openpond 0.2.66 → 0.2.67

package/dist/chunks/{chunk-QCJY6FTC.js → chunk-FCXDPVGJ.js}

@@ -1,3 +1,3 @@
 import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{a as le,b as De,c as Be}from"./chunk-UTIELP6F.js";import{k as re,q as oe,t as se}from"./chunk-6S2MWJVD.js";import{d as ie,f as Ee,g as He,h as Oe,i as ze,l as B,n as Ve}from"./chunk-VS3AMJEK.js";import{a as Pe}from"./chunk-AF3AVDPP.js";import{i as D}from"./chunk-OKEC2EIE.js";import{B as Ae,Bb as Me,k as te,m as ne,oa as Ce,ta as we,wa as ae}from"./chunk-WLZJOCLA.js";import{A as Te}from"./chunk-ZWW227M3.js";import{h as je}from"./chunk-QAFXQHHG.js";import{f as V}from"./chunk-HNQPEOJD.js";import{c as n,d as o,e as c,g as r,h as g}from"./chunk-RCT3O3HQ.js";import{A as e}from"./chunk-X2J2P5ZT.js";import{a as v}from"./chunk-FTQLQJQV.js";var L=e.string().trim().min(1).max(240),N=e.string().trim().min(1),me=e.string().trim().min(8).max(256),Le=e.record(e.string(),e.unknown()).default({}),Ct=e.enum(["created","inspecting","awaiting_source_review","awaiting_mapping","awaiting_materialization_approval","materializing","validating","ready","cancelling","cancelled","failed"]),qe=e.object({schemaVersion:e.literal("openpond.huggingFaceDatasetLocator.v1"),repositoryId:e.string().trim().regex(/^[^/\s]+\/[^/\s]+$/).max(500),repositoryUrl:e.string().url().max(2e3),requestedRevision:e.string().trim().min(1).max(256).nullable()}),wt=e.object({path:e.string().trim().min(1).max(1e3),logicalType:e.string().trim().min(1).max(500),nullable:e.boolean(),sampleValues:e.array(e.unknown()).max(5)}),Mt=e.object({configuration:e.string().trim().min(1).max(500),split:e.string().trim().min(1).max(500),rowCount:e.number().int().nonnegative().nullable(),sizeBytes:e.number().int().nonnegative().nullable()}),Pt=e.object({schemaVersion:e.literal("openpond.huggingFaceDatasetInspection.v1"),id:L,locator:qe,resolvedRevision:e.string().trim().min(7).max(256),title:e.string().trim().min(1).max(500),description:e.string().max(2e4).nullable(),declaredLicense:e.string().trim().min(1).max(500).nullable(),gated:e.boolean(),private:e.boolean(),configurations:e.array(e.string().trim().min(1).max(500)).min(1).max(1e3),splits:e.array(Mt).min(1).max(1e4),columns:e.array(wt).min(1).max(1e4),previewRows:e.array(e.record(e.string(),e.unknown())).max(25),sourceFiles:e.array(e.object({path:e.string().trim().min(1).max(2e3),configuration:e.string().trim().min(1).max(500),split:e.string().trim().min(1).max(500),revision:e.string().trim().min(7).max(256),sizeBytes:e.number().int().nonnegative().nullable(),contentHash:me.nullable()})).max(1e5),inspectedAt:N,metadata:Le}),Et=e.enum(["row_id","cluster_id","prompt","messages","demonstration","chosen","rejected","expected_output","privileged_context","reward","feedback","tag","metadata"]),Ht=e.object({sourcePath:e.string().trim().min(1).max(1e3),target:Et,transform:e.enum(["identity","string","json","messages","numeric","math_final_answer"]),policy:e.enum(["visible","privileged","metadata"]),required:e.boolean()}),Ue=e.object({schemaVersion:e.literal("openpond.datasetImportMapping.v1"),sourceSchemaHash:me,configuration:e.string().trim().min(1).max(500),upstreamSplits:e.array(e.string().trim().min(1).max(500)).min(1).max(1e3),preset:e.enum(["prompt_only","prompt_completion","messages","preference","prompt_expected_answer","prompt_privileged_solution","prompt_feedback"]),bindings:e.array(Ht).min(1).max(1e4),nullPolicy:e.enum(["reject","drop","preserve"]),invalidRowPolicy:e.enum(["reject_import","drop_with_receipt"]),splitPolicy:e.object({seed:e.number().int(),assignments:e.record(e.string(),te),validationPercent:e.number().min(0).max(100),frozenEvalPercent:e.number().min(0).max(100)}),importerVersion:e.string().trim().min(1).max(200),mappingHash:me}),Ot=e.object({phase:e.enum(["idle","metadata","download","mapping","parquet_write","verification","publish"]),completedBytes:e.number().int().nonnegative(),totalBytes:e.number().int().nonnegative().nullable(),completedRows:e.number().int().nonnegative(),totalRows:e.number().int().nonnegative().nullable(),message:e.string().trim().min(1).max(2e3).nullable()}),Ne=e.object({schemaVersion:e.literal("openpond.datasetImportJob.v1"),id:L,profileId:L,sourceKind:e.enum(["huggingface","uploaded_file"]),status:Ct,locator:qe.nullable(),inspection:Pt.nullable(),mapping:Ue.nullable(),progress:Ot,targetStorageRoot:e.string().trim().min(1).max(4e3).nullable(),tasksetId:L.nullable(),tasksetRevision:e.number().int().positive().nullable(),artifactId:L.nullable(),error:e.string().trim().min(1).max(2e4).nullable(),cancellationRequested:e.boolean(),createdAt:N,updatedAt:N,completedAt:N.nullable(),metadata:Le}),ya=e.object({profileId:L,url:e.string().trim().min(1).max(2e3)}),_a=e.object({name:e.string().trim().min(1).max(500),objective:e.string().trim().min(1).max(2e4),mapping:Ue,targetStorageRoot:e.string().trim().min(1).max(4e3).nullable(),licenseApproved:e.boolean().default(!1)});var y=e.string().trim().min(1).max(240),w=e.string().trim().min(1),$e=e.record(e.string(),e.unknown()).default({}),Fe=e.enum(["nominated","needs_review","approved_for_creation","creating","needs_task_review","baselining","ready_for_export","rejected","dismissed","retired","blocked"]),zt=e.object({id:y,kind:e.enum(["repeated_success","accepted_correction","agent_action_recurrence","outcome_linked_change","frontier_cost","expert_label","runtime_feedback"]),sourceRefIds:e.array(y).min(1).max(1e4),occurredAt:w,signature:e.string().trim().min(1).max(2e3),summary:e.string().trim().min(1).max(5e3),confidence:e.number().min(0).max(1),consented:e.boolean(),metadata:$e}),Vt=e.object({frequency:e.number().min(0).max(1),businessValue:e.number().min(0).max(1),frontierCost:e.number().min(0).max(1),signalQuality:e.number().min(0).max(1),verifiability:e.number().min(0).max(1),repeatability:e.number().min(0).max(1),privacyRisk:e.number().min(0).max(1),overall:e.number().min(0).max(1)}),Dt=e.object({tactic:Pe,eligible:e.boolean(),reasons:e.array(e.string().trim().min(1).max(5e3)).min(1).max(100),blockers:e.array(e.string().trim().min(1).max(5e3)).max(100).default([]),requiredSignals:e.array(e.string().trim().min(1).max(500)).max(100).default([]),generatedBy:e.enum(["decision_table","baseline_reassessment"])}),ce=e.object({schemaVersion:e.literal("openpond.taskCandidate.v1"),id:y,profileId:y,status:Fe,fingerprint:e.string().trim().min(1).max(512),title:e.string().trim().min(1).max(500),summary:e.string().trim().min(1).max(1e4),workflowSignature:e.string().trim().min(1).max(2e3),evidence:e.array(zt).min(1).max(1e5),scorecard:Vt,recommendation:Dt,mergedIntoId:y.nullable(),createdAt:w,updatedAt:w,metadata:$e}),q=e.object({schemaVersion:e.literal("openpond.taskMinerConfig.v1"),enabled:e.boolean().default(!1),localOnly:e.boolean().default(!0),observationWindowDays:e.number().int().min(1).max(365).default(30),minimumRecurrence:e.number().int().min(2).max(100).default(3),clustering:e.literal("hybrid_deterministic_first").default("hybrid_deterministic_first"),consentRequired:e.boolean().default(!0)}),ja=e.object({items:e.array(ce),generatedAt:w,config:q}),Ta=e.object({status:Fe.optional(),mergeIntoId:y.nullable().optional()}),Aa=e.object({profileId:y,sourceIds:e.array(y).max(1e5).default([]),sessionIds:e.array(y).max(1e5).default([]),config:q.optional()}),Ge=e.object({schemaVersion:e.literal("openpond.taskMinerRun.v1"),id:y,profileId:y,status:e.enum(["queued","running","cancelling","cancelled","succeeded","failed"]),config:q,sourceIds:e.array(y).max(1e5),sessionIds:e.array(y).max(1e5).default([]),progress:e.object({stage:e.enum(["queued","ingesting","preparing","clustering","persisting","complete"]),processedSources:e.number().int().nonnegative(),totalSources:e.number().int().nonnegative(),candidatesFound:e.number().int().nonnegative(),skippedSources:e.number().int().nonnegative().default(0)}),candidateIds:e.array(y).max(1e5),cancelRequested:e.boolean(),error:e.string().trim().min(1).max(2e4).nullable(),createdAt:w,startedAt:w.nullable(),completedAt:w.nullable(),updatedAt:w});var Bt=e.enum(["draft","active","completed","archived"]),Lt=e.enum(["draft","ready","queued","running","candidate","accepted","rejected","no_signal","failed","cancelled"]),Je=e.enum(["seed","daily_residual","rank_candidate","weekly_rollup","full_refresh"]),qt=e.enum(["base_model","previous_release","seed_release","accepted_daily_head","accepted_seed"]),Ut=e.enum(["seed_taskset","nightly_selection","daily_cohort_union","eligible_task_pool"]),Nt=e.enum(["live_client","evaluation_attempt","replay_evidence","manual"]),$t=e.object({id:n,ordinal:e.number().int().nonnegative(),label:e.string().trim().min(1).max(100),role:Je,parentRule:qt,taskSource:Ut,trainableRank:e.number().int().positive().max(1024),minimumTasks:e.number().int().nonnegative().max(1e5),maximumTasks:e.number().int().positive().max(1e5)}).strict().superRefine((t,i)=>{t.minimumTasks>t.maximumTasks&&i.addIssue({code:"custom",path:["minimumTasks"],message:"The minimum task count cannot exceed the maximum task count."});let s={seed:{parentRules:["base_model"],taskSource:"seed_taskset"},daily_residual:{parentRules:["previous_release","accepted_daily_head"],taskSource:"nightly_selection"},rank_candidate:{parentRules:["seed_release"],taskSource:"eligible_task_pool"},weekly_rollup:{parentRules:["seed_release","accepted_seed"],taskSource:"daily_cohort_union"},full_refresh:{parentRules:["base_model"],taskSource:"eligible_task_pool"}},{parentRules:l,taskSource:f}=s[t.role];l.includes(t.parentRule)||i.addIssue({code:"custom",path:["parentRule"],message:`${t.role} requires parent rule ${l.join(" or ")}.`}),t.taskSource!==f&&i.addIssue({code:"custom",path:["taskSource"],message:`${t.role} requires task source ${f}.`}),t.ordinal===0&&t.role!=="seed"&&i.addIssue({code:"custom",path:["role"],message:"The first series entry must be the seed."})}),Ft=e.object({id:n,version:e.number().int().positive(),requireCheckpoint:e.boolean(),requireAppliedOptimizerUpdate:e.boolean(),minimumCurrentCohortMeanImprovement:e.number().finite(),maximumRetainedMeanRegression:e.number().finite().nonnegative(),blockCriticalInvariantRegression:e.boolean(),automaticDailyAdvancement:e.boolean()}).strict(),We=e.object({schemaVersion:e.literal("openpond.modelComparisonSeries.v1"),id:n,profileId:n,modelProjectId:n,name:e.string().trim().min(1).max(500),objective:e.string().trim().min(1).max(2e4),status:Bt,revision:e.number().int().positive(),productionBinding:e.object({role:e.literal("chat_manual"),roleTargetId:n}).strict(),baseModel:e.object({id:n,revision:e.string().trim().min(1).max(256)}).strict(),seedTaskset:g,eligibleTaskPool:g,evaluationTasksets:e.object({development:g,retained:g,frozenFinal:g}).strict(),grader:r,benchmarkProtocol:Be.nullable().default(null),automaticEvaluation:e.object({enabled:e.boolean()}).strict().default({enabled:!1}),residualProfile:e.object({profileId:n,serializedEnvelopeRank:e.number().int().positive().max(1024),maximumEnabledRank:e.number().int().positive().max(1024),topology:e.literal("uniform_block_masked")}).strict(),schedule:e.array($t).min(1).max(1e3),scheduleSealedAt:c.nullable(),advancementPolicy:Ft,executionPolicy:e.object({startWhenReady:e.literal(!1)}).strict(),acceptedSeedEntryId:n.nullable(),acceptedDailyHeadEntryId:n.nullable(),promotedBindingId:n.nullable(),createdBy:n,createdAt:c,updatedAt:c}).strict().superRefine((t,i)=>{[...t.schedule].sort((d,j)=>d.ordinal-j.ordinal).some((d,j)=>d.ordinal!==j)&&i.addIssue({code:"custom",path:["schedule"],message:"Schedule ordinals must be unique and contiguous from zero."}),(new Set(t.schedule.map(d=>d.id)).size!==t.schedule.length||new Set(t.schedule.map(d=>d.label)).size!==t.schedule.length)&&i.addIssue({code:"custom",path:["schedule"],message:"Schedule ids and labels must be unique."}),t.residualProfile.maximumEnabledRank>t.residualProfile.serializedEnvelopeRank&&i.addIssue({code:"custom",path:["residualProfile"],message:"Maximum enabled rank cannot exceed the serialized envelope."});let l=Object.values(t.evaluationTasksets);new Set(l.map(d=>`${d.id}:${d.revision}:${d.contentHash}`)).size!==l.length&&i.addIssue({code:"custom",path:["evaluationTasksets"],message:"Development, retained, and frozen-final cohorts require distinct immutable Taskset releases."});let f=t.schedule.find(d=>d.role==="seed");if(!f||t.schedule.filter(d=>d.role==="seed").length!==1){i.addIssue({code:"custom",path:["schedule"],message:"A series requires exactly one seed entry."});return}let h=t.schedule.filter(d=>d.role==="rank_candidate");h.length===1&&i.addIssue({code:"custom",path:["schedule"],message:"A causal rank comparison requires at least two rank candidates."}),new Set(h.map(d=>d.trainableRank)).size!==h.length&&i.addIssue({code:"custom",path:["schedule"],message:"Causal rank candidates must use distinct trainable ranks."}),[f.trainableRank+t.schedule.filter(d=>d.role==="daily_residual").reduce((d,j)=>d+j.trainableRank,0),...t.schedule.filter(d=>d.role==="weekly_rollup").map(d=>f.trainableRank+d.trainableRank),...t.schedule.filter(d=>d.role==="rank_candidate").map(d=>f.trainableRank+d.trainableRank),...t.schedule.filter(d=>d.role==="full_refresh").map(d=>d.trainableRank)].some(d=>d>t.residualProfile.maximumEnabledRank)&&i.addIssue({code:"custom",path:["schedule"],message:"A declared branch exceeds the maximum enabled rank."})}),Ye=e.object({source:Nt,taskIds:e.array(n).min(1).max(1e5),observedFrom:c,observedTo:c,reviewedAt:c,reviewedBy:n,sourceTaskset:g}).strict(),Gt=Ye.extend({derivedFamilyKeys:e.array(n).min(1).max(1e5),familyDerivation:e.object({algorithm:n,sourceTasksetHash:o}).strict(),selectionHash:o}).strict(),de=e.discriminatedUnion("kind",[e.object({kind:e.literal("base_model"),id:n,revision:e.string().trim().min(1).max(256)}).strict(),e.object({kind:e.literal("model_version"),id:n,contentHash:o}).strict()]),Jt=e.object({id:n,branchOrdinal:e.number().int().nonnegative(),rank:e.number().int().positive().max(1024),offsetStart:e.number().int().nonnegative(),offsetEnd:e.number().int().positive(),optimizationRole:e.enum(["frozen","trainable"]),artifactLineageId:n.nullable()}).strict().superRefine((t,i)=>{t.offsetEnd-t.offsetStart!==t.rank&&i.addIssue({code:"custom",path:["offsetEnd"],message:"Block offsets must span exactly the declared rank."})}),Wt=e.object({disposition:e.enum(["advance","hold","no_signal"]),policy:e.object({id:n,version:e.number().int().positive()}).strict(),reasonCodes:e.array(n).min(1).max(100),summary:e.string().trim().min(1).max(5e3),decidedBy:n,decidedAt:c}).strict(),Yt=e.object({evaluationRunId:n,modelVersionId:n,taskset:g,grader:r,cohortRole:e.enum(["current","correction","sibling_verification","cumulative_known","development","retained","prior_disclosed","frozen_final"]),panelId:n.nullable().default(null),protocol:e.object({id:n,revision:e.number().int().positive(),contentHash:o}).strict().nullable().default(null)}).strict(),Kt=e.object({attemptOrdinal:e.number().int().positive(),trainingPlanId:n.nullable(),modelRunId:n.nullable(),terminalStatus:e.enum(["failed","cancelled"]),queuedAt:c.nullable(),startedAt:c.nullable(),completedAt:c}).strict(),A=e.object({seriesId:n,entryId:n,scheduleEntryId:n,ordinal:e.number().int().nonnegative(),releaseHash:o}).strict(),Ke=e.object({schemaVersion:e.literal("openpond.modelComparisonSeriesEntry.v1"),id:n,seriesId:n,profileId:n,modelProjectId:n,scheduleEntryId:n,releaseHash:o,ordinal:e.number().int().nonnegative(),label:e.string().trim().min(1).max(100),role:Je,branch:e.enum(["daily","rank_candidate","weekly_rollup","full_refresh"]),status:Lt,parent:de,taskset:g,sourceTasksets:e.array(g).min(1).max(1e3),taskSelection:Gt.nullable(),trainableRank:e.number().int().positive().max(1024),serializedEnvelopeRank:e.number().int().positive().max(1024),enabledCumulativeRank:e.number().int().positive().max(1024),trainableBlockId:n,residualBlocks:e.array(Jt).min(1).max(1024),attemptOrdinal:e.number().int().positive().default(1),priorRunAttempts:e.array(Kt).max(100).default([]),trainingPlanId:n.nullable(),modelRunId:n.nullable(),modelVersionId:n.nullable(),evaluations:e.array(Yt).max(1e5),decision:Wt.nullable(),promotionBindingId:n.nullable(),queuedAt:c.nullable(),startedAt:c.nullable(),completedAt:c.nullable(),createdAt:c,updatedAt:c}).strict().superRefine((t,i)=>{let s=[...t.residualBlocks].sort((x,k)=>x.offsetStart-k.offsetStart);s.some((x,k)=>x.offsetStart!==(k===0?0:s[k-1].offsetEnd))&&i.addIssue({code:"custom",path:["residualBlocks"],message:"Residual block offsets must be contiguous from zero."});let l=s.at(-1)?.offsetEnd??0;(l!==t.enabledCumulativeRank||l>t.serializedEnvelopeRank)&&i.addIssue({code:"custom",path:["enabledCumulativeRank"],message:"Enabled rank must match the block inventory and fit the envelope."});let f=t.residualBlocks.filter(x=>x.optimizationRole==="trainable");(f.length!==1||f[0]?.id!==t.trainableBlockId||f[0]?.rank!==t.trainableRank)&&i.addIssue({code:"custom",path:["trainableBlockId"],message:"Exactly one block must be trainable during the Run and match trainableRank."});let h=t.priorRunAttempts.map(x=>x.attemptOrdinal);(t.priorRunAttempts.length!==t.attemptOrdinal-1||h.some((x,k)=>x!==k+1))&&i.addIssue({code:"custom",path:["priorRunAttempts"],message:"Prior Run attempts must be a contiguous history before the active attempt."})}),Ea=e.object({seriesId:n,scheduleEntryId:n,taskSelection:Ye.nullable(),expectedSeriesRevision:e.number().int().positive()}).strict();var M=n.nullable(),C=e.number().int().nonnegative(),Qt=e.string().regex(/^[a-f0-9]{40}$/),Zt=e.object({name:e.string().trim().min(1).max(160),startedAt:c,completedAt:c,durationMs:e.number().nonnegative(),clock:e.enum(["monotonic","provider","wall"]),outcome:e.enum(["succeeded","failed","cancelled"])}).strict(),Qe=e.object({schemaVersion:e.literal("openpond.correlatedTelemetryReceipt.v1"),stage:e.enum(["training","evaluation","deployment","inference"]),correlation:e.object({modelRunId:M,modelVersionId:M,policyVersion:C.nullable(),taskId:M,rolloutGroupId:M,providerResourceId:M,deploymentId:M,inferenceRequestId:M}).strict(),spans:e.array(Zt).max(1e4),usage:e.object({promptTokens:C.nullable(),generatedTokens:C.nullable(),gpuSeconds:e.number().nonnegative().nullable(),workerActiveSeconds:e.number().nonnegative().nullable(),optimizerSteps:C.nullable(),rolloutGroups:C.nullable(),successfulTrajectories:C.nullable(),failedTrajectories:C.nullable(),peakGpuMemoryBytes:C.nullable(),peakGpuUtilizationPercent:e.number().min(0).max(100).nullable()}).strict(),resource:e.object({provider:n,resourceIds:e.array(n).max(1e3),gpuType:e.string().trim().min(1).max(300).nullable(),gpuCount:C.nullable(),baseProfileId:M,baseRepository:e.string().trim().min(1).max(512).nullable(),baseRevision:Qt.nullable(),adapterContentHash:o.nullable()}).strict(),cost:e.object({currency:e.literal("USD"),providerReportedUsd:e.number().nonnegative().nullable(),quotedHourlyUsd:e.number().nonnegative().nullable(),estimatedUsd:e.number().nonnegative().nullable(),methodologyVersion:e.string().trim().min(1).max(160),pricingInputs:e.record(e.string().trim().min(1).max(160),e.number().finite()),unitEstimates:e.record(e.string().trim().min(1).max(160),e.number().nonnegative())}).strict(),recordedAt:c,contentHash:o}).strict();var Xt=e.enum(["base_reference","lora_adapter"]),ue=e.object({baseModel:e.object({source:e.literal("huggingface"),repoId:n,revision:e.string().regex(/^[a-f0-9]{40}$/),configHash:o,tokenizerHash:o,licenseId:n,gated:e.boolean()}).strict(),processor:e.object({repository:n,revision:e.string().regex(/^[a-f0-9]{40}$/),configHash:o}).strict()}).strict().superRefine((t,i)=>{(t.baseModel.repoId!==t.processor.repository||t.baseModel.revision!==t.processor.revision)&&i.addIssue({code:"custom",path:["processor"],message:"Reward Model processor must match the pinned scorer base model."})}),tt=e.object({schemaVersion:e.literal("openpond.modelVersion.v1"),id:n,modelId:n,profileId:n,version:e.number().int().nonnegative(),kind:Xt,status:e.enum(["available","failed"]),baseModel:B,taskset:g,comparisonSeriesEntry:A.nullable().optional(),evaluationTasksetRef:g.nullable().optional(),releaseGraph:e.object({resolvedBundleHash:o,profileRelease:g,harnessRelease:r,agentRelease:r.nullable(),grader:e.object({id:n,contentHash:o}).strict()}).strict(),artifactLineageId:n.nullable(),adapterStatus:e.enum(["not_trained","trained"]),createdAt:c,contentHash:o}).strict().superRefine((t,i)=>{t.kind==="base_reference"&&(t.version!==0||t.adapterStatus!=="not_trained"||t.artifactLineageId!==null)&&i.addIssue({code:"custom",message:"A base-reference Model Version must be version 0 without adapter lineage."}),t.kind==="lora_adapter"&&(t.version===0||t.adapterStatus!=="trained"||!t.artifactLineageId)&&i.addIssue({code:"custom",message:"A LoRA Model Version must be version 1+ with adapter lineage."})}),nt=e.object({schemaVersion:e.literal("openpond.rewardModelVersion.v1"),id:n,modelId:n,profileId:n,version:e.number().int().positive(),role:e.literal("reward"),status:e.enum(["available","failed"]),scope:e.enum(["synthetic_smoke","human_preference"]),baseModel:B,runtime:ue.nullable().default(null),taskset:g,preferenceDatasetRelease:r,releaseGraph:e.object({resolvedBundleHash:o,profileRelease:g,harnessRelease:r,grader:e.object({id:n,contentHash:o}).strict()}).strict(),artifacts:e.object({checkpoint:r.extend({objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),files:e.array(e.object({path:e.string().trim().min(1).max(512),sizeBytes:e.number().int().positive(),sha256:o}).strict()).min(4).max(128)}).strict(),adapter:r,scalarHead:r,bucketHead:r.nullable(),processorRelease:r}).strict(),qualificationReport:r.nullable(),createdAt:c,contentHash:o}).strict(),en=e.object({schemaVersion:e.literal("openpond.rewardModelRunReceipt.v1"),provider:n,providerRunId:n,resolvedBundleHash:o,finalCheckpoint:r.extend({objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/)}).strict(),adapter:r,scalarHead:r,bucketHead:r.nullable(),processorRelease:r,optimizerEvidence:r,managedExecutionReceipt:r.nullable().default(null),parameterDeltaHash:o,cleanup:e.object({computeReleased:e.boolean(),providerTerminalObserved:e.boolean()}).strict(),contentHash:o}).strict(),at=e.preprocess(t=>{if(!t||typeof t!="object"||Array.isArray(t))return t;let i=t;if(i.tasksetRelease!==void 0)return t;let s=i.taskset;if(!s||typeof s!="object"||Array.isArray(s))return t;let l=s;return typeof l.id!="string"||typeof l.contentHash!="string"?t:{...i,tasksetRelease:{id:l.id,contentHash:l.contentHash}}},e.object({schemaVersion:e.literal("openpond.rewardModelRun.v1"),id:n,rewardModelId:n,rewardModelVersionId:n.nullable(),profileId:n,role:e.literal("reward"),scope:e.enum(["synthetic_smoke","human_preference"]),status:e.enum(["prepared","running","succeeded","failed","cancelled"]),taskset:g,tasksetRelease:r,preferenceDatasetRelease:r,recipeRelease:r,destinationId:n,quote:e.object({maximumSpendUsd:e.number().positive(),hourlyCostUsd:e.number().nonnegative().nullable()}).strict(),managedRunId:n.nullable(),progress:e.object({completedSteps:e.number().int().nonnegative(),totalSteps:e.number().int().positive(),latestLoss:e.number().finite().nullable()}).strict(),receipt:en.nullable(),qualificationReport:r.nullable(),accruedSpendUsd:e.number().nonnegative().nullable(),failureOwner:e.enum(["authoring","admission","provider","runner","artifact","qualification","cleanup"]).nullable(),failure:e.string().trim().min(1).max(5e3).nullable(),startedAt:c,completedAt:c.nullable(),updatedAt:c}).strict()).superRefine((t,i)=>{t.status==="succeeded"&&!t.receipt&&i.addIssue({code:"custom",message:"A successful Reward Model Run requires a canonical receipt."}),t.status!=="succeeded"&&t.rewardModelVersionId!==null&&i.addIssue({code:"custom",message:"Only a successful Reward Model Run can publish a Reward Model Version."