sBAAsB,IACvD,qBAAqB,MAAM;AAAA,IAC3B,GAAG;AAAA,IACH,aAAa,YAAY,cAAc;AAAA,EACzC,CAAC;AACH,QAAM,iBAAiB,MAAM,qBACzB,sBAAsB,EAAE,OAAO,oBAAoB,mBAAmB,CAAC,IACvE,6BAA6B;AAAA,IAC7B,SAAS;AAAA,IACT,aAAa;AAAA,IACb,aAAa;AAAA,EACf,CAAC;AACH,SAAO,EAAE,oBAAoB,oBAAoB,eAAe;AAClE;AAOO,SAAS,mCAAmC,OAGnB;AAC9B,MAAI,CAAC,kCAAkC,MAAM,oBAAoB,GAAG;AAClE,UAAM,IAAI,MAAM,sEAAsE;AAAA,EACxF;AACA,QAAM,UAAU,MAAM;AACtB,MACE,QAAQ,eAAe,OAAO,MAAM,eAAe,MAChD,QAAQ,eAAe,gBAAgB,MAAM,eAAe,aAC/D;AACA,UAAM,IAAI,MAAM,qFAAqF;AAAA,EACvG;AACA,SAAO;AACT;AAEO,SAAS,2BAA2B,SAAuC;AAChF,SAAO,oBAAoB,OAAO;AACpC;AAEO,SAAS,qBAAqB,SAAqC;AACxE,SAAO,cAAc,OAAO;AAC9B;AAGO,SAAS,0BAA0B,MAAsB;AAC9D,QAAM,WAAW,KAAK,SAAS;AAC/B,MAAI,aAAa,QAAW;AAC1B,UAAM,WAAW,iBAAiB,MAAM,QAAQ;AAChD,UAAM,YAAY,iBAAiB,MAAM;AAAA,MACvC,GAAG;AAAA,MACH,IAAI,KAAK;AAAA,MACT,YAAY,KAAK;AAAA,MACjB,OAAO,KAAK;AAAA,MACZ,OAAO,KAAK;AAAA,MACZ,gBAAgB,uBAAuB,IAAI;AAAA,MAC3C,sBAAsB,KAAK;AAAA,MAC3B,sBAAsB,KAAK;AAAA,MAC3B,MAAM,KAAK;AAAA,MACX,cAAc,KAAK,WAAW,SAAY,SAAS,eAAe,KAAK,OAAO,IAAI,CAAC,UAAU;AAC3F,cAAM,WAAW,SAAS,aAAa,KAAK,UAAQ,KAAK,OAAO,MAAM,EAAE;AACxE,eAAO;AAAA,UACL,IAAI,MAAM;AAAA,UACV,MAAM,UAAU,QAAQ,SAAS,QAAQ,KAAK,EAAE,CAAC,IAAI,QAAQ,MAAM,QAAQ,CAAC;AAAA,UAC5E,aAAa,KAAK,MAAM,MAAM;AAAA,UAC9B,WAAW,MAAM;AAAA,UACjB,WAAW,MAAM;AAAA,UACjB,YAAY,UAAU,cAAc;AAAA,QACtC;AAAA,MACF,CAAC;AAAA,MACD,iBAAiB,KAAK,oBAAoB,SAAY,SAAS,kBAAkB,KAAK,gBAAgB,IAAI,CAAC,WAAW;AACpH,cAAM,WAAW,SAAS,iBAAiB,KAAK,UAAQ,KAAK,SAAS,OAAO,IAAI;AACjF,eAAO;AAAA,UACL,MAAM,OAAO;AAAA,UACb,WAAW,OAAO;AAAA,UAClB,WAAW,UAAU,aAAa,SAAS,UAAU,OAAO,OAAO,YAAY,SAAS,YAAY;AAAA,UACpG,UAAU,OAAO,YAAY;AAAA,UAC7B,UAAU;AAAA,YACR,GAAG,OAAO;AAAA,YACV,GAAI,OAAO,aAAa,UAAU,WAAW,OAAO,OAAO,YACvD,EAAE,iBAAiB,OAAO,UAAU,IACpC,CAAC;AAAA,UACP;AAAA,QACF;AAAA,MACF,CAAC;AAAA,IACH,CAAC;AACD,QAAI,UAAU,oBAAoB,OAAW,QAAO,UAAU;AAC9D,WAAO;AAAA,EACT;AACA,SAAO;AAAA,IACL,IAAI,KAAK;AAAA,IACT,YAAY,KAAK;AAAA,IACjB,OAAO,KAAK;AAAA,IACZ,OAAO,KAAK;AAAA,IACZ,gBAAgB,uBAAuB,IAAI;AAAA,IAC3C,sBAAsB,KAAK;AAAA,IAC3B,sBAAsB,KAAK;AAAA,IAC3B,eAAe,KAAK,UAAU,CAAC,GAAG,IAAI,CAAC,UAAU;AAAA,MAC/C,IAAI,KAAK;AAAA,MACT,MAAM,SAAS,QAAQ,KAAK,EAAE,CAAC,IAAI,QAAQ,KAAK,QAAQ,CAAC;AAAA,MACzD,aAAa,KAAK,KAAK,MAAM;AAAA,MAC7B,WAAW,KAAK;AAAA,MAChB,WAAW,KAAK;AAAA,MAChB,YAAY;AAAA,IACd,EAAE;AAAA,IACF,kBAAkB,KAAK,mBAAmB,CAAC,GAAG,IAAI,CAAC,YAAY;AAAA,MAC7D,MAAM,OAAO;AAAA,MACb,WAAW,OAAO;AAAA,MAClB,WAAW;AAAA,MACX,UAAU,OAAO,YAAY;AAAA,MAC7B,UAAU;AAAA,QACR,GAAG,OAAO;AAAA,QACV,iBAAiB,OAAO,aAAa;AAAA,MACvC;AAAA,IACF,EAAE;AAAA,IACF,MAAM,KAAK;AAAA,EACb;AACF;AAEA,SAAS,uBACP,MACgC;AAChC,MAAI,CAAC,KAAK,eAAgB,QAAO;AACjC,MAAI,sBAAsB,KAAK,gBAAgB;AAC7C,UAAM,IAAI;AAAA,MACR,QAAQ,KAAK,EAAE;AAAA,IACjB;AAAA,EACF;AACA,SAAO,KAAK;AACd;AAEA,SAAS,oBAAoB,SAAuC;AAClE,QAAM,OAAoC,QAAQ,YAAY,SAAS,SACnE,SACA,QAAQ,YAAY,SAAS,SAAS,SACpC,QAAQ,YAAY,SAAS,YAAY,mBAAmB;AAClE,SAAO;AAAA,IACL,iBAAiB;AAAA,IACjB;AAAA,IACA,YAAY,QAAQ,YAAY;AAAA,IAChC,UAAU,QAAQ,YAAY;AAAA,IAC9B,oBAAoB,QAAQ,YAAY;AAAA,IACxC,WAAW,CAAC,UAAU,SAAS,QAAQ,WAAW,SAAS;AAAA,IAC3D,eAAe,QAAQ,YAAY;AAAA,IACnC,kBAAkB,QAAQ,YAAY;AAAA,EACxC;AACF;AAEA,SAAS,cAAc,SAAqC;AAC1D,QAAM,WAAW,QAAQ,YAAY,SAAS;AAC9C,MAAI,aAAa,QAAW;AAC1B,WAAO,sBAAsB,MAAM,EAAE,IAAI,GAAG,EAAE,MAAM,QAAQ;AAAA,EAC9D;AACA,QAAM,WAAW,QAAQ,YAAY,kBAAkB,CAAC;AACxD,MAAI,SAAS,OAAQ,QAAO,SAAS,IAAI,CAAC,aAAa;AAAA,IACrD,MAAM,QAAQ;AAAA,IACd,aAAa,QAAQ;AAAA,IACrB,aAAa,QAAQ;AAAA,IACrB,iBAAiB,KAAK,QAAQ,gBAAgB;AAAA,IAC9C,YAAY,QAAQ;AAAA,IACpB,WAAW,QAAQ;AAAA,EACrB,EAAE;AACF,SAAO,QAAQ,YAAY,UAAU,IAAI,CAAC,UAAU;AAAA,IAClD;AAAA,IACA,aAAa,iBAAiB,IAAI;AAAA,IAClC,aAAa,CAAC;AAAA,IACd,iBAAiB,YAAY,CAAC,CAAC;AAAA,IAC/B,YAAY;AAAA,IACZ,WAAW,QAAQ,YAAY;AAAA,EACjC,EAAE;AACJ;AAEA,SAAS,eAAe,QAAwC;AAC9D,QAAM,OAAO;AAAA,IACX,IAAI,OAAO;AAAA,IACX,SAAS,OAAO;AAAA,IAChB,QAAQ,OAAO;AAAA,IACf,UAAU,OAAO;AAAA,IACjB,gBAAgB,OAAO;AAAA,IACvB,YAAY,OAAO;AAAA,EACrB;AACA,MAAI,OAAO,SAAS,cAAe,QAAO;AAAA,IACxC,GAAG;AAAA,IACH,MAAM;AAAA,IACN,WAAW,YAAY,MAAM;AAAA,IAC7B,OAAO,cAAc,MAAM;AAAA,IAC3B,aAAa,OAAO;AAAA,IACpB,mBAAmB,OAAO;AAAA,EAC5B;AACA,MAAI,OAAO,SAAS,mBAAmB;AACrC,QAAI,OAAO,SAAS,wBAAwB,QAAW;AACrD,YAAM,IAAI,MAAM,mBAAmB,OAAO,EAAE,8DAA8D;AAAA,IAC5G;AACA,UAAM,WAAW,yBAAyB,UAAU,OAAO,SAAS,cAAc;AAGlF,UAAM,kBAAkB,SAAS,WAAW,SAAS,KAAK,SAAS,qBAC9D,SAAS,KAAK,eAAe,UAAa,OAAO,eAAe;AACrE,WAAO;AAAA,MACL,GAAG;AAAA,MACH,MAAM;AAAA,MACN,GAAI,OAAO,YAAY,SAAY,CAAC,IAAI,EAAE,SAAS,OAAO,QAAQ;AAAA,MAClE,aAAa,wBAAwB,MAAM,OAAO,SAAS,mBAAmB;AAAA,MAC9E,WAAW,OAAO;AAAA,MAClB,GAAI,kBAAkB,CAAC,IAAI,EAAE,YAAY,OAAO,WAAW;AAAA,MAC3D,eAAe;AAAA,IACjB;AAAA,EACF;AACA,MAAI,OAAO,SAAS,QAAS,QAAO;AAAA,IAClC,GAAG;AAAA,IACH,MAAM;AAAA,IACN,WAAW,YAAY,MAAM;AAAA,IAC7B,cAAc,OAAO;AAAA,EACvB;AACA,SAAO;AAAA,IACL,GAAG;AAAA,IACH,GAAG,2BAA2B,MAAM;AAAA,EACtC;AACF;AAEA,SAAS,qBAAqB,SAAkB;AAC9C,QAAM,WAAW,QAAQ,SAAS;AAClC,MAAI,aAAa,QAAW;AAC1B,WAAO,4BAA4B,MAAM,EAAE,IAAI,GAAG,EAAE,MAAM,QAAQ;AAAA,EACpE;AACA,SAAO;AAAA,IACL,GAAI,QAAQ,aAAa,gBACrB,CAAC,EAAE,IAAI,SAAS,UAAU,MAAM,QAAQ,QAAQ,YAAY,UAAU,CAAC,IACvE,CAAC;AAAA,IACL,GAAI,QAAQ,aAAa,gBACrB,CAAC,EAAE,IAAI,eAAe,UAAU,MAAM,QAAQ,CAAC,EAAE,CAAC,IAClD,CAAC;AAAA,IACL,GAAI,QAAQ,aAAa,4BACrB,CAAC,EAAE,IAAI,oBAAoB,UAAU,MAAM,QAAQ,QAAQ,OAAO,iBAAiB,CAAC,IACpF,CAAC;AAAA,IACL,GAAG,QAAQ,OAAO,mBAAmB,IAAI,CAAC,WAAW;AAAA,MACnD,IAAI,iBAAiB,QAAQ,KAAK,CAAC;AAAA,MACnC,UAAU;AAAA,MACV,QAAQ,CAAC,KAAK;AAAA,IAChB,EAAE;AAAA,EACJ,EAAE,IAAI,CAAC,iBAAiB;AAAA,IACtB,GAAG;AAAA,IACH,aAAa,YAAY,OAAO,gBAAgB,iBAA0B;AAAA,EAC5E,EAAE;AACJ;AAEA,SAAS,sBAAsB,OAIZ;AACjB,MACE,MAAM,MAAM,oBAAoB,OAAO,MAAM,mBAAmB,MAC7D,MAAM,MAAM,mBAAmB,gBAAgB,MAAM,mBAAmB,aAC3E;AACA,UAAM,IAAI,MAAM,iEAAiE;AAAA,EACnF;AACA,MACE,MAAM,MAAM,oBAAoB,OAAO,MAAM,mBAAmB,MAC7D,MAAM,MAAM,mBAAmB,gBAAgB,MAAM,mBAAmB,aAC3E;AACA,UAAM,IAAI,MAAM,kEAAkE;AAAA,EACpF;AACA,MAAI,YAAY,MAAM,MAAM,WAAW,MAAM,YAAY,MAAM,mBAAmB,QAAQ,GAAG;AAC3F,UAAM,IAAI,MAAM,0EAA0E;AAAA,EAC5F;AACA,MAAI,YAAY,MAAM,MAAM,OAAO,MAAM,YAAY,MAAM,mBAAmB,OAAO,GAAG;AACtF,UAAM,IAAI,MAAM,sEAAsE;AAAA,EACxF;AACA,SAAO,MAAM;AACf;AAEA,SAAS,cAAc,QAAsD;AAC3E,QAAM,WAAW,OAAO,SAAS,mBAAmB,SAAY,OAAO,yBAAyB,MAAM,OAAO,SAAS,cAAc;AACpI,QAAM,WAAW,UAAU,SAAS,iBAAiB,SAAS,OAAO,eAAe,OAAO,MAAM,cAAc,SAAS,MAAM,YAAY,OAAO,MAAM,UAAU,SAAS,MAAM,WAAW;AAC3L,SAAO,EAAE,GAAG,OAAO,OAAO,SAAS;AACrC;AAEA,SAAS,YAAY,QAAmF;AACtG,QAAMA,eAAc,OAAO,OAAO,MAAM;AACxC,QAAM,YAAY,IAAI,YAAY,EAAE,OAAO,OAAO,MAAM,EAAE;AAC1D,QAAM,YAAY,OAAO,SAAS;AAClC,MAAI,cAAc,QAAW;AAC3B,UAAM,QAAQ,wBAAwB,MAAM,SAAS;AACrD,QAAI,MAAM,gBAAgBA,gBAAe,MAAM,cAAc,aAAa,MAAM,eAAe,SAAU,OAAM,IAAI,MAAM,2DAA2D;AACpL,WAAO;AAAA,EACT;AACA,SAAO,EAAE,IAAI,UAAU,OAAO,EAAE,IAAI,MAAM,WAAW,QAAQ,OAAO,EAAE,CAAC,cAAc,aAAAA,cAAa,WAAW,WAAW,iBAAiB,YAAY,WAAW;AAClK;AAEA,SAAS,KAAK,OAAuB;AACnC,SAAO,iBAAiB,KAAK,KAAK,IAAI,QAAQ,YAAY,KAAK;AACjE;AAEA,SAAS,QAAQ,OAAuB;AACtC,QAAM,aAAa,MAAM,KAAK,EAAE,QAAQ,qBAAqB,GAAG,EAC7D,QAAQ,YAAY,EAAE;AACzB,SAAO,cAAc;AACvB;",
