openpond-sdk 0.11.4 → 0.11.6

package/README.md

@@ -372,3 +372,18 @@ identity and all snapshot hashes. Use `acceptanceGroupVerdict` from
 These methods only read evidence. They do not dispatch checks, bypass server-side
 acceptance, start training or activate serving. An empty page means this Job has
 no recorded independent group; its existing trainer validation is separate.
+
+### Experiment spending defaults and comparisons
+
+`OpenPondExperimentsClient.defaults()` returns the hosted global spend defaults.
+`prepareHarness()` and `run()` accept an omitted `maximumCostUsd`; explicit
+per-run limits and retained run limits take precedence. Token allocation is
+managed within model capacity and the available monetary reservation.
+
+Use `createComparisonBudget({ id, operationIds, maximumSpendUsd? })` before
+admitting any member runs to enforce one aggregate cap across a comparison.
+Register fresh, unique operation IDs and retain the returned grant. Membership
+is immutable and owner-scoped. `comparisonBudget(id)` returns each operation's
+run ID, known settled spend, pending reservations and remaining allowance.
+Registering a comparison does not start any work. Independent experiments do
+not implicitly share a spending cap.

package/dist/advanced-refiner-evaluations.js

@@ -12,18 +12,18 @@ import {
   sealAdvancedRefinerEvaluationPin,
   sealAdvancedRefinerReviewReceipt,
   verifyAdvancedRefinerEvaluationPin
-} from "./chunk-WPPHGWFA.js";
-import "./chunk-5LFLXFPT.js";
+} from "./chunk-DG5CRENU.js";
+import "./chunk-D4WTE6ZA.js";
 import "./chunk-YCQVVFE5.js";
-import "./chunk-6ZDQ7HSA.js";
+import "./chunk-KAE7AMYE.js";
+import "./chunk-B7VYE7YQ.js";
 import "./chunk-T4APHDES.js";
-import "./chunk-GMVYYRMR.js";
-import "./chunk-RGGI5O4R.js";
-import "./chunk-Z3PPZRK2.js";
+import "./chunk-L5PQPAU6.js";
+import "./chunk-7WUJBFCH.js";
+import "./chunk-CK3ONRRC.js";
 import "./chunk-TCHAWJMB.js";
 import "./chunk-3KXDH6OB.js";
-import "./chunk-4NZW4LFB.js";
-import "./chunk-QUOW6URQ.js";
+import "./chunk-COPVBJPL.js";
 export {
   AdvancedRefinerEvaluationAccountingSchema,
   AdvancedRefinerEvaluationCommandSchema,

package/dist/candidate-evaluations.js

@@ -5,15 +5,14 @@ import {
   CandidateEvaluationPreparationSchema,
   CandidateEvaluationRequestSchema,
   createCandidateEvaluationClient
-} from "./chunk-OWWDSG27.js";
-import "./chunk-KJ5VENYN.js";
-import "./chunk-GMVYYRMR.js";
-import "./chunk-RGGI5O4R.js";
-import "./chunk-Z3PPZRK2.js";
+} from "./chunk-DBRRQASP.js";
+import "./chunk-YSKUXQ2N.js";
+import "./chunk-L5PQPAU6.js";
+import "./chunk-7WUJBFCH.js";
+import "./chunk-CK3ONRRC.js";
 import "./chunk-TCHAWJMB.js";
 import "./chunk-3KXDH6OB.js";
-import "./chunk-4NZW4LFB.js";
-import "./chunk-QUOW6URQ.js";
+import "./chunk-COPVBJPL.js";
 export {
   CandidateEvaluationControlSchema,
   CandidateEvaluationOptionSchema,