})}),tn=e.object({schemaVersion:e.literal("openpond.modelRunReceipt.v1"),provider:n,providerRunId:n,assignmentHash:o,resultHash:o,transcriptHash:o,traceHash:o.nullable(),resolvedBundleHash:o,artifactPath:e.string().trim().min(1).max(2e3),cleanup:e.object({computeReleased:e.boolean(),tunnelClosed:e.boolean()}).strict(),telemetry:Qe.nullable().default(null),contentHash:o}).strict(),P=e.object({inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),totalTokens:e.number().int().nonnegative(),costUsd:e.number().nonnegative().nullable()}).strict(),Ze=e.string().regex(/^(?:[a-f0-9]{40}|[a-f0-9]{64})$/),pe=e.object({phase:e.enum(["baseline","adaptation","candidate_adaptation","candidate"]),taskId:n,attemptId:n,sessionId:n.nullable(),turnId:n.nullable(),passed:e.boolean(),score:e.number().min(0).max(1).nullable(),failureClass:ie.nullable(),inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),totalTokens:e.number().int().nonnegative(),latencyMs:e.number().int().nonnegative(),costUsd:e.number().nonnegative().nullable(),startedAt:c}).strict(),Xe=e.object({targetTaskCount:e.number().int().nonnegative(),comparedTaskCount:e.number().int().nonnegative(),passedTaskCount:e.number().int().nonnegative(),failedTaskCount:e.number().int().nonnegative(),lowerTaskCount:e.number().int().nonnegative(),higherTaskCount:e.number().int().nonnegative(),unchangedTaskCount:e.number().int().nonnegative()}).strict(),it=e.object({target:e.literal("all_tasks_lower"),targetTaskCount:e.number().int().nonnegative(),comparedTaskCount:e.number().int().nonnegative(),passedTaskCount:e.number().int().nonnegative(),failedTaskCount:e.number().int().nonnegative(),lowerTaskCount:e.number().int().nonnegative(),higherTaskCount:e.number().int().nonnegative(),unchangedTaskCount:e.number().int().nonnegative(),baselineTokens:e.number().int().nonnegative(),refinedTokens:e.number().int().nonnegative(),tokenDelta:e.number().int(),tokenDeltaPercent:e.number().finite().nullable(),complete:e.boolean(),passed:e.boolean(),cohorts:e.object({adaptation:Xe,heldOut:Xe}).strict()}).strict(),be=e.object({baseline:P,adaptation:P,candidateAdaptation:P,candidate:P,refiner:P,grader:P}).strict(),nn=e.object({usage:be,observedSpendUsd:e.number().nonnegative(),attempts:e.array(pe).max(1e4)}).strict(),an=e.object({id:n,contentHash:o,artifactPath:e.string().trim().min(1).max(2e3)}).strict(),rn=e.object({schemaVersion:e.literal("openpond.modelEvaluationReceipt.v1"),benchmarkId:n,resultManifest:e.object({id:n,contentHash:o,artifactPath:e.string().trim().min(1).max(2e3)}).strict(),stages:e.object({baseline:r,adaptation:r,refiner:r,candidateAdaptation:r,candidate:r,comparison:r}).strict(),usage:be,quality:e.object({baselinePassRate:e.number().min(0).max(1),candidatePassRate:e.number().min(0).max(1),adaptationBaselinePassRate:e.number().min(0).max(1),adaptationCandidatePassRate:e.number().min(0).max(1),adaptationCandidatePassed:e.boolean(),heldOutCandidatePassed:e.boolean(),passed:e.boolean()}).strict(),foregroundTokenDelta:e.number().int(),foregroundTokenDeltaPercent:e.number().finite().nullable(),taskEfficiency:it.optional(),efficiency:e.object({grossForegroundTokenSavings:e.number().int(),overheadTokens:e.number().int().nonnegative(),firstPassNetTokenSavings:e.number().int(),breakEvenReuseCount:e.number().int().nonnegative().nullable(),amortizedTokenSavings:e.number().int(),amortizedReuseCount:e.number().int().positive()}).strict(),budget:e.object({maximumSpendUsd:e.number().nonnegative(),observedSpendUsd:e.number().nonnegative(),enforced:e.boolean()}).strict(),evidenceSnapshot:r,lineage:e.object({adaptationEvidenceHash:o,refinerInputHash:o,refinerOutcomeHash:o,validationHash:o,applyReceiptHash:o,candidateRelease:r,valid:e.boolean()}).strict(),invalidReasons:e.array(e.string().trim().min(1).max(2e3)).max(100),attempts:e.array(pe).max(1e4).default([]),terminalClassification:e.enum(["improved","no_improvement","regressed","inconclusive","cancelled","infrastructure_failure"]),profileGit:e.object({ref:e.string().trim().min(1).max(2e3),commit:Ze,baseCommit:Ze}).strict().nullable(),contentHash:o}).strict(),on=e.object({schemaVersion:e.literal("openpond.modelEvaluationStopReceipt.v1"),benchmarkId:n,terminalClassification:e.literal("inconclusive"),stopReason:e.enum(["candidate_harness_unchanged","candidate_lineage_invalid"]),reason:e.string().trim().min(1).max(5e3),stoppedAfter:e.literal("refiner"),baselineHarness:r,candidateHarness:r,refiner:e.object({id:n,contentHash:o,outcomeCount:e.number().int().nonnegative()}).strict(),usage:be,budget:e.object({maximumSpendUsd:e.number().nonnegative(),observedSpendUsd:e.number().nonnegative(),enforced:e.boolean()}).strict(),evidenceSnapshot:r,attempts:e.array(pe).max(1e4),contentHash:o}).strict(),et=e.object({level:e.literal(.95),lower:e.number().finite(),upper:e.number().finite()}).strict().superRefine((t,i)=>{t.lower>t.upper&&i.addIssue({code:"custom",message:"A confidence interval's lower bound cannot exceed its upper bound."})}),sn=e.object({schemaVersion:e.literal("openpond.modelComparisonBenchmarkReceipt.v1"),benchmarkId:e.literal("model-comparison"),target:e.object({kind:e.enum(["base_model","model_version","external_reference"]),label:e.string().trim().min(1).max(300),modelVersionId:n.nullable(),model:V.nullable()}).strict(),taskset:g,grader:r,sampling:e.object({seeds:e.array(e.number().int()).min(1).max(100),repetitions:e.number().int().positive().max(20)}).strict(),deterministic:e.object({attemptedTaskCount:e.number().int().nonnegative(),completedTaskCount:e.number().int().nonnegative(),passedTaskCount:e.number().int().nonnegative(),failedTaskCount:e.number().int().nonnegative(),meanScore:e.number().finite().nullable(),passRate:e.number().min(0).max(1).nullable(),passRateCi95:et.nullable()}).strict(),judge:e.object({judgeModel:V,judgeRelease:r,rubricRelease:r,calibrationRelease:r,score:e.number().min(0).max(100),scoreCi95:et,comparisonCount:e.number().int().positive(),wins:e.number().int().nonnegative(),ties:e.number().int().nonnegative(),losses:e.number().int().nonnegative(),referenceLabel:e.string().trim().min(1).max(300)}).strict().superRefine((t,i)=>{t.wins+t.ties+t.losses!==t.comparisonCount&&i.addIssue({code:"custom",message:"Judge win, tie, and loss counts must equal comparisonCount."})}).nullable(),attempts:e.array(e.object({attemptId:n.nullable().default(null),taskId:n,seed:e.number().int(),repetition:e.number().int().nonnegative(),status:e.enum(["succeeded","failed"]),deterministicScore:e.number().finite().nullable(),passed:e.boolean().nullable(),judgeScore:e.number().min(0).max(100).nullable(),judgePreference:e.enum(["win","tie","loss"]).nullable(),transcriptHash:o.nullable(),traceHash:o.nullable(),transcriptArtifact:e.object({artifactPath:e.string().trim().min(1).max(2e3),jsonPointer:e.string().trim().min(1).max(1e3)}).strict().nullable().default(null),traceArtifact:e.object({artifactPath:e.string().trim().min(1).max(2e3),jsonPointer:e.string().trim().min(1).max(1e3)}).strict().nullable().default(null),latencyMs:e.number().int().nonnegative().nullable().default(null),failureClass:ie.nullable()}).strict()).max(1e5),usage:e.object({policy:P.nullable(),judge:P.nullable(),observedSpendUsd:e.number().nonnegative().nullable(),evaluationGpuSeconds:e.number().nonnegative().nullable().default(null)}).strict(),managedServing:e.object({jobId:n,terminalState:e.enum(["completed","cancelled","failed","budget_exhausted"]),sourcePolicyVersion:e.number().int().nonnegative(),sourceAdapterSha256:o.nullable(),servedPolicyVersion:e.number().int().nonnegative().nullable(),servedAdapterSha256:o.nullable(),accruedSpendUsd:e.number().nonnegative(),cleanupAttestationHash:o,resourceCount:e.number().int().nonnegative(),activeResourceCount:e.literal(0)}).strict().nullable().default(null),evidenceSnapshot:e.object({id:n,contentHash:o,artifactPath:e.string().trim().min(1).max(2e3)}).strict(),completedAt:c,contentHash:o}).strict().superRefine((t,i)=>{let s=t.deterministic;s.completedTaskCount>s.attemptedTaskCount&&i.addIssue({code:"custom",message:"Completed task count cannot exceed attempted task count."}),s.passedTaskCount+s.failedTaskCount!==s.completedTaskCount&&i.addIssue({code:"custom",message:"Passed and failed task counts must equal completed task count."}),s.completedTaskCount===0!=(s.meanScore===null)&&i.addIssue({code:"custom",message:"Deterministic mean score is available exactly when tasks completed."}),s.completedTaskCount===0!=(s.passRate===null)&&i.addIssue({code:"custom",message:"Deterministic pass rate is available exactly when tasks completed."})}),ln=e.object({advancedEvaluation:re.optional(),benchmarkId:e.literal("harness-refiner"),model:V,upstreamModel:e.object({providerId:n,modelId:e.string().trim().min(1).max(300),revision:e.string().trim().min(1).max(300),pricing:e.object({version:e.string().trim().min(1).max(300),source:e.string().trim().min(1).max(300),effectiveAt:c,inputUsdPerMillionTokens:e.number().nonnegative(),cachedInputUsdPerMillionTokens:e.number().nonnegative(),outputUsdPerMillionTokens:e.number().nonnegative()}).strict().optional()}).strict(),reasoningEffort:e.string().trim().min(1).max(100).nullable(),seeds:e.array(e.number().int()).min(1).max(100),repetitions:e.number().int().positive().max(20),maximumSpendUsd:e.number().nonnegative(),attemptPlan:e.array(e.object({stage:e.enum(["baseline","adaptation","candidate_adaptation","candidate"]),split:e.string().trim().min(1).max(100),taskIds:e.array(n).min(1).max(1e5),attemptCount:e.number().int().positive()}).strict()).length(4)}).strict(),mn=e.object({benchmarkId:e.literal("model-comparison"),target:e.object({kind:e.enum(["base_model","model_version","external_reference"]),label:e.string().trim().min(1).max(300),modelVersionId:n.nullable(),model:V.nullable()}).strict(),grader:r,judge:e.object({model:V,release:r,rubricRelease:r,calibrationRelease:r,referenceLabel:e.string().trim().min(1).max(300)}).strict().nullable(),seeds:e.array(e.number().int()).min(1).max(100),repetitions:e.number().int().positive().max(20),maximumSpendUsd:e.number().nonnegative(),series:e.object({id:n,protocol:e.object({id:n,revision:e.number().int().positive(),contentHash:o}).strict()}).strict().nullable().default(null),panel:e.object({id:n,role:e.enum(["correction","sibling_verification","cumulative_known","development","retained","frozen_final"]),passLabel:n.nullable()}).strict().nullable().default(null),comparisonPair:e.object({entryId:n,parent:de,candidateModelVersionId:n}).strict().nullable().default(null),attemptPlan:e.array(e.object({stage:e.literal("comparison"),split:e.string().trim().min(1).max(100),taskIds:e.array(n).min(1).max(1e5),attemptCount:e.number().int().positive()}).strict()).length(1)}).strict(),cn=e.union([ln,mn]),rt=e.object({schemaVersion:e.literal("openpond.modelRun.v1"),id:n,modelId:n,modelVersionId:n.nullable(),profileId:n,kind:e.enum(["rollout_smoke","training","evaluation"]),status:e.enum(["prepared","running","succeeded","failed","cancelled"]),method:e.enum(["sft","dpo","grpo","ppo","sdft","opd","opsd","sdpo"]).nullable(),destinationId:n.nullable(),taskset:g,comparisonSeriesEntry:A.nullable().optional(),evaluationTasksetRef:g.nullable().optional(),harnessRelease:r.nullable().optional(),quote:e.object({maximumSpendUsd:e.number().nonnegative(),hourlyCostUsd:e.number().nonnegative().nullable()}).strict().nullable(),evaluation:cn.nullable().default(null),evaluationProgress:e.object({stage:e.enum(["baseline","adaptation","refiner","candidate_adaptation","candidate","comparison"]),completedAttempts:e.number().int().nonnegative(),totalAttempts:e.number().int().positive(),accounting:nn.nullable().default(null),advancedAccounting:se.optional(),evidenceSnapshot:an.nullable().default(null)}).strict().nullable().default(null),reward:e.object({raw:e.number(),components:e.record(e.string(),e.number())}).strict().nullable(),receipt:e.union([tn,rn,on,sn,oe]).nullable(),adapterArtifactLineageId:n.nullable(),failure:e.string().trim().min(1).max(5e3).nullable(),startedAt:c,completedAt:c.nullable(),updatedAt:c}).strict().superRefine((t,i)=>{t.status==="succeeded"&&!t.receipt&&i.addIssue({code:"custom",message:"A successful Model Run requires a canonical receipt."}),t.kind==="evaluation"&&(t.method!==null||t.destinationId!==null||!t.evaluation||t.adapterArtifactLineageId!==null)&&i.addIssue({code:"custom",message:"An evaluation Model Run requires evaluation configuration and cannot claim training method, destination, or adapter lineage."}),t.kind==="evaluation"&&t.receipt&&t.receipt.schemaVersion!=="openpond.modelEvaluationReceipt.v1"&&t.receipt.schemaVersion!=="openpond.modelEvaluationStopReceipt.v1"&&t.receipt.schemaVersion!=="openpond.modelComparisonBenchmarkReceipt.v1"&&t.receipt.schemaVersion!=="openpond.advancedRefinerReviewReceipt.v1"&&i.addIssue({code:"custom",message:"An evaluation Model Run requires an evaluation receipt."}),t.kind!=="evaluation"&&(t.method===null||t.destinationId===null||t.quote===null)&&i.addIssue({code:"custom",message:"A training or rollout Model Run requires method, destination, and quote."}),t.kind==="rollout_smoke"&&t.adapterArtifactLineageId!==null&&i.addIssue({code:"custom",message:"A rollout smoke cannot claim adapter artifacts."})});function ai(t){let i=new Map;for(let p of t.attempts){let T=`${p.phase}\0${p.taskId}`,S=i.get(T);(!S||p.startedAt>S.startedAt||p.startedAt===S.startedAt&&p.attemptId>S.attemptId)&&i.set(T,p)}let s=[...i.values()],l=[["adaptation","adaptation","candidate_adaptation"],["held_out","baseline","candidate"]].flatMap(([p,T,S])=>{let At=new Map(s.filter(R=>R.phase===S).map(R=>[R.taskId,R]));return s.filter(R=>R.phase===T).flatMap(R=>{let ee=At.get(R.taskId);return ee?[{cohort:p,taskId:R.taskId,baseline:R,refined:ee,tokenDelta:ee.totalTokens-R.totalTokens}]:[]})}),f=new Set(s.filter(p=>p.phase==="baseline"||p.phase==="adaptation").map(p=>p.taskId)).size,h=t.targetTaskCount??f;if(!Number.isInteger(h)||h<0)throw new Error("Benchmark task-efficiency target must be a non-negative integer.");let x={adaptation:new Set(s.filter(p=>p.phase==="adaptation").map(p=>p.taskId)).size,held_out:new Set(s.filter(p=>p.phase==="baseline").map(p=>p.taskId)).size},k=v((p,T)=>({targetTaskCount:T,comparedTaskCount:p.length,passedTaskCount:p.filter(S=>S.tokenDelta<0).length,failedTaskCount:p.filter(S=>S.tokenDelta>=0).length,lowerTaskCount:p.filter(S=>S.tokenDelta<0).length,higherTaskCount:p.filter(S=>S.tokenDelta>0).length,unchangedTaskCount:p.filter(S=>S.tokenDelta===0).length}),"summarizePairs"),d=k(l,h),j=l.reduce((p,T)=>p+T.baseline.totalTokens,0),ke=l.reduce((p,T)=>p+T.refined.totalTokens,0),Re=ke-j,Ie=l.length===h;return{summary:it.parse({target:"all_tasks_lower",...d,baselineTokens:j,refinedTokens:ke,tokenDelta:Re,tokenDeltaPercent:j>0?Re/j*100:null,complete:Ie,passed:Ie&&d.passedTaskCount===h,cohorts:{adaptation:k(l.filter(p=>p.cohort==="adaptation"),x.adaptation),heldOut:k(l.filter(p=>p.cohort==="held_out"),x.held_out)}}),pairs:l}}v(ai,"summarizeModelEvaluationTaskEfficiency");var dn=e.enum(["measuring","up_to_date","needs_attention"]),un=e.enum(["fixed","retained","regressed","unresolved","unavailable"]),ot=e.object({evaluationRunId:n,attemptKey:n,artifactPath:e.string().trim().min(1).max(2e3),jsonPointer:e.string().trim().min(1).max(1e3),transcriptHash:o.nullable(),traceHash:o.nullable()}).strict(),ge=e.object({level:e.literal(.95),lower:e.number().finite(),upper:e.number().finite()}).strict(),pn=e.object({panelId:n,panelRole:le,passLabel:n.nullable(),taskset:g,attempted:e.number().int().nonnegative(),available:e.number().int().nonnegative(),passed:e.number().int().nonnegative(),strictSuccess:e.number().min(0).max(1).nullable(),strictSuccessCi95:ge.nullable(),judgeScore:e.number().min(0).max(100).nullable(),judgeScoreCi95:ge.nullable()}).strict(),st=e.object({schemaVersion:e.literal("openpond.modelCurrencySnapshot.v1"),id:n,contentHash:o,seriesId:n,protocol:e.object({id:n,revision:e.number().int().positive(),contentHash:o}).strict(),entryId:n,passLabel:n,parent:e.discriminatedUnion("kind",[e.object({kind:e.literal("base_model"),id:n,revision:e.string().trim().min(1).max(500)}).strict(),e.object({kind:e.literal("model_version"),id:n,contentHash:o}).strict()]),candidate:e.object({kind:e.literal("model_version"),id:n,contentHash:o}).strict(),sources:e.object({evaluationRunIds:e.array(n).max(1e5),attemptIds:e.array(n).max(1e6),tasksets:e.array(g).max(1e4),grader:r,judge:r.nullable(),calibration:r.nullable()}).strict(),matches:e.array(e.object({key:n,panelId:n,panelRole:le,taskId:n,issueFamilyId:n,seed:e.number().int(),repetition:e.number().int().nonnegative(),classification:un,parentPassed:e.boolean().nullable(),candidatePassed:e.boolean().nullable(),parentAttempt:ot.nullable(),candidateAttempt:ot.nullable()}).strict()).max(1e6),taskIds:e.object({fixed:e.array(n).max(1e6),retained:e.array(n).max(1e6),regressed:e.array(n).max(1e6),unresolved:e.array(n).max(1e6),unavailable:e.array(n).max(1e6)}).strict(),panels:e.array(pn).max(1e4),metrics:e.object({knownIssueCoverage:e.number().min(0).max(1).nullable(),issueFamilyGeneralization:e.number().min(0).max(1).nullable(),behavioralRetention:e.number().min(0).max(1).nullable(),currentAcquisitionDelta:e.number().finite().nullable(),retainedDeltaPoints:e.number().finite().nullable(),frontierStrictDelta:e.number().finite().nullable(),pairedPValue:e.number().min(0).max(1).nullable(),wins:e.number().int().nonnegative(),ties:e.number().int().nonnegative(),losses:e.number().int().nonnegative()}).strict(),statistics:e.object({matchedAttemptCount:e.number().int().nonnegative(),strictDelta:e.number().finite().nullable(),strictDeltaCi95:ge.nullable(),exactPairedBinaryPValue:e.number().min(0).max(1).nullable(),pairedBootstrapSamples:e.number().int().positive()}).strict(),criteria:e.object({allRequiredAttemptsTerminal:e.boolean(),criticalCorrectionPassRate:e.number().min(0).max(1).nullable(),siblingPassRate:e.number().min(0).max(1).nullable(),behavioralRetentionRate:e.number().min(0).max(1).nullable(),retainedRegressionPoints:e.number().finite().nullable(),criticalPriorRegressionCount:e.number().int().nonnegative(),thresholdRelease:r}).strict(),efficiency:e.object({inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),totalTokens:e.number().int().nonnegative(),costUsd:e.number().nonnegative().nullable(),latencyMs:e.number().int().nonnegative(),tokensPerSuccess:e.number().nonnegative().nullable(),throughputPerHour:e.number().nonnegative().nullable(),gpuSeconds:e.number().nonnegative().nullable()}).strict(),invariants:e.object({systemPromptHash:o,toolSchema:r,application:r,harness:r,runtime:r,grader:r,autoRefinerEnabled:e.boolean()}).strict(),evidenceState:dn,evidenceReasons:e.array(n).max(100),projectedAt:c}).strict();var $=e.discriminatedUnion("kind",[e.object({kind:e.literal("base_model"),id:n,revision:e.string().trim().min(1).max(500),label:e.string().trim().min(1).max(300)}).strict(),e.object({kind:e.literal("model_version"),id:n,label:e.string().trim().min(1).max(300)}).strict(),e.object({kind:e.literal("captured_model"),id:n,label:e.string().trim().min(1).max(300)}).strict()]),U=e.union([e.object({source:e.literal("evaluation_run").default("evaluation_run"),evaluationRunId:n,attemptId:n,modelLabel:e.string().trim().min(1).max(300).nullable().default(null),modelVersionId:n.nullable().default(null),reward:e.number().finite().nullable(),components:e.record(e.string(),e.number().finite()).default({})}).strict(),e.object({source:e.literal("imported_response"),modelLabel:e.string().trim().min(1).max(300),modelVersionId:n.nullable().default(null),response:e.unknown().refine(t=>t!==void 0,"An imported model response is required."),reward:e.number().finite().nullable().default(null),components:e.record(e.string(),e.number().finite()).default({})}).strict()]),bn=e.union([e.object({taskId:n,target:$,attempt:U}).strict(),e.object({taskId:n,target:e.enum(["base","current"]).default("current"),attempt:U}).strict(),e.object({taskId:n,target:e.enum(["base","current"]).default("current"),evaluationRunId:n,attemptId:n,reward:e.number().finite().nullable(),components:e.record(e.string(),e.number().finite()).default({})}).strict().transform(({taskId:t,target:i,...s})=>({taskId:t,target:i,attempt:{source:"evaluation_run",modelLabel:null,modelVersionId:null,...s}}))]),bi=e.object({schemaVersion:e.literal("openpond.continualLearningDailyBatchManifest.v1"),id:n,seriesId:n,scheduleEntryId:n,dayOrdinal:e.number().int().positive().max(1e4),sourceTaskset:g,taskIds:e.array(n).min(1).max(1e3),observedAttempts:e.array(bn).max(1e3).default([]),sourceFileName:e.string().trim().min(1).max(500).nullable().default(null),availableAt:c}).strict().superRefine((t,i)=>{new Set(t.taskIds).size!==t.taskIds.length&&i.addIssue({code:"custom",path:["taskIds"],message:"A daily batch cannot repeat a task."}),new Set(t.observedAttempts.map(l=>`${l.taskId}:${gn(l.target)}`)).size!==t.observedAttempts.length&&i.addIssue({code:"custom",path:["observedAttempts"],message:"An intake batch cannot repeat a task and Policy target pair."});let s=new Set(t.taskIds);t.observedAttempts.some(l=>!s.has(l.taskId))&&i.addIssue({code:"custom",path:["observedAttempts"],message:"Observed attempts must belong to tasks in the daily batch."})}),lt=e.object({schemaVersion:e.literal("openpond.continualLearningDailyBatch.v1"),id:n,seriesId:n,scheduleEntryId:n,dayOrdinal:e.number().int().positive().max(1e4),label:e.string().trim().min(1).max(100),runMode:e.enum(["automatic","manual"]).optional(),source:e.enum(["json_upload","sealed_fixture"]),sourceFileName:e.string().trim().min(1).max(500).nullable(),sourceTaskset:g,tasks:e.array(e.object({taskId:n,taskContentHash:o,familyKey:n,disposition:e.enum(["include","defer","exclude"]).nullable(),oracleReview:e.enum(["pending","confirmed","needs_correction"]),note:e.string().trim().max(2e3),observedAttempt:U.nullable(),baselineAttempt:U.nullable().default(null),responses:e.array(e.object({target:$,attempt:U}).strict()).max(100).default([]),stagedAt:c.nullable().default(null),queuedEntry:A.nullable().default(null)}).strict()).min(1).max(1e3),intakeEvaluation:e.object({status:e.enum(["awaiting","running","ready","failed"]),requestedTargets:e.array(e.enum(["base","current"])).max(2).default(["base","current"]),requestedPolicies:e.array($).max(100).default([]),runs:e.array(e.object({target:$,runId:n}).strict()).max(100).default([]),baselineRunId:n.nullable(),currentRunId:n.nullable(),currentPolicy:e.discriminatedUnion("kind",[e.object({kind:e.literal("base_model"),id:n}).strict(),e.object({kind:e.literal("model_version"),id:n}).strict()]).nullable(),failure:e.string().trim().min(1).max(2e3).nullable()}).strict().default({status:"awaiting",requestedTargets:["base","current"],requestedPolicies:[],runs:[],baselineRunId:null,currentRunId:null,currentPolicy:null,failure:null}),status:e.enum(["pending","reviewed","queued"]),queuedEntry:A.nullable(),reviewedBy:n.nullable(),reviewedAt:c.nullable(),availableAt:c,revision:e.number().int().positive(),createdAt:c,updatedAt:c}).strict().superRefine((t,i)=>{new Set(t.tasks.map(l=>l.taskId)).size!==t.tasks.length&&i.addIssue({code:"custom",path:["tasks"],message:"A daily batch cannot repeat a task."});for(let[l,f]of t.tasks.entries()){let h=f.responses.map(x=>ct(x.target));new Set(h).size!==h.length&&i.addIssue({code:"custom",path:["tasks",l,"responses"],message:"A task cannot repeat a model response target."})}let s=t.tasks.every(l=>l.disposition!==null&&l.oracleReview!=="pending"&&(l.oracleReview!=="needs_correction"||l.note.length>0));t.status!=="pending"&&!s&&i.addIssue({code:"custom",path:["tasks"],message:"Every task requires a decision and oracle review before the day completes."}),t.status==="queued"!=(t.queuedEntry!==null)&&i.addIssue({code:"custom",path:["queuedEntry"],message:"Only a queued day carries an exact series-entry reference."}),t.status==="pending"&&(t.reviewedAt!==null||t.reviewedBy!==null)&&i.addIssue({code:"custom",path:["reviewedAt"],message:"A pending day cannot carry reviewer completion identity."}),t.status!=="pending"&&(t.reviewedAt===null||t.reviewedBy===null)&&i.addIssue({code:"custom",path:["reviewedAt"],message:"Reviewed identity is required after a day review completes."})}),mt=e.object({schemaVersion:e.literal("openpond.continualBenchIssueReview.v1"),id:n,seriesId:n,scheduleEntryId:n,passLabel:n,packet:De,splitManifest:e.object({id:n,contentHash:o}).strict(),decisions:e.array(e.object({taskId:n,disposition:e.enum(["include","defer","exclude"]).nullable(),note:e.string().trim().max(2e3)}).strict()).min(2).max(1e4),status:e.enum(["pending","reviewed","queued"]),queuedEntry:A.nullable(),reviewedBy:n.nullable(),reviewedAt:c.nullable(),revision:e.number().int().positive(),createdAt:c,updatedAt:c}).strict().superRefine((t,i)=>{let s=[...t.packet.cases.map(h=>h.taskId)].sort(),l=[...t.decisions.map(h=>h.taskId)].sort();JSON.stringify(s)!==JSON.stringify(l)&&i.addIssue({code:"custom",path:["decisions"],message:"Review decisions must cover the exact issue-packet cases."});let f=t.decisions.every(h=>h.disposition!==null&&h.note.length>0);t.status!