6
+ "names": ["contentHash"]
7
7
  }
@@ -5,70 +5,89 @@ import {
5
5
  } from "./chunk-TCHAWJMB.js";
6
6
 
7
7
  // src/model-starter-attempts.ts
8
+ import { z as z2 } from "zod";
9
+
10
+ // src/evaluation-project-context.ts
8
11
  import { z } from "zod";
12
+ var EvaluationProjectContextSchema = z.object({
13
+ id: z.string().trim().min(1).max(240),
14
+ revision: z.number().int().positive(),
15
+ contentHash: z.string().regex(/^[a-f0-9]{64}$/),
16
+ targetId: z.string().trim().min(1).max(240).nullable()
17
+ }).strict();
18
+ var EvaluationMessagesSchema = z.array(z.object({
19
+ role: z.enum(["system", "user", "assistant"]),
20
+ content: z.string().max(1e5)
21
+ }).strict()).max(100).superRefine((messages, context) => {
22
+ if (messages.reduce((total, message) => total + message.content.length, 0) > 1e5)
23
+ context.addIssue({ code: "custom", message: "Evaluation prompt exceeds 100,000 characters." });
24
+ });
25
+
26
+ // src/model-starter-attempts.ts
9
27
  import { RewardCompositionSchema } from "@openpond/evals/rewards";
10
28
  import { TaskGradeSchema } from "@openpond/evals/graders";
11
- var IdSchema = z.string().trim().min(1).max(200);
12
- var HashSchema = z.string().regex(/^[a-f0-9]{64}$/);
13
- var ModelStarterAttemptPolicySchema = z.discriminatedUnion("kind", [
14
- z.object({ kind: z.literal("hosted_chat"), modelId: IdSchema, maxOutputTokens: z.number().int().min(1).max(4096).default(1024), temperature: z.number().min(0).max(2).default(0), topP: z.number().gt(0).max(1).default(1) }).strict(),
15
- z.object({ kind: z.literal("fixture"), fixtureId: IdSchema }).strict()
29
+ var IdSchema = z2.string().trim().min(1).max(200);
30
+ var HashSchema = z2.string().regex(/^[a-f0-9]{64}$/);
31
+ var ModelStarterAttemptPolicySchema = z2.discriminatedUnion("kind", [
32
+ z2.object({ kind: z2.literal("hosted_chat"), modelId: IdSchema, messages: EvaluationMessagesSchema.optional(), maxOutputTokens: z2.number().int().min(1).max(4096).default(1024), temperature: z2.number().min(0).max(2).default(0), topP: z2.number().gt(0).max(1).default(1) }).strict(),
33
+ z2.object({ kind: z2.literal("fixture"), fixtureId: IdSchema }).strict()
16
34
  ]);
17
- var ModelStarterAttemptRequestSchema = z.object({
18
- schemaVersion: z.literal("openpond.modelStarterAttemptRequest.v1"),
35
+ var ModelStarterAttemptRequestSchema = z2.object({
36
+ schemaVersion: z2.literal("openpond.modelStarterAttemptRequest.v1"),
19
37
  operationId: IdSchema,
20
38
  teamId: IdSchema,
21
- modelProjectId: IdSchema,
39
+ modelProjectId: IdSchema.nullable(),
40
+ project: EvaluationProjectContextSchema.optional(),
22
41
  taskset: ModelProjectVersionedRefSchema,
23
42
  taskId: IdSchema,
24
- environmentSeed: z.number().int().min(0).max(2147483647).default(0),
43
+ environmentSeed: z2.number().int().min(0).max(2147483647).default(0),
25
44
  policy: ModelStarterAttemptPolicySchema
26
45
  }).strict();
27
- var ModelStarterAttemptChoicesQuerySchema = z.object({ modelProjectId: IdSchema, taskset: ModelProjectVersionedRefSchema, afterTaskId: IdSchema.optional(), limit: z.number().int().min(1).max(100).default(25) }).strict();
28
- var ModelStarterAttemptChoicesSchema = z.object({
29
- modelProjectId: IdSchema,
46
+ var ModelStarterAttemptChoicesQuerySchema = z2.object({ modelProjectId: IdSchema.nullable(), taskset: ModelProjectVersionedRefSchema, afterTaskId: IdSchema.optional(), limit: z2.number().int().min(1).max(100).default(25) }).strict();
47
+ var ModelStarterAttemptChoicesSchema = z2.object({
48
+ modelProjectId: IdSchema.nullable(),
30
49
  taskset: ModelProjectVersionedRefSchema,
31
- available: z.boolean(),
32
- unavailableReason: z.string().max(1e3).nullable(),
33
- models: z.array(z.object({ id: IdSchema, name: z.string().max(500) }).strict()).max(200),
34
- tasks: z.array(z.object({ id: IdSchema, split: z.enum(["train", "validation", "frozen_eval"]), inputPreview: z.string().max(500), fixtures: z.array(z.object({ id: IdSchema, label: z.string().max(200) }).strict()).max(1e3) }).strict()).max(100),
50
+ available: z2.boolean(),
51
+ unavailableReason: z2.string().max(1e3).nullable(),
52
+ models: z2.array(z2.object({ id: IdSchema, name: z2.string().max(500) }).strict()).max(200),
53
+ tasks: z2.array(z2.object({ id: IdSchema, split: z2.enum(["train", "validation", "frozen_eval"]), inputPreview: z2.string().max(500), fixtures: z2.array(z2.object({ id: IdSchema, label: z2.string().max(200) }).strict()).max(1e3) }).strict()).max(100),
35
54
  nextCursor: IdSchema.nullable()
36
55
  }).strict();
37
- var ModelStarterAttemptSummarySchema = z.object({
38
- schemaVersion: z.literal("openpond.modelStarterAttemptSummary.v1"),
56
+ var ModelStarterAttemptSummarySchema = z2.object({
57
+ schemaVersion: z2.literal("openpond.modelStarterAttemptSummary.v1"),
39
58
  id: IdSchema,
40
- revision: z.number().int().positive(),
59
+ revision: z2.number().int().positive(),
41
60
  request: ModelStarterAttemptRequestSchema,
42
- status: z.enum(["queued", "running", "cancelling", "completed", "failed", "cancelled"]),
43
- policySnapshot: z.object({ modelId: IdSchema, provider: IdSchema, upstreamModelId: z.string().min(1).max(500), configurationHash: HashSchema }).strict().nullable(),
44
- createdAt: z.iso.datetime(),
45
- startedAt: z.iso.datetime().nullable(),
46
- completedAt: z.iso.datetime().nullable(),
47
- cleanupComplete: z.boolean(),
48
- resultAvailable: z.boolean(),
49
- score: z.number().min(0).max(1).nullable(),
50
- gradingStatus: z.enum(["not_started", "scored", "unscorable", "not_configured"]).default("not_started"),
51
- passed: z.boolean().nullable(),
52
- outputPreview: z.string().max(500).nullable(),
53
- error: z.object({ code: IdSchema, message: z.string().max(2e3) }).strict().nullable()
61
+ status: z2.enum(["queued", "running", "cancelling", "completed", "failed", "cancelled"]),
62
+ policySnapshot: z2.object({ modelId: IdSchema, provider: IdSchema, upstreamModelId: z2.string().min(1).max(500), configurationHash: HashSchema }).strict().nullable(),
63
+ createdAt: z2.iso.datetime(),
64
+ startedAt: z2.iso.datetime().nullable(),
65
+ completedAt: z2.iso.datetime().nullable(),
66
+ cleanupComplete: z2.boolean(),