package/dist/chunk-2NBI5UOK.js

@@ -0,0 +1,100 @@
+import {
+  GeneratedTaskFileSchema,
+  TasksetReadinessReportSchema,
+  TrainingPathRecommendationSchema
+} from "./chunk-IS3P4OOW.js";
+import {
+  GraderFixtureLabelSchema,
+  GraderSpecSchema,
+  TaskPolicyBoundarySchema,
+  TasksetCapabilityManifestSchema,
+  TasksetSplitSchema
+} from "./chunk-OLHLJ2X3.js";
+
+// src/task-design-contracts.ts
+import { z } from "zod";
+var Id = z.string().trim().min(1).max(240);
+var Text = z.string().trim().min(1).max(5e3);
+var TrainingTacticSchema = z.enum([
+  "no_training",
+  "prompting",
+  "retrieval",
+  "sft",
+  "preference",
+  "grpo_rft",
+  "sdft_opsd",
+  "sdpo",
+  "agentic_rl"
+]);
+var TaskDesignFixtureTemplateSchema = z.object({
+  id: Id,
+  taskIndex: z.number().int().nonnegative(),
+  label: GraderFixtureLabelSchema,
+  output: z.record(z.string(), z.unknown()),
+  infrastructureError: z.string().trim().min(1).max(1e4).nullable(),
+  expectedPassed: z.boolean(),
+  expectedRewardEligible: z.boolean(),
+  metadata: z.record(z.string(), z.unknown()).default({})
+});
+var CapabilityDiagnosisSchema = z.object({
+  schemaVersion: z.literal("openpond.capabilityDiagnosis.v1"),
+  summary: z.string().trim().min(1).max(1e4),
+  stableBehavior: z.array(Text).max(100).default([]),
+  changingKnowledge: z.array(Text).max(100).default([]),
+  requiredContext: z.array(Text).max(100).default([]),
+  requiredTools: z.array(Id).max(100).default([]),
+  intervention: TrainingTacticSchema,
+  trainingEligible: z.boolean(),
+  rationale: z.array(Text).min(1).max(100),
+  confidence: z.number().min(0).max(1)
+});
+var TaskExampleProposalSchema = z.object({
+  id: Id,
+  sourceId: Id,
+  sourceTurnId: Id.nullable(),
+  split: TasksetSplitSchema,
+  origin: z.enum(["extracted", "corrected", "synthetic", "expert_authored"]),
+  inputPrompt: z.string().trim().min(1).max(1e5),
+  expectedOutputText: z.string().trim().min(1).max(2e5).nullable(),
+  rationale: Text
+});
+var TaskDesignProposalSchema = z.object({
+  schemaVersion: z.literal("openpond.taskDesignProposal.v1"),
+  id: Id,
+  name: z.string().trim().min(1).max(500),
+  objective: z.string().trim().min(1).max(2e4),
+  diagnosis: CapabilityDiagnosisSchema.default({
+    schemaVersion: "openpond.capabilityDiagnosis.v1",
+    summary: "Reproduce the selected approved behavior.",
+    stableBehavior: [],
+    changingKnowledge: [],
+    requiredContext: [],
+    requiredTools: [],
+    intervention: "sft",
+    trainingEligible: true,
+    rationale: ["The selected examples were supplied as demonstrations."],
+    confidence: 0.5
+  }),
+  taskKind: TasksetCapabilityManifestSchema.shape.taskKind,
+  sourceIds: z.array(Id).min(1).max(1e5),
+  assumptions: z.array(Text).max(1e3),
+  successCriteria: z.array(Text).min(1).max(1e3),
+  proposedGraders: z.array(GraderSpecSchema).max(1e3).default([]),
+  graderFixtures: z.array(TaskDesignFixtureTemplateSchema).max(1e5).default([]),
+  generatedFiles: z.array(GeneratedTaskFileSchema).max(1e3).default([]),
+  proposedExamples: z.array(TaskExampleProposalSchema).max(1e5).default([]),
+  proposedMethod: TasksetReadinessReportSchema.shape.recommendedMethod,
+  trainingPath: TrainingPathRecommendationSchema.nullable().default(null),
+  policy: TaskPolicyBoundarySchema,
+  warnings: z.array(z.string().trim().min(1)).default([]),
+  createdAt: z.string().trim().min(1)
+});
+
+export {
+  TrainingTacticSchema,
+  TaskDesignFixtureTemplateSchema,
+  CapabilityDiagnosisSchema,
+  TaskExampleProposalSchema,
+  TaskDesignProposalSchema
+};
+//# sourceMappingURL=chunk-2NBI5UOK.js.map