=="pending"&&!f&&i.addIssue({code:"custom",path:["decisions"],message:"Every case requires a disposition and note before review completes."}),t.status==="queued"!=(t.queuedEntry!==null)&&i.addIssue({code:"custom",path:["queuedEntry"],message:"Only a queued review carries an exact series-entry reference."})}).readonly();function ct(t){return`${t.kind}:${t.id}`}v(ct,"responseTargetKey");function gn(t){return typeof t=="string"?t:ct(t)}v(gn,"observedTargetKey");var I=e.string().trim().min(1).max(240),_=e.string().trim().min(8).max(256),F=e.enum(["grpo","dapo","gspo-token"]),H=e.object({id:I,revision:e.string().trim().min(1).max(256),tokenizerRevision:e.string().trim().min(1).max(256),chatTemplateHash:_}),dt=e.object({maxRollouts:e.number().int().positive().max(1e6),maxEnvironmentExecutions:e.number().int().positive().max(1e6),maxInputTokens:e.number().int().positive(),maxOutputTokens:e.number().int().positive(),maxOptimizerSteps:e.number().int().positive().max(1e6),wallTimeMs:e.number().int().positive(),maximumCostUsd:e.number().nonnegative().nullable()}),ut=e.object({method:e.literal("grpo"),groupSize:e.number().int().min(2).max(128),normalization:e.literal("group_standardized"),advantageEpsilon:e.number().positive().max(.01).default(1e-8),loss:F,clipRange:e.number().positive().max(1).default(.2),iterations:e.number().int().min(2).max(16).default(2),microbatchSize:e.number().int().positive().max(128).default(1),gradientAccumulationSteps:e.number().int().positive().max(128).default(1),adamw:D}),G=e.object({method:e.literal("ppo"),valueModel:H,gamma:e.number().min(0).max(1),gaeLambda:e.number().min(0).max(1),policyClip:e.number().positive().max(1),valueClip:e.number().positive().max(10),valueLossCoefficient:e.number().nonnegative().max(10),ppoEpochs:e.number().int().positive().max(100),minibatchSize:e.number().int().positive().max(1e5)}),pt=e.discriminatedUnion("method",[ut,G]),J=e.object({rewardModelVersion:e.object({id:I,contentHash:_}).strict(),qualificationReport:e.object({id:I,contentHash:_}).strict().nullable().default(null),evaluationReferences:e.array(e.object({id:I,contentHash:_}).strict()).max(1e3).default([]),checkpoint:e.object({id:I,contentHash:_,objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),files:e.array(e.object({path:e.string().trim().min(1).max(512),sizeBytes:e.number().int().positive(),sha256:_}).strict()).min(4).max(128)}).strict(),runtime:ue,processorRelease:e.object({id:I,contentHash:_}).strict(),rewardComposerRelease:e.object({id:I,contentHash:_}).strict(),executionReceipt:e.object({id:I,contentHash:_}).strict()}).strict(),W=e.object({schemaVersion:e.literal("openpond.policyOptimization.v1"),policyModel:H,referenceModel:H,dataset:e.object({tasksetId:I,tasksetHash:_,split:e.literal("train"),selectionStrategy:e.enum(["stable_hash_top_n","rft_easy_curriculum_v1"]),selectionSeed:e.number().int(),maxExamples:e.number().int().positive().max(1e5)}),sampler:e.object({temperature:e.number().min(0).max(2),topP:e.number().min(0).max(1),maxOutputTokens:e.number().int().positive().max(32768),maxTurns:e.number().int().positive().max(1e3),concurrency:e.number().int().positive().max(1e3)}),environment:e.object({id:I,version:e.string().trim().min(1).max(256),toolContractHash:_}),reward:e.object({graderId:I,graderHash:_,learnedPreference:J.nullable().default(null)}),kl:e.object({coefficient:e.number().nonnegative().nullable(),referenceConstraint:e.literal("fixed_reference")}),budgets:dt,checkpointEverySteps:e.number().int().positive(),seed:e.number().int(),evaluationSplit:e.literal("frozen_eval"),optimizer:pt});var a=e.string().trim().min(1).max(240),b=e.string().trim().min(1),m=e.string().trim().min(8).max(256),O=e.record(e.string(),e.unknown()).default({});var he=e.object({schemaVersion:e.literal("openpond.localModelChatConfiguration.v1").default("openpond.localModelChatConfiguration.v1"),profile:e.enum(["efficient","full_harness","custom"]).default("efficient"),systemPromptMode:e.enum(["lean","full_harness","custom"]).default("lean"),customSystemPrompt:e.string().max(2e4).nullable().default(null),contextWindowTokens:e.number().int().min(128).max(32768).default(1024),maxOutputTokens:e.number().int().min(1).max(512).default(64),temperature:e.number().min(0).max(2).default(0),repetitionPenalty:e.number().min(.5).max(2).default(1.1),noRepeatNgramSize:e.number().int().min(0).max(10).default(3),compaction:e.enum(["off","when_needed"]).default("when_needed"),keepWarmSeconds:e.number().int().min(0).max(3600).default(300),updatedAt:b.nullable().default(null)}),bt=he.parse({}),gt=e.object({schemaVersion:e.literal("openpond.managedAdapterServingProjection.v1"),teamId:a.nullable().default(null),source:e.literal("sandbox_managed_rl"),sourceRef:a,canonicalArtifactId:a.nullable(),canonicalModelName:e.string().trim().min(1).max(256).nullable().default(null),canonicalArtifactState:e.enum(["imported_unvalidated","evaluating","promotable","rejected","deleted"]).nullable(),canonicalDeploymentId:a.nullable(),canonicalDeploymentState:e.enum(["requested","deploying","ready","degraded","deleting","deleted","failed"]).nullable(),state:e.enum(["pending","imported","ready","failed"]),customerBindingAllowed:e.boolean().default(!1),artifactContentHash:m.nullable().default(null),baseProfileId:a.nullable().default(null),publishedAt:b.nullable(),lastSyncedAt:b,lastError:e.string().trim().min(1).max(5e3).nullable()});function hn(t){return t.state==="ready"&&t.canonicalArtifactState==="promotable"&&t.canonicalDeploymentState==="ready"&&t.customerBindingAllowed&&!!t.teamId&&!!t.canonicalArtifactId&&!!t.canonicalDeploymentId}v(hn,"managedAdapterProjectionReady");function ji(t){let i=t.managedServing;return i?hn(i)&&i.canonicalArtifactId&&i.canonicalDeploymentId?{kind:"sandbox_serving",canonicalArtifactId:i.canonicalArtifactId,canonicalDeploymentId:i.canonicalDeploymentId}:null:{kind:"source_artifact",canonicalArtifactId:null,canonicalDeploymentId:null}}v(ji,"resolveModelBindingEligibility");var ye="openpond.crossSystemOperations.v1",wi="1.0.0",ft="openpond.crossSystemTools.v1",fn=["search_crm","query_billing","search_support","run_python"],vn=e.enum(["train","validation","frozen_eval"]),xn=e.enum(["easy","medium","hard"]),Sn=e.enum(["renewal_risk_v2"]),Mi=e.enum(["renewal_exposure","collections_prioritization","invoice_reconciliation","sla_escalation","contract_billing_mismatch"]),Pi=e.object({seed:e.number().int().nonnegative(),split:vn,difficulty:xn,scenarioProfile:Sn.optional()});var fe={anyOf:[{type:"string",minLength:1,maxLength:256},{type:"null"}],default:null},ve={type:"integer",minimum:1,maximum:50,default:25},vt=[{name:"search_crm",description:"Search synthetic CRM accounts and contacts. Results are scoped, projected, paginated, and budgeted.",parameters:{type:"object",additionalProperties:!1,required:["query","fields","cursor","limit"],properties:{query:{type:"string",minLength:1,maxLength:500},fields:{type:"array",minItems:1,maxItems:12,uniqueItems:!0,items:{type:"string",enum:["account_id","name","aliases","contact_ids","contract_value_usd_cents","renewal_date","tier","active_contract_id","contract_term_months"]}},cursor:fe,limit:ve}}},{name:"query_billing",description:"Query synthetic invoices and payments for explicit account IDs and a bounded date range.",parameters:{type:"object",additionalProperties:!1,required:["account_ids","date_range","status","cursor","limit"],properties:{account_ids:{type:"array",minItems:1,maxItems:50,uniqueItems:!0,items:{type:"string",minLength:1,maxLength:100}},date_range:{type:"object",additionalProperties:!1,required:["from","to"],properties:{from:{type:"string",format:"date"},to:{type:"string",format:"date"}}},status:{type:"array",minItems:1,maxItems:6,uniqueItems:!0,items:{type:"string",enum:["open","overdue","paid","disputed","scheduled","void"]}},cursor:fe,limit:ve}}},{name:"search_support",description:"Search synthetic support cases for explicit account IDs, severity, and state filters.",parameters:{type:"object",additionalProperties:!1,required:["account_ids","severity","state","cursor","limit"],properties:{account_ids:{type:"array",minItems:1,maxItems:50,uniqueItems:!0,items:{type:"string",minLength:1,maxLength:100}},severity:{type:"array",minItems:1,maxItems:4,uniqueItems:!0,items:{type:"string",enum:["P1","P2","P3","P4"]}},state:{type:"array",minItems:1,maxItems:5,uniqueItems:!0,items:{type:"string",enum:["new","investigating","waiting_customer","resolved","closed"]}},cursor:fe,limit:ve}}},{name:"run_python",description:"Run standard-library-only Python in the persistent synthetic attempt sandbox with strict limits and no network.",parameters:{type:"object",additionalProperties:!1,required:["code"],properties:{code:{type:"string",minLength:1,maxLength:1e4}}}}],Z=Cn({schemaVersion:ft,tools:vt}),yn=[`Use only the four registered synthetic Cross-System Operations tools. Contract ${Z}.`,"Treat every filter as exact. An overdue-only query_billing call returns invoice rows; payment rows are included only when paid is requested.","For cross-system joins, retrieve the bounded projected rows and use run_python for date filtering, set intersection, and arithmetic before answering.","Finish with ANSWER: JSON."].join(" "),_n=15;function kn(t="auto"){return[yn,"LOCAL TOOL PROTOCOL (strict):","Use only the registered functions below. Never invent, rename, or call any other tool.","For one tool call, output exactly one JSON object and no prose:",'{"type":"tool_call","name":"registered_name","arguments":{}}',"Tool results arrive as exactly one JSON object:",'{"type":"tool_result","name":"registered_name","ok":true,"result":{},"error":null}','After one tool result, either call one registered tool again or answer normally. A JSON final may use {"type":"final","content":"..."}.',`Tool choice: ${JSON.stringify(t)}. Maximum tool turns: ${_n}.`,`Registered tool signatures: ${vt.map(Rn).join("; ")}`].join(`
+import{a as le,b as De,c as Be}from"./chunk-UTIELP6F.js";import{k as re,q as oe,t as se}from"./chunk-5TPYBXWM.js";import{d as ie,f as Ee,g as He,h as Oe,i as ze,l as B,n as Ve}from"./chunk-6LFTQBXP.js";import{a as Pe}from"./chunk-VPAEFG7X.js";import{i as D}from"./chunk-TS5HNRZB.js";import{B as Ae,Bb as Me,k as te,m as ne,oa as Ce,ta as we,wa as ae}from"./chunk-5ZNE43XC.js";import{A as Te}from"./chunk-263WS3TB.js";import{h as je}from"./chunk-QAFXQHHG.js";import{f as V}from"./chunk-HNQPEOJD.js";import{c as n,d as o,e as c,g as r,h as g}from"./chunk-RCT3O3HQ.js";import{A as e}from"./chunk-X2J2P5ZT.js";import{a as v}from"./chunk-FTQLQJQV.js";var L=e.string().trim().min(1).max(240),N=e.string().trim().min(1),me=e.string().trim().min(8).max(256),Le=e.record(e.string(),e.unknown()).default({}),Ct=e.enum(["created","inspecting","awaiting_source_review","awaiting_mapping","awaiting_materialization_approval","materializing","validating","ready","cancelling","cancelled","failed"]),qe=e.object({schemaVersion:e.literal("openpond.huggingFaceDatasetLocator.v1"),repositoryId:e.string().trim().regex(/^[^/\s]+\/[^/\s]+$/).max(500),repositoryUrl:e.string().url().max(2e3),requestedRevision:e.string().trim().min(1).max(256).nullable()}),wt=e.object({path:e.string().trim().min(1).max(1e3),logicalType:e.string().trim().min(1).max(500),nullable:e.boolean(),sampleValues:e.array(e.unknown()).max(5)}),Mt=e.object({configuration:e.string().trim().min(1).max(500),split:e.string().trim().min(1).max(500),rowCount:e.number().int().nonnegative().nullable(),sizeBytes:e.number().int().nonnegative().nullable()}),Pt=e.object({schemaVersion:e.literal("openpond.huggingFaceDatasetInspection.v1"),id:L,locator:qe,resolvedRevision:e.string().trim().min(7).max(256),title:e.string().trim().min(1).max(500),description:e.string().max(2e4).nullable(),declaredLicense:e.string().trim().min(1).max(500).nullable(),gated:e.boolean(),private:e.boolean(),configurations:e.array(e.string().trim().min(1).max(500)).min(1).max(1e3),splits:e.array(Mt).min(1).max(1e4),columns:e.array(wt).min(1).max(1e4),previewRows:e.array(e.record(e.string(),e.unknown())).max(25),sourceFiles:e.array(e.object({path:e.string().trim().min(1).max(2e3),configuration:e.string().trim().min(1).max(500),split:e.string().trim().min(1).max(500),revision:e.string().trim().min(7).max(256),sizeBytes:e.number().int().nonnegative().nullable(),contentHash:me.nullable()})).max(1e5),inspectedAt:N,metadata:Le}),Et=e.enum(["row_id","cluster_id","prompt","messages","demonstration","chosen","rejected","expected_output","privileged_context","reward","feedback","tag","metadata"]),Ht=e.object({sourcePath:e.string().trim().min(1).max(1e3),target:Et,transform:e.enum(["identity","string","json","messages","numeric","math_final_answer"]),policy:e.enum(["visible","privileged","metadata"]),required:e.boolean()}),Ue=e.object({schemaVersion:e.literal("openpond.datasetImportMapping.v1"),sourceSchemaHash:me,configuration:e.string().trim().min(1).max(500),upstreamSplits:e.array(e.string().trim().min(1).max(500)).min(1).max(1e3),preset:e.enum(["prompt_only","prompt_completion","messages","preference","prompt_expected_answer","prompt_privileged_solution","prompt_feedback"]),bindings:e.array(Ht).min(1).max(1e4),nullPolicy:e.enum(["reject","drop","preserve"]),invalidRowPolicy:e.enum(["reject_import","drop_with_receipt"]),splitPolicy:e.object({seed:e.number().int(),assignments:e.record(e.string(),te),validationPercent:e.number().min(0).max(100),frozenEvalPercent:e.number().min(0).max(100)}),importerVersion:e.string().trim().min(1).max(200),mappingHash:me}),Ot=e.object({phase:e.enum(["idle","metadata","download","mapping","parquet_write","verification","publish"]),completedBytes:e.number().int().nonnegative(),totalBytes:e.number().int().nonnegative().nullable(),completedRows:e.number().int().nonnegative(),totalRows:e.number().int().nonnegative().nullable(),message:e.string().trim().min(1).max(2e3).nullable()}),Ne=e.object({schemaVersion:e.literal("openpond.datasetImportJob.v1"),id:L,profileId:L,sourceKind:e.enum(["huggingface","uploaded_file"]),status:Ct,locator:qe.nullable(),inspection:Pt.nullable(),mapping:Ue.nullable(),progress:Ot,targetStorageRoot:e.string().trim().min(1).max(4e3).nullable(),tasksetId:L.nullable(),tasksetRevision:e.number().int().positive().nullable(),artifactId:L.nullable(),error:e.string().trim().min(1).max(2e4).nullable(),cancellationRequested:e.boolean(),createdAt:N,updatedAt:N,completedAt:N.nullable(),metadata:Le}),ya=e.object({profileId:L,url:e.string().trim().min(1).max(2e3)}),_a=e.object({name:e.string().trim().min(1).max(500),objective:e.string().trim().min(1).max(2e4),mapping:Ue,targetStorageRoot:e.string().trim().min(1).max(4e3).nullable(),licenseApproved:e.boolean().default(!1)});var y=e.string().trim().min(1).max(240),w=e.string().trim().min(1),$e=e.record(e.string(),e.unknown()).default({}),Fe=e.enum(["nominated","needs_review","approved_for_creation","creating","needs_task_review","baselining","ready_for_export","rejected","dismissed","retired","blocked"]),zt=e.object({id:y,kind:e.enum(["repeated_success","accepted_correction","agent_action_recurrence","outcome_linked_change","frontier_cost","expert_label","runtime_feedback"]),sourceRefIds:e.array(y).min(1).max(1e4),occurredAt:w,signature:e.string().trim().min(1).max(2e3),summary:e.string().trim().min(1).max(5e3),confidence:e.number().min(0).max(1),consented:e.boolean(),metadata:$e}),Vt=e.object({frequency:e.number().min(0).max(1),businessValue:e.number().min(0).max(1),frontierCost:e.number().min(0).max(1),signalQuality:e.number().min(0).max(1),verifiability:e.number().min(0).max(1),repeatability:e.number().min(0).max(1),privacyRisk:e.number().min(0).max(1),overall:e.number().min(0).max(1)}),Dt=e.object({tactic:Pe,eligible:e.boolean(),reasons:e.array(e.string().trim().min(1).max(5e3)).min(1).max(100),blockers:e.array(e.string().trim().min(1).max(5e3)).max(100).default([]),requiredSignals:e.array(e.string().trim().min(1).max(500)).max(100).default([]),generatedBy:e.enum(["decision_table","baseline_reassessment"])}),ce=e.object({schemaVersion:e.literal("openpond.taskCandidate.v1"),id:y,profileId:y,status:Fe,fingerprint:e.string().trim().min(1).max(512),title:e.string().trim().min(1).max(500),summary:e.string().trim().min(1).max(1e4),workflowSignature:e.string().trim().min(1).max(2e3),evidence:e.array(zt).min(1).max(1e5),scorecard:Vt,recommendation:Dt,mergedIntoId:y.nullable(),createdAt:w,updatedAt:w,metadata:$e}),q=e.object({schemaVersion:e.literal("openpond.taskMinerConfig.v1"),enabled:e.boolean().default(!1),localOnly:e.boolean().default(!0),observationWindowDays:e.number().int().min(1).max(365).default(30),minimumRecurrence:e.number().int().min(2).max(100).default(3),clustering:e.literal("hybrid_deterministic_first").default("hybrid_deterministic_first"),consentRequired:e.boolean().default(!0)}),ja=e.object({items:e.array(ce),generatedAt:w,config:q}),Ta=e.object({status:Fe.optional(),mergeIntoId:y.nullable().optional()}),Aa=e.object({profileId:y,sourceIds:e.array(y).max(1e5).default([]),sessionIds:e.array(y).max(1e5).default([]),config:q.optional()}),Ge=e.object({schemaVersion:e.literal("openpond.taskMinerRun.v1"),id:y,profileId:y,status:e.enum(["queued","running","cancelling","cancelled","succeeded","failed"]),config:q,sourceIds:e.array(y).max(1e5),sessionIds:e.array(y).max(1e5).default([]),progress:e.object({stage:e.enum(["queued","ingesting","preparing","clustering","persisting","complete"]),processedSources:e.number().int().nonnegative(),totalSources:e.number().int().nonnegative(),candidatesFound:e.number().int().nonnegative(),skippedSources:e.number().int().nonnegative().default(0)}),candidateIds:e.array(y).max(1e5),cancelRequested:e.boolean(),error:e.string().trim().min(1).max(2e4).nullable(),createdAt:w,startedAt:w.nullable(),completedAt:w.nullable(),updatedAt:w});var Bt=e.enum(["draft","active","completed","archived"]),Lt=e.enum(["draft","ready","queued","running","candidate","accepted","rejected","no_signal","failed","cancelled"]),Je=e.enum(["seed","daily_residual","rank_candidate","weekly_rollup","full_refresh"]),qt=e.enum(["base_model","previous_release","seed_release","accepted_daily_head","accepted_seed"]),Ut=e.enum(["seed_taskset","nightly_selection","daily_cohort_union","eligible_task_pool"]),Nt=e.enum(["live_client","evaluation_attempt","replay_evidence","manual"]),$t=e.object({id:n,ordinal:e.number().int().nonnegative(),label:e.string().trim().min(1).max(100),role:Je,parentRule:qt,taskSource:Ut,trainableRank:e.number().int().positive().max(1024),minimumTasks:e.number().int().nonnegative().max(1e5),maximumTasks:e.number().int().positive().max(1e5)}).strict().superRefine((t,i)=>{t.minimumTasks>t.maximumTasks&&i.addIssue({code:"custom",path:["minimumTasks"],message:"The minimum task count cannot exceed the maximum task count."});let s={seed:{parentRules:["base_model"],taskSource:"seed_taskset"},daily_residual:{parentRules:["previous_release","accepted_daily_head"],taskSource:"nightly_selection"},rank_candidate:{parentRules:["seed_release"],taskSource:"eligible_task_pool"},weekly_rollup:{parentRules:["seed_release","accepted_seed"],taskSource:"daily_cohort_union"},full_refresh:{parentRules:["base_model"],taskSource:"eligible_task_pool"}},{parentRules:l,taskSource:f}=s[t.role];l.includes(t.parentRule)||i.addIssue({code:"custom",path:["parentRule"],message:`${t.role} requires parent rule ${l.join(" or ")}.`}),t.taskSource!==f&&i.addIssue({code:"custom",path:["taskSource"],message:`${t.role} requires task source ${f}.`}),t.ordinal===0&&t.role!=="seed"&&i.addIssue({code:"custom",path:["role"],message:"The first series entry must be the seed."})}),Ft=e.object({id:n,version:e.number().int().positive(),requireCheckpoint:e.boolean(),requireAppliedOptimizerUpdate:e.boolean(),minimumCurrentCohortMeanImprovement:e.number().finite(),maximumRetainedMeanRegression:e.number().finite().nonnegative(),blockCriticalInvariantRegression:e.boolean(),automaticDailyAdvancement:e.boolean()}).strict(),We=e.object({schemaVersion:e.literal("openpond.modelComparisonSeries.v1"),id:n,profileId:n,modelProjectId:n,name:e.string().trim().min(1).max(500),objective:e.string().trim().min(1).max(2e4),status:Bt,revision:e.number().int().positive(),productionBinding:e.object({role:e.literal("chat_manual"),roleTargetId:n}).strict(),baseModel:e.object({id:n,revision:e.string().trim().min(1).max(256)}).strict(),seedTaskset:g,eligibleTaskPool:g,evaluationTasksets:e.object({development:g,retained:g,frozenFinal:g}).strict(),grader:r,benchmarkProtocol:Be.nullable().default(null),automaticEvaluation:e.object({enabled:e.boolean()}).strict().default({enabled:!1}),residualProfile:e.object({profileId:n,serializedEnvelopeRank:e.number().int().positive().max(1024),maximumEnabledRank:e.number().int().positive().max(1024),topology:e.literal("uniform_block_masked")}).strict(),schedule:e.array($t).min(1).max(1e3),scheduleSealedAt:c.nullable(),advancementPolicy:Ft,executionPolicy:e.object({startWhenReady:e.literal(!1)}).strict(),acceptedSeedEntryId:n.nullable(),acceptedDailyHeadEntryId:n.nullable(),promotedBindingId:n.nullable(),createdBy:n,createdAt:c,updatedAt:c}).strict().superRefine((t,i)=>{[...t.schedule].sort((d,j)=>d.ordinal-j.ordinal).some((d,j)=>d.ordinal!==j)&&i.addIssue({code:"custom",path:["schedule"],message:"Schedule ordinals must be unique and contiguous from zero."}),(new Set(t.schedule.map(d=>d.id)).size!==t.schedule.length||new Set(t.schedule.map(d=>d.label)).size!==t.schedule.length)&&i.addIssue({code:"custom",path:["schedule"],message:"Schedule ids and labels must be unique."}),t.residualProfile.maximumEnabledRank>t.residualProfile.serializedEnvelopeRank&&i.addIssue({code:"custom",path:["residualProfile"],message:"Maximum enabled rank cannot exceed the serialized envelope."});let l=Object.values(t.evaluationTasksets);new Set(l.map(d=>`${d.id}:${d.revision}:${d.contentHash}`)).size!==l.length&&i.addIssue({code:"custom",path:["evaluationTasksets"],message:"Development, retained, and frozen-final cohorts require distinct immutable Taskset releases."});let f=t.schedule.find(d=>d.role==="seed");if(!f||t.schedule.filter(d=>d.role==="seed").length!==1){i.addIssue({code:"custom",path:["schedule"],message:"A series requires exactly one seed entry."});return}let h=t.schedule.filter(d=>d.role==="rank_candidate");h.length===1&&i.addIssue({code:"custom",path:["schedule"],message:"A causal rank comparison requires at least two rank candidates."}),new Set(h.map(d=>d.trainableRank)).size!==h.length&&i.addIssue({code:"custom",path:["schedule"],message:"Causal rank candidates must use distinct trainable ranks."}),[f.trainableRank+t.schedule.filter(d=>d.role==="daily_residual").reduce((d,j)=>d+j.trainableRank,0),...t.schedule.filter(d=>d.role==="weekly_rollup").map(d=>f.trainableRank+d.trainableRank),...t.schedule.filter(d=>d.role==="rank_candidate").map(d=>f.trainableRank+d.trainableRank),...t.schedule.filter(d=>d.role==="full_refresh").map(d=>d.trainableRank)].some(d=>d>t.residualProfile.maximumEnabledRank)&&i.addIssue({code:"custom",path:["schedule"],message:"A declared branch exceeds the maximum enabled rank."})}),Ye=e.object({source:Nt,taskIds:e.array(n).min(1).max(1e5),observedFrom:c,observedTo:c,reviewedAt:c,reviewedBy:n,sourceTaskset:g}).strict(),Gt=Ye.extend({derivedFamilyKeys:e.array(n).min(1).max(1e5),familyDerivation:e.object({algorithm:n,sourceTasksetHash:o}).strict(),selectionHash:o}).strict(),de=e.discriminatedUnion("kind",[e.object({kind:e.literal("base_model"),id:n,revision:e.string().trim().min(1).max(256)}).strict(),e.object({kind:e.literal("model_version"),id:n,contentHash:o}).strict()]),Jt=e.object({id:n,branchOrdinal:e.number().int().nonnegative(),rank:e.number().int().positive().max(1024),offsetStart:e.number().int().nonnegative(),offsetEnd:e.number().int().positive(),optimizationRole:e.enum(["frozen","trainable"]),artifactLineageId:n.nullable()}).strict().superRefine((t,i)=>{t.offsetEnd-t.offsetStart!==t.rank&&i.addIssue({code:"custom",path:["offsetEnd"],message:"Block offsets must span exactly the declared rank."