67
+ resultAvailable: z2.boolean(),
68
+ score: z2.number().min(0).max(1).nullable(),
69
+ gradingStatus: z2.enum(["not_started", "scored", "unscorable", "not_configured"]).default("not_started"),
70
+ passed: z2.boolean().nullable(),
71
+ outputPreview: z2.string().max(500).nullable(),
72
+ error: z2.object({ code: IdSchema, message: z2.string().max(2e3) }).strict().nullable()
54
73
  }).strict().superRefine((value, context) => {
55
74
  if (value.request.policy.kind === "fixture" !== (value.policySnapshot === null)) context.addIssue({ code: "custom", path: ["policySnapshot"], message: "Fixture execution must not claim a model identity." });
56
75
  if (value.status === "completed" && (!value.cleanupComplete || !value.resultAvailable || !value.completedAt)) context.addIssue({ code: "custom", message: "Completed attempts require retained results and confirmed cleanup." });
57
76
  });
58
- var ModelStarterAttemptListQuerySchema = z.object({ modelProjectId: IdSchema, afterId: IdSchema.optional(), limit: z.number().int().min(1).max(100).default(25) }).strict();
59
- var ModelStarterAttemptPageSchema = z.object({ items: z.array(ModelStarterAttemptSummarySchema).max(100), nextCursor: IdSchema.nullable() }).strict();
60
- var ModelStarterAttemptResultSchema = z.object({
61
- schemaVersion: z.literal("openpond.modelStarterAttemptResult.v1"),
77
+ var ModelStarterAttemptListQuerySchema = z2.object({ modelProjectId: IdSchema, afterId: IdSchema.optional(), limit: z2.number().int().min(1).max(100).default(25) }).strict();
78
+ var ModelStarterAttemptPageSchema = z2.object({ items: z2.array(ModelStarterAttemptSummarySchema).max(100), nextCursor: IdSchema.nullable() }).strict();
79
+ var ModelStarterAttemptResultSchema = z2.object({
80
+ schemaVersion: z2.literal("openpond.modelStarterAttemptResult.v1"),
62
81
  attempt: ModelStarterAttemptSummarySchema,
63
- output: z.string().max(1048576).nullable(),
82
+ output: z2.string().max(1048576).nullable(),
64
83
  // Only model-facing messages, never the private environment snapshot.
65
- messages: z.array(z.record(z.string(), z.unknown())).max(4002),
84
+ messages: z2.array(z2.record(z2.string(), z2.unknown())).max(4002),
66
85
  composition: RewardCompositionSchema.nullable(),
67
86
  // Ordinary Tasksets have no Reward binding. Absence preserves the hash of
68
87
  // already retained v1 results; new producers explicitly return null or a grade.
69
88
  grade: TaskGradeSchema.nullable().optional(),
70
- environment: z.object({ status: z.enum(["completed", "budget_exhausted", "cancelled", "timed_out", "policy_failure", "environment_failure"]), collected: z.boolean(), definition: ModelProjectVersionedRefSchema, initialStateHash: HashSchema.nullable(), finalStateHash: HashSchema.nullable(), attemptHash: HashSchema }).strict(),
71
- providerRequestIds: z.array(IdSchema).max(1001),
89
+ environment: z2.object({ status: z2.enum(["completed", "budget_exhausted", "cancelled", "timed_out", "policy_failure", "environment_failure"]), collected: z2.boolean(), definition: ModelProjectVersionedRefSchema, initialStateHash: HashSchema.nullable(), finalStateHash: HashSchema.nullable(), attemptHash: HashSchema }).strict(),
90
+ providerRequestIds: z2.array(IdSchema).max(1001),
72
91
  contentHash: HashSchema
73
92
  }).strict().superRefine((value, context) => {
74
93
  if (!value.grade) return;
@@ -104,7 +123,7 @@ var OpenPondModelStarterAttemptsClient = class {
104
123
  }
105
124
  async choices(value, options = {}) {
106
125
  const query = ModelStarterAttemptChoicesQuerySchema.parse(value);
107
- const params = new URLSearchParams({ modelProjectId: query.modelProjectId, tasksetId: query.taskset.id, revision: String(query.taskset.revision), contentHash: query.taskset.contentHash, limit: String(query.limit) });
126
+ const params = new URLSearchParams({ ...query.modelProjectId ? { modelProjectId: query.modelProjectId } : {}, tasksetId: query.taskset.id, revision: String(query.taskset.revision), contentHash: query.taskset.contentHash, limit: String(query.limit) });
108
127
  if (query.afterTaskId) params.set("afterTaskId", query.afterTaskId);
109
128
  const result = ModelStarterAttemptChoicesSchema.parse(await this.#request(`/choices?${params}`, "GET", void 0, options.signal));
110
129
  if (result.modelProjectId !== query.modelProjectId || canonicalJson(result.taskset) !== canonicalJson(query.taskset) || result.tasks.length > query.limit || new Set(result.tasks.map((task) => task.id)).size !== result.tasks.length) throw new OpenPondModelStarterAttemptError(502, "attempt_choices_mismatch", "Attempt choices differ from the selected model or Taskset.");
@@ -155,7 +174,7 @@ var OpenPondModelStarterAttemptsClient = class {
155
174
  }
156
175
  const value = JSON.parse(text);
157
176
  if (!response.ok) {
158
- const error = z.object({ code: IdSchema, message: z.string().max(2e3) }).safeParse(value);
177
+ const error = z2.object({ code: IdSchema, message: z2.string().max(2e3) }).safeParse(value);
159
178
  throw new OpenPondModelStarterAttemptError(response.status, error.success ? error.data.code : "attempt_request_failed", error.success ? error.data.message : "Attempt request failed.");
160
179
  }
161
180
  return value;
@@ -163,6 +182,8 @@ var OpenPondModelStarterAttemptsClient = class {
163
182
  };
164
183
 
165
184
  export {
185
+ EvaluationProjectContextSchema,
186
+ EvaluationMessagesSchema,
166
187
  ModelStarterAttemptPolicySchema,
167
188
  ModelStarterAttemptRequestSchema,
168
189
  ModelStarterAttemptChoicesQuerySchema,
@@ -174,4 +195,4 @@ export {
174
195
  OpenPondModelStarterAttemptError,
175
196
  OpenPondModelStarterAttemptsClient
176
197
  };
177
- //# sourceMappingURL=chunk-ECTMEJ5N.js.map
198
+ //# sourceMappingURL=chunk-IKTZ2CD4.js.map
@@ -0,0 +1,7 @@
1
+ {
2
+ "version": 3,
3
+ "sources": ["../src/model-starter-attempts.ts", "../src/evaluation-project-context.ts"],
4