package/dist/chunk-2NBI5UOK.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../src/task-design-contracts.ts"],
+  "sourcesContent": ["import { z } from \"zod\";\nimport {\n  TasksetSplitSchema,\n  TasksetCapabilityManifestSchema,\n  TaskPolicyBoundarySchema,\n  GraderSpecSchema,\n  GraderFixtureLabelSchema,\n} from \"./taskset-draft-core.js\";\nimport {\n  GeneratedTaskFileSchema,\n  TasksetReadinessReportSchema,\n  TrainingPathRecommendationSchema,\n} from \"./taskset-authored-contracts.js\";\n\nconst Id = z.string().trim().min(1).max(240);\nconst Text = z.string().trim().min(1).max(5_000);\nexport const TrainingTacticSchema = z.enum([\n  \"no_training\",\n  \"prompting\",\n  \"retrieval\",\n  \"sft\",\n  \"preference\",\n  \"grpo_rft\",\n  \"sdft_opsd\",\n  \"sdpo\",\n  \"agentic_rl\",\n]);\nexport const TaskDesignFixtureTemplateSchema = z.object({\n  id: Id,\n  taskIndex: z.number().int().nonnegative(),\n  label: GraderFixtureLabelSchema,\n  output: z.record(z.string(), z.unknown()),\n  infrastructureError: z.string().trim().min(1).max(10_000).nullable(),\n  expectedPassed: z.boolean(),\n  expectedRewardEligible: z.boolean(),\n  metadata: z.record(z.string(), z.unknown()).default({}),\n});\nexport const CapabilityDiagnosisSchema = z.object({\n  schemaVersion: z.literal(\"openpond.capabilityDiagnosis.v1\"),\n  summary: z.string().trim().min(1).max(10_000),\n  stableBehavior: z.array(Text).max(100).default([]),\n  changingKnowledge: z.array(Text).max(100).default([]),\n  requiredContext: z.array(Text).max(100).default([]),\n  requiredTools: z.array(Id).max(100).default([]),\n  intervention: TrainingTacticSchema,\n  trainingEligible: z.boolean(),\n  rationale: z.array(Text).min(1).max(100),\n  confidence: z.number().min(0).max(1),\n});\nexport const TaskExampleProposalSchema = z.object({\n  id: Id,\n  sourceId: Id,\n  sourceTurnId: Id.nullable(),\n  split: TasksetSplitSchema,\n  origin: z.enum([\"extracted\", \"corrected\", \"synthetic\", \"expert_authored\"]),\n  inputPrompt: z.string().trim().min(1).max(100_000),\n  expectedOutputText: z.string().trim().min(1).max(200_000).nullable(),\n  rationale: Text,\n});\nexport const TaskDesignProposalSchema = z.object({\n  schemaVersion: z.literal(\"openpond.taskDesignProposal.v1\"),\n  id: Id,\n  name: z.string().trim().min(1).max(500),\n  objective: z.string().trim().min(1).max(20_000),\n  diagnosis: CapabilityDiagnosisSchema.default({\n    schemaVersion: \"openpond.capabilityDiagnosis.v1\",\n    summary: \"Reproduce the selected approved behavior.\",\n    stableBehavior: [],\n    changingKnowledge: [],\n    requiredContext: [],\n    requiredTools: [],\n    intervention: \"sft\",\n    trainingEligible: true,\n    rationale: [\"The selected examples were supplied as demonstrations.\"],\n    confidence: 0.5,\n  }),\n  taskKind: TasksetCapabilityManifestSchema.shape.taskKind,\n  sourceIds: z.array(Id).min(1).max(100_000),\n  assumptions: z.array(Text).max(1_000),\n  successCriteria: z.array(Text).min(1).max(1_000),\n  proposedGraders: z.array(GraderSpecSchema).max(1_000).default([]),\n  graderFixtures: z.array(TaskDesignFixtureTemplateSchema).max(100_000).default([]),\n  generatedFiles: z.array(GeneratedTaskFileSchema).max(1_000).default([]),\n  proposedExamples: z.array(TaskExampleProposalSchema).max(100_000).default([]),\n  proposedMethod: TasksetReadinessReportSchema.shape.recommendedMethod,\n  trainingPath: TrainingPathRecommendationSchema.nullable().default(null),\n  policy: TaskPolicyBoundarySchema,\n  warnings: z.array(z.string().trim().min(1)).default([]),\n  createdAt: z.string().trim().min(1),\n});\nexport type TaskDesignFixtureTemplate = z.infer<typeof TaskDesignFixtureTemplateSchema>;\nexport type CapabilityDiagnosis = z.infer<typeof CapabilityDiagnosisSchema>;\nexport type TaskExampleProposal = z.infer<typeof TaskExampleProposalSchema>;\nexport type TaskDesignProposal = z.infer<typeof TaskDesignProposalSchema>;\nexport type TrainingTactic = z.infer<typeof TrainingTacticSchema>;\n"],