})}),Wt=e.object({disposition:e.enum(["advance","hold","no_signal"]),policy:e.object({id:n,version:e.number().int().positive()}).strict(),reasonCodes:e.array(n).min(1).max(100),summary:e.string().trim().min(1).max(5e3),decidedBy:n,decidedAt:c}).strict(),Yt=e.object({evaluationRunId:n,modelVersionId:n,taskset:g,grader:r,cohortRole:e.enum(["current","correction","sibling_verification","cumulative_known","development","retained","prior_disclosed","frozen_final"]),panelId:n.nullable().default(null),protocol:e.object({id:n,revision:e.number().int().positive(),contentHash:o}).strict().nullable().default(null)}).strict(),Kt=e.object({attemptOrdinal:e.number().int().positive(),trainingPlanId:n.nullable(),modelRunId:n.nullable(),terminalStatus:e.enum(["failed","cancelled"]),queuedAt:c.nullable(),startedAt:c.nullable(),completedAt:c}).strict(),A=e.object({seriesId:n,entryId:n,scheduleEntryId:n,ordinal:e.number().int().nonnegative(),releaseHash:o}).strict(),Ke=e.object({schemaVersion:e.literal("openpond.modelComparisonSeriesEntry.v1"),id:n,seriesId:n,profileId:n,modelProjectId:n,scheduleEntryId:n,releaseHash:o,ordinal:e.number().int().nonnegative(),label:e.string().trim().min(1).max(100),role:Je,branch:e.enum(["daily","rank_candidate","weekly_rollup","full_refresh"]),status:Lt,parent:de,taskset:g,sourceTasksets:e.array(g).min(1).max(1e3),taskSelection:Gt.nullable(),trainableRank:e.number().int().positive().max(1024),serializedEnvelopeRank:e.number().int().positive().max(1024),enabledCumulativeRank:e.number().int().positive().max(1024),trainableBlockId:n,residualBlocks:e.array(Jt).min(1).max(1024),attemptOrdinal:e.number().int().positive().default(1),priorRunAttempts:e.array(Kt).max(100).default([]),trainingPlanId:n.nullable(),modelRunId:n.nullable(),modelVersionId:n.nullable(),evaluations:e.array(Yt).max(1e5),decision:Wt.nullable(),promotionBindingId:n.nullable(),queuedAt:c.nullable(),startedAt:c.nullable(),completedAt:c.nullable(),createdAt:c,updatedAt:c}).strict().superRefine((t,i)=>{let s=[...t.residualBlocks].sort((x,k)=>x.offsetStart-k.offsetStart);s.some((x,k)=>x.offsetStart!==(k===0?0:s[k-1].offsetEnd))&&i.addIssue({code:"custom",path:["residualBlocks"],message:"Residual block offsets must be contiguous from zero."});let l=s.at(-1)?.offsetEnd??0;(l!==t.enabledCumulativeRank||l>t.serializedEnvelopeRank)&&i.addIssue({code:"custom",path:["enabledCumulativeRank"],message:"Enabled rank must match the block inventory and fit the envelope."});let f=t.residualBlocks.filter(x=>x.optimizationRole==="trainable");(f.length!==1||f[0]?.id!==t.trainableBlockId||f[0]?.rank!==t.trainableRank)&&i.addIssue({code:"custom",path:["trainableBlockId"],message:"Exactly one block must be trainable during the Run and match trainableRank."});let h=t.priorRunAttempts.map(x=>x.attemptOrdinal);(t.priorRunAttempts.length!==t.attemptOrdinal-1||h.some((x,k)=>x!==k+1))&&i.addIssue({code:"custom",path:["priorRunAttempts"],message:"Prior Run attempts must be a contiguous history before the active attempt."})}),Ea=e.object({seriesId:n,scheduleEntryId:n,taskSelection:Ye.nullable(),expectedSeriesRevision:e.number().int().positive()}).strict();var M=n.nullable(),C=e.number().int().nonnegative(),Qt=e.string().regex(/^[a-f0-9]{40}$/),Zt=e.object({name:e.string().trim().min(1).max(160),startedAt:c,completedAt:c,durationMs:e.number().nonnegative(),clock:e.enum(["monotonic","provider","wall"]),outcome:e.enum(["succeeded","failed","cancelled"])}).strict(),Qe=e.object({schemaVersion:e.literal("openpond.correlatedTelemetryReceipt.v1"),stage:e.enum(["training","evaluation","deployment","inference"]),correlation:e.object({modelRunId:M,modelVersionId:M,policyVersion:C.nullable(),taskId:M,rolloutGroupId:M,providerResourceId:M,deploymentId:M,inferenceRequestId:M}).strict(),spans:e.array(Zt).max(1e4),usage:e.object({promptTokens:C.nullable(),generatedTokens:C.nullable(),gpuSeconds:e.number().nonnegative().nullable(),workerActiveSeconds:e.number().nonnegative().nullable(),optimizerSteps:C.nullable(),rolloutGroups:C.nullable(),successfulTrajectories:C.nullable(),failedTrajectories:C.nullable(),peakGpuMemoryBytes:C.nullable(),peakGpuUtilizationPercent:e.number().min(0).max(100).nullable()}).strict(),resource:e.object({provider:n,resourceIds:e.array(n).max(1e3),gpuType:e.string().trim().min(1).max(300).nullable(),gpuCount:C.nullable(),baseProfileId:M,baseRepository:e.string().trim().min(1).max(512).nullable(),baseRevision:Qt.nullable(),adapterContentHash:o.nullable()}).strict(),cost:e.object({currency:e.literal("USD"),providerReportedUsd:e.number().nonnegative().nullable(),quotedHourlyUsd:e.number().nonnegative().nullable(),estimatedUsd:e.number().nonnegative().nullable(),methodologyVersion:e.string().trim().min(1).max(160),pricingInputs:e.record(e.string().trim().min(1).max(160),e.number().finite()),unitEstimates:e.record(e.string().trim().min(1).max(160),e.number().nonnegative())}).strict(),recordedAt:c,contentHash:o}).strict();var Xt=e.enum(["base_reference","lora_adapter"]),ue=e.object({baseModel:e.object({source:e.literal("huggingface"),repoId:n,revision:e.string().regex(/^[a-f0-9]{40}$/),configHash:o,tokenizerHash:o,licenseId:n,gated:e.boolean()}).strict(),processor:e.object({repository:n,revision:e.string().regex(/^[a-f0-9]{40}$/),configHash:o}).strict()}).strict().superRefine((t,i)=>{(t.baseModel.repoId!==t.processor.repository||t.baseModel.revision!==t.processor.revision)&&i.addIssue({code:"custom",path:["processor"],message:"Reward Model processor must match the pinned scorer base model."})}),tt=e.object({schemaVersion:e.literal("openpond.modelVersion.v1"),id:n,modelId:n,profileId:n,version:e.number().int().nonnegative(),kind:Xt,status:e.enum(["available","failed"]),baseModel:B,taskset:g,comparisonSeriesEntry:A.nullable().optional(),evaluationTasksetRef:g.nullable().optional(),releaseGraph:e.object({resolvedBundleHash:o,profileRelease:g,harnessRelease:r,agentRelease:r.nullable(),grader:e.object({id:n,contentHash:o}).strict()}).strict(),artifactLineageId:n.nullable(),adapterStatus:e.enum(["not_trained","trained"]),createdAt:c,contentHash:o}).strict().superRefine((t,i)=>{t.kind==="base_reference"&&(t.version!==0||t.adapterStatus!=="not_trained"||t.artifactLineageId!==null)&&i.addIssue({code:"custom",message:"A base-reference Model Version must be version 0 without adapter lineage."}),t.kind==="lora_adapter"&&(t.version===0||t.adapterStatus!=="trained"||!t.artifactLineageId)&&i.addIssue({code:"custom",message:"A LoRA Model Version must be version 1+ with adapter lineage."})}),nt=e.object({schemaVersion:e.literal("openpond.rewardModelVersion.v1"),id:n,modelId:n,profileId:n,version:e.number().int().positive(),role:e.literal("reward"),status:e.enum(["available","failed"]),scope:e.enum(["synthetic_smoke","human_preference"]),baseModel:B,runtime:ue.nullable().default(null),taskset:g,preferenceDatasetRelease:r,releaseGraph:e.object({resolvedBundleHash:o,profileRelease:g,harnessRelease:r,grader:e.object({id:n,contentHash:o}).strict()}).strict(),artifacts:e.object({checkpoint:r.extend({objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),files:e.array(e.object({path:e.string().trim().min(1).max(512),sizeBytes:e.number().int().positive(),sha256:o}).strict()).min(4).max(128)}).strict(),adapter:r,scalarHead:r,bucketHead:r.nullable(),processorRelease:r}).strict(),qualificationReport:r.nullable(),createdAt:c,contentHash:o}).strict(),en=e.object({schemaVersion:e.literal("openpond.rewardModelRunReceipt.v1"),provider:n,providerRunId:n,resolvedBundleHash:o,finalCheckpoint:r.extend({objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/)}).strict(),adapter:r,scalarHead:r,bucketHead:r.nullable(),processorRelease:r,optimizerEvidence:r,managedExecutionReceipt:r.nullable().default(null),parameterDeltaHash:o,cleanup:e.object({computeReleased:e.boolean(),providerTerminalObserved:e.boolean()}).strict(),contentHash:o}).strict(),at=e.preprocess(t=>{if(!t||typeof t!="object"||Array.isArray(t))return t;let i=t;if(i.tasksetRelease!==void 0)return t;let s=i.taskset;if(!s||typeof s!="object"||Array.isArray(s))return t;let l=s;return typeof l.id!="string"||typeof l.contentHash!="string"?t:{...i,tasksetRelease:{id:l.id,contentHash:l.contentHash}}},e.object({schemaVersion:e.literal("openpond.rewardModelRun.v1"),id:n,rewardModelId:n,rewardModelVersionId:n.nullable(),profileId:n,role:e.literal("reward"),scope:e.enum(["synthetic_smoke","human_preference"]),status:e.enum(["prepared","running","succeeded","failed","cancelled"]),taskset:g,tasksetRelease:r,preferenceDatasetRelease:r,recipeRelease:r,destinationId:n,quote:e.object({maximumSpendUsd:e.number().positive(),hourlyCostUsd:e.number().nonnegative().nullable()}).strict(),managedRunId:n.nullable(),progress:e.object({completedSteps:e.number().int().nonnegative(),totalSteps:e.number().int().positive(),latestLoss:e.number().finite().nullable()}).strict(),receipt:en.nullable(),qualificationReport:r.nullable(),accruedSpendUsd:e.number().nonnegative().nullable(),failureOwner:e.enum(["authoring","admission","provider","runner","artifact","qualification","cleanup"]).nullable(),failure:e.string().trim().min(1).max(5e3).nullable(),startedAt:c,completedAt:c.nullable(),updatedAt:c}).strict()).superRefine((t,i)=>{t.status==="succeeded"&&!t.receipt&&i.addIssue({code:"custom",message:"A successful Reward Model Run requires a canonical receipt."}),t.status!=="succeeded"&&t.rewardModelVersionId!==null&&i.addIssue({code:"custom",message:"Only a successful Reward Model Run can publish a Reward Model Version."})}),tn=e.object({schemaVersion:e.literal("openpond.modelRunReceipt.v1"),provider:n,providerRunId:n,assignmentHash:o,resultHash:o,transcriptHash:o,traceHash:o.nullable(),resolvedBundleHash:o,artifactPath:e.string().trim().min(1).max(2e3),cleanup:e.object({computeReleased:e.boolean(),tunnelClosed:e.boolean()}).strict(),telemetry:Qe.nullable().default(null),contentHash:o}).strict(),P=e.object({inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),totalTokens:e.number().int().nonnegative(),costUsd:e.number().nonnegative().nullable()}).strict(),Ze=e.string().regex(/^(?:[a-f0-9]{40}|[a-f0-9]{64})$/),pe=e.object({phase:e.enum(["baseline","adaptation","candidate_adaptation","candidate"]),taskId:n,attemptId:n,sessionId:n.nullable(),turnId:n.nullable(),passed:e.boolean(),score:e.number().min(0).max(1).nullable(),failureClass:ie.nullable(),inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),totalTokens:e.number().int().nonnegative(),latencyMs:e.number().int().nonnegative(),costUsd:e.number().nonnegative().nullable(),startedAt:c}).strict(),Xe=e.object({targetTaskCount:e.number().int().nonnegative(),comparedTaskCount:e.number().int().nonnegative(),passedTaskCount:e.number().int().nonnegative(),failedTaskCount:e.number().int().nonnegative(),lowerTaskCount:e.number().int().nonnegative(),higherTaskCount:e.number().int().nonnegative(),unchangedTaskCount:e.number().int().nonnegative()}).strict(),it=e.object({target:e.literal("all_tasks_lower"),targetTaskCount:e.number().int().nonnegative(),comparedTaskCount:e.number().int().nonnegative(),passedTaskCount:e.number().int().nonnegative(),failedTaskCount:e.number().int().nonnegative(),lowerTaskCount:e.number().int().nonnegative(),higherTaskCount:e.number().int().nonnegative(),unchangedTaskCount:e.number().int().nonnegative(),baselineTokens:e.number().int().nonnegative(),refinedTokens:e.number().int().nonnegative(),tokenDelta:e.number().int(),tokenDeltaPercent:e.number().finite().nullable(),complete:e.boolean(),passed:e.boolean(),cohorts:e.object({adaptation:Xe,heldOut:Xe}).strict()}).strict(),be=e.object({baseline:P,adaptation:P,candidateAdaptation:P,candidate:P,refiner:P,grader:P}).strict(),nn=e.object({usage:be,observedSpendUsd:e.number().nonnegative(),attempts:e.array(pe).max(1e4)}).strict(),an=e.object({id:n,contentHash:o,artifactPath:e.string().trim().min(1).max(2e3)}).strict(),rn=e.object({schemaVersion:e.literal("openpond.modelEvaluationReceipt.v1"),benchmarkId:n,resultManifest:e.object({id:n,contentHash:o,artifactPath:e.string().trim().min(1).max(2e3)}).strict(),stages:e.object({baseline:r,adaptation:r,refiner:r,candidateAdaptation:r,candidate:r,comparison:r}).strict(),usage:be,quality:e.object({baselinePassRate:e.number().min(0).max(1),candidatePassRate:e.number().min(0).max(1),adaptationBaselinePassRate:e.number().min(0).max(1),adaptationCandidatePassRate:e.number().min(0).max(1),adaptationCandidatePassed:e.boolean(),heldOutCandidatePassed:e.boolean(),passed:e.boolean()}).strict(),foregroundTokenDelta:e.number().int(),foregroundTokenDeltaPercent:e.number().finite().nullable(),taskEfficiency:it.optional(),efficiency:e.object({grossForegroundTokenSavings:e.number().int(),overheadTokens:e.number().int().nonnegative(),firstPassNetTokenSavings:e.number().int(),breakEvenReuseCount:e.number().int().nonnegative().nullable(),amortizedTokenSavings:e.number().int(),amortizedReuseCount:e.number().int().positive()}).strict(),budget:e.object({maximumSpendUsd:e.number().nonnegative(),observedSpendUsd:e.number().nonnegative(),enforced:e.boolean()}).strict(),evidenceSnapshot:r,lineage:e.object({adaptationEvidenceHash:o,refinerInputHash:o,refinerOutcomeHash:o,validationHash:o,applyReceiptHash:o,candidateRelease:r,valid:e.boolean()}).strict(),invalidReasons:e.array(e.string().trim().min(1).max(2e3)).max(100),attempts:e.array(pe).max(1e4).default([]),terminalClassification:e.enum(["improved","no_improvement","regressed","inconclusive","cancelled","infrastructure_failure"]),profileGit:e.object({ref:e.string().trim().min(1).max(2e3),commit:Ze,baseCommit:Ze}).strict().nullable(),contentHash:o}).strict(),on=e.object({schemaVersion:e.literal("openpond.modelEvaluationStopReceipt.v1"),benchmarkId:n,terminalClassification:e.literal("inconclusive"),stopReason:e.enum(["candidate_harness_unchanged","candidate_lineage_invalid"]),reason:e.string().trim().min(1).max(5e3),stoppedAfter:e.literal("refiner"),baselineHarness:r,candidateHarness:r,refiner:e.object({id:n,contentHash:o,outcomeCount:e.number().int().nonnegative()}).strict(),usage:be,budget:e.object({maximumSpendUsd:e.number().nonnegative(),observedSpendUsd:e.number().nonnegative(),enforced:e.boolean()}).strict(),evidenceSnapshot:r,attempts:e.array(pe).max(1e4),contentHash:o}).strict(),et=e.object({level:e.literal(.95),lower:e.number().finite(),upper:e.number().finite()}).strict().superRefine((t,i)=>{t.lower>t.upper&&i.addIssue({code:"custom",message:"A confidence interval's lower bound cannot exceed its upper bound."})}),sn=e.object({schemaVersion:e.literal("openpond.modelComparisonBenchmarkReceipt.v1"),benchmarkId:e.literal("model-comparison"),target:e.object({kind:e.enum(["base_model","model_version","external_reference"]),label:e.string().trim().min(1).max(300),modelVersionId:n.nullable(),model:V.nullable()}).strict(),taskset:g,grader:r,sampling:e.object({seeds:e.array(e.number().int()).min(1).max(100),repetitions:e.number().int().positive().max(20)}).strict(),deterministic:e.object({attemptedTaskCount:e.number().int().nonnegative(),completedTaskCount:e.number().int().nonnegative(),passedTaskCount:e.number().int().nonnegative(),failedTaskCount:e.number().int().nonnegative(),meanScore:e.number().finite().nullable(),passRate:e.number().min(0).max(1).nullable(),passRateCi95:et.nullable()}).strict(),judge:e.object({judgeModel:V,judgeRelease:r,rubricRelease:r,calibrationRelease:r,score:e.number().min(0).max(100),scoreCi95:et,comparisonCount:e.number().int().positive(),wins:e.number().int().nonnegative(),ties:e.number().int().nonnegative(),losses:e.number().int().nonnegative(),referenceLabel:e.string().trim().min(1).max(300)}).strict().superRefine((t,i)=>{t.wins+t.ties+t.losses!==t.comparisonCount&&i.addIssue({code:"custom",message:"Judge win, tie, and loss counts must equal comparisonCount."})}).nullable(),attempts:e.array(e.object({attemptId:n.nullable().default(null),taskId:n,seed:e.number().int(),repetition:e.number().int().nonnegative(),status:e.enum(["succeeded","failed"]),deterministicScore:e.number().finite().nullable(),passed:e.boolean().nullable(),judgeScore:e.number().min(0).max(100).nullable(),judgePreference:e.enum(["win","tie","loss"]).nullable(),transcriptHash:o.nullable(),traceHash:o.nullable(),transcriptArtifact:e.object({artifactPath:e.string().trim().min(1).max(2e3),jsonPointer:e.string().trim().min(1).max(1e3)}).strict().nullable().default(null),traceArtifact:e.object({artifactPath:e.string().trim().min(1).max(2e3),jsonPointer:e.string().trim().min(1).max(1e3)}).strict().nullable().default(null),latencyMs:e.number().int().nonnegative().nullable().default(null),failureClass:ie.nullable()}).strict()).max(1e5),usage:e.object({policy:P.nullable(),judge:P.nullable(),observedSpendUsd:e.number().nonnegative().nullable(),evaluationGpuSeconds:e.number().nonnegative().nullable().default(null)}).strict(),managedServing:e.object({jobId:n,terminalState:e.enum(["completed","cancelled","failed","budget_exhausted"]),sourcePolicyVersion:e.number().int().nonnegative(),sourceAdapterSha256:o.nullable(),servedPolicyVersion:e.number().int().nonnegative().nullable(),servedAdapterSha256:o.nullable(),accruedSpendUsd:e.number().nonnegative(),cleanupAttestationHash:o,resourceCount:e.number().int().nonnegative(),activeResourceCount:e.literal(0)}).strict().nullable().default(null),evidenceSnapshot:e.object({id:n,contentHash:o,artifactPath:e.string().trim().min(1).max(2e3)}).strict(),completedAt:c,contentHash:o}).strict().superRefine((t,i)=>{let s=t.deterministic;s.completedTaskCount>s.attemptedTaskCount&&i.addIssue({code:"custom",message:"Completed task count cannot exceed attempted task count."}),s.passedTaskCount+s.failedTaskCount!==s.completedTaskCount&&i.addIssue({code:"custom",message:"Passed and failed task counts must equal completed task count."}),s.completedTaskCount===0!=(s.meanScore===null)&&i.addIssue({code:"custom",message:"Deterministic mean score is available exactly when tasks completed."}),s.completedTaskCount===0!=(s.passRate===null)&&i.addIssue({code:"custom",message:"Deterministic pass rate is available exactly when tasks completed."})}),ln=e.object({advancedEvaluation:re.optional(),benchmarkId:e.literal("harness-refiner"),model:V,upstreamModel:e.object({providerId:n,modelId:e.string().trim().min(1).max(300),revision:e.string().trim().min(1).max(300),pricing:e.object({version:e.string().trim().min(1).max(300),source:e.string().trim().min(1).max(300),effectiveAt:c,inputUsdPerMillionTokens:e.number().nonnegative(),cachedInputUsdPerMillionTokens:e.number().nonnegative(),outputUsdPerMillionTokens:e.number().nonnegative()}).strict().optional()}).strict(),reasoningEffort:e.string().trim().min(1).max(100).nullable(),seeds:e.array(e.number().int()).min(1).max(100),repetitions:e.number().int().positive().max(20),maximumSpendUsd:e.number().nonnegative(),attemptPlan:e.array(e.object({stage:e.enum(["baseline","adaptation","candidate_adaptation","candidate"]),split:e.string().trim().min(1).max(100),taskIds:e.array(n).min(1).max(1e5),attemptCount:e.number().int().positive()}).strict()).length(4)}).strict(),mn=e.object({benchmarkId:e.literal("model-comparison"),target:e.object({kind:e.enum(["base_model","model_version","external_reference"]),label:e.string().trim().min(1).max(300),modelVersionId:n.nullable(),model:V.nullable()}).strict(),grader:r,judge:e.object({model:V,release:r,rubricRelease:r,calibrationRelease:r,referenceLabel:e.string().trim().min(1).max(300)}).strict().nullable(),seeds:e.array(e.number().int()).min(1).max(100),repetitions:e.number().int().positive().max(20),maximumSpendUsd:e.number().nonnegative(),series:e.object({id:n,protocol:e.object({id:n,revision:e.number().int().positive(),contentHash:o}).strict()}).strict().nullable().default(null),panel:e.object({id:n,role:e.enum(["correction","sibling_verification","cumulative_known","development","retained","frozen_final"]),passLabel:n.nullable()}).strict().nullable().default(null),comparisonPair:e.object({entryId:n,parent:de,candidateModelVersionId:n}).strict().nullable().default(null),attemptPlan:e.array(e.object({stage:e.literal("comparison"),split:e.string().trim().min(1).max(100),taskIds:e.array(n).min(1).max(1e5),attemptCount:e.number().int().positive()}).strict()).length(1)}).strict(),cn=e.union([ln,mn]),rt=e.object({schemaVersion:e.literal("openpond.modelRun.v1"),id:n,modelId:n,modelVersionId:n.nullable(),profileId:n,kind:e.enum(["rollout_smoke","training","evaluation"]),status:e.enum(["prepared","running","succeeded","failed","cancelled"]),method:e.enum(["sft","dpo","grpo","ppo","sdft","opd","opsd","sdpo"]).nullable(),destinationId:n.nullable(),taskset:g,comparisonSeriesEntry:A.nullable().optional(),evaluationTasksetRef:g.nullable().optional(),harnessRelease:r.nullable().optional(),quote:e.object({maximumSpendUsd:e.number().nonnegative(),hourlyCostUsd:e.number().nonnegative().nullable()}).strict().nullable(),evaluation:cn.nullable().default(null),evaluationProgress:e.object({stage:e.enum(["baseline","adaptation","refiner","candidate_adaptation","candidate","comparison"]),completedAttempts:e.number().int().nonnegative(),totalAttempts:e.number().int().positive(),accounting:nn.nullable().default(null),advancedAccounting:se.optional(),evidenceSnapshot:an.nullable().default(null)}).strict().nullable().default(null),reward:e.object({raw:e.number(),components:e.record(e.string(),e.number())}).strict().nullable(),receipt:e.union([tn,rn,on,sn,oe]).nullable(),adapterArtifactLineageId:n.nullable(),failure:e.string().trim().min(1).max(5e3).nullable(),startedAt:c,completedAt:c.nullable(),updatedAt:c}).strict().superRefine((t,i)=>{t.status==="succeeded"&&!t.receipt&&i.addIssue({code:"custom",message:"A successful Model Run requires a canonical receipt."}),t.kind==="evaluation"&&(t.method!==null||t.destinationId!==null||!t.evaluation||t.adapterArtifactLineageId!==null)&&i.addIssue({code:"custom",message:"An evaluation Model Run requires evaluation configuration and cannot claim training method, destination, or adapter lineage."}),t.kind==="evaluation"&&t.receipt&&t.receipt.schemaVersion!=="openpond.modelEvaluationReceipt.v1"&&t.receipt.schemaVersion!=="openpond.modelEvaluationStopReceipt.v1"&&t.receipt.schemaVersion!=="openpond.modelComparisonBenchmarkReceipt.v1"&&t.receipt.schemaVersion!=="openpond.advancedRefinerReviewReceipt.v1"&&i.addIssue({code:"custom",message:"An evaluation Model Run requires an evaluation receipt."}),t.kind!=="evaluation"&&(t.method===null||t.destinationId===null||t.quote===null)&&i.addIssue({code:"custom",message:"A training or rollout Model Run requires method, destination, and quote."}),t.kind==="rollout_smoke"&&t.adapterArtifactLineageId!==null&&i.addIssue({code:"custom",message:"A rollout smoke cannot claim adapter artifacts."})});function ai(t){let i=new Map;for(let p of t.attempts){let T=`${p.phase}\0${p.taskId}`,S=i.get(T);(!S||p.startedAt>S.startedAt||p.startedAt===S.startedAt&&p.attemptId>S.attemptId)&&i.set(T,p)}let s=[...i.values()],l=[["adaptation","adaptation","candidate_adaptation"],["held_out","baseline","candidate"]].flatMap(([p,T,S])=>{let At=new Map(s.filter(R=>R.phase===S).map(R=>[R.taskId,R]));return s.filter(R=>R.phase===T).flatMap(R=>{let ee=At.get(R.taskId);return ee?