+ "sourcesContent": ["import { z } from \"zod\";\nimport { EvaluationProjectContextSchema, EvaluationMessagesSchema } from \"./evaluation-project-context.js\";\nexport { EvaluationProjectContextSchema, EvaluationMessagesSchema, type EvaluationProjectContext } from \"./evaluation-project-context.js\";\nimport { RewardCompositionSchema } from \"@openpond/evals/rewards\";\nimport { TaskGradeSchema } from \"@openpond/evals/graders\";\nimport { ModelProjectVersionedRefSchema } from \"./model-projects.js\";\nimport { canonicalJson, canonicalSha256 } from \"./protocol.js\";\n\nconst IdSchema = z.string().trim().min(1).max(200);\nconst HashSchema = z.string().regex(/^[a-f0-9]{64}$/);\nexport const ModelStarterAttemptPolicySchema = z.discriminatedUnion(\"kind\", [\n z.object({ kind: z.literal(\"hosted_chat\"), modelId: IdSchema, messages: EvaluationMessagesSchema.optional(), maxOutputTokens: z.number().int().min(1).max(4_096).default(1_024), temperature: z.number().min(0).max(2).default(0), topP: z.number().gt(0).max(1).default(1) }).strict(),\n z.object({ kind: z.literal(\"fixture\"), fixtureId: IdSchema }).strict(),\n]);\n/** The server resolves task input, private state, verifier and fixture script.\n * A caller can select releases, but cannot submit replacement execution bytes. */\nexport const ModelStarterAttemptRequestSchema = z.object({\n schemaVersion: z.literal(\"openpond.modelStarterAttemptRequest.v1\"),\n operationId: IdSchema, teamId: IdSchema, modelProjectId: IdSchema.nullable(),\n project: EvaluationProjectContextSchema.optional(),\n taskset: ModelProjectVersionedRefSchema, taskId: IdSchema,\n environmentSeed: z.number().int().min(0).max(2_147_483_647).default(0),\n policy: ModelStarterAttemptPolicySchema,\n}).strict();\nexport type ModelStarterAttemptRequest = z.infer<typeof ModelStarterAttemptRequestSchema>;\nexport const ModelStarterAttemptChoicesQuerySchema = z.object({ modelProjectId: IdSchema.nullable(), taskset: ModelProjectVersionedRefSchema, afterTaskId: IdSchema.optional(), limit: z.number().int().min(1).max(100).default(25) }).strict();\nexport const ModelStarterAttemptChoicesSchema = z.object({\n modelProjectId: IdSchema.nullable(), taskset: ModelProjectVersionedRefSchema, available: z.boolean(), unavailableReason: z.string().max(1_000).nullable(),\n models: z.array(z.object({ id: IdSchema, name: z.string().max(500) }).strict()).max(200),\n tasks: z.array(z.object({ id: IdSchema, split: z.enum([\"train\", \"validation\", \"frozen_eval\"]), inputPreview: z.string().max(500), fixtures: z.array(z.object({ id: IdSchema, label: z.string().max(200) }).strict()).max(1_000) }).strict()).max(100),\n nextCursor: IdSchema.nullable(),\n}).strict();\nexport const ModelStarterAttemptSummarySchema = z.object({\n schemaVersion: z.literal(\"openpond.modelStarterAttemptSummary.v1\"),\n id: IdSchema, revision: z.number().int().positive(), request: ModelStarterAttemptRequestSchema,\n status: z.enum([\"queued\", \"running\", \"cancelling\", \"completed\", \"failed\", \"cancelled\"]),\n policySnapshot: z.object({ modelId: IdSchema, provider: IdSchema, upstreamModelId: z.string().min(1).max(500), configurationHash: HashSchema }).strict().nullable(),\n createdAt: z.iso.datetime(), startedAt: z.iso.datetime().nullable(), completedAt: z.iso.datetime().nullable(),\n cleanupComplete: z.boolean(), resultAvailable: z.boolean(), score: z.number().min(0).max(1).nullable(),\n gradingStatus: z.enum([\"not_started\", \"scored\", \"unscorable\", \"not_configured\"]).default(\"not_started\"),\n passed: z.boolean().nullable(), outputPreview: z.string().max(500).nullable(),\n error: z.object({ code: IdSchema, message: z.string().max(2_000) }).strict().nullable(),\n}).strict().superRefine((value, context) => {\n if ((value.request.policy.kind === \"fixture\") !== (value.policySnapshot === null)) context.addIssue({ code: \"custom\", path: [\"policySnapshot\"], message: \"Fixture execution must not claim a model identity.\" });\n if (value.status === \"completed\" && (!value.cleanupComplete || !value.resultAvailable || !value.completedAt)) context.addIssue({ code: \"custom\", message: \"Completed attempts require retained results and confirmed cleanup.\" });\n});\nexport type ModelStarterAttemptSummary = z.infer<typeof ModelStarterAttemptSummarySchema>;\nexport const ModelStarterAttemptListQuerySchema = z.object({ modelProjectId: IdSchema, afterId: IdSchema.optional(), limit: z.number().int().min(1).max(100).default(25) }).strict();\nexport const ModelStarterAttemptPageSchema = z.object({ items: z.array(ModelStarterAttemptSummarySchema).max(100), nextCursor: IdSchema.nullable() }).strict();\nexport const ModelStarterAttemptResultSchema = z.object({\n schemaVersion: z.literal(\"openpond.modelStarterAttemptResult.v1\"),\n attempt: ModelStarterAttemptSummarySchema,\n output: z.string().max(1_048_576).nullable(),\n // Only model-facing messages, never the private environment snapshot.\n messages: z.array(z.record(z.string(), z.unknown())).max(4_002),\n composition: RewardCompositionSchema.nullable(),\n // Ordinary Tasksets have no Reward binding. Absence preserves the hash of\n // already retained v1 results; new producers explicitly return null or a grade.\n grade: TaskGradeSchema.nullable().optional(),\n environment: z.object({ status: z.enum([\"completed\", \"budget_exhausted\", \"cancelled\", \"timed_out\", \"policy_failure\", \"environment_failure\"]), collected: z.boolean(), definition: ModelProjectVersionedRefSchema, initialStateHash: HashSchema.nullable(), finalStateHash: HashSchema.nullable(), attemptHash: HashSchema }).strict(),\n providerRequestIds: z.array(IdSchema).max(1_001),\n contentHash: HashSchema,\n}).strict().superRefine((value, context) => {\n if (!value.grade) return;\n if (value.composition) context.addIssue({ code: \"custom\", path: [\"grade\"], message: \"An ordinary grade cannot also claim a Reward composition.\" });\n if (![\"completed\", \"failed\"].includes(value.attempt.status) || !value.attempt.resultAvailable || value.attempt.score !== value.grade.score || value.attempt.gradingStatus !== value.grade.gradingStatus || value.attempt.passed !== (value.grade.score === null ? null : value.grade.passed)) context.addIssue({ code: \"custom\", path: [\"grade\"], message: \"Ordinary grade differs from its terminal attempt summary.\" });\n});\nexport type ModelStarterAttemptResult = z.infer<typeof ModelStarterAttemptResultSchema>;\n\nexport class OpenPondModelStarterAttemptError extends Error {\n constructor(readonly status: number, readonly code: string, message: string) { super(message); this.name = \"OpenPondModelStarterAttemptError\"; }\n}\n\nexport class OpenPondModelStarterAttemptsClient {\n readonly #options: { baseUrl: string; apiKey: string; teamId: string; fetch?: typeof globalThis.fetch };\n constructor(options: { baseUrl: string; apiKey: string; teamId: string; fetch?: typeof globalThis.fetch }) {\n const url = new URL(options.baseUrl);\n if (!options.apiKey.trim() || !options.teamId.trim() || ![\"https:\", \"http:\"].includes(url.protocol) || url.username || url.password || url.search || url.hash) throw new Error(\"Starter attempts require an HTTP(S) endpoint and workspace credentials.