+  "mappings": ";;;;;;;;;;;;;;AAAA,SAAS,SAAS;AAclB,IAAM,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,EAAE,IAAI,GAAG;AAC3C,IAAM,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,EAAE,IAAI,GAAK;AACxC,IAAM,uBAAuB,EAAE,KAAK;AAAA,EACzC;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF,CAAC;AACM,IAAM,kCAAkC,EAAE,OAAO;AAAA,EACtD,IAAI;AAAA,EACJ,WAAW,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY;AAAA,EACxC,OAAO;AAAA,EACP,QAAQ,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC;AAAA,EACxC,qBAAqB,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,EAAE,IAAI,GAAM,EAAE,SAAS;AAAA,EACnE,gBAAgB,EAAE,QAAQ;AAAA,EAC1B,wBAAwB,EAAE,QAAQ;AAAA,EAClC,UAAU,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAAE,QAAQ,CAAC,CAAC;AACxD,CAAC;AACM,IAAM,4BAA4B,EAAE,OAAO;AAAA,EAChD,eAAe,EAAE,QAAQ,iCAAiC;AAAA,EAC1D,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,EAAE,IAAI,GAAM;AAAA,EAC5C,gBAAgB,EAAE,MAAM,IAAI,EAAE,IAAI,GAAG,EAAE,QAAQ,CAAC,CAAC;AAAA,EACjD,mBAAmB,EAAE,MAAM,IAAI,EAAE,IAAI,GAAG,EAAE,QAAQ,CAAC,CAAC;AAAA,EACpD,iBAAiB,EAAE,MAAM,IAAI,EAAE,IAAI,GAAG,EAAE,QAAQ,CAAC,CAAC;AAAA,EAClD,eAAe,EAAE,MAAM,EAAE,EAAE,IAAI,GAAG,EAAE,QAAQ,CAAC,CAAC;AAAA,EAC9C,cAAc;AAAA,EACd,kBAAkB,EAAE,QAAQ;AAAA,EAC5B,WAAW,EAAE,MAAM,IAAI,EAAE,IAAI,CAAC,EAAE,IAAI,GAAG;AAAA,EACvC,YAAY,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC;AACrC,CAAC;AACM,IAAM,4BAA4B,EAAE,OAAO;AAAA,EAChD,IAAI;AAAA,EACJ,UAAU;AAAA,EACV,cAAc,GAAG,SAAS;AAAA,EAC1B,OAAO;AAAA,EACP,QAAQ,EAAE,KAAK,CAAC,aAAa,aAAa,aAAa,iBAAiB,CAAC;AAAA,EACzE,aAAa,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,EAAE,IAAI,GAAO;AAAA,EACjD,oBAAoB,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,EAAE,IAAI,GAAO,EAAE,SAAS;AAAA,EACnE,WAAW;AACb,CAAC;AACM,IAAM,2BAA2B,EAAE,OAAO;AAAA,EAC/C,eAAe,EAAE,QAAQ,gCAAgC;AAAA,EACzD,IAAI;AAAA,EACJ,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,EAAE,IAAI,GAAG;AAAA,EACtC,WAAW,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,EAAE,IAAI,GAAM;AAAA,EAC9C,WAAW,0BAA0B,QAAQ;AAAA,IAC3C,eAAe;AAAA,IACf,SAAS;AAAA,IACT,gBAAgB,CAAC;AAAA,IACjB,mBAAmB,CAAC;AAAA,IACpB,iBAAiB,CAAC;AAAA,IAClB,eAAe,CAAC;AAAA,IAChB,cAAc;AAAA,IACd,kBAAkB;AAAA,IAClB,WAAW,CAAC,wDAAwD;AAAA,IACpE,YAAY;AAAA,EACd,CAAC;AAAA,EACD,UAAU,gCAAgC,MAAM;AAAA,EAChD,WAAW,EAAE,MAAM,EAAE,EAAE,IAAI,CAAC,EAAE,IAAI,GAAO;AAAA,EACzC,aAAa,EAAE,MAAM,IAAI,EAAE,IAAI,GAAK;AAAA,EACpC,iBAAiB,EAAE,MAAM,IAAI,EAAE,IAAI,CAAC,EAAE,IAAI,GAAK;AAAA,EAC/C,iBAAiB,EAAE,MAAM,gBAAgB,EAAE,IAAI,GAAK,EAAE,QAAQ,CAAC,CAAC;AAAA,EAChE,gBAAgB,EAAE,MAAM,+BAA+B,EAAE,IAAI,GAAO,EAAE,QAAQ,CAAC,CAAC;AAAA,EAChF,gBAAgB,EAAE,MAAM,uBAAuB,EAAE,IAAI,GAAK,EAAE,QAAQ,CAAC,CAAC;AAAA,EACtE,kBAAkB,EAAE,MAAM,yBAAyB,EAAE,IAAI,GAAO,EAAE,QAAQ,CAAC,CAAC;AAAA,EAC5E,gBAAgB,6BAA6B,MAAM;AAAA,EACnD,cAAc,iCAAiC,SAAS,EAAE,QAAQ,IAAI;AAAA,EACtE,QAAQ;AAAA,EACR,UAAU,EAAE,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC;AAAA,EACtD,WAAW,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC;AACpC,CAAC;",
+  "names": []
+}

package/dist/{chunk-K4C5QOSV.js → chunk-3LP7DNLQ.js}

@@ -1,6 +1,6 @@
 import {
   ReleaseIdSchema
-} from "./chunk-QUOW6URQ.js";
+} from "./chunk-COPVBJPL.js";
 
 // src/learning-credentials.ts
 import { z } from "zod";
@@ -233,4 +233,4 @@ export {
   OpenPondLearningError,
   OpenPondLearningClient
 };
-//# sourceMappingURL=chunk-K4C5QOSV.js.map
+//# sourceMappingURL=chunk-3LP7DNLQ.js.map