[{cohort:p,taskId:R.taskId,baseline:R,refined:ee,tokenDelta:ee.totalTokens-R.totalTokens}]:[]})}),f=new Set(s.filter(p=>p.phase==="baseline"||p.phase==="adaptation").map(p=>p.taskId)).size,h=t.targetTaskCount??f;if(!Number.isInteger(h)||h<0)throw new Error("Benchmark task-efficiency target must be a non-negative integer.");let x={adaptation:new Set(s.filter(p=>p.phase==="adaptation").map(p=>p.taskId)).size,held_out:new Set(s.filter(p=>p.phase==="baseline").map(p=>p.taskId)).size},k=v((p,T)=>({targetTaskCount:T,comparedTaskCount:p.length,passedTaskCount:p.filter(S=>S.tokenDelta<0).length,failedTaskCount:p.filter(S=>S.tokenDelta>=0).length,lowerTaskCount:p.filter(S=>S.tokenDelta<0).length,higherTaskCount:p.filter(S=>S.tokenDelta>0).length,unchangedTaskCount:p.filter(S=>S.tokenDelta===0).length}),"summarizePairs"),d=k(l,h),j=l.reduce((p,T)=>p+T.baseline.totalTokens,0),ke=l.reduce((p,T)=>p+T.refined.totalTokens,0),Re=ke-j,Ie=l.length===h;return{summary:it.parse({target:"all_tasks_lower",...d,baselineTokens:j,refinedTokens:ke,tokenDelta:Re,tokenDeltaPercent:j>0?Re/j*100:null,complete:Ie,passed:Ie&&d.passedTaskCount===h,cohorts:{adaptation:k(l.filter(p=>p.cohort==="adaptation"),x.adaptation),heldOut:k(l.filter(p=>p.cohort==="held_out"),x.held_out)}}),pairs:l}}v(ai,"summarizeModelEvaluationTaskEfficiency");var dn=e.enum(["measuring","up_to_date","needs_attention"]),un=e.enum(["fixed","retained","regressed","unresolved","unavailable"]),ot=e.object({evaluationRunId:n,attemptKey:n,artifactPath:e.string().trim().min(1).max(2e3),jsonPointer:e.string().trim().min(1).max(1e3),transcriptHash:o.nullable(),traceHash:o.nullable()}).strict(),ge=e.object({level:e.literal(.95),lower:e.number().finite(),upper:e.number().finite()}).strict(),pn=e.object({panelId:n,panelRole:le,passLabel:n.nullable(),taskset:g,attempted:e.number().int().nonnegative(),available:e.number().int().nonnegative(),passed:e.number().int().nonnegative(),strictSuccess:e.number().min(0).max(1).nullable(),strictSuccessCi95:ge.nullable(),judgeScore:e.number().min(0).max(100).nullable(),judgeScoreCi95:ge.nullable()}).strict(),st=e.object({schemaVersion:e.literal("openpond.modelCurrencySnapshot.v1"),id:n,contentHash:o,seriesId:n,protocol:e.object({id:n,revision:e.number().int().positive(),contentHash:o}).strict(),entryId:n,passLabel:n,parent:e.discriminatedUnion("kind",[e.object({kind:e.literal("base_model"),id:n,revision:e.string().trim().min(1).max(500)}).strict(),e.object({kind:e.literal("model_version"),id:n,contentHash:o}).strict()]),candidate:e.object({kind:e.literal("model_version"),id:n,contentHash:o}).strict(),sources:e.object({evaluationRunIds:e.array(n).max(1e5),attemptIds:e.array(n).max(1e6),tasksets:e.array(g).max(1e4),grader:r,judge:r.nullable(),calibration:r.nullable()}).strict(),matches:e.array(e.object({key:n,panelId:n,panelRole:le,taskId:n,issueFamilyId:n,seed:e.number().int(),repetition:e.number().int().nonnegative(),classification:un,parentPassed:e.boolean().nullable(),candidatePassed:e.boolean().nullable(),parentAttempt:ot.nullable(),candidateAttempt:ot.nullable()}).strict()).max(1e6),taskIds:e.object({fixed:e.array(n).max(1e6),retained:e.array(n).max(1e6),regressed:e.array(n).max(1e6),unresolved:e.array(n).max(1e6),unavailable:e.array(n).max(1e6)}).strict(),panels:e.array(pn).max(1e4),metrics:e.object({knownIssueCoverage:e.number().min(0).max(1).nullable(),issueFamilyGeneralization:e.number().min(0).max(1).nullable(),behavioralRetention:e.number().min(0).max(1).nullable(),currentAcquisitionDelta:e.number().finite().nullable(),retainedDeltaPoints:e.number().finite().nullable(),frontierStrictDelta:e.number().finite().nullable(),pairedPValue:e.number().min(0).max(1).nullable(),wins:e.number().int().nonnegative(),ties:e.number().int().nonnegative(),losses:e.number().int().nonnegative()}).strict(),statistics:e.object({matchedAttemptCount:e.number().int().nonnegative(),strictDelta:e.number().finite().nullable(),strictDeltaCi95:ge.nullable(),exactPairedBinaryPValue:e.number().min(0).max(1).nullable(),pairedBootstrapSamples:e.number().int().positive()}).strict(),criteria:e.object({allRequiredAttemptsTerminal:e.boolean(),criticalCorrectionPassRate:e.number().min(0).max(1).nullable(),siblingPassRate:e.number().min(0).max(1).nullable(),behavioralRetentionRate:e.number().min(0).max(1).nullable(),retainedRegressionPoints:e.number().finite().nullable(),criticalPriorRegressionCount:e.number().int().nonnegative(),thresholdRelease:r}).strict(),efficiency:e.object({inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),totalTokens:e.number().int().nonnegative(),costUsd:e.number().nonnegative().nullable(),latencyMs:e.number().int().nonnegative(),tokensPerSuccess:e.number().nonnegative().nullable(),throughputPerHour:e.number().nonnegative().nullable(),gpuSeconds:e.number().nonnegative().nullable()}).strict(),invariants:e.object({systemPromptHash:o,toolSchema:r,application:r,harness:r,runtime:r,grader:r,autoRefinerEnabled:e.boolean()}).strict(),evidenceState:dn,evidenceReasons:e.array(n).max(100),projectedAt:c}).strict();var $=e.discriminatedUnion("kind",[e.object({kind:e.literal("base_model"),id:n,revision:e.string().trim().min(1).max(500),label:e.string().trim().min(1).max(300)}).strict(),e.object({kind:e.literal("model_version"),id:n,label:e.string().trim().min(1).max(300)}).strict(),e.object({kind:e.literal("captured_model"),id:n,label:e.string().trim().min(1).max(300)}).strict()]),U=e.union([e.object({source:e.literal("evaluation_run").default("evaluation_run"),evaluationRunId:n,attemptId:n,modelLabel:e.string().trim().min(1).max(300).nullable().default(null),modelVersionId:n.nullable().default(null),reward:e.number().finite().nullable(),components:e.record(e.string(),e.number().finite()).default({})}).strict(),e.object({source:e.literal("imported_response"),modelLabel:e.string().trim().min(1).max(300),modelVersionId:n.nullable().default(null),response:e.unknown().refine(t=>t!==void 0,"An imported model response is required."),reward:e.number().finite().nullable().default(null),components:e.record(e.string(),e.number().finite()).default({})}).strict()]),bn=e.union([e.object({taskId:n,target:$,attempt:U}).strict(),e.object({taskId:n,target:e.enum(["base","current"]).default("current"),attempt:U}).strict(),e.object({taskId:n,target:e.enum(["base","current"]).default("current"),evaluationRunId:n,attemptId:n,reward:e.number().finite().nullable(),components:e.record(e.string(),e.number().finite()).default({})}).strict().transform(({taskId:t,target:i,...s})=>({taskId:t,target:i,attempt:{source:"evaluation_run",modelLabel:null,modelVersionId:null,...s}}))]),bi=e.object({schemaVersion:e.literal("openpond.continualLearningDailyBatchManifest.v1"),id:n,seriesId:n,scheduleEntryId:n,dayOrdinal:e.number().int().positive().max(1e4),sourceTaskset:g,taskIds:e.array(n).min(1).max(1e3),observedAttempts:e.array(bn).max(1e3).default([]),sourceFileName:e.string().trim().min(1).max(500).nullable().default(null),availableAt:c}).strict().superRefine((t,i)=>{new Set(t.taskIds).size!==t.taskIds.length&&i.addIssue({code:"custom",path:["taskIds"],message:"A daily batch cannot repeat a task."}),new Set(t.observedAttempts.map(l=>`${l.taskId}:${gn(l.target)}`)).size!==t.observedAttempts.length&&i.addIssue({code:"custom",path:["observedAttempts"],message:"An intake batch cannot repeat a task and Policy target pair."});let s=new Set(t.taskIds);t.observedAttempts.some(l=>!s.has(l.taskId))&&i.addIssue({code:"custom",path:["observedAttempts"],message:"Observed attempts must belong to tasks in the daily batch."})}),lt=e.object({schemaVersion:e.literal("openpond.continualLearningDailyBatch.v1"),id:n,seriesId:n,scheduleEntryId:n,dayOrdinal:e.number().int().positive().max(1e4),label:e.string().trim().min(1).max(100),runMode:e.enum(["automatic","manual"]).optional(),source:e.enum(["json_upload","sealed_fixture"]),sourceFileName:e.string().trim().min(1).max(500).nullable(),sourceTaskset:g,tasks:e.array(e.object({taskId:n,taskContentHash:o,familyKey:n,disposition:e.enum(["include","defer","exclude"]).nullable(),oracleReview:e.enum(["pending","confirmed","needs_correction"]),note:e.string().trim().max(2e3),observedAttempt:U.nullable(),baselineAttempt:U.nullable().default(null),responses:e.array(e.object({target:$,attempt:U}).strict()).max(100).default([]),stagedAt:c.nullable().default(null),queuedEntry:A.nullable().default(null)}).strict()).min(1).max(1e3),intakeEvaluation:e.object({status:e.enum(["awaiting","running","ready","failed"]),requestedTargets:e.array(e.enum(["base","current"])).max(2).default(["base","current"]),requestedPolicies:e.array($).max(100).default([]),runs:e.array(e.object({target:$,runId:n}).strict()).max(100).default([]),baselineRunId:n.nullable(),currentRunId:n.nullable(),currentPolicy:e.discriminatedUnion("kind",[e.object({kind:e.literal("base_model"),id:n}).strict(),e.object({kind:e.literal("model_version"),id:n}).strict()]).nullable(),failure:e.string().trim().min(1).max(2e3).nullable()}).strict().default({status:"awaiting",requestedTargets:["base","current"],requestedPolicies:[],runs:[],baselineRunId:null,currentRunId:null,currentPolicy:null,failure:null}),status:e.enum(["pending","reviewed","queued"]),queuedEntry:A.nullable(),reviewedBy:n.nullable(),reviewedAt:c.nullable(),availableAt:c,revision:e.number().int().positive(),createdAt:c,updatedAt:c}).strict().superRefine((t,i)=>{new Set(t.tasks.map(l=>l.taskId)).size!==t.tasks.length&&i.addIssue({code:"custom",path:["tasks"],message:"A daily batch cannot repeat a task."});for(let[l,f]of t.tasks.entries()){let h=f.responses.map(x=>ct(x.target));new Set(h).size!==h.length&&i.addIssue({code:"custom",path:["tasks",l,"responses"],message:"A task cannot repeat a model response target."})}let s=t.tasks.every(l=>l.disposition!==null&&l.oracleReview!=="pending"&&(l.oracleReview!=="needs_correction"||l.note.length>0));t.status!=="pending"&&!s&&i.addIssue({code:"custom",path:["tasks"],message:"Every task requires a decision and oracle review before the day completes."}),t.status==="queued"!=(t.queuedEntry!==null)&&i.addIssue({code:"custom",path:["queuedEntry"],message:"Only a queued day carries an exact series-entry reference."}),t.status==="pending"&&(t.reviewedAt!==null||t.reviewedBy!==null)&&i.addIssue({code:"custom",path:["reviewedAt"],message:"A pending day cannot carry reviewer completion identity."}),t.status!=="pending"&&(t.reviewedAt===null||t.reviewedBy===null)&&i.addIssue({code:"custom",path:["reviewedAt"],message:"Reviewed identity is required after a day review completes."})}),mt=e.object({schemaVersion:e.literal("openpond.continualBenchIssueReview.v1"),id:n,seriesId:n,scheduleEntryId:n,passLabel:n,packet:De,splitManifest:e.object({id:n,contentHash:o}).strict(),decisions:e.array(e.object({taskId:n,disposition:e.enum(["include","defer","exclude"]).nullable(),note:e.string().trim().max(2e3)}).strict()).min(2).max(1e4),status:e.enum(["pending","reviewed","queued"]),queuedEntry:A.nullable(),reviewedBy:n.nullable(),reviewedAt:c.nullable(),revision:e.number().int().positive(),createdAt:c,updatedAt:c}).strict().superRefine((t,i)=>{let s=[...t.packet.cases.map(h=>h.taskId)].sort(),l=[...t.decisions.map(h=>h.taskId)].sort();JSON.stringify(s)!==JSON.stringify(l)&&i.addIssue({code:"custom",path:["decisions"],message:"Review decisions must cover the exact issue-packet cases."});let f=t.decisions.every(h=>h.disposition!==null&&h.note.length>0);t.status!=="pending"&&!f&&i.addIssue({code:"custom",path:["decisions"],message:"Every case requires a disposition and note before review completes."}),t.status==="queued"!=(t.queuedEntry!==null)&&i.addIssue({code:"custom",path:["queuedEntry"],message:"Only a queued review carries an exact series-entry reference."})}).readonly();function ct(t){return`${t.kind}:${t.id}`}v(ct,"responseTargetKey");function gn(t){return typeof t=="string"?t:ct(t)}v(gn,"observedTargetKey");var I=e.string().trim().min(1).max(240),_=e.string().trim().min(8).max(256),F=e.enum(["grpo","dapo","gspo-token"]),H=e.object({id:I,revision:e.string().trim().min(1).max(256),tokenizerRevision:e.string().trim().min(1).max(256),chatTemplateHash:_}),dt=e.object({maxRollouts:e.number().int().positive().max(1e6),maxEnvironmentExecutions:e.number().int().positive().max(1e6),maxInputTokens:e.number().int().positive(),maxOutputTokens:e.number().int().positive(),maxOptimizerSteps:e.number().int().positive().max(1e6),wallTimeMs:e.number().int().positive(),maximumCostUsd:e.number().nonnegative().nullable()}),ut=e.object({method:e.literal("grpo"),groupSize:e.number().int().min(2).max(128),normalization:e.literal("group_standardized"),advantageEpsilon:e.number().positive().max(.01).default(1e-8),loss:F,clipRange:e.number().positive().max(1).default(.2),iterations:e.number().int().min(2).max(16).default(2),microbatchSize:e.number().int().positive().max(128).default(1),gradientAccumulationSteps:e.number().int().positive().max(128).default(1),adamw:D}),G=e.object({method:e.literal("ppo"),valueModel:H,gamma:e.number().min(0).max(1),gaeLambda:e.number().min(0).max(1),policyClip:e.number().positive().max(1),valueClip:e.number().positive().max(10),valueLossCoefficient:e.number().nonnegative().max(10),ppoEpochs:e.number().int().positive().max(100),minibatchSize:e.number().int().positive().max(1e5)}),pt=e.discriminatedUnion("method",[ut,G]),J=e.object({rewardModelVersion:e.object({id:I,contentHash:_}).strict(),qualificationReport:e.object({id:I,contentHash:_}).strict().nullable().default(null),evaluationReferences:e.array(e.object({id:I,contentHash:_}).strict()).max(1e3).default([]),checkpoint:e.object({id:I,contentHash:_,objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),files:e.array(e.object({path:e.string().trim().min(1).max(512),sizeBytes:e.number().int().positive(),sha256:_}).strict()).min(4).max(128)}).strict(),runtime:ue,processorRelease:e.object({id:I,contentHash:_}).strict(),rewardComposerRelease:e.object({id:I,contentHash:_}).strict(),executionReceipt:e.object({id:I,contentHash:_}).strict()}).strict(),W=e.object({schemaVersion:e.literal("openpond.policyOptimization.v1"),policyModel:H,referenceModel:H,dataset:e.object({tasksetId:I,tasksetHash:_,split:e.literal("train"),selectionStrategy:e.enum(["stable_hash_top_n","rft_easy_curriculum_v1"]),selectionSeed:e.number().int(),maxExamples:e.number().int().positive().max(1e5)}),sampler:e.object({temperature:e.number().min(0).max(2),topP:e.number().min(0).max(1),maxOutputTokens:e.number().int().positive().max(32768),maxTurns:e.number().int().positive().max(1e3),concurrency:e.number().int().positive().max(1e3)}),environment:e.object({id:I,version:e.string().trim().min(1).max(256),toolContractHash:_}),reward:e.object({graderId:I,graderHash:_,learnedPreference:J.nullable().default(null)}),kl:e.object({coefficient:e.number().nonnegative().nullable(),referenceConstraint:e.literal("fixed_reference")}),budgets:dt,checkpointEverySteps:e.number().int().positive(),seed:e.number().int(),evaluationSplit:e.literal("frozen_eval"),optimizer:pt});var a=e.string().trim().min(1).max(240),b=e.string().trim().min(1),m=e.string().trim().min(8).max(256),O=e.record(e.string(),e.unknown()).default({});var he=e.object({schemaVersion:e.literal("openpond.localModelChatConfiguration.v1").default("openpond.localModelChatConfiguration.v1"),profile:e.enum(["efficient","full_harness","custom"]).default("efficient"),systemPromptMode:e.enum(["lean","full_harness","custom"]).default("lean"),customSystemPrompt:e.string().max(2e4).nullable().default(null),contextWindowTokens:e.number().int().min(128).max(32768).default(1024),maxOutputTokens:e.number().int().min(1).max(512).default(64),temperature:e.number().min(0).max(2).default(0),repetitionPenalty:e.number().min(.5).max(2).default(1.1),noRepeatNgramSize:e.number().int().min(0).max(10).default(3),compaction:e.enum(["off","when_needed"]).default("when_needed"),keepWarmSeconds:e.number().int().min(0).max(3600).default(300),updatedAt:b.nullable().default(null)}),bt=he.parse({}),gt=e.object({schemaVersion:e.literal("openpond.managedAdapterServingProjection.v1"),teamId:a.nullable().default(null),source:e.literal("sandbox_managed_rl"),sourceRef:a,canonicalArtifactId:a.nullable(),canonicalModelName:e.string().trim().min(1).max(256).nullable().default(null),canonicalArtifactState:e.enum(["imported_unvalidated","evaluating","promotable","rejected","deleted"]).nullable(),canonicalDeploymentId:a.nullable(),canonicalDeploymentState:e.enum(["requested","deploying","ready","degraded","deleting","deleted","failed"]).nullable(),state:e.enum(["pending","imported","ready","failed"]),customerBindingAllowed:e.boolean().default(!1),artifactContentHash:m.nullable().default(null),baseProfileId:a.nullable().default(null),publishedAt:b.nullable(),lastSyncedAt:b,lastError:e.string().trim().min(1).max(5e3).nullable()});function hn(t){return t.state==="ready"&&t.canonicalArtifactState==="promotable"&&t.canonicalDeploymentState==="ready"&&t.customerBindingAllowed&&!!t.teamId&&!!t.canonicalArtifactId&&!!t.canonicalDeploymentId}v(hn,"managedAdapterProjectionReady");function ji(t){let i=t.managedServing;return i?hn(i)&&i.canonicalArtifactId&&i.canonicalDeploymentId?{kind:"sandbox_serving",canonicalArtifactId:i.canonicalArtifactId,canonicalDeploymentId:i.canonicalDeploymentId}:null:{kind:"source_artifact",canonicalArtifactId:null,canonicalDeploymentId:null}}v(ji,"resolveModelBindingEligibility");var ye="openpond.crossSystemOperations.v1",wi="1.0.0",ft="openpond.crossSystemTools.v1",fn=["search_crm","query_billing","search_support","run_python"],vn=e.enum(["train","validation","frozen_eval"]),xn=e.enum(["easy","medium","hard"]),Sn=e.enum(["renewal_risk_v2"]),Mi=e.enum(["renewal_exposure","collections_prioritization","invoice_reconciliation","sla_escalation","contract_billing_mismatch"]),Pi=e.object({seed:e.number().int().nonnegative(),split:vn,difficulty:xn,scenarioProfile:Sn.optional()});var fe={anyOf:[{type:"string",minLength:1,maxLength:256},{type:"null"}],default:null},ve={type:"integer",minimum:1,maximum:50,default:25},vt=[{name:"search_crm",description:"Search synthetic CRM accounts and contacts. Results are scoped, projected, paginated, and budgeted.",parameters:{type:"object",additionalProperties:!1,required:["query","fields","cursor","limit"],properties:{query:{type:"string",minLength:1,maxLength:500},fields:{type:"array",minItems:1,maxItems:12,uniqueItems:!0,items:{type:"string",enum:["account_id","name","aliases","contact_ids","contract_value_usd_cents","renewal_date","tier","active_contract_id","contract_term_months"]}},cursor:fe,limit:ve}}},{name:"query_billing",description:"Query synthetic invoices and payments for explicit account IDs and a bounded date range.",parameters:{type:"object",additionalProperties:!1,required:["account_ids","date_range","status","cursor","limit"],properties:{account_ids:{type:"array",minItems:1,maxItems:50,uniqueItems:!0,items:{type:"string",minLength:1,maxLength:100}},date_range:{type:"object",additionalProperties:!1,required:["from","to"],properties:{from:{type:"string",format:"date"},to:{type:"string",format:"date"}}},status:{type:"array",minItems:1,maxItems:6,uniqueItems:!0,items:{type:"string",enum:["open","overdue","paid","disputed","scheduled","void"]}},cursor:fe,limit:ve}}},{name:"search_support",description:"Search synthetic support cases for explicit account IDs, severity, and state filters.",parameters:{type:"object",additionalProperties:!1,required:["account_ids","severity","state","cursor","limit"],properties:{account_ids:{type:"array",minItems:1,maxItems:50,uniqueItems:!0,items:{type:"string",minLength:1,maxLength:100}},severity:{type:"array",minItems:1,maxItems:4,uniqueItems:!0,items:{type:"string",enum:["P1","P2","P3","P4"]}},state:{type:"array",minItems:1,maxItems:5,uniqueItems:!0,items:{type:"string",enum:["new","investigating","waiting_customer","resolved","closed"]}},cursor:fe,limit:ve}}},{name:"run_python",description:"Run standard-library-only Python in the persistent synthetic attempt sandbox with strict limits and no network.",parameters:{type:"object",additionalProperties:!1,required:["code"],properties:{code:{type:"string",minLength:1,maxLength:1e4}}}}],Z=Cn({schemaVersion:ft,tools:vt}),yn=[`Use only the four registered synthetic Cross-System Operations tools. Contract ${Z}.`,"Treat every filter as exact. An overdue-only query_billing call returns invoice rows; payment rows are included only when paid is requested.","For cross-system joins, retrieve the bounded projected rows and use run_python for date filtering, set intersection, and arithmetic before answering.","Finish with ANSWER: JSON."].join(" "),_n=15;function kn(t="auto"){return[yn,"LOCAL TOOL PROTOCOL (strict):","Use only the registered functions below. Never invent, rename, or call any other tool.","For one tool call, output exactly one JSON object and no prose:",'{"type":"tool_call","name":"registered_name","arguments":{}}',"Tool results arrive as exactly one JSON object:",'{"type":"tool_result","name":"registered_name","ok":true,"result":{},"error":null}','After one tool result, either call one registered tool again or answer normally. A JSON final may use {"type":"final","content":"..."}.',`Tool choice: ${JSON.stringify(t)}. Maximum tool turns: ${_n}.`,`Registered tool signatures: ${vt.map(Rn).join("; ")}`].join(`
 `)}v(kn,"crossSystemLocalToolSystemPrompt");var Ei=kn();function Rn(t){let i=K(t.parameters),s=K(i.properties),l=xt(i.required),f=Object.fromEntries(l.map(h=>[h,Y(s[h])]));return`${t.name}(${JSON.stringify(f)})`}v(Rn,"compactToolSignature");function Y(t){let i=K(t),s=xe(i.anyOf);if(s.length)return s.map(Y).join("|");let l=xe(i.enum);if(l.length)return l.join("|");if(i.type==="array")return[Y(i.items)];if(i.type==="object"){let f=K(i.properties),h=xt(i.required);return Object.fromEntries(h.map(x=>[x,Y(f[x])]))}return typeof i.type=="string"?i.type:"unknown"}v(Y,"compactSchema");function K(t){return t&&typeof t=="object"&&!Array.isArray(t)?t:{}}v(K,"recordValue");function xe(t){return Array.isArray(t)?t:[]}v(xe,"arrayValue");function xt(t){return xe(t).filter(i=>typeof i=="string")}v(xt,"stringValues");var Q=e.enum(fn),In=e.discriminatedUnion("kind",[e.object({kind:e.literal("model"),turn:e.number().int().nonnegative(),content:e.string().max(1e5)}),e.object({kind:e.literal("tool_call"),turn:e.number().int().nonnegative(),callId:e.string().trim().min(1).max(240),name:Q,arguments:e.record(e.string(),e.unknown())}),e.object({kind:e.literal("tool_result"),turn:e.number().int().nonnegative(),callId:e.string().trim().min(1).max(240),name:Q,ok:e.boolean(),result:e.unknown(),rows:e.number().int().nonnegative(),bytes:e.number().int().nonnegative(),durationMs:e.number().int().nonnegative(),error:e.string().trim().min(1).max(2e4).nullable()}),e.object({kind:e.literal("final"),turn:e.number().int().nonnegative(),content:e.string().max(1e5)})]),St=e.object({schemaVersion:e.literal(ye),id:e.string().trim().min(1).max(240),worldId:e.string().trim().min(1).max(240),taskId:e.string().trim().min(1).max(240),toolContractHash:e.literal(Z),modelRef:e.object({providerId:e.string().trim().min(1),modelId:e.string().trim().min(1)}).nullable(),status:e.enum(["completed","budget_exhausted","cancelled","infrastructure_failure"]),steps:e.array(In).max(1e3),startedAt:e.string().trim().min(1),completedAt:e.string().trim().min(1),infrastructureError:e.string().trim().min(1).max(2e4).nullable(),metadata:e.record(e.string(),e.unknown()).default({})}),yt=e.object({schemaVersion:e.literal(ye),trajectoryId:e.string().trim().min(1).max(240),outcome:e.enum(["correct","incorrect","parse_failure","budget_exhausted","tool_schema_violation","infrastructure_failure","cancelled"]),reward:e.number().min(0).max(1.15).nullable(),rewardEligible:e.boolean(),exactAnswer:e.boolean(),components:e.object({exactAnswer:e.number().min(0).max(1),efficiency:e.number().min(0).max(.1),conciseOutput:e.number().min(0).max(.05)}),metrics:e.object({toolCalls:e.number().int().nonnegative(),rowsRead:e.number().int().nonnegative(),bytesRead:e.number().int().nonnegative(),consecutiveRetrievalCalls:e.number().int().nonnegative(),wallTimeMs:e.number().int().nonnegative(),parseFailures:e.number().int().nonnegative(),budgetExhausted:e.boolean()}),parsedAnswer:e.unknown().nullable(),feedback:e.array(e.string().trim().min(1).max(5e3)).max(100)}),jn=e.object({id:e.string().trim().min(1).max(240),type:e.literal("function"),function:e.object({name:Q,arguments:e.string().max(2e4)})}),_t=e.object({role:e.enum(["system","user","assistant","tool"]),content:e.string().max(2e5).nullable(),tool_call_id:e.string().trim().min(1).max(240).optional(),tool_calls:e.array(jn).max(4).optional()}),Hi=e.object({schemaVersion:e.literal(ye),id:e.string().trim().min(1).max(240),taskId:e.string().trim().min(1).max(240),worldId:e.string().trim().min(1).max(240),trajectoryId:e.string().trim().min(1).max(240),toolContractHash:e.literal(Z),approval:e.object({status:e.literal("approved"),approvedBy:e.string().trim().min(1).max(240),approvedAt:e.string().trim().min(1)}),messages:e.array(_t).min(3).max(1e3)}),Tn=e.object({tasksetTaskId:e.string().trim().min(1).max(240),environmentTaskId:e.string().trim().min(1).max(240),family:e.enum(["renewal_exposure","collections_prioritization","invoice_reconciliation","sla_escalation","contract_billing_mismatch"]),prompt:e.string().trim().min(1).max(1e5),finalAnswer:e.string().trim().min(1).max(2e5),trajectoryId:e.string().trim().min(1).max(240),trajectoryHash:e.string().trim().min(8).max(256),toolNames:e.array(Q).min(1).max(100),toolCallCount:e.number().int().positive(),messageCount:e.number().int().min(3),reward:e.number().min(0).max(1.15),messages:e.array(_t).min(3).max(1e3)}),An=e.object({status:e.literal("approved"),approvedBy:e.string().trim().min(1).max(240),approvedAt:e.string().trim().min(1),previewHash:e.string().trim().min(8).max(256),trajectoryCount:e.number().int().positive()}),Oi=e.object({schemaVersion:e.literal("openpond.crossSystemExpertBootstrapPreview.v1"),tasksetId:e.string().trim().min(1).max(240),tasksetHash:e.string().trim().min(8).max(256),tasksetRevision:e.number().int().positive(),previewHash:e.string().trim().min(8).max(256),toolContractHash:e.literal(Z),status:e.enum(["ready_for_review","approved"]),approval:An.nullable(),tasks:e.array(Tn).min(1).max(100)});function Cn(t){let i=Se(t),s=2166136261,l=2654435769;for(let f=0;f<i.length;f+=1){let h=i.charCodeAt(f);s=Math.imul(s^h,16777619),l=Math.imul(l^h,2246822507)}return`${ft}:${ht(s)}${ht(l)}`}v(Cn,"contractHash");function Se(t){return Array.isArray(t)?