\");\n this.#options = { ...options, baseUrl: url.toString().replace(/\\/+$/, \"\") };\n }\n async create(value: z.input<typeof ModelStarterAttemptRequestSchema>, options: { signal?: AbortSignal } = {}) {\n const request = ModelStarterAttemptRequestSchema.parse(value);\n if (request.teamId !== this.#options.teamId) throw new Error(\"Attempt belongs to another workspace.\");\n const result = this.#summary(await this.#request(\"\", \"POST\", request, options.signal));\n if (canonicalJson(result.request) !== canonicalJson(request)) throw new OpenPondModelStarterAttemptError(502, \"attempt_request_mismatch\", \"Attempt receipt differs from the submitted request.\");\n return result;\n }\n async get(id: string, options: { signal?: AbortSignal } = {}) { return this.#summary(await this.#request(`/${encodeURIComponent(IdSchema.parse(id))}`, \"GET\", undefined, options.signal), id); }\n async choices(value: z.input<typeof ModelStarterAttemptChoicesQuerySchema>, options: { signal?: AbortSignal } = {}) {\n const query = ModelStarterAttemptChoicesQuerySchema.parse(value);\n const params = new URLSearchParams({ ...(query.modelProjectId ? { modelProjectId: query.modelProjectId } : {}), tasksetId: query.taskset.id, revision: String(query.taskset.revision), contentHash: query.taskset.contentHash, limit: String(query.limit) });\n if (query.afterTaskId) params.set(\"afterTaskId\", query.afterTaskId);\n const result = ModelStarterAttemptChoicesSchema.parse(await this.#request(`/choices?${params}`, \"GET\", undefined, options.signal));\n if (result.modelProjectId !== query.modelProjectId || canonicalJson(result.taskset) !== canonicalJson(query.taskset) || result.tasks.length > query.limit || new Set(result.tasks.map(task => task.id)).size !== result.tasks.length) throw new OpenPondModelStarterAttemptError(502, \"attempt_choices_mismatch\", \"Attempt choices differ from the selected model or Taskset.\");\n return result;\n }\n async cancel(id: string, options: { signal?: AbortSignal } = {}) { return this.#summary(await this.#request(`/${encodeURIComponent(IdSchema.parse(id))}/cancel`, \"POST\", undefined, options.signal), id); }\n async result(id: string, options: { signal?: AbortSignal } = {}) {\n const value = ModelStarterAttemptResultSchema.parse(await this.#request(`/${encodeURIComponent(IdSchema.parse(id))}/result`, \"GET\", undefined, options.signal));\n this.#summary(value.attempt, id);\n const { contentHash, ...content } = value;\n if (!value.attempt.resultAvailable || await canonicalSha256(content) !== contentHash) throw new OpenPondModelStarterAttemptError(502, \"attempt_result_integrity\", \"Attempt result integrity failed.\");\n return value;\n }\n async list(value: z.input<typeof ModelStarterAttemptListQuerySchema>, options: { signal?: AbortSignal } = {}) {\n const query = ModelStarterAttemptListQuerySchema.parse(value);\n const params = new URLSearchParams({ modelProjectId: query.modelProjectId, limit: String(query.limit) });\n if (query.afterId) params.set(\"afterId\", query.afterId);\n const page = ModelStarterAttemptPageSchema.parse(await this.#request(`?${params}`, \"GET\", undefined, options.signal));\n if (page.items.length > query.limit || new Set(page.items.map(item => item.id)).size !== page.items.length || page.items.some(item => this.#summary(item).request.modelProjectId !== query.modelProjectId)) throw new OpenPondModelStarterAttemptError(502, \"attempt_page_mismatch\", \"Attempt page differs from the selected model.\");\n return page;\n }\n #summary(value: unknown, id?: string): ModelStarterAttemptSummary {\n const result = ModelStarterAttemptSummarySchema.parse(value);\n if (result.request.teamId !== this.#options.teamId || (id && result.id !== id)) throw new OpenPondModelStarterAttemptError(502, \"attempt_identity_mismatch\", \"Attempt receipt belongs to another identity or workspace.\");\n return result;\n }\n async #request(path: string, method: string, body: unknown, signal?: AbortSignal): Promise<unknown> {\n const response = await (this.#options.fetch ?? globalThis.fetch)(`${this.#options.baseUrl}/v1/model-starter-attempts${path}`, { method, headers: { Authorization: `Bearer ${this.#options.apiKey}`, \"X-OpenPond-Team-Id\": this.#options.teamId, \"Content-Type\": \"application/json\", Accept: \"application/json\" }, ...(body === undefined ? {} : { body: JSON.stringify(body) }), signal, redirect: \"error\" });\n const reader = response.body?.getReader();\n if (!reader) throw new OpenPondModelStarterAttemptError(response.status, \"attempt_empty_response\", \"Attempt response was empty.\");\n let bytes = 0; let text = \"\"; const decoder = new TextDecoder();\n try {\n while (true) {\n const chunk = await reader.read(); if (chunk.done) break;\n bytes += chunk.value.byteLength;\n if (bytes > 16_777_216) throw new OpenPondModelStarterAttemptError(502, \"attempt_response_too_large\", \"Attempt response exceeded its byte limit.\");\n text += decoder.decode(chunk.value, { stream: true });\n }\n text += decoder.decode();\n } finally { await reader.cancel(); reader.releaseLock(); }\n const value: unknown = JSON.parse(text);\n if (!response.ok) {\n const error = z.object({ code: IdSchema, message: z.string().max(2_000) }).safeParse(value);\n throw new OpenPondModelStarterAttemptError(response.status, error.success ? error.data.code : \"attempt_request_failed\", error.success ? error.data.message : \"Attempt request failed.\");\n }\n return value;\n }\n}\n", "import { z } from \"zod\";\n\n/** The selected Project revision is retained independently of model artifacts.\n * Hosts resolve and authorize this reference before admitting execution. */\nexport const EvaluationProjectContextSchema = z.object({\n id: z.string().trim().min(1).max(240),\n revision: z.number().int().positive(),\n contentHash: z.string().regex(/^[a-f0-9]{64}$/),\n targetId: z.string().trim().min(1).max(240).nullable(),\n}).strict();\nexport type EvaluationProjectContext = z.infer<typeof EvaluationProjectContextSchema>;\n\nexport const EvaluationMessagesSchema = z.array(z.object({\n role: z.enum([\"system\", \"user\", \"assistant\"]),\n content: z.string().max(100_000),\n}).strict()).max(100).superRefine((messages, context) => {\n if (messages.reduce((total, message) => total + message.content.length, 0) > 100_000)\n context.addIssue({ code: \"custom\", message: \"Evaluation prompt exceeds 100,000 characters.\" });\n});\n"],