`[${t.map(Se).join(",")}]`:t&&typeof t=="object"?`{${Object.entries(t).sort(([i],[s])=>i.localeCompare(s)).map(([i,s])=>`${JSON.stringify(i)}:${Se(s)}`).join(",")}}`:JSON.stringify(t)}v(Se,"stableJson");function ht(t){return(t>>>0).toString(16).padStart(8,"0")}v(ht,"hex");var wn=e.object({schemaVersion:e.literal("openpond.singleTurnPolicyTrajectory.v1"),id:a,taskId:a,status:e.enum(["completed","infrastructure_failure"]),promptHash:m,responseText:e.string().max(2e6),infrastructureError:e.string().trim().min(1).max(1e4).nullable(),startedAt:b,completedAt:b,metadata:O}),Mn=e.object({schemaVersion:e.literal("openpond.ppoTrajectory.v1"),id:a,taskId:a,policyModelId:a,referenceModelId:a,valueModelId:a,steps:e.array(e.object({index:e.number().int().nonnegative(),observationHash:m,actionTokenIds:e.array(e.number().int().nonnegative()).min(1),terminated:e.boolean(),truncated:e.boolean(),reward:e.number(),policyLogProbability:e.number(),referenceLogProbability:e.number(),valuePrediction:e.number(),return:e.number(),advantage:e.number(),mask:e.number().min(0).max(1)})).min(1).max(1e5),createdAt:b}),Pn=e.object({schemaVersion:e.literal("openpond.exactAnswerVerifierResult.v1"),outcome:e.enum(["correct","incorrect","parse_failure","infrastructure_failure"]),graderSetHash:m,score:e.number().min(0).max(1).nullable(),passed:e.boolean(),rewardEligible:e.boolean(),expectedAnswerHash:m,extractedAnswer:e.string().max(2e4).nullable(),feedback:e.array(e.string().trim().min(1).max(2e4)).max(1e3)}),kt=e.object({schemaVersion:e.literal("openpond.rolloutTrajectoryReceipt.v1"),id:a,jobId:a,planId:a,tasksetId:a,tasksetHash:m,taskId:a,split:e.literal("train"),correlationId:a,provider:a,providerTrace:e.object({invocationId:a,experimentId:a,rolloutId:a,runId:a,rowId:a}),optimizerMethod:e.enum(["grpo","ppo"]).default("grpo"),evidenceLevels:e.object({requested:e.enum(["trajectory","aggregate","provider_reported"]),observed:e.enum(["trajectory","aggregate","provider_reported"]),providerReported:e.enum(["trajectory","aggregate","provider_reported"])}).default({requested:"trajectory",observed:"trajectory",providerReported:"provider_reported"}),budgetUsage:e.object({rollouts:e.number().int().nonnegative(),environmentExecutions:e.number().int().nonnegative(),inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),optimizerSteps:e.number().int().nonnegative(),costUsd:e.number().nonnegative()}).default({rollouts:0,environmentExecutions:0,inputTokens:0,outputTokens:0,optimizerSteps:0,costUsd:0}),environment:e.object({id:a,version:e.string().trim().min(1).max(256),worldId:a,worldHash:m,toolContractHash:m}),policy:e.object({modelId:a,checkpointId:a.nullable(),completionParametersHash:m}),status:e.enum(["received","running","succeeded","failed"]),failureClass:e.enum(["policy_failure","parse_failure","tool_schema_violation","budget_exhausted","cancelled","environment_failure","infrastructure_failure"]).nullable(),reward:e.object({eligible:e.boolean(),raw:e.number().min(0).max(1.15).nullable(),normalized:e.number().min(0).max(1).nullable(),components:e.record(e.string(),e.number())}),trajectory:e.union([St,wn,Mn]).nullable(),verifier:e.union([yt,Pn]).nullable(),providerStatus:e.record(e.string(),e.unknown()).default({}),receivedAt:b,startedAt:b.nullable(),completedAt:b.nullable(),updatedAt:b});function u(t){let i;return new e.ZodObject({type:"object",get shape(){return i??=t()}})}v(u,"deferredObject");var E=e.enum(["sft","dpo","grpo","ppo","sdft","opd","opsd","sdpo"]),X=e.enum(["lora","full"]),z=e.enum(["openpond_managed"]),dr=u(()=>({id:a,input:e.record(e.string(),e.unknown()),expectedOutput:e.record(e.string(),e.unknown()),tags:e.array(a).max(100).default([])})),ur=u(()=>({id:a,prompt:e.string().min(1).max(5e5),chosen:e.string().min(1).max(5e5),rejected:e.string().min(1).max(5e5),sourceRefs:e.array(a).min(1).max(1e4)})),pr=u(()=>({id:a,input:e.record(e.string(),e.unknown()),tags:e.array(a).max(100).default([])})),br=u(()=>({schemaVersion:e.literal("openpond.rewardModelRecipe.v1"),method:e.literal("reward_model"),parameterization:e.literal("lora_with_scalar_head"),runScope:e.enum(["synthetic_smoke","human_preference"]),baseModel:e.object({id:a,revision:e.string().trim().min(1).max(256),tokenizerRevision:e.string().trim().min(1).max(256),processorRevision:e.string().trim().min(1).max(256),chatTemplateHash:m}).strict(),tasksetRelease:r,preferenceDatasetRelease:r,processorRelease:r,input:e.object({kind:e.literal("structured_text"),serialization:e.enum(["scenario_input_and_candidate_json_v1","visible_agent_trajectory_v1"]),maxCharacters:e.number().int().positive().max(5e5)}).strict().default({kind:"structured_text",serialization:"scenario_input_and_candidate_json_v1",maxCharacters:32e3}),lora:e.object({rank:e.number().int().positive().max(256),alpha:e.number().positive().max(1024),dropout:e.number().min(0).max(1),targetModules:e.array(a).min(1).max(100)}).strict(),heads:e.object({scalar:e.literal("pooled_hidden_state_linear"),bucket:e.enum(["none","three_class"]).default("three_class")}).strict(),loss:e.object({ranking:e.literal("bradley_terry"),rankingWeight:e.number().positive(),bucketWeight:e.number().nonnegative(),tieWeight:e.number().nonnegative()}).strict(),optimizer:e.object({learningRate:e.number().positive(),maxSteps:e.number().int().positive().max(1e5),batchSize:e.number().int().positive().max(1e4),gradientAccumulationSteps:e.number().int().positive().max(1e4),seed:e.number().int(),checkpointEverySteps:e.number().int().positive().max(1e5)}).strict(),resourceLimits:e.object({wallTimeMs:e.number().int().positive().max(1440*60*1e3),maxExamples:e.number().int().positive().max(1e5),maxInputCharacters:e.number().int().positive().max(5e5).default(32e3),maxImagePixels:e.number().int().positive().max(1e8).optional(),maximumSpendUsd:e.number().nonnegative().max(1e5)}).strict()})).superRefine((t,i)=>{t.runScope==="synthetic_smoke"&&(t.optimizer.maxSteps>100||t.resourceLimits.maximumSpendUsd>10)&&i.addIssue({code:"custom",path:["resourceLimits"],message:"Synthetic-smoke Reward Model runs are limited to 100 steps and USD 10."})}),En=u(()=>({schemaVersion:e.literal("openpond.rftRecipe.v1"),method:e.literal("grpo"),parameterization:e.literal("lora"),baseModel:e.object({id:a,revision:e.string().trim().min(1).max(256),tokenizerRevision:e.string().trim().min(1).max(256),chatTemplateHash:m}),dataset:e.object({trainSplit:e.literal("train"),validationSplit:e.enum(["validation","frozen_eval"]),maxPromptTokens:e.number().int().positive().max(32768),maxExamples:e.number().int().positive().max(1e5).default(1e3),selectionStrategy:e.enum(["stable_hash_top_n","rft_easy_curriculum_v1"]).default("stable_hash_top_n")}),lora:e.object({rank:e.number().int().positive().max(256)}),rollout:e.object({groupSize:e.number().int().min(2).max(64),concurrency:e.number().int().positive().max(32),maxTurns:e.number().int().positive().max(100),maxOutputTokens:e.number().int().positive().max(8192),temperature:e.number().min(0).max(2),topP:e.number().positive().max(1),seed:e.number().int()}),optimizer:e.object({learningRate:e.number().positive(),maxSteps:e.number().int().positive().max(1e5),clipRange:e.number().positive().max(1).default(.2),iterations:e.number().int().min(2).max(16).default(2),microbatchSize:e.number().int().positive().max(64).default(1),gradientAccumulationSteps:e.number().int().positive().max(64).default(1),advantageEpsilon:e.number().positive().max(.01).default(1e-8),adamw:D}),loss:e.object({method:F.default("grpo"),klBeta:e.number().min(0).nullable().default(null)}).default({method:"grpo",klBeta:null}),reward:e.object({graderId:a,graderHash:m,environmentId:a,environmentVersion:e.string().trim().min(1).max(256),toolContractHash:m,learnedPreference:J.nullable().optional()}),resourceLimits:e.object({wallTimeMs:e.number().int().positive().max(1440*60*1e3),maxGpuSeconds:e.number().int().positive().max(21600).optional(),warmupDeadlineSeconds:e.number().int().positive().max(3600).optional(),maxRollouts:e.number().int().positive().max(1e5),maxPayloadBytes:e.number().int().positive().max(10*1024*1024)}),policyOptimization:W.nullable().default(null),continuation:e.object({schemaVersion:e.literal("openpond.crossJobContinuationRequest.v1"),parentArtifact:r,sourceArtifact:e.object({jobId:a,artifactId:a,checkpointId:a,contentHash:m}).strict(),optimizerMode:e.enum(["continue","reset"]),adapterMode:e.enum(["residual_append","shared","sequential_merge"]).optional()}).strict().optional()})),Hn=u(()=>({schemaVersion:e.literal("openpond.dpoRecipe.v1"),method:e.literal("dpo"),parameterization:e.literal("lora"),policyModel:H,referenceModel:H,dataset:e.object({trainSplit:e.literal("train"),validationSplit:e.enum(["validation","frozen_eval"]),maxPairs:e.number().int().positive().max(1e5),maxPromptTokens:e.number().int().positive().max(32768),maxCompletionTokens:e.number().int().positive().max(32768),selectionStrategy:e.literal("stable_hash_top_n"),selectionSeed:e.number().int()}),lora:e.object({rank:e.number().int().positive().max(256),alpha:e.number().positive().max(1024),dropout:e.number().min(0).max(1),targetModules:e.array(a).min(1).max(100)}),loss:e.object({variant:e.literal("sigmoid"),beta:e.number().positive().max(10),labelSmoothing:e.number().min(0).max(.5).default(0)}),optimizer:e.object({learningRate:e.number().positive(),epochs:e.number().positive().max(100),maxSteps:e.number().int().positive().max(1e6),batchSize:e.number().int().positive().max(1e4),gradientAccumulationSteps:e.number().int().positive().max(1e4),seed:e.number().int()}),referenceLogprobs:e.object({cacheSchemaVersion:e.literal("openpond.dpoReferenceLogprobs.v1"),cacheKey:m,invalidationHash:m}),resourceLimits:e.object({cpuThreads:e.number().int().positive().max(256),memoryBytes:e.number().int().positive(),wallTimeMs:e.number().int().positive()})})),On=u(()=>({schemaVersion:e.literal("openpond.ppoRecipe.v1"),method:e.literal("ppo"),parameterization:e.literal("lora"),policyOptimization:W.extend({optimizer:G}),lora:e.object({rank:e.number().int().positive().max(256),alpha:e.number().positive().max(1024),dropout:e.number().min(0).max(1),targetModules:e.array(a).min(1).max(100)}),valueHead:e.object({initialization:e.literal("policy_hidden_state_linear"),optimizerLearningRate:e.number().positive(),artifactName:e.literal("value_head.safetensors")}),policyLearningRate:e.number().positive(),resume:e.object({checkpointId:a.nullable(),policyHash:m,referenceHash:m,valueModelHash:m,optimizerStateHash:m.nullable()}),resourceLimits:e.object({cpuThreads:e.number().int().positive().max(256),memoryBytes:e.number().int().positive(),wallTimeMs:e.number().int().positive()})})),zn=u(()=>({schemaVersion:e.literal("openpond.unsupportedRecipe.v1"),method:E.exclude(["sft","dpo","grpo","ppo"]),parameterization:X,unsupportedReason:e.string().trim().min(1).max(5e3)})),Vn=e.union([Me,Hn,En,On,zn]),Dn=u(()=>({schemaVersion:e.literal("openpond.trainingDestinationCapabilities.v1"),destinationId:z,available:e.boolean(),methods:e.array(E),parameterizations:e.array(X),modelAllowlist:e.array(a).default([]),maxDatasetBytes:e.number().int().nonnegative().nullable(),environmentPlacements:e.array(e.enum(["none","local","remote","colocated","provider_native"])),nonProduction:e.boolean(),unavailableReason:e.string().trim().min(1).max(5e3).nullable(),checkedAt:b})),Bn=u(()=>({destinationId:z,available:e.boolean(),methods:e.array(E),parameterizations:e.array(X),nonProduction:e.boolean(),unavailableReason:e.string().trim().min(1).max(5e3).nullable()})),Ln=u(()=>({schemaVersion:e.literal("openpond.baseModelCandidate.v1"),selectionKey:a,label:e.string().trim().min(1).max(500),sourceLabel:e.string().trim().min(1).max(500),preference:B,available:e.boolean(),nonProduction:e.boolean(),unavailableReason:e.string().trim().min(1).max(5e3).nullable(),methods:e.array(E),executionOptions:e.array(Bn).min(1)})),qn=u(()=>({code:a,severity:e.enum(["warning","error"]),path:e.string().trim().max(2e3).nullable(),message:e.string().trim().min(1).max(5e3)})),Un=u(()=>({schemaVersion:e.literal("openpond.trainingCompatibility.v1"),compatible:e.boolean(),destinationId:z,tasksetId:a,recipeMethod:E,issues:e.array(qn),checkedAt:b})),Nn=e.union([we,e.enum(["destination_unavailable","destination_method_unsupported","destination_model_unsupported","destination_parameterization_unsupported","experimental_destination"])]),gr=u(()=>({method:E,state:e.enum(["recommended","compatible","needs_dataset_work","destination_unavailable","experimental_destination"]),reasonCodes:e.array(Nn).default([]),reasons:e.array(e.string().trim().min(1).max(5e3)).default([]),destinationId:z.nullable().default(null)})),hr=e.enum(["small","standard","custom","small_experiment"]);var $n=Te,jt=u(()=>({schemaVersion:e.literal("openpond.trainingPlan.v1"),id:a,modelId:a,tasksetId:a,tasksetHash:m,comparisonSeriesEntry:A.nullable().optional(),evaluationTasksetRef:g.nullable().optional(),harnessRelease:r.nullable().optional(),modelImprovementQualification:r.nullable().optional(),destinationId:z,recipe:Vn,environmentPlacement:e.enum(["none","local","remote","colocated","provider_native"]),compatibility:Un,dataPolicy:e.object({exportApproved:e.boolean(),approvedSourceIds:e.array(a),retentionDays:e.number().int().nonnegative().nullable(),region:e.string().trim().min(1).max(200).nullable()}),estimatedCostUsd:e.number().nonnegative().nullable(),createdAt:b,contentHash:m})),Fn=u(()=>({path:e.string().trim().min(1).max(2e3),sha256:m,sizeBytes:e.number().int().nonnegative(),role:e.enum(["manifest","task_data","grader","environment","recipe","policy","provenance"])})),_e=u(()=>({schemaVersion:e.literal("openpond.trainingBundle.v1"),id:a,planId:a,tasksetId:a,tasksetHash:m,recipeHash:m,files:e.array(Fn).min(1).max(1e6),totalSizeBytes:e.number().int().nonnegative(),sourceIds:e.array(a),excludedSourceIds:e.array(a),containsRawChats:e.literal(!1),containsSecrets:e.literal(!1),containsHiddenGraderAssets:e.literal(!1),createdAt:b,contentHash:m})),fr=u(()=>({schemaVersion:e.literal("openpond.trainingBundleExport.v1"),manifest:_e,files:e.array(e.object({path:e.string().trim().min(1).max(2e3),sha256:m,sizeBytes:e.number().int().nonnegative(),encoding:e.literal("base64"),content:e.string()})).min(1),contentHash:m})),vr=u(()=>({schemaVersion:e.literal("openpond.trainingPreparedStart.v1"),plan:jt,bundle:_e,approvalActor:a.nullable(),preparedAt:b})),xr=u(()=>({schemaVersion:e.literal("openpond.trainingApproval.v1"),evaluationTasksetRef:g.nullable().optional(),id:a,planId:a,bundleHash:m,harnessRelease:r.nullable().optional(),modelImprovementQualification:r.nullable().optional(),destinationId:z,modelId:a,method:E,parameterization:X,maximumCostUsd:e.number().nonnegative().nullable(),approvedBy:a,approvedAt:b})),Sr=u(()=>({schemaVersion:e.literal("openpond.trainingJobSourceSnapshot.v1"),evaluationTasksetRef:g.nullable().optional(),modelProjectId:a,sourceProjectRevision:e.number().int().positive(),profileId:a,taskset:e.object({id:a,revision:e.number().int().positive(),contentHash:m}).strict(),tasksetRelease:r,harnessRelease:r,baseModel:B,method:E})).strict(),Gn=e.enum(["queued","starting","running","cancelling","cancelled","succeeded","failed","reconciling"]),Tt=u(()=>({schemaVersion:e.literal("openpond.trainingJob.v1"),id:a,planId:a,bundleHash:m,approvalId:a,destinationId:z,status:Gn,nonProduction:e.boolean(),workerPid:e.number().int().positive().nullable(),startedAt:b.nullable(),completedAt:b.nullable(),error:e.string().trim().min(1).max(2e4).nullable(),createdAt:b,updatedAt:b,metadata:O})),Jn=u(()=>({schemaVersion:e.literal("openpond.trainingJobEvent.v1"),id:a,jobId:a,sequence:e.number().int().nonnegative(),type:e.enum(["queued","start","progress","metric","checkpoint","cancel","complete","failure","reconcile"]),timestamp:b,payload:O})),Wn=u(()=>({schemaVersion:e.literal("openpond.sftStepMetric.v1"),step:e.number().int().nonnegative(),maxSteps:e.number().int().positive(),timestamp:b,epoch:e.number().nonnegative().nullable(),loss:e.number().nullable(),learningRate:e.number().nonnegative().nullable(),gradientNorm:e.number().nonnegative().nullable(),entropy:e.number().nonnegative().nullable(),meanTokenAccuracy:e.number().min(0).max(1).nullable(),preferenceAccuracy:e.number().min(0).max(1).nullable().default(null),preferenceMargin:e.number().nullable().default(null),chosenReward:e.number().nullable().default(null),rejectedReward:e.number().nullable().default(null),chosenLogProbability:e.number().nullable().default(null),rejectedLogProbability:e.number().nullable().default(null),reward:e.number().nullable().default(null),policyLoss:e.number().nullable().default(null),advantageLoss:e.number().nullable().default(null),inputTokensSeen:e.number().int().nonnegative().nullable(),memoryBytes:e.number().int().nonnegative().nullable(),elapsedSeconds:e.number().nonnegative().nullable()})),Yn=u(()=>({schemaVersion:e.literal("openpond.policyOptimizationMetric.v1"),method:e.enum(["grpo","ppo"]),step:e.number().int().nonnegative(),timestamp:b,learningRate:e.number().nonnegative().nullable().default(null),policyLoss:e.number().nullable(),valueLoss:e.number().nullable(),gradientNorm:e.number().nonnegative().nullable().default(null),meanReward:e.number().nullable(),meanReturn:e.number().nullable(),kl:e.number().nullable(),behaviorPolicyKlPreUpdate:e.number().nonnegative().nullable().default(null),entropy:e.number().nullable(),policyClipFraction:e.number().min(0).max(1).nullable(),behaviorPolicyClipFractionPreUpdate:e.number().min(0).max(1).nullable().default(null),valueClipFraction:e.number().min(0).max(1).nullable(),explainedVariance:e.number().nullable(),rolloutLearnerLag:e.number().int().nonnegative().nullable(),inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),environmentExecutions:e.number().int().nonnegative(),trajectoryCount:e.number().int().nonnegative().nullable().default(null),costUsd:e.number().nonnegative().nullable()})),Kn=u(()=>({schemaVersion:e.literal("openpond.managedTrainingRunEvidence.v2"),provider:a,providerRunId:a,state:a,progress:e.object({targetOptimizerSteps:e.number().int().nonnegative(),committedOptimizerSteps:e.number().int().nonnegative(),skippedOptimizerSteps:e.number().int().nonnegative()}),reward:e.object({finalMean:e.number().nullable(),variance:e.number().nonnegative().nullable(),minimum:e.number().nullable(),maximum:e.number().nullable(),distinctValueCount:e.number().int().nonnegative(),noSignalGroupCount:e.number().int().nonnegative(),trajectoryCount:e.number().int().nonnegative(),eligibleTrajectoryCount:e.number().int().nonnegative()}),usage:e.object({inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),environmentExecutions:e.number().int().nonnegative()}),resource:e.object({provider:a,gpuType:e.string().trim().min(1).max(300).nullable(),gpuCount:e.number().int().nonnegative().nullable(),hourlyCostUsd:e.number().nonnegative().nullable(),durationSeconds:e.number().nonnegative().nullable(),gpuSeconds:e.number().nonnegative().nullable()}),cost:e.object({totalUsd:e.number().nonnegative().nullable()}),checkpoint:e.object({id:a,policyVersion:e.number().int().nonnegative(),sha256:m.nullable(),sizeBytes:e.number().int().nonnegative().nullable()}).nullable(),movement:e.object({adapterDeltaNorm:e.number().nonnegative().nullable()}),evaluations:e.array(e.object({reference:e.object({id:a,contentHash:m}).nullable().optional(),kind:e.enum(["baseline","candidate"]),policyVersion:e.number().int().nonnegative(),taskCount:e.number().int().nonnegative().nullable().optional(),targetTaskCount:e.number().int().nonnegative().nullable().optional(),score:e.number().nullable(),threshold:e.number().nullable(),passed:e.boolean().nullable()})),canonicalPublication:e.object({state:a.nullable(),artifactId:a.nullable()}),syncedAt:b})),Rt=u(()=>({count:e.number().int().nonnegative(),scoredCount:e.number().int().nonnegative(),meanScore:e.number().min(0).max(1).nullable(),passedCount:e.number().int().nonnegative(),passRate:e.number().min(0).max(1).nullable()})),It=u(()=>({status:e.enum(["scored","unavailable"]),score:e.number().min(0).max(1).nullable(),passed:e.boolean(),rewardEligible:e.boolean(),failureClass:e.enum(["policy_failure","grader_failure","environment_failure","infrastructure_failure","timeout","cancelled"]).nullable(),feedback:e.array(e.string().trim().min(1).max(2e4)).max(1e3),components:e.array(e.object({graderId:a,score:e.number().min(0).max(1).nullable(),passed:e.boolean(),feedback:e.string().trim().max(2e4).nullable()})).max(1e3)})),Qn=u(()=>({taskId:a,input:e.record(e.string(),e.unknown()),baseOutput:e.record(e.string(),e.unknown()).nullable(),trainedOutput:e.record(e.string(),e.unknown()).nullable(),baseGrade:It.nullable(),trainedGrade:It.nullable()})),Zn=u(()=>({schemaVersion:e.literal("openpond.trainingEvaluationSummary.v1"),jobId:a,tasksetId:a,base:Rt,trained:Rt,meanScoreDelta:e.number().min(-1).max(1).nullable(),examples:e.array(Qn).max(1e6)})),yr=u(()=>({schemaVersion:e.literal("openpond.trainingRunDetail.v1"),job:Tt,events:e.array(Jn),stepMetrics:e.array(Wn),policyMetrics:e.array(Yn).default([]),managedEvidence:Kn.nullable().default(null),evaluation:Zn.nullable(),generatedAt:b})),Xn=u(()=>({schemaVersion:e.literal("openpond.trainingArtifact.v1"),id:a,jobId:a,kind:e.enum(["adapter","checkpoint","metrics","log","manifest","evaluation"]),path:e.string().trim().min(1).max(4e3),sha256:m,sizeBytes:e.number().int().nonnegative(),baseModelId:a.nullable(),baseModelRevision:e.string().trim().min(1).max(256).nullable(),tokenizerRevision:e.string().trim().min(1).max(256).nullable(),chatTemplateHash:m.nullable(),nonProduction:e.boolean(),createdAt:b,metadata:O})),ea=u(()=>({schemaVersion:e.literal("openpond.modelArtifactLineage.v1"),id:a,modelId:a,artifactId:a,jobId:a,tasksetId:a,tasksetHash:m,graderHash:m,planHash:m,bundleHash:m,recipeHash:m,workerVersion:e.string().trim().min(1).max(256),trainerVersion:e.string().trim().min(1).max(256),importedAt:b,frozenEvaluationArtifactId:a.nullable(),promotable:e.boolean(),pinned:e.boolean().default(!1),status:e.enum(["imported","rejected"]).default("imported"),rejectedAt:b.nullable().default(null),rejectionReason:e.string().trim().min(1).max(5e3).nullable().default(null),chatConfiguration:he.default(bt),managedServing:gt.nullable().default(null)})),ta=e.enum(["chat_manual","agent","extension","authoring_optimizer"]),na=u(()=>({schemaVersion:e.literal("openpond.modelBinding.v1"),id:a,profileId:a,role:ta,roleTargetId:a,modelArtifactLineageId:a,tasksetId:a,evaluationArtifactId:a.nullable(),status:e.enum(["active","rolled_back"]),priorBindingId:a.nullable(),rollbackTargetBindingId:a.nullable(),promotedBy:a,promotedAt:b,rolledBackAt:b.nullable(),metadata:O})),_r=u(()=>({schemaVersion:e.literal("openpond.trainingActivity.v1"),profileId:a,active:e.boolean(),activeCounts:e.object({jobs:e.number().int().nonnegative(),creations:e.number().int().nonnegative(),minerRuns:e.number().int().nonnegative(),datasetImports:e.number().int().nonnegative()}),revision:m,generatedAt:b})),kr=u(()=>({schemaVersion:e.literal("openpond.tasksetOperationalState.v1"),tasksetId:a,attempts:e.array(Ee),artifacts:e.array(He),grades:e.array(Oe),evaluationResults:e.array(je).default([]),generatedAt:b})),Rr=u(()=>({schemaVersion:e.literal("openpond.trainingState.v1"),profileId:a,sources:e.array(Ae),creations:e.array(Ve),tasksetDrafts:e.array(Ce).default([]),tasksets:e.array(ae),benchmarkRuns:e.array(e.custom()).default([]),benchmarkComparisons:e.array(e.custom()).default([]),datasetImports:e.array(Ne).default([]),datasetArtifacts:e.array(ne).default([]),graderAuditReports:e.array(ze),evaluationResults:e.array(e.custom()).default([]),candidates:e.array(ce),minerConfig:q,minerRuns:e.array(Ge).default([]),modelProjects:e.array($n).default([]),modelVersions:e.array(tt).default([]),modelRuns:e.array(rt).default([]),comparisonSeries:e.array(We).default([]),comparisonSeriesEntries:e.array(Ke).default([]),modelCurrencySnapshots:e.array(st).default([]),continualBenchIssueReviews:e.array(mt).default([]),continualLearningDailyBatches:e.array(lt).default([]),rewardModelVersions:e.array(nt).default([]),rewardModelRuns:e.array(at).default([]),modelTasksets:e.array(ae).default([]),plans:e.array(jt),bundles:e.array(_e),jobs:e.array(Tt),artifacts:e.array(Xn),models:e.array(ea),rolloutReceipts:e.array(kt).default([]),modelBindings:e.array(na).default([]),destinations:e.array(Dn),baseModelCandidates:e.array(Ln).default([]),activityRevision:m.optional(),generatedAt:b}));export{qe as a,Pt as b,Ue as c,Ne as d,ya as e,_a as f,Dt as g,ce as h,q as i,Ta as j,Aa as k,Ge as l,Lt as m,We as n,Gt as o,Wt as p,Yt as q,A as r,Ke as s,Ea as t,tt as u,nt as v,en as w,at as x,rn as y,sn as z,rt as A,ai as B,st as C,$ as D,bi as E,lt as F,mt as G,J as H,he as I,gt as J,hn as K,ji as L,ye as M,wi as N,ft as O,fn as P,Pi as Q,vt as R,Z as S,Ei as T,St as U,yt as V,_t as W,Hi as X,An as Y,Oi as Z,kt as _,z as $,br as aa,Vn as ba,Dn as ca,Ln as da,Un as ea,$n as fa,jt as ga,_e as ha,fr as ia,vr as ja,xr as ka,Sr as la,Tt as ma,Jn as na,Wn as oa,Yn as pa,Kn as qa,Zn as ra,yr as sa,Xn as ta,ea as ua,ta as va,na as wa,kr as xa};