5
+ "mappings": ";;;;;;;AAAA,SAAS,KAAAA,UAAS;;;ACAlB,SAAS,SAAS;AAIX,IAAM,iCAAiC,EAAE,OAAO;AAAA,EACrD,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,EAAE,IAAI,GAAG;AAAA,EACpC,UAAU,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS;AAAA,EACpC,aAAa,EAAE,OAAO,EAAE,MAAM,gBAAgB;AAAA,EAC9C,UAAU,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,EAAE,IAAI,GAAG,EAAE,SAAS;AACvD,CAAC,EAAE,OAAO;AAGH,IAAM,2BAA2B,EAAE,MAAM,EAAE,OAAO;AAAA,EACvD,MAAM,EAAE,KAAK,CAAC,UAAU,QAAQ,WAAW,CAAC;AAAA,EAC5C,SAAS,EAAE,OAAO,EAAE,IAAI,GAAO;AACjC,CAAC,EAAE,OAAO,CAAC,EAAE,IAAI,GAAG,EAAE,YAAY,CAAC,UAAU,YAAY;AACvD,MAAI,SAAS,OAAO,CAAC,OAAO,YAAY,QAAQ,QAAQ,QAAQ,QAAQ,CAAC,IAAI;AAC3E,YAAQ,SAAS,EAAE,MAAM,UAAU,SAAS,gDAAgD,CAAC;AACjG,CAAC;;;ADfD,SAAS,+BAA+B;AACxC,SAAS,uBAAuB;AAIhC,IAAM,WAAWC,GAAE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,EAAE,IAAI,GAAG;AACjD,IAAM,aAAaA,GAAE,OAAO,EAAE,MAAM,gBAAgB;AAC7C,IAAM,kCAAkCA,GAAE,mBAAmB,QAAQ;AAAA,EAC1EA,GAAE,OAAO,EAAE,MAAMA,GAAE,QAAQ,aAAa,GAAG,SAAS,UAAU,UAAU,yBAAyB,SAAS,GAAG,iBAAiBA,GAAE,OAAO,EAAE,IAAI,EAAE,IAAI,CAAC,EAAE,IAAI,IAAK,EAAE,QAAQ,IAAK,GAAG,aAAaA,GAAE,OAAO,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC,EAAE,QAAQ,CAAC,GAAG,MAAMA,GAAE,OAAO,EAAE,GAAG,CAAC,EAAE,IAAI,CAAC,EAAE,QAAQ,CAAC,EAAE,CAAC,EAAE,OAAO;AAAA,EACtRA,GAAE,OAAO,EAAE,MAAMA,GAAE,QAAQ,SAAS,GAAG,WAAW,SAAS,CAAC,EAAE,OAAO;AACvE,CAAC;AAGM,IAAM,mCAAmCA,GAAE,OAAO;AAAA,EACvD,eAAeA,GAAE,QAAQ,wCAAwC;AAAA,EACjE,aAAa;AAAA,EAAU,QAAQ;AAAA,EAAU,gBAAgB,SAAS,SAAS;AAAA,EAC3E,SAAS,+BAA+B,SAAS;AAAA,EACjD,SAAS;AAAA,EAAgC,QAAQ;AAAA,EACjD,iBAAiBA,GAAE,OAAO,EAAE,IAAI,EAAE,IAAI,CAAC,EAAE,IAAI,UAAa,EAAE,QAAQ,CAAC;AAAA,EACrE,QAAQ;AACV,CAAC,EAAE,OAAO;AAEH,IAAM,wCAAwCA,GAAE,OAAO,EAAE,gBAAgB,SAAS,SAAS,GAAG,SAAS,gCAAgC,aAAa,SAAS,SAAS,GAAG,OAAOA,GAAE,OAAO,EAAE,IAAI,EAAE,IAAI,CAAC,EAAE,IAAI,GAAG,EAAE,QAAQ,EAAE,EAAE,CAAC,EAAE,OAAO;AACvO,IAAM,mCAAmCA,GAAE,OAAO;AAAA,EACvD,gBAAgB,SAAS,SAAS;AAAA,EAAG,SAAS;AAAA,EAAgC,WAAWA,GAAE,QAAQ;AAAA,EAAG,mBAAmBA,GAAE,OAAO,EAAE,IAAI,GAAK,EAAE,SAAS;AAAA,EACxJ,QAAQA,GAAE,MAAMA,GAAE,OAAO,EAAE,IAAI,UAAU,MAAMA,GAAE,OAAO,EAAE,IAAI,GAAG,EAAE,CAAC,EAAE,OAAO,CAAC,EAAE,IAAI,GAAG;AAAA,EACvF,OAAOA,GAAE,MAAMA,GAAE,OAAO,EAAE,IAAI,UAAU,OAAOA,GAAE,KAAK,CAAC,SAAS,cAAc,aAAa,CAAC,GAAG,cAAcA,GAAE,OAAO,EAAE,IAAI,GAAG,GAAG,UAAUA,GAAE,MAAMA,GAAE,OAAO,EAAE,IAAI,UAAU,OAAOA,GAAE,OAAO,EAAE,IAAI,GAAG,EAAE,CAAC,EAAE,OAAO,CAAC,EAAE,IAAI,GAAK,EAAE,CAAC,EAAE,OAAO,CAAC,EAAE,IAAI,GAAG;AAAA,EACpP,YAAY,SAAS,SAAS;AAChC,CAAC,EAAE,OAAO;AACH,IAAM,mCAAmCA,GAAE,OAAO;AAAA,EACvD,eAAeA,GAAE,QAAQ,wCAAwC;AAAA,EACjE,IAAI;AAAA,EAAU,UAAUA,GAAE,OAAO,EAAE,IAAI,EAAE,SAAS;AAAA,EAAG,SAAS;AAAA,EAC9D,QAAQA,GAAE,KAAK,CAAC,UAAU,WAAW,cAAc,aAAa,UAAU,WAAW,CAAC;AAAA,EACtF,gBAAgBA,GAAE,OAAO,EAAE,SAAS,UAAU,UAAU,UAAU,iBAAiBA,GAAE,OAAO,EAAE,IAAI,CAAC,EAAE,IAAI,GAAG,GAAG,mBAAmB,WAAW,CAAC,EAAE,OAAO,EAAE,SAAS;AAAA,EAClK,WAAWA,GAAE,IAAI,SAAS;AAAA,EAAG,WAAWA,GAAE,IAAI,SAAS,EAAE,SAAS;AAAA,EAAG,aAAaA,GAAE,IAAI,SAAS,EAAE,SAAS;AAAA,EAC5G,iBAAiBA,GAAE,QAAQ;AAAA,EAAG,iBAAiBA,GAAE,QAAQ;AAAA,EAAG,OAAOA,GAAE,OAAO,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC,EAAE,SAAS;AAAA,EACrG,eAAeA,GAAE,KAAK,CAAC,eAAe,UAAU,cAAc,gBAAgB,CAAC,EAAE,QAAQ,aAAa;AAAA,EACtG,QAAQA,GAAE,QAAQ,EAAE,SAAS;AAAA,EAAG,eAAeA,GAAE,OAAO,EAAE,IAAI,GAAG,EAAE,SAAS;AAAA,EAC5E,OAAOA,GAAE,OAAO,EAAE,MAAM,UAAU,SAASA,GAAE,OAAO,EAAE,IAAI,GAAK,EAAE,CAAC,EAAE,OAAO,EAAE,SAAS;AACxF,CAAC,EAAE,OAAO,EAAE,YAAY,CAAC,OAAO,YAAY;AAC1C,MAAK,MAAM,QAAQ,OAAO,SAAS,eAAgB,MAAM,mBAAmB,MAAO,SAAQ,SAAS,EAAE,MAAM,UAAU,MAAM,CAAC,gBAAgB,GAAG,SAAS,qDAAqD,CAAC;AAC/M,MAAI,MAAM,WAAW,gBAAgB,CAAC,MAAM,mBAAmB,CAAC,MAAM,mBAAmB,CAAC,MAAM,aAAc,SAAQ,SAAS,EAAE,MAAM,UAAU,SAAS,qEAAqE,CAAC;AAClO,CAAC;AAEM,IAAM,qCAAqCA,GAAE,OAAO,EAAE,gBAAgB,UAAU,SAAS,SAAS,SAAS,GAAG,OAAOA,GAAE,OAAO,EAAE,IAAI,EAAE,IAAI,CAAC,EAAE,IAAI,GAAG,EAAE,QAAQ,EAAE,EAAE,CAAC,EAAE,OAAO;AAC5K,IAAM,gCAAgCA,GAAE,OAAO,EAAE,OAAOA,GAAE,MAAM,gCAAgC,EAAE,IAAI,GAAG,GAAG,YAAY,SAAS,SAAS,EAAE,CAAC,EAAE,OAAO;AACtJ,IAAM,kCAAkCA,GAAE,OAAO;AAAA,EACtD,eAAeA,GAAE,QAAQ,uCAAuC;AAAA,EAChE,SAAS;AAAA,EACT,QAAQA,GAAE,OAAO,EAAE,IAAI,OAAS,EAAE,SAAS;AAAA;AAAA,EAE3C,UAAUA,GAAE,MAAMA,GAAE,OAAOA,GAAE,OAAO,GAAGA,GAAE,QAAQ,CAAC,CAAC,EAAE,IAAI,IAAK;AAAA,EAC9D,aAAa,wBAAwB,SAAS;AAAA;AAAA;AAAA,EAG9C,OAAO,gBAAgB,SAAS,EAAE,SAAS;AAAA,EAC3C,aAAaA,GAAE,OAAO,EAAE,QAAQA,GAAE,KAAK,CAAC,aAAa,oBAAoB,aAAa,aAAa,kBAAkB,qBAAqB,CAAC,GAAG,WAAWA,GAAE,QAAQ,GAAG,YAAY,gCAAgC,kBAAkB,WAAW,SAAS,GAAG,gBAAgB,WAAW,SAAS,GAAG,aAAa,WAAW,CAAC,EAAE,OAAO;AAAA,EACpU,oBAAoBA,GAAE,MAAM,QAAQ,EAAE,IAAI,IAAK;AAAA,EAC/C,aAAa;AACf,CAAC,EAAE,OAAO,EAAE,YAAY,CAAC,OAAO,YAAY;AAC1C,MAAI,CAAC,MAAM,MAAO;AAClB,MAAI,MAAM,YAAa,SAAQ,SAAS,EAAE,MAAM,UAAU,MAAM,CAAC,OAAO,GAAG,SAAS,4DAA4D,CAAC;AACjJ,MAAI,CAAC,CAAC,aAAa,QAAQ,EAAE,SAAS,MAAM,QAAQ,MAAM,KAAK,CAAC,MAAM,QAAQ,mBAAmB,MAAM,QAAQ,UAAU,MAAM,MAAM,SAAS,MAAM,QAAQ,kBAAkB,MAAM,MAAM,iBAAiB,MAAM,QAAQ,YAAY,MAAM,MAAM,UAAU,OAAO,OAAO,MAAM,MAAM,QAAS,SAAQ,SAAS,EAAE,MAAM,UAAU,MAAM,CAAC,OAAO,GAAG,SAAS,4DAA4D,CAAC;AAC1Z,CAAC;AAGM,IAAM,mCAAN,cAA+C,MAAM;AAAA,EAC1D,YAAqB,QAAyB,MAAc,SAAiB;AAAE,UAAM,OAAO;AAAvE;AAAyB;AAAiD,SAAK,OAAO;AAAA,EAAoC;AAAA,EAA1H;AAAA,EAAyB;AAChD;AAEO,IAAM,qCAAN,MAAyC;AAAA,EACrC;AAAA,EACT,YAAY,SAA+F;AACzG,UAAM,MAAM,IAAI,IAAI,QAAQ,OAAO;AACnC,QAAI,CAAC,QAAQ,OAAO,KAAK,KAAK,CAAC,QAAQ,OAAO,KAAK,KAAK,CAAC,CAAC,UAAU,OAAO,EAAE,SAAS,IAAI,QAAQ,KAAK,IAAI,YAAY,IAAI,YAAY,IAAI,UAAU,IAAI,KAAM,OAAM,IAAI,MAAM,yEAAyE;AACxP,SAAK,WAAW,EAAE,GAAG,SAAS,SAAS,IAAI,SAAS,EAAE,QAAQ,QAAQ,EAAE,EAAE;AAAA,EAC5E;AAAA,EACA,MAAM,OAAO,OAAyD,UAAoC,CAAC,GAAG;AAC5G,UAAM,UAAU,iCAAiC,MAAM,KAAK;AAC5D,QAAI,QAAQ,WAAW,KAAK,SAAS,OAAQ,OAAM,IAAI,MAAM,uCAAuC;AACpG,UAAM,SAAS,KAAK,SAAS,MAAM,KAAK,SAAS,IAAI,QAAQ,SAAS,QAAQ,MAAM,CAAC;AACrF,QAAI,cAAc,OAAO,OAAO,MAAM,cAAc,OAAO,EAAG,OAAM,IAAI,iCAAiC,KAAK,4BAA4B,qDAAqD;AAC/L,WAAO;AAAA,EACT;AAAA,EACA,MAAM,IAAI,IAAY,UAAoC,CAAC,GAAG;AAAE,WAAO,KAAK,SAAS,MAAM,KAAK,SAAS,IAAI,mBAAmB,SAAS,MAAM,EAAE,CAAC,CAAC,IAAI,OAAO,QAAW,QAAQ,MAAM,GAAG,EAAE;AAAA,EAAG;AAAA,EAC/L,MAAM,QAAQ,OAA8D,UAAoC,CAAC,GAAG;AAClH,UAAM,QAAQ,sCAAsC,MAAM,KAAK;AAC/D,UAAM,SAAS,IAAI,gBAAgB,EAAE,GAAI,MAAM,iBAAiB,EAAE,gBAAgB,MAAM,eAAe,IAAI,CAAC,GAAI,WAAW,MAAM,QAAQ,IAAI,UAAU,OAAO,MAAM,QAAQ,QAAQ,GAAG,aAAa,MAAM,QAAQ,aAAa,OAAO,OAAO,MAAM,KAAK,EAAE,CAAC;AAC3P,QAAI,MAAM,YAAa,QAAO,IAAI,eAAe,MAAM,WAAW;AAClE,UAAM,SAAS,iCAAiC,MAAM,MAAM,KAAK,SAAS,YAAY,MAAM,IAAI,OAAO,QAAW,QAAQ,MAAM,CAAC;AACjI,QAAI,OAAO,mBAAmB,MAAM,kBAAkB,cAAc,OAAO,OAAO,MAAM,cAAc,MAAM,OAAO,KAAK,OAAO,MAAM,SAAS,MAAM,SAAS,IAAI,IAAI,OAAO,MAAM,IAAI,UAAQ,KAAK,EAAE,CAAC,EAAE,SAAS,OAAO,MAAM,OAAQ,OAAM,IAAI,iCAAiC,KAAK,4BAA4B,4DAA4D;AAC9W,WAAO;AAAA,EACT;AAAA,EACA,MAAM,OAAO,IAAY,UAAoC,CAAC,GAAG;AAAE,WAAO,KAAK,SAAS,MAAM,KAAK,SAAS,IAAI,mBAAmB,SAAS,MAAM,EAAE,CAAC,CAAC,WAAW,QAAQ,QAAW,QAAQ,MAAM,GAAG,EAAE;AAAA,EAAG;AAAA,EAC1M,MAAM,OAAO,IAAY,UAAoC,CAAC,GAAG;AAC/D,UAAM,QAAQ,gCAAgC,MAAM,MAAM,KAAK,SAAS,IAAI,mBAAmB,SAAS,MAAM,EAAE,CAAC,CAAC,WAAW,OAAO,QAAW,QAAQ,MAAM,CAAC;AAC9J,SAAK,SAAS,MAAM,SAAS,EAAE;AAC/B,UAAM,EAAE,aAAa,GAAG,QAAQ,IAAI;AACpC,QAAI,CAAC,MAAM,QAAQ,mBAAmB,MAAM,gBAAgB,OAAO,MAAM,YAAa,OAAM,IAAI,iCAAiC,KAAK,4BAA4B,kCAAkC;AACpM,WAAO;AAAA,EACT;AAAA,EACA,MAAM,KAAK,OAA2D,UAAoC,CAAC,GAAG;AAC5G,UAAM,QAAQ,mCAAmC,MAAM,KAAK;AAC5D,UAAM,SAAS,IAAI,gBAAgB,EAAE,gBAAgB,MAAM,gBAAgB,OAAO,OAAO,MAAM,KAAK,EAAE,CAAC;AACvG,QAAI,MAAM,QAAS,QAAO,IAAI,WAAW,MAAM,OAAO;AACtD,UAAM,OAAO,8BAA8B,MAAM,MAAM,KAAK,SAAS,IAAI,MAAM,IAAI,OAAO,QAAW,QAAQ,MAAM,CAAC;AACpH,QAAI,KAAK,MAAM,SAAS,MAAM,SAAS,IAAI,IAAI,KAAK,MAAM,IAAI,UAAQ,KAAK,EAAE,CAAC,EAAE,SAAS,KAAK,MAAM,UAAU,KAAK,MAAM,KAAK,UAAQ,KAAK,SAAS,IAAI,EAAE,QAAQ,mBAAmB,MAAM,cAAc,EAAG,OAAM,IAAI,iCAAiC,KAAK,yBAAyB,+CAA+C;AACpU,WAAO;AAAA,EACT;AAAA,EACA,SAAS,OAAgB,IAAyC;AAChE,UAAM,SAAS,iCAAiC,MAAM,KAAK;AAC3D,QAAI,OAAO,QAAQ,WAAW,KAAK,SAAS,UAAW,MAAM,OAAO,OAAO,GAAK,OAAM,IAAI,iCAAiC,KAAK,6BAA6B,2DAA2D;AACxN,WAAO;AAAA,EACT;AAAA,EACA,MAAM,SAAS,MAAc,QAAgB,MAAe,QAAwC;AAClG,UAAM,WAAW,OAAO,KAAK,SAAS,SAAS,WAAW,OAAO,GAAG,KAAK,SAAS,OAAO,6BAA6B,IAAI,IAAI,EAAE,QAAQ,SAAS,EAAE,eAAe,UAAU,KAAK,SAAS,MAAM,IAAI,sBAAsB,KAAK,SAAS,QAAQ,gBAAgB,oBAAoB,QAAQ,mBAAmB,GAAG,GAAI,SAAS,SAAY,CAAC,IAAI,EAAE,MAAM,KAAK,UAAU,IAAI,EAAE,GAAI,QAAQ,UAAU,QAAQ,CAAC;AAC5Y,UAAM,SAAS,SAAS,MAAM,UAAU;AACxC,QAAI,CAAC,OAAQ,OAAM,IAAI,iCAAiC,SAAS,QAAQ,0BAA0B,6BAA6B;AAChI,QAAI,QAAQ;AAAG,QAAI,OAAO;AAAI,UAAM,UAAU,IAAI,YAAY;AAC9D,QAAI;AACF,aAAO,MAAM;AACX,cAAM,QAAQ,MAAM,OAAO,KAAK;AAAG,YAAI,MAAM,KAAM;AACnD,iBAAS,MAAM,MAAM;AACrB,YAAI,QAAQ,SAAY,OAAM,IAAI,iCAAiC,KAAK,8BAA8B,2CAA2C;AACjJ,gBAAQ,QAAQ,OAAO,MAAM,OAAO,EAAE,QAAQ,KAAK,CAAC;AAAA,MACtD;AACA,cAAQ,QAAQ,OAAO;AAAA,IACzB,UAAE;AAAU,YAAM,OAAO,OAAO;AAAG,aAAO,YAAY;AAAA,IAAG;AACzD,UAAM,QAAiB,KAAK,MAAM,IAAI;AACtC,QAAI,CAAC,SAAS,IAAI;AAChB,YAAM,QAAQA,GAAE,OAAO,EAAE,MAAM,UAAU,SAASA,GAAE,OAAO,EAAE,IAAI,GAAK,EAAE,CAAC,EAAE,UAAU,KAAK;AAC1F,YAAM,IAAI,iCAAiC,SAAS,QAAQ,MAAM,UAAU,MAAM,KAAK,OAAO,0BAA0B,MAAM,UAAU,MAAM,KAAK,UAAU,yBAAyB;AAAA,IACxL;AACA,WAAO;AAAA,EACT;AACF;",
6
+ "names": ["z", "z"]
7
+ }
@@ -3,10 +3,7 @@ import {
3
3
  validateModelStarterExecution,
4
4
  validateModelTasksetPackage,
5
5
  validateTasksetPackage
6
- } from "./chunk-42YOB36J.js";
7
- import {
8
- contentHash
9
- } from "./chunk-QUOW6URQ.js";
6
+ } from "./chunk-WOPM32SE.js";
10
7
  import {
11
8
  ModelProjectBaseModelSchema,
12
9
  ModelProjectTrainingMethodSchema,
@@ -17,6 +14,9 @@ import {
17
14
  import {
18
15
  ModelInitialLearningSchema
19
16
  } from "./chunk-3KXDH6OB.js";
17
+ import {
18
+ contentHash
19
+ } from "./chunk-QUOW6URQ.js";
20
20
 
21
21
  // src/model-starters.ts
22
22
  import { z } from "zod";
@@ -308,4 +308,4 @@ export {
308
308
  previewModelStarter,
309
309
  validateResolvedModelStarter
310
310
  };
311
- //# sourceMappingURL=chunk-DNEPLQXN.js.map
311
+ //# sourceMappingURL=chunk-KYCQ7LZ4.js.map
@@ -1,6 +1,6 @@
1
1
  import {
2
2
  ChatModelRefSchema
3
- } from "./chunk-EAEYAU4R.js";
3
+ } from "./chunk-Y4NGWGTA.js";
4
4
  import {
5
5
  ImmutableReleaseRefSchema,
6
6
  ReleaseHashSchema,
@@ -1226,4 +1226,4 @@ export {
1226
1226
  computeTasksetHash,
1227
1227
  validatePortability
1228
1228
  };
1229
- //# sourceMappingURL=chunk-TXQHWOFM.js.map
1229
+ //# sourceMappingURL=chunk-LGEF4S5I.js.map
@@ -0,0 +1,171 @@
1
+ import {
2
+ MAX_TASKSET_PACKAGE_BYTES,
3
+ TasksetPackageSchema,
4
+ tasksetPackageRewardBinding,
5
+ validateTasksetPackage
6
+ } from "./chunk-WOPM32SE.js";
7
+ import {
8
+ HostedModelProjectSummarySchema,
9
+ HostedModelProjectSyncSchema,
10
+ HostedModelProjectTrainingSetupSchema
11
+ } from "./chunk-TCHAWJMB.js";
12
+ import {
13
+ TasksetCatalogReleaseRefSchema
14
+ } from "./chunk-A4ZDOZKP.js";
15
+ import {
16
+ contentHash
17
+ } from "./chunk-QUOW6URQ.js";
18
+
19
+ // src/taskset-package-client.ts
20
+ import { z } from "zod";
21
+ import { learningRef } from "@openpond/evals/learning";
22
+ var IdSchema = z.string().trim().min(1).max(500);
23
+ var HashSchema = z.string().regex(/^[a-f0-9]{64}$/);
24
+ var TasksetPackageModelConfigurationSchema = HostedModelProjectSyncSchema.pick({
25
+ portableProjectId: true,
26
+ name: true,
27
+ objective: true,
28
+ defaultBaseModel: true,
29
+ defaultDestinationId: true,
30
+ sourceRevision: true,
31
+ sourceUpdatedAt: true
32
+ }).extend({ trainingSetup: HostedModelProjectTrainingSetupSchema.omit({ tasksetRef: true, rewardBindingRef: true, tasksetRelease: true }) }).strict();
33
+ var TasksetPackagePublicationSchema = z.object({
34
+ schemaVersion: z.literal("openpond.tasksetPackagePublication.v1"),
35
+ operationId: IdSchema,
36
+ modelProjectId: IdSchema,
37
+ expectedProjectEtag: HashSchema.nullable(),
38
+ name: z.string().trim().min(1).max(200),
39
+ description: z.string().max(5e3),
40
+ buildIntent: z.enum(["demonstrations", "preferences", "verifiable_reward", "rubric", "discovery"]),
41
+ methodHint: z.enum(["sft", "dpo", "grpo", "ppo"]).nullable(),
42
+ package: TasksetPackageSchema,
43
+ modelConfiguration: TasksetPackageModelConfigurationSchema.optional(),
44
+ evaluationPackage: TasksetPackageSchema.optional(),
45
+ selection: z.enum(["select", "attach"]).optional()
46
+ }).strict().refine((value) => value.selection !== "attach" || value.modelConfiguration === void 0, "Attachment-only publication cannot replace Model configuration.").refine(
47
+ (value) => !value.evaluationPackage || value.modelConfiguration !== void 0 && value.modelConfiguration.trainingSetup.evaluationTasksetRef != null && contentHash(value.modelConfiguration.trainingSetup.evaluationTasksetRef) === contentHash(learningRef(value.evaluationPackage.taskset)),
48
+ "Evaluation package must match the selected immutable evaluation in Model configuration."