package/dist/chunks/{chunk-FEQ7HNJR.js → chunk-FE2ENE3U.js}

@@ -1,2 +1,2 @@
 import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{Ab as _,Bb as E,zb as F}from"./chunk-MVPKU72J.js";import{a as T,b as A,c as H,d as w,e as b,f as S,g,l as k}from"./chunk-HVM2GUM3.js";import{e as C,f as R}from"./chunk-SH7POUVX.js";import{l as x,n as O}from"./chunk-RCT3O3HQ.js";import{h as y,p as N,r as L,s as j}from"./chunk-UGOK7VTC.js";import{a as r}from"./chunk-FTQLQJQV.js";import{promises as z}from"node:fs";import d from"node:path";import{randomUUID as J}from"node:crypto";import{promises as v}from"node:fs";import U from"node:path";function u(e){return{binding:r(()=>{let i=L(e.home,"refiner_bindings","active");return i?w.parse(i.value):null},"binding"),releases:r(async()=>{let i=(await v.readdir(e.releases)).filter(n=>n.endsWith(".json"));return Promise.all(i.map(async n=>H.parse(JSON.parse(await v.readFile(U.join(e.releases,n),"utf8")))))},"releases"),transitions:r(()=>Object.values(j(e.home,"refiner_transitions")).map(i=>b.parse(i.value)),"transitions"),release:r(async i=>{let n=H.parse(JSON.parse(await v.readFile(U.join(e.releases,`${i.contentHash}.json`),"utf8")));if(n.id!==i.id||n.contentHash!==i.contentHash)throw new Error("Refiner release reference does not match immutable release content.");return n},"release"),persistRelease:r(async i=>V(e.releases+`/${i.contentHash}.json`,x(i),!0),"persistRelease"),writeSource:r(async i=>V(e.source,i),"writeSource"),transition:r(async(i,n)=>{N(e.home,t=>{t.exec("BEGIN IMMEDIATE");try{let a=t.prepare("SELECT payload FROM refiner_bindings WHERE id = 'active'").get(),o=a?w.parse(JSON.parse(a.payload)):null,s=new Date().toISOString(),l={id:i.id,contentHash:i.contentHash},c={schemaVersion:"openpond.refinerTransitionReceipt.v1",id:`refiner-transition-${J()}`,operation:n.operation,bindingChanged:n.bindingChanged,previousRelease:o?.release??null,nextRelease:l,actor:n.actor,reason:n.reason,authoringSkillHash:n.authoringSkillHash,validation:{valid:!0,messages:[]},createdAt:s},p=b.parse({...c,contentHash:O(c)});if(t.prepare("INSERT INTO refiner_transitions VALUES (?, ?, 1)").run(p.contentHash,JSON.stringify(p)),n.bindingChanged){let m=w.parse({schemaVersion:"openpond.refinerBinding.v1",channel:"active",revision:(o?.revision??-1)+1,release:l,updatedAt:s});t.prepare("INSERT INTO refiner_bindings VALUES ('active', ?, 1) ON CONFLICT(id) DO UPDATE SET payload=excluded.payload, revision=revision+1").run(JSON.stringify(m))}t.exec("COMMIT")}catch(a){throw t.exec("ROLLBACK"),a}})},"transition")}}r(u,"createLocalRefinerProfileRepository");async function V(e,i,n=!1){await R(e,async()=>{if(n){let t=await v.readFile(e,"utf8").catch(()=>null);if(t!==null){if(t!==i)throw new Error("Immutable Refiner release bytes changed.");return}}await C(e,i)})}r(V,"atomicWrite");var B="OpenPond Refiner Core: evidence admission, privacy, ownership, validation, and immutable activation boundaries.";function P(e){let i=d.join(e,"library","refiners");return{home:e,root:i,source:d.join(i,"source","openpond.review.json"),releases:d.join(i,"releases")}}r(P,"refinerProfilePaths");function Re(e){let i=r(()=>f(e),"inspect"),n=r(o=>M(e,o),"update"),t=r(o=>q(e,o),"activate"),a=r(o=>G(e,o),"rollback");return{refinerHistoryPayload:i,updateRefinerProfilePayload:n,activateRefinerReleasePayload:t,rollbackRefinerReleasePayload:a,inspectRefiner:i,updateRefiner:n,activateRefiner:t,rollbackRefiner:a}}r(Re,"createRefinerProfileRoutePayloads");async function f(e){let i=P(e),n=u(i),t=n.binding();if(!t)throw new Error("Active Refiner binding is unavailable.");let a=(await n.releases()).sort((l,c)=>c.createdAt.localeCompare(l.createdAt)),o=a.find(l=>l.id===t.release.id&&l.contentHash===t.release.contentHash);if(!o)throw new Error("Active Refiner release is unavailable.");let s=n.transitions().sort((l,c)=>c.createdAt.localeCompare(l.createdAt));return F.parse({rootPath:i.root,sourcePath:i.source,binding:t,currentRelease:o,releases:a,transitions:s})}r(f,"inspectRefinerProfile");async function ue(e){return(await f(e)).currentRelease}r(ue,"loadActiveRefinerRelease");async function M(e,i){return R(d.join(y(e).runtime,"refiner-edit"),()=>W(e,i))}r(M,"updateRefinerProfile");async function W(e,i){let n=_.parse(i),t=P(e),a=u(t);await I(t);let o=T.parse(n.profile),s=S({profile:o,coreVersion:k,corePrompt:B}),c=(await a.releases()).find(m=>m.composedPromptHash===s.composedPromptHash&&m.profileHash===s.profileHash)??s;await a.persistRelease(c),await a.writeSource(g(o));let p=a.binding();if(!p)throw new Error("Active Refiner binding is unavailable.");return n.activate&&p.release.contentHash===c.contentHash||!n.activate&&a.transitions().some(h=>!h.bindingChanged&&h.nextRelease.contentHash===c.contentHash&&h.actor===n.actor&&h.reason===n.reason&&h.authoringSkillHash===n.authoringSkillHash)||await a.transition(c,{operation:"update",bindingChanged:n.activate,actor:n.actor,reason:n.reason,authoringSkillHash:n.authoringSkillHash}),f(e)}r(W,"updateRefinerProfileUnlocked");async function q(e,i){return R(d.join(y(e).runtime,"refiner-edit"),()=>$(e,i))}r(q,"activateRefinerRelease");async function $(e,i){let n=E.parse(i),t=P(e),a=u(t);await I(t);let o=await a.release(n.release),s=a.binding();if(!s)throw new Error("Active Refiner binding is unavailable.");return s.release.contentHash===o.contentHash||(await a.transition(o,{operation:"activate",bindingChanged:!0,actor:n.actor,reason:n.reason,authoringSkillHash:null}),await a.writeSource(g(o.profile))),f(e)}r($,"activateRefinerReleaseUnlocked");async function G(e,i){return R(d.join(y(e).runtime,"refiner-edit"),()=>K(e,i))}r(G,"rollbackRefinerRelease");async function K(e,i){let n=E.parse(i),t=P(e),a=u(t);await I(t);let o=await a.release(n.release),s=a.binding();if(!s)throw new Error("Active Refiner binding is unavailable.");return s.release.contentHash===o.contentHash||(await a.transition(o,{operation:"rollback",bindingChanged:!0,actor:n.actor,reason:n.reason,authoringSkillHash:null}),await a.writeSource(g(o.profile))),f(e)}r(K,"rollbackRefinerReleaseUnlocked");async function Y(e){await R(d.join(y(e).runtime,"refiner-initialize"),()=>Q(P(e)))}r(Y,"initializeRefinerProfile");async function I(e){await Y(e.home)}r(I,"ensureInitialized");async function Q(e){let i=u(e);if(await Promise.all([z.mkdir(d.dirname(e.source),{recursive:!0,mode:448}),z.mkdir(e.releases,{recursive:!0,mode:448})]),i.binding())return;let n=S({profile:A,coreVersion:k,corePrompt:B});await i.persistRelease(n),await i.writeSource(g(n.profile)),await i.transition(n,{operation:"initialize",bindingChanged:!0,actor:"openpond",reason:"Initialize the default Refiner Review Profile.",authoringSkillHash:null})}r(Q,"initialize");export{u as a,P as b,Re as c,f as d,ue as e,M as f,q as g,G as h,Y as i};
+import{Ab as _,Bb as E,zb as F}from"./chunk-CJDGLPNX.js";import{a as T,b as A,c as H,d as w,e as b,f as S,g,l as k}from"./chunk-HVM2GUM3.js";import{e as C,f as R}from"./chunk-SH7POUVX.js";import{l as x,n as O}from"./chunk-RCT3O3HQ.js";import{h as y,p as N,r as L,s as j}from"./chunk-UGOK7VTC.js";import{a as r}from"./chunk-FTQLQJQV.js";import{promises as z}from"node:fs";import d from"node:path";import{randomUUID as J}from"node:crypto";import{promises as v}from"node:fs";import U from"node:path";function u(e){return{binding:r(()=>{let i=L(e.home,"refiner_bindings","active");return i?w.parse(i.value):null},"binding"),releases:r(async()=>{let i=(await v.readdir(e.releases)).filter(n=>n.endsWith(".json"));return Promise.all(i.map(async n=>H.parse(JSON.parse(await v.readFile(U.join(e.releases,n),"utf8")))))},"releases"),transitions:r(()=>Object.values(j(e.home,"refiner_transitions")).map(i=>b.parse(i.value)),"transitions"),release:r(async i=>{let n=H.parse(JSON.parse(await v.readFile(U.join(e.releases,`${i.contentHash}.json`),"utf8")));if(n.id!==i.id||n.contentHash!==i.contentHash)throw new Error("Refiner release reference does not match immutable release content.");return n},"release"),persistRelease:r(async i=>V(e.releases+`/${i.contentHash}.json`,x(i),!0),"persistRelease"),writeSource:r(async i=>V(e.source,i),"writeSource"),transition:r(async(i,n)=>{N(e.home,t=>{t.exec("BEGIN IMMEDIATE");try{let a=t.prepare("SELECT payload FROM refiner_bindings WHERE id = 'active'").get(),o=a?w.parse(JSON.parse(a.payload)):null,s=new Date().toISOString(),l={id:i.id,contentHash:i.contentHash},c={schemaVersion:"openpond.refinerTransitionReceipt.v1",id:`refiner-transition-${J()}`,operation:n.operation,bindingChanged:n.bindingChanged,previousRelease:o?.release??null,nextRelease:l,actor:n.actor,reason:n.reason,authoringSkillHash:n.authoringSkillHash,validation:{valid:!0,messages:[]},createdAt:s},p=b.parse({...c,contentHash:O(c)});if(t.prepare("INSERT INTO refiner_transitions VALUES (?, ?, 1)").run(p.contentHash,JSON.stringify(p)),n.bindingChanged){let m=w.parse({schemaVersion:"openpond.refinerBinding.v1",channel:"active",revision:(o?.revision??-1)+1,release:l,updatedAt:s});t.prepare("INSERT INTO refiner_bindings VALUES ('active', ?, 1) ON CONFLICT(id) DO UPDATE SET payload=excluded.payload, revision=revision+1").run(JSON.stringify(m))}t.exec("COMMIT")}catch(a){throw t.exec("ROLLBACK"),a}})},"transition")}}r(u,"createLocalRefinerProfileRepository");async function V(e,i,n=!1){await R(e,async()=>{if(n){let t=await v.readFile(e,"utf8").catch(()=>null);if(t!==null){if(t!==i)throw new Error("Immutable Refiner release bytes changed.");return}}await C(e,i)})}r(V,"atomicWrite");var B="OpenPond Refiner Core: evidence admission, privacy, ownership, validation, and immutable activation boundaries.";function P(e){let i=d.join(e,"library","refiners");return{home:e,root:i,source:d.join(i,"source","openpond.review.json"),releases:d.join(i,"releases")}}r(P,"refinerProfilePaths");function Re(e){let i=r(()=>f(e),"inspect"),n=r(o=>M(e,o),"update"),t=r(o=>q(e,o),"activate"),a=r(o=>G(e,o),"rollback");return{refinerHistoryPayload:i,updateRefinerProfilePayload:n,activateRefinerReleasePayload:t,rollbackRefinerReleasePayload:a,inspectRefiner:i,updateRefiner:n,activateRefiner:t,rollbackRefiner:a}}r(Re,"createRefinerProfileRoutePayloads");async function f(e){let i=P(e),n=u(i),t=n.binding();if(!t)throw new Error("Active Refiner binding is unavailable.");let a=(await n.releases()).sort((l,c)=>c.createdAt.localeCompare(l.createdAt)),o=a.find(l=>l.id===t.release.id&&l.contentHash===t.release.contentHash);if(!o)throw new Error("Active Refiner release is unavailable.");let s=n.transitions().sort((l,c)=>c.createdAt.localeCompare(l.createdAt));return F.parse({rootPath:i.root,sourcePath:i.source,binding:t,currentRelease:o,releases:a,transitions:s})}r(f,"inspectRefinerProfile");async function ue(e){return(await f(e)).currentRelease}r(ue,"loadActiveRefinerRelease");async function M(e,i){return R(d.join(y(e).runtime,"refiner-edit"),()=>W(e,i))}r(M,"updateRefinerProfile");async function W(e,i){let n=_.parse(i),t=P(e),a=u(t);await I(t);let o=T.parse(n.profile),s=S({profile:o,coreVersion:k,corePrompt:B}),c=(await a.releases()).find(m=>m.composedPromptHash===s.composedPromptHash&&m.profileHash===s.profileHash)??s;await a.persistRelease(c),await a.writeSource(g(o));let p=a.binding();if(!p)throw new Error("Active Refiner binding is unavailable.");return n.activate&&p.release.contentHash===c.contentHash||!n.activate&&a.transitions().some(h=>!h.bindingChanged&&h.nextRelease.contentHash===c.contentHash&&h.actor===n.actor&&h.reason===n.reason&&h.authoringSkillHash===n.authoringSkillHash)||await a.transition(c,{operation:"update",bindingChanged:n.activate,actor:n.actor,reason:n.reason,authoringSkillHash:n.authoringSkillHash}),f(e)}r(W,"updateRefinerProfileUnlocked");async function q(e,i){return R(d.join(y(e).runtime,"refiner-edit"),()=>$(e,i))}r(q,"activateRefinerRelease");async function $(e,i){let n=E.parse(i),t=P(e),a=u(t);await I(t);let o=await a.release(n.release),s=a.binding();if(!s)throw new Error("Active Refiner binding is unavailable.");return s.release.contentHash===o.contentHash||(await a.transition(o,{operation:"activate",bindingChanged:!0,actor:n.actor,reason:n.reason,authoringSkillHash:null}),await a.writeSource(g(o.profile))),f(e)}r($,"activateRefinerReleaseUnlocked");async function G(e,i){return R(d.join(y(e).runtime,"refiner-edit"),()=>K(e,i))}r(G,"rollbackRefinerRelease");async function K(e,i){let n=E.parse(i),t=P(e),a=u(t);await I(t);let o=await a.release(n.release),s=a.binding();if(!s)throw new Error("Active Refiner binding is unavailable.");return s.release.contentHash===o.contentHash||(await a.transition(o,{operation:"rollback",bindingChanged:!0,actor:n.actor,reason:n.reason,authoringSkillHash:null}),await a.writeSource(g(o.profile))),f(e)}r(K,"rollbackRefinerReleaseUnlocked");async function Y(e){await R(d.join(y(e).runtime,"refiner-initialize"),()=>Q(P(e)))}r(Y,"initializeRefinerProfile");async function I(e){await Y(e.home)}r(I,"ensureInitialized");async function Q(e){let i=u(e);if(await Promise.all([z.mkdir(d.dirname(e.source),{recursive:!0,mode:448}),z.mkdir(e.releases,{recursive:!0,mode:448})]),i.binding())return;let n=S({profile:A,coreVersion:k,corePrompt:B});await i.persistRelease(n),await i.writeSource(g(n.profile)),await i.transition(n,{operation:"initialize",bindingChanged:!0,actor:"openpond",reason:"Initialize the default Refiner Review Profile.",authoringSkillHash:null})}r(Q,"initialize");export{u as a,P as b,Re as c,f as d,ue as e,M as f,q as g,G as h,Y as i};

package/dist/chunks/{chunk-6TRVEA3T.js → chunk-FVK4W27G.js}

@@ -1,2 +1,2 @@
 import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{Ga as E,_ as m,aa as T,ja as p}from"./chunk-WLZJOCLA.js";import{ca as y}from"./chunk-G4E3WGLM.js";import{n as w}from"./chunk-RCT3O3HQ.js";import{a as k}from"./chunk-FTQLQJQV.js";import{constants as u}from"node:fs";import{link as O,lstat as F,mkdir as h,mkdtemp as P,open as N,readFile as x,readdir as B,rename as A,rm as v,writeFile as I}from"node:fs/promises";import i from"node:path";async function M(r,a){let s=a.metadata.importedPackageHash;if(s===void 0)return;if(typeof s!="string"||!r)throw new Error("Imported batch requires its package storage directory.");let t=await S(r,s);if(!t.learningResources||t.taskset.id!==a.id||t.taskset.revision!==a.revision||w(y(t.taskset))!==w(a.metadata.learning))throw new Error("Imported batch differs from its pinned package.");return t}k(M,"readImportedLearningTasksetPackage");async function J(r,a,s=m){let t=p(a,s),o=i.join(r,"training","portable-packages");await h(o,{recursive:!0});let d=i.join(o,`${t.contentHash}.json`),c=await P(i.join(o,".package-"));try{let n=i.join(c,"package.json");await I(n,JSON.stringify(t),{flag:"wx",mode:384});try{await O(n,d)}catch(e){if(e.code!=="EEXIST")throw e;await S(r,t.contentHash,s)}}finally{await v(c,{recursive:!0,force:!0})}}k(J,"cacheTasksetPackage");async function S(r,a,s=m){if(!/^[a-f0-9]{64}$/.test(a))throw new Error("Invalid Taskset package cache identity.");let t=i.join(r,"training","portable-packages",`${a}.json`),o=await N(t,u.O_RDONLY|u.O_NOFOLLOW|u.O_NONBLOCK);try{let d=await o.stat();if(!d.isFile()||d.size>s)throw new Error("Taskset package cache must be a bounded regular file.");let c=Buffer.alloc(d.size),n=0;for(;n<c.length;){let f=await o.read(c,n,c.length-n,n);if(!f.bytesRead)throw new Error("Taskset package cache changed during read.");n+=f.bytesRead}if((await o.read(Buffer.alloc(1),0,1,n)).bytesRead)throw new Error("Taskset package cache changed during read.");let e=p(JSON.parse(c.toString("utf8")),s);if(e.contentHash!==a)throw new Error("Cached Taskset package differs from its saved receipt.");return e}finally{await o.close()}}k(S,"readCachedTasksetPackage");async function $(r){let a=p(r.package),s=E(r.taskset.profileId,a.contentHash);if(r.taskset.environment.metadata.runtimeSourceTasksetId!==s)throw new Error("Imported Taskset directory differs from its package identity.");let t=new Map,o=k((e,f)=>{if(i.isAbsolute(e)||e.includes("\\")||e.split("/").some(l=>!l||l==="."||l===".."))throw new Error("Taskset package file requires a safe relative path.");let g=t.get(e);if(g&&!Buffer.from(g).equals(Buffer.from(f)))throw new Error(`Taskset package files collide at ${e}.`);t.set(e,f)},"add");for(let e of a.files)o(e.asset.path,T(e));for(let e of r.generatedFiles)o(e.path,Buffer.from(e.content,"utf8"));o("taskset.json",Buffer.from(JSON.stringify(r.taskset),"utf8"));let d=i.join(r.home,"training","tasksets");await h(d,{recursive:!0});let c=await P(i.join(d,".package-import-")),n=i.join(d,s);try{for(let[e,f]of t)await h(i.dirname(i.join(c,e)),{recursive:!0}),await I(i.join(c,e),f,{flag:"wx",mode:384});try{await A(c,n)}catch(e){if(!["EEXIST","ENOTEMPTY"].includes(e.code??""))throw e;let f=await F(n);if(!f.isDirectory()||f.isSymbolicLink())throw new Error("Imported Taskset destination must be a regular directory.");let g=await j(n);if(JSON.stringify(g.sort())!==JSON.stringify([...t.keys()].sort()))throw new Error("Imported Taskset file inventory changed.");for(let[l,b]of t)if(!(await x(i.join(n,l))).equals(Buffer.from(b)))throw new Error("Imported Taskset files differ from the pinned package.")}await J(r.home,a)}finally{await v(c,{recursive:!0,force:!0})}}k($,"materializeImportedTasksetPackage");async function j(r,a=""){let s=[];for(let t of await B(i.join(r,a),{withFileTypes:!0})){let o=i.posix.join(a,t.name);if(t.isSymbolicLink())throw new Error("Imported Taskset directories cannot contain symbolic links.");if(t.isDirectory())s.push(...await j(r,o));else if(t.isFile())s.push(o);else throw new Error("Imported Taskset packages require regular files.")}return s}k(j,"regularFiles");export{M as a,J as b,S as c,$ as d};
+import{Ga as E,_ as m,aa as T,ja as p}from"./chunk-5ZNE43XC.js";import{ca as y}from"./chunk-XHU4BZQY.js";import{n as w}from"./chunk-RCT3O3HQ.js";import{a as k}from"./chunk-FTQLQJQV.js";import{constants as u}from"node:fs";import{link as O,lstat as F,mkdir as h,mkdtemp as P,open as N,readFile as x,readdir as B,rename as A,rm as v,writeFile as I}from"node:fs/promises";import i from"node:path";async function M(r,a){let s=a.metadata.importedPackageHash;if(s===void 0)return;if(typeof s!="string"||!r)throw new Error("Imported batch requires its package storage directory.");let t=await S(r,s);if(!t.learningResources||t.taskset.id!==a.id||t.taskset.revision!==a.revision||w(y(t.taskset))!==w(a.metadata.learning))throw new Error("Imported batch differs from its pinned package.");return t}k(M,"readImportedLearningTasksetPackage");async function J(r,a,s=m){let t=p(a,s),o=i.join(r,"training","portable-packages");await h(o,{recursive:!0});let d=i.join(o,`${t.contentHash}.json`),c=await P(i.join(o,".package-"));try{let n=i.join(c,"package.json");await I(n,JSON.stringify(t),{flag:"wx",mode:384});try{await O(n,d)}catch(e){if(e.code!=="EEXIST")throw e;await S(r,t.contentHash,s)}}finally{await v(c,{recursive:!0,force:!0})}}k(J,"cacheTasksetPackage");async function S(r,a,s=m){if(!/^[a-f0-9]{64}$/.test(a))throw new Error("Invalid Taskset package cache identity.");let t=i.join(r,"training","portable-packages",`${a}.json`),o=await N(t,u.O_RDONLY|u.O_NOFOLLOW|u.O_NONBLOCK);try{let d=await o.stat();if(!d.isFile()||d.size>s)throw new Error("Taskset package cache must be a bounded regular file.");let c=Buffer.alloc(d.size),n=0;for(;n<c.length;){let f=await o.read(c,n,c.length-n,n);if(!f.bytesRead)throw new Error("Taskset package cache changed during read.");n+=f.bytesRead}if((await o.read(Buffer.alloc(1),0,1,n)).bytesRead)throw new Error("Taskset package cache changed during read.");let e=p(JSON.parse(c.toString("utf8")),s);if(e.contentHash!==a)throw new Error("Cached Taskset package differs from its saved receipt.");return e}finally{await o.close()}}k(S,"readCachedTasksetPackage");async function $(r){let a=p(r.package),s=E(r.taskset.profileId,a.contentHash);if(r.taskset.environment.metadata.runtimeSourceTasksetId!==s)throw new Error("Imported Taskset directory differs from its package identity.");let t=new Map,o=k((e,f)=>{if(i.isAbsolute(e)||e.includes("\\")||e.split("/").some(l=>!l||l==="."||l===".."))throw new Error("Taskset package file requires a safe relative path.");let g=t.get(e);if(g&&!Buffer.from(g).equals(Buffer.from(f)))throw new Error(`Taskset package files collide at ${e}.`);t.set(e,f)},"add");for(let e of a.files)o(e.asset.path,T(e));for(let e of r.generatedFiles)o(e.path,Buffer.from(e.content,"utf8"));o("taskset.json",Buffer.from(JSON.stringify(r.taskset),"utf8"));let d=i.join(r.home,"training","tasksets");await h(d,{recursive:!0});let c=await P(i.join(d,".package-import-")),n=i.join(d,s);try{for(let[e,f]of t)await h(i.dirname(i.join(c,e)),{recursive:!0}),await I(i.join(c,e),f,{flag:"wx",mode:384});try{await A(c,n)}catch(e){if(!["EEXIST","ENOTEMPTY"].includes(e.code??""))throw e;let f=await F(n);if(!f.isDirectory()||f.isSymbolicLink())throw new Error("Imported Taskset destination must be a regular directory.");let g=await j(n);if(JSON.stringify(g.sort())!==JSON.stringify([...t.keys()].sort()))throw new Error("Imported Taskset file inventory changed.");for(let[l,b]of t)if(!(await x(i.join(n,l))).equals(Buffer.from(b)))throw new Error("Imported Taskset files differ from the pinned package.")}await J(r.home,a)}finally{await v(c,{recursive:!0,force:!0})}}k($,"materializeImportedTasksetPackage");async function j(r,a=""){let s=[];for(let t of await B(i.join(r,a),{withFileTypes:!0})){let o=i.posix.join(a,t.name);if(t.isSymbolicLink())throw new Error("Imported Taskset directories cannot contain symbolic links.");if(t.isDirectory())s.push(...await j(r,o));else if(t.isFile())s.push(o);else throw new Error("Imported Taskset packages require regular files.")}return s}k(j,"regularFiles");export{M as a,J as b,S as c,$ as d};

package/dist/chunks/{chunk-7FBWUVX7.js → chunk-GC6RJBRZ.js}

@@ -1,5 +1,5 @@
 import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{wa as n}from"./chunk-WLZJOCLA.js";import{a}from"./chunk-FTQLQJQV.js";function l(i,e,s){i.exec("BEGIN IMMEDIATE");try{let o=i.get("SELECT content_hash, profile_id FROM taskset_revisions WHERE taskset_id = ? AND revision = ?",[e.id,e.revision]);if(o&&o.content_hash!==e.contentHash)throw new Error(`Taskset ${e.id}@${e.revision} is immutable and already has another content hash.`);let t=i.get("SELECT payload FROM tasksets WHERE id = ?",[e.id]),d=t?n.parse(JSON.parse(t.payload)):null;if(o&&o.profile_id!==e.profileId||d&&d.profileId!==e.profileId)throw new Error(`Taskset ${e.id} belongs to another Profile.`);let r=JSON.stringify(e);i.run(`INSERT INTO taskset_revisions (taskset_id, revision, content_hash, profile_id, status, payload, created_at, updated_at)
+import{wa as n}from"./chunk-5ZNE43XC.js";import{a}from"./chunk-FTQLQJQV.js";function l(i,e,s){i.exec("BEGIN IMMEDIATE");try{let o=i.get("SELECT content_hash, profile_id FROM taskset_revisions WHERE taskset_id = ? AND revision = ?",[e.id,e.revision]);if(o&&o.content_hash!==e.contentHash)throw new Error(`Taskset ${e.id}@${e.revision} is immutable and already has another content hash.`);let t=i.get("SELECT payload FROM tasksets WHERE id = ?",[e.id]),d=t?n.parse(JSON.parse(t.payload)):null;if(o&&o.profile_id!==e.profileId||d&&d.profileId!==e.profileId)throw new Error(`Taskset ${e.id} belongs to another Profile.`);let r=JSON.stringify(e);i.run(`INSERT INTO taskset_revisions (taskset_id, revision, content_hash, profile_id, status, payload, created_at, updated_at)
        VALUES (?, ?, ?, ?, ?, ?, ?, ?)