49
+ );
50
+ var TasksetPackageReceiptSchema = z.object({
51
+ schemaVersion: z.literal("openpond.tasksetPackageReceipt.v1"),
52
+ teamId: IdSchema,
53
+ modelProjectId: IdSchema,
54
+ operationId: IdSchema,
55
+ taskset: TasksetCatalogReleaseRefSchema,
56
+ packageHash: HashSchema,
57
+ hostedTasksetId: IdSchema,
58
+ projectEtag: HashSchema,
59
+ project: HostedModelProjectSummarySchema.optional(),
60
+ selection: z.enum(["select", "attach"]).optional(),
61
+ evaluation: z.object({ taskset: TasksetCatalogReleaseRefSchema, packageHash: HashSchema, hostedTasksetId: IdSchema }).strict().optional()
62
+ }).strict();
63
+ var TasksetPackageReadbackSchema = z.object({
64
+ schemaVersion: z.literal("openpond.tasksetPackageReadback.v1"),
65
+ teamId: IdSchema,
66
+ modelProjectId: IdSchema,
67
+ package: TasksetPackageSchema
68
+ }).strict();
69
+ var OpenPondTasksetPackageError = class extends Error {
70
+ constructor(status, code, message) {
71
+ super(message);
72
+ this.status = status;
73
+ this.code = code;
74
+ this.name = "OpenPondTasksetPackageError";
75
+ }
76
+ status;
77
+ code;
78
+ };
79
+ var OpenPondTasksetPackageClient = class {
80
+ #options;
81
+ constructor(options) {
82
+ const url = new URL(options.baseUrl);
83
+ if (!["http:", "https:"].includes(url.protocol) || url.username || url.password || url.search || url.hash || !options.apiKey.trim() || !options.teamId.trim()) throw new Error("Taskset package client requires a clean HTTP(S) endpoint and workspace credentials.");
84
+ this.#options = { ...options, baseUrl: url.toString().replace(/\/+$/, "") };
85
+ }
86
+ async publish(input, options = {}) {
87
+ const request = TasksetPackagePublicationSchema.parse(input);
88
+ validateTasksetPackage(request.package);
89
+ if (request.evaluationPackage) validateTasksetPackage(request.evaluationPackage);
90
+ const receipt = TasksetPackageReceiptSchema.parse(await this.#request("", "POST", request, options.signal));
91
+ const release = request.package.taskset;
92
+ if (receipt.teamId !== this.#options.teamId || receipt.modelProjectId !== request.modelProjectId || receipt.operationId !== request.operationId || receipt.packageHash !== request.package.contentHash || receipt.taskset.id !== release.id || receipt.taskset.revision !== release.revision || receipt.taskset.contentHash !== release.contentHash) {
93
+ throw new OpenPondTasksetPackageError(502, "package_receipt_mismatch", "Taskset publication receipt did not match the requested package and owner.");
94
+ }
95
+ if (request.selection !== void 0 && receipt.selection !== request.selection) throw new OpenPondTasksetPackageError(502, "package_receipt_mismatch", "Taskset publication did not retain its selection mode.");
96
+ if (request.evaluationPackage && (!receipt.evaluation || receipt.evaluation.packageHash !== request.evaluationPackage.contentHash || contentHash(receipt.evaluation.taskset) !== contentHash(learningRef(request.evaluationPackage.taskset)))) {
97
+ throw new OpenPondTasksetPackageError(502, "package_receipt_mismatch", "Taskset publication did not retain the complete selected evaluation package.");
98
+ }
99
+ if (request.selection === "attach" && receipt.projectEtag !== request.expectedProjectEtag) throw new OpenPondTasksetPackageError(502, "package_receipt_mismatch", "Attaching a historical Taskset changed the Model configuration.");
100
+ if (request.modelConfiguration) {
101
+ const expected = request.modelConfiguration;
102
+ const project = receipt.project;
103
+ const binding = tasksetPackageRewardBinding(request.package);
104
+ const setup = HostedModelProjectTrainingSetupSchema.parse({
105
+ ...expected.trainingSetup,
106
+ tasksetRef: learningRef(release),
107
+ rewardBindingRef: binding ? learningRef(binding) : null,
108
+ tasksetRelease: null
109
+ });
110
+ if (!project || project.teamId !== this.#options.teamId || project.etag !== receipt.projectEtag || ![project.id, project.portableProjectId].includes(request.modelProjectId) || project.portableProjectId !== expected.portableProjectId || project.name !== expected.name || project.objective !== expected.objective || project.sourceRevision !== expected.sourceRevision || Date.parse(project.sourceUpdatedAt) !== Date.parse(expected.sourceUpdatedAt) || contentHash(project.defaultBaseModel) !== contentHash(expected.defaultBaseModel) || project.defaultDestinationId !== expected.defaultDestinationId || contentHash(project.trainingSetup) !== contentHash(setup)) {
111
+ throw new OpenPondTasksetPackageError(502, "package_receipt_mismatch", "Taskset publication did not retain the requested Model configuration.");
112
+ }
113
+ }
114
+ return receipt;
115
+ }
116
+ async get(modelProjectId, reference, options = {}) {
117
+ const project = IdSchema.parse(modelProjectId);
118
+ const ref = TasksetCatalogReleaseRefSchema.parse(reference);
119
+ const expectedPackageHash = options.expectedPackageHash === void 0 ? void 0 : HashSchema.parse(options.expectedPackageHash);
120
+ const response = TasksetPackageReadbackSchema.parse(await this.#request(`/${encodeURIComponent(project)}/${encodeURIComponent(ref.id)}/${ref.revision}/${ref.contentHash}`, "GET", void 0, options.signal));
121
+ const taskset = response.package.taskset;
122
+ if (response.teamId !== this.#options.teamId || response.modelProjectId !== project || taskset.id !== ref.id || taskset.revision !== ref.revision || taskset.contentHash !== ref.contentHash) throw new OpenPondTasksetPackageError(502, "package_readback_mismatch", "Taskset package did not match the requested release and owner.");
123
+ if (expectedPackageHash !== void 0 && response.package.contentHash !== expectedPackageHash) throw new OpenPondTasksetPackageError(502, "package_readback_mismatch", "Taskset package did not match the saved publication receipt.");
124
+ return validateTasksetPackage(response.package);
125
+ }
126
+ async #request(path, method, body, signal) {
127
+ const serialized = body === void 0 ? void 0 : JSON.stringify(body);
128
+ if (serialized && new TextEncoder().encode(serialized).byteLength > MAX_TASKSET_PACKAGE_BYTES) throw new OpenPondTasksetPackageError(413, "package_too_large", "Taskset publication exceeds the 64 MiB envelope.");
129
+ const response = await (this.#options.fetch ?? globalThis.fetch)(`${this.#options.baseUrl}/v1/taskset-packages${path}`, {
130
+ method,
131
+ headers: { Authorization: `Bearer ${this.#options.apiKey}`, "X-OpenPond-Team-Id": this.#options.teamId, Accept: "application/json", ...serialized ? { "Content-Type": "application/json" } : {} },
132
+ body: serialized,
133
+ signal,
134
+ redirect: "error"
135
+ });
136
+ const reader = response.body?.getReader();
137
+ if (!reader) throw new OpenPondTasksetPackageError(response.status, "empty_response", "Taskset package response was empty.");
138
+ const decoder = new TextDecoder("utf-8", { fatal: true });
139
+ let bytes = 0;
140
+ let text = "";
141
+ try {
142
+ while (true) {
143
+ const chunk = await reader.read();
144
+ if (chunk.done) break;
145
+ bytes += chunk.value.byteLength;
146
+ if (bytes > MAX_TASKSET_PACKAGE_BYTES) throw new OpenPondTasksetPackageError(413, "package_too_large", "Taskset response exceeds the 64 MiB envelope.");
147
+ text += decoder.decode(chunk.value, { stream: true });
148
+ }
149
+ text += decoder.decode();
150
+ } finally {
151
+ await reader.cancel();
152
+ reader.releaseLock();
153
+ }
154
+ const payload = JSON.parse(text);
155
+ if (!response.ok) {
156
+ const error = z.object({ code: z.string(), message: z.string() }).safeParse(payload);
157
+ throw new OpenPondTasksetPackageError(response.status, error.success ? error.data.code : "package_request_failed", error.success ? error.data.message : `Taskset package request failed (${response.status}).`);
158
+ }
159
+ return payload;
160
+ }
161
+ };
162
+
163
+ export {
164
+ TasksetPackageModelConfigurationSchema,
165
+ TasksetPackagePublicationSchema,
166
+ TasksetPackageReceiptSchema,
167
+ TasksetPackageReadbackSchema,
168
+ OpenPondTasksetPackageError,
169
+ OpenPondTasksetPackageClient
170
+ };
171
+ //# sourceMappingURL=chunk-QABCKPW6.js.map