        ON CONFLICT(taskset_id, revision) DO UPDATE SET status = excluded.status, payload = excluded.payload, updated_at = excluded.updated_at`,[e.id,e.revision,e.contentHash,e.profileId,e.status,r,e.createdAt,e.updatedAt]),(!d||d.revision<=e.revision)&&i.run(`INSERT INTO tasksets (id, profile_id, status, payload, created_at, updated_at)
          VALUES (?, ?, ?, ?, ?, ?)

package/dist/chunks/{chunk-5Z3AU5CL.js → chunk-H2XG5N7D.js}

@@ -1,5 +1,5 @@
 import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{c as _}from"./chunk-MHPCVQOE.js";import{a as J}from"./chunk-55Q64WWF.js";import{b as U}from"./chunk-RHMULF56.js";import{I as z,K as V,ja as M,na as G,v as O}from"./chunk-MVPKU72J.js";import{aa as N}from"./chunk-WLZJOCLA.js";import{Pb as F,Rb as q,qb as C,ub as A}from"./chunk-QAFXQHHG.js";import{d as H,e as b,ga as x,ja as j}from"./chunk-ZY32SZ2P.js";import{f as T}from"./chunk-HNQPEOJD.js";import{c as I,d as S,e as W,j as $,m as L,n as c}from"./chunk-RCT3O3HQ.js";import{A as P}from"./chunk-X2J2P5ZT.js";import{a as h}from"./chunk-FTQLQJQV.js";import Y from"node:path";import{constants as K}from"node:fs";import{open as ae,realpath as X}from"node:fs/promises";async function Q(e){let i=e.attempt.profileNative,p=P.array($).max(16).parse(e.attempt.artifactRefs??[]);if(!p.length)return[];if(!i||!e.storeDir)throw new Error("The actual retained artifact owner is unavailable.");await e.authorize();let t=await e.events(i.turnId);if(c(t)!==i.traceHash||t.some(r=>r.sessionId!==i.sessionId||r.turnId!==i.turnId))throw new Error("The saved artifacts differ from their actual retained case trace.");let o=[],l=0;for(let r of p){let f=t.filter(d=>(d.action==="work_output_save"||d.action==="sandbox_save_output")&&d.status==="completed").flatMap(d=>{let k=P.object({outputRef:M}).passthrough().safeParse(d.data);return k.success?[k.data.outputRef]:[]}).filter(d=>`${i.sessionId}/${d.id}/${d.revision}/${d.title}`===r.id&&d.sourceTaskId===i.sessionId&&d.sourceTurnId===i.turnId&&d.sha256===r.contentHash&&d.sizeBytes===r.sizeBytes&&d.contentType===r.mediaType);if(f.length!==1)throw new Error("The selected artifact has no unique actual saved output.");let s=f[0];if(s.location.kind!=="local"&&(s.location.kind!=="managed"||!e.readManaged)||s.sizeBytes>G)throw new Error("The retained artifact has no supported local byte owner.");if((l+=s.sizeBytes)>2e7)throw new Error("The retained artifacts exceed their aggregate grading byte limit.");await e.authorize(),e.signal?.throwIfAborted();let a;if(s.location.kind==="managed")a=await e.readManaged(s,e.signal);else if(s.location.kind==="local"){let d=await X(Y.join(e.storeDir,"work","outputs")),k=await X(s.location.path);if(!k.startsWith(`${d}${Y.sep}`))throw new Error("The artifact escaped canonical output storage.");let w=await ae(k,K.O_RDONLY|K.O_NOFOLLOW);try{let g=await w.stat();if(!g.isFile()||g.size!==s.sizeBytes)throw new Error("The artifact bytes changed.");a=await w.readFile()}finally{await w.close()}}else throw new Error("The saved artifact has no admitted byte owner.");if(a.length!==s.sizeBytes||L(a)!==s.sha256)throw new Error("The artifact bytes differ from their sealed receipt.");await e.authorize(),o.push({reference:r,title:s.title,contentType:s.contentType,base64:a.toString("base64"),...s.contentType.startsWith("text/")||s.contentType==="application/json"?{text:new TextDecoder("utf8",{fatal:!0}).decode(a)}:{}})}return await e.authorize(),o}h(Q,"readLocalProfileArtifacts");import oe from"node:path";async function Z(e,i){if(e.limits.maximumSpendUsd===null)return;let p=process.env.OPENPOND_API_KEY?.trim(),t=process.env.OPENPOND_OPCHAT_API_URL?.trim();if(!p||!t)throw new Error("Profile evaluation requires an admitted hosted spend ceiling.");let o=new URL(`${t.replace(/\/+$/,"")}/experiment-budget`);if(o.protocol!=="https:"||o.username||o.password||o.search||o.hash)throw new Error("Profile evaluation spend authority URL is invalid.");o.searchParams.set("runId",e.id),o.searchParams.set("manifestHash",e.contentHash);let l=await fetch(o,{headers:{Authorization:`Bearer ${p}`},signal:i});if(!l.ok)throw new Error("The exact Profile evaluation spend ceiling is unavailable.");let r=P.object({runId:P.string(),manifestHash:P.string(),maximumCostUsd:P.number().positive()}).strict().parse(await l.json());if(r.runId!==e.id||r.manifestHash!==e.contentHash||r.maximumCostUsd!==e.limits.maximumSpendUsd)throw new Error("Profile evaluation spend authority differs from its admitted manifest.")}h(Z,"assertProfileEvaluationSpendAuthority");async function ee(e){e.signal?.throwIfAborted();let i,p,t=h(()=>{},"rejectCancellation"),o=new Promise((s,a)=>{t=a}),l=h((s,a)=>{p||(p=new Error(s),e.onInterrupt?.(a),e.interrupt?(i=e.interrupt(s),i.catch(()=>{})):t(p))},"cancel"),r=h(()=>l("Profile evaluation run cancelled.","cancelled"),"aborted"),f=setTimeout(()=>l(`Profile evaluation case exceeded its ${e.timeoutMs}ms timeout.`,"timed_out"),e.timeoutMs);e.signal?.addEventListener("abort",r,{once:!0});try{e.signal?.aborted&&r();let s=await Promise.race([e.send(),o]);if(i){let a=await i;if(a.status==="in_progress"||!a.completedAt)throw new Error("Profile evaluation cleanup remains unconfirmed.")}if(s.status==="in_progress"||!s.completedAt)throw new Error("Workflow evaluation turn did not settle.");return s}catch(s){if(i){let a=await i;if(a.status==="in_progress"||!a.completedAt)throw new Error("Profile evaluation cleanup remains unconfirmed.",{cause:s})}throw s}finally{clearTimeout(f),e.signal?.removeEventListener("abort",r)}}h(ee,"awaitProfileEvaluationTurn");function te(e){let i=e.manifest.profileEvaluation,p=e.manifest.policy;if(!i||p.kind!=="model"||e.manifest.execution.kind!=="harness")throw new Error("Profile evaluation requires a released source and model policy.");if(i.profileId!==e.profileRef.profileId||i.profileId!==e.binding.profileId||i.sourceRevision!==e.binding.sourceRevision||i.harnessRelease.id!==e.binding.harnessRelease.id||i.harnessRelease.contentHash!==e.binding.harnessRelease.contentHash||(e.binding.schemaVersion==="openpond.profileWorkflowBinding.v1"?i.target.kind!=="workflow"||i.target.workflowId!==e.binding.workflowId:i.target.kind==="workflow"||c(i.target)!==c(e.binding.target)))throw new Error("Profile evaluation binding differs from its admitted source.");if(p.model.provider!==e.modelRef.providerId||p.model.model!==e.modelRef.modelId||p.configurationHash!==e.modelConfigurationHash)throw new Error("Workflow evaluation model differs from its admitted configuration.");return async t=>{if(c(t.source)!==c(i))throw new Error("Workflow evaluation case differs from its admitted Profile source.");t.signal?.throwIfAborted(),await(e.assertSpendAuthority??Z)(e.manifest,t.signal);let o=!!(e.attachments?.length||e.requiredOutputs?.length),l=o&&!e.ownedSessionMetadata?.localProfileExperiment&&e.profileRef.source==="local"&&(e.modelRef.providerId==="codex"||U(e.modelRef.providerId)),r=await e.createSession({...o?{experience:"work"}:{},provider:e.modelRef.providerId,modelRef:e.modelRef,currentProfile:e.profileRef,...e.binding.schemaVersion==="openpond.profileWorkflowBinding.v1"?{profileWorkflowBinding:e.binding}:{profileComponentBinding:e.binding},hiddenFromDefaultSidebar:!0,metadata:{profileEvaluationRun:{id:e.manifest.id,contentHash:e.manifest.contentHash},taskId:t.task.id,seed:t.seed,...o&&e.profileRef.source==="local"?{workspaceTarget:"local"}:{},...o?{profileEvaluationOutputs:e.requiredOutputs??[]}:{},...e.ownedSessionMetadata}}),f=[...o?[l?`This is a local evaluation run. Complete the requested task using its public inputs. Save these required files at their declared paths relative to the current task workspace: ${JSON.stringify(e.requiredOutputs??[])}. The app preserves those declared files automatically when the turn finishes; work_save_output is not a shell command and is not needed here. For headless tools that write user preferences, use writable configuration, cache and profile directories within this task workspace and a bounded command timeout. Send the requested final response to finish the case. Private grading runs afterward; do not wait for grader feedback.`:"This is an evaluation run. Complete the requested task using its public inputs. work_exec starts in writable Work storage; inputs are ../inputs and released policy source is ../harness. When work_finish is available, create every declared output in Work and call work_finish with the path of your complete working response file, then send a short final answer. Otherwise save each required output with work_save_output and send your final response. Private grading runs after your final response; do not wait for grader feedback."]:[],typeof t.task.input=="string"?t.task.input:JSON.stringify(t.task.input),...Object.keys(t.task.policyVisibleContext).length?[`Policy-visible task context:
+import{c as _}from"./chunk-ZWEY66RT.js";import{a as J}from"./chunk-XD4ZIQZ2.js";import{b as U}from"./chunk-RHMULF56.js";import{I as z,K as V,ja as M,na as G,v as O}from"./chunk-CJDGLPNX.js";import{aa as N}from"./chunk-5ZNE43XC.js";import{Pb as F,Rb as q,qb as C,ub as A}from"./chunk-QAFXQHHG.js";import{d as H,e as b,ga as x,ja as j}from"./chunk-ZY32SZ2P.js";import{f as T}from"./chunk-HNQPEOJD.js";import{c as I,d as S,e as W,j as $,m as L,n as c}from"./chunk-RCT3O3HQ.js";import{A as P}from"./chunk-X2J2P5ZT.js";import{a as h}from"./chunk-FTQLQJQV.js";import Y from"node:path";import{constants as K}from"node:fs";import{open as ae,realpath as X}from"node:fs/promises";async function Q(e){let i=e.attempt.profileNative,p=P.array($).max(16).parse(e.attempt.artifactRefs??[]);if(!p.length)return[];if(!i||!e.storeDir)throw new Error("The actual retained artifact owner is unavailable.");await e.authorize();let t=await e.events(i.turnId);if(c(t)!==i.traceHash||t.some(r=>r.sessionId!==i.sessionId||r.turnId!==i.turnId))throw new Error("The saved artifacts differ from their actual retained case trace.");let o=[],l=0;for(let r of p){let f=t.filter(d=>(d.action==="work_output_save"||d.action==="sandbox_save_output")&&d.status==="completed").flatMap(d=>{let k=P.object({outputRef:M}).passthrough().safeParse(d.data);return k.success?[k.data.outputRef]:[]}).filter(d=>`${i.sessionId}/${d.id}/${d.revision}/${d.title}`===r.id&&d.sourceTaskId===i.sessionId&&d.sourceTurnId===i.turnId&&d.sha256===r.contentHash&&d.sizeBytes===r.sizeBytes&&d.contentType===r.mediaType);if(f.length!==1)throw new Error("The selected artifact has no unique actual saved output.");let s=f[0];if(s.location.kind!=="local"&&(s.location.kind!=="managed"||!e.readManaged)||s.sizeBytes>G)throw new Error("The retained artifact has no supported local byte owner.");if((l+=s.sizeBytes)>2e7)throw new Error("The retained artifacts exceed their aggregate grading byte limit.");await e.authorize(),e.signal?.throwIfAborted();let a;if(s.location.kind==="managed")a=await e.readManaged(s,e.signal);else if(s.location.kind==="local"){let d=await X(Y.join(e.storeDir,"work","outputs")),k=await X(s.location.path);if(!k.startsWith(`${d}${Y.sep}`))throw new Error("The artifact escaped canonical output storage.");let w=await ae(k,K.O_RDONLY|K.O_NOFOLLOW);try{let g=await w.stat();if(!g.isFile()||g.size!==s.sizeBytes)throw new Error("The artifact bytes changed.");a=await w.readFile()}finally{await w.close()}}else throw new Error("The saved artifact has no admitted byte owner.");if(a.length!==s.sizeBytes||L(a)!==s.sha256)throw new Error("The artifact bytes differ from their sealed receipt.");await e.authorize(),o.push({reference:r,title:s.title,contentType:s.contentType,base64:a.toString("base64"),...s.contentType.startsWith("text/")||s.contentType==="application/json"?{text:new TextDecoder("utf8",{fatal:!0}).decode(a)}:{}})}return await e.authorize(),o}h(Q,"readLocalProfileArtifacts");import oe from"node:path";async function Z(e,i){if(e.limits.maximumSpendUsd===null)return;let p=process.env.OPENPOND_API_KEY?.trim(),t=process.env.OPENPOND_OPCHAT_API_URL?.trim();if(!p||!t)throw new Error("Profile evaluation requires an admitted hosted spend ceiling.");let o=new URL(`${t.replace(/\/+$/,"")}/experiment-budget`);if(o.protocol!=="https:"||o.username||o.password||o.search||o.hash)throw new Error("Profile evaluation spend authority URL is invalid.");o.searchParams.set("runId",e.id),o.searchParams.set("manifestHash",e.contentHash);let l=await fetch(o,{headers:{Authorization:`Bearer ${p}`},signal:i});if(!l.ok)throw new Error("The exact Profile evaluation spend ceiling is unavailable.");let r=P.object({runId:P.string(),manifestHash:P.string(),maximumCostUsd:P.number().positive()}).strict().parse(await l.json());if(r.runId!==e.id||r.manifestHash!==e.contentHash||r.maximumCostUsd!==e.limits.maximumSpendUsd)throw new Error("Profile evaluation spend authority differs from its admitted manifest.")}h(Z,"assertProfileEvaluationSpendAuthority");async function ee(e){e.signal?.throwIfAborted();let i,p,t=h(()=>{},"rejectCancellation"),o=new Promise((s,a)=>{t=a}),l=h((s,a)=>{p||(p=new Error(s),e.onInterrupt?.(a),e.interrupt?(i=e.interrupt(s),i.catch(()=>{})):t(p))},"cancel"),r=h(()=>l("Profile evaluation run cancelled.","cancelled"),"aborted"),f=setTimeout(()=>l(`Profile evaluation case exceeded its ${e.timeoutMs}ms timeout.`,"timed_out"),e.timeoutMs);e.signal?.addEventListener("abort",r,{once:!0});try{e.signal?.aborted&&r();let s=await Promise.race([e.send(),o]);if(i){let a=await i;if(a.status==="in_progress"||!a.completedAt)throw new Error("Profile evaluation cleanup remains unconfirmed.")}if(s.status==="in_progress"||!s.completedAt)throw new Error("Workflow evaluation turn did not settle.");return s}catch(s){if(i){let a=await i;if(a.status==="in_progress"||!a.completedAt)throw new Error("Profile evaluation cleanup remains unconfirmed.",{cause:s})}throw s}finally{clearTimeout(f),e.signal?.removeEventListener("abort",r)}}h(ee,"awaitProfileEvaluationTurn");function te(e){let i=e.manifest.profileEvaluation,p=e.manifest.policy;if(!i||p.kind!=="model"||e.manifest.execution.kind!=="harness")throw new Error("Profile evaluation requires a released source and model policy.");if(i.profileId!==e.profileRef.profileId||i.profileId!==e.binding.profileId||i.sourceRevision!==e.binding.sourceRevision||i.harnessRelease.id!==e.binding.harnessRelease.id||i.harnessRelease.contentHash!==e.binding.harnessRelease.contentHash||(e.binding.schemaVersion==="openpond.profileWorkflowBinding.v1"?i.target.kind!=="workflow"||i.target.workflowId!==e.binding.workflowId:i.target.kind==="workflow"||c(i.target)!==c(e.binding.target)))throw new Error("Profile evaluation binding differs from its admitted source.");if(p.model.provider!==e.modelRef.providerId||p.model.model!==e.modelRef.modelId||p.configurationHash!==e.modelConfigurationHash)throw new Error("Workflow evaluation model differs from its admitted configuration.");return async t=>{if(c(t.source)!==c(i))throw new Error("Workflow evaluation case differs from its admitted Profile source.");t.signal?.throwIfAborted(),await(e.assertSpendAuthority??Z)(e.manifest,t.signal);let o=!!(e.attachments?.length||e.requiredOutputs?.length),l=o&&!e.ownedSessionMetadata?.localProfileExperiment&&e.profileRef.source==="local"&&(e.modelRef.providerId==="codex"||U(e.modelRef.providerId)),r=await e.createSession({...o?{experience:"work"}:{},provider:e.modelRef.providerId,modelRef:e.modelRef,currentProfile:e.profileRef,...e.binding.schemaVersion==="openpond.profileWorkflowBinding.v1"?{profileWorkflowBinding:e.binding}:{profileComponentBinding:e.binding},hiddenFromDefaultSidebar:!0,metadata:{profileEvaluationRun:{id:e.manifest.id,contentHash:e.manifest.contentHash},taskId:t.task.id,seed:t.seed,...o&&e.profileRef.source==="local"?{workspaceTarget:"local"}:{},...o?{profileEvaluationOutputs:e.requiredOutputs??[]}:{},...e.ownedSessionMetadata}}),f=[...o?[l?`This is a local evaluation run. Complete the requested task using its public inputs. Save these required files at their declared paths relative to the current task workspace: ${JSON.stringify(e.requiredOutputs??[])}. The app preserves those declared files automatically when the turn finishes; work_save_output is not a shell command and is not needed here. For headless tools that write user preferences, use writable configuration, cache and profile directories within this task workspace and a bounded command timeout. Send the requested final response to finish the case. Private grading runs afterward; do not wait for grader feedback.`:"This is an evaluation run. Complete the requested task using its public inputs. work_exec starts in writable Work storage; inputs are ../inputs and released policy source is ../harness. When work_finish is available, create every declared output in Work and call work_finish with the path of your complete working response file, then send a short final answer. Otherwise save each required output with work_save_output and send your final response. Private grading runs after your final response; do not wait for grader feedback."]:[],typeof t.task.input=="string"?t.task.input:JSON.stringify(t.task.input),...Object.keys(t.task.policyVisibleContext).length?[`Policy-visible task context:
 ${JSON.stringify(t.task.policyVisibleContext)}`]:[]].join(`
 
 `)||" ";await e.admitSession?.(r);try{let s=!1,a=await ee({signal:t.signal,timeoutMs:e.manifest.limits.timeoutMs,onInterrupt:h(n=>{s=n==="timed_out",e.onInterrupt?.(n)},"onInterrupt"),...e.interruptSessionTurn?{interrupt:h(n=>e.interruptSessionTurn(r.id,n),"interrupt")}:{},send:h(()=>{let n={prompt:f,...e.binding.schemaVersion==="openpond.profileWorkflowBinding.v1"?{workflowInput:{}}:e.binding.target.kind==="agent_action"?{workflowInput:t.task.input}:{},modelRef:e.modelRef,approvalPolicy:"on-request",...o?{sandbox:"workspace-write",codexPermissionMode:"default"}:{},...e.attachments?.length?{attachments:e.attachments}:{}};return e.modelRef.providerId==="codex"?e.sendTurn(r.id,n,{nativeTimeoutMs:e.manifest.limits.timeoutMs}):e.sendTurn(r.id,n)},"send")});if(a.status==="in_progress"||!a.completedAt)throw new Error("Workflow evaluation turn did not settle.");if(await e.validateTerminalTurn?.(r,a),!a.harnessSnapshot||a.harnessSnapshot.harnessRelease.id!==i.harnessRelease.id||a.harnessSnapshot.harnessRelease.contentHash!==i.harnessRelease.contentHash)throw new Error("Workflow evaluation turn executed a different Harness release.");if(!a.modelRef||c(a.modelRef)!==c(e.modelRef))throw new Error("Workflow evaluation turn executed a different model.");let d=await e.runtimeEventsForTurn(a.id),k=d.filter(n=>n.name==="workspace_action_result"&&(n.action==="sandbox_save_output"||n.action==="work_output_save")&&n.status==="completed"&&n.turnId===a.id).flatMap(n=>D(n.data)).filter(n=>n.sourceTaskId===r.id&&n.sourceTurnId===a.id&&n.location.kind!=="external"),g=(e.requiredOutputs??[]).flatMap(n=>{let y=[...k].reverse().find(B=>B.title===n.path&&B.contentType===n.mediaType&&(n.maxBytes===null||B.sizeBytes<=n.maxBytes));return y?[y]:[]}).map(n=>({id:`${r.id}/${n.id}/${n.revision}/${n.title}`,contentHash:n.sha256,mediaType:n.contentType,sizeBytes:n.sizeBytes})),v=-1;for(let n=d.length-1;n>=0;n-=1)if(d[n]?.name.endsWith(".delta")){v=n;break}let R=d.flatMap((n,y)=>n.name.endsWith(".delta")&&y!==v?[]:[n.id]);if(R.length+g.length>1e4)throw new Error("Profile evaluation produced more than 10,000 substantive evidence references.");let u=J(d,i.target.kind,a.status),m=Math.min(e.manifest.limits.maxOutputBytes,e.maximumOutputBytes??e.manifest.limits.maxOutputBytes);if(!Number.isSafeInteger(m)||m<1)throw new Error("Profile evaluation requires a positive bounded output limit.");let E=Buffer.byteLength(u,"utf8")>m;if(E&&!e.retainOutputLimitEvidence)throw new Error(`Profile evaluation case exceeded its ${m}-byte output limit.`);return E&&e.onOutputLimit?.(c(u)),{evidence:{output:{text:E?"":u},runtimeEventRefs:R,artifactRefs:g.map(n=>n.id),...E?{infrastructureError:`Profile evaluation case exceeded its ${m}-byte output limit.`}:a.status!=="completed"?{infrastructureError:a.error??`Workflow evaluation turn ${a.status}.`}:{}},traceHash:c(d),retainedEvidenceRef:{sessionId:r.id,turnId:a.id},artifactRefs:g,startedAt:a.startedAt,completedAt:a.completedAt,latencyMs:Math.max(0,Date.parse(a.completedAt)-Date.parse(a.startedAt)),costUsd:null,terminal:a.status==="completed"&&!E,failureClass:a.status==="completed"&&!E?null:a.status==="interrupted"?s?"timeout":"cancelled":"infrastructure_failure"}}finally{await e.settleSession?.(r.id)}}}h(te,"createProfileWorkflowEvaluationExecutor");function D(e,i=0){if(e==null||i>8)return[];let p=M.safeParse(e);return p.success?[p.data]:Array.isArray(e)?e.flatMap(t=>D(t,i+1)):typeof e!="object"?[]:Object.values(e).flatMap(t=>D(t,i+1))}h(D,"findFileOutputs");var re=P.object({manifest:C,taskset:x,profileRef:O,binding:P.union([H,b]),modelRef:T,modelConfigurationHash:S,taskId:I,seed:P.string().trim().min(1).max(500)}).strict();function De(e){return async(i,p)=>{let t=re.parse(i),o=await e.selectedProfile();if(!o||c(o.ref)!==c(t.profileRef)||o.sourceRevision!==t.binding.sourceRevision)throw new Error("Evaluation Profile differs from the app-server's authorized selection.");let l=await e.loadCatalog({ref:o.ref,sourceRevision:o.sourceRevision,harnessRelease:t.binding.harnessRelease}),r=t.manifest.profileEvaluation,f=await _({manifest:t.manifest,taskset:t.taskset,selected:{profileRef:o.ref,sourceRevision:o.sourceRevision,harnessRelease:t.binding.harnessRelease},loadCatalog:e.loadCatalog,resolveExternalDataset:e.resolveExternalDataset}),s=f?.definition??l.definitions.find(u=>u.id===r?.definitionId);if(A(t.manifest,t.taskset,f?.catalog??{schemaVersion:"openpond.profileEvaluations.v1",definitions:l.definitions,suites:l.suites}),!r||!s||(f?c(f.catalog):l.catalogHash)!==r.catalogHash||c(s)!==r.definitionHash||c(s.target)!==c(r.target)||s.tasksetRelease.id!==t.manifest.tasksetRelease.id||s.tasksetRelease.contentHash!==t.manifest.tasksetRelease.contentHash||!s.taskIds.includes(t.taskId)||!s.seeds.includes(t.seed)||t.manifest.population.filter(u=>u.taskId===t.taskId&&u.seed===t.seed).length!==1)throw new Error("Evaluation case differs from its released definition or admitted population.");let a=t.taskset.tasks.find(u=>u.id===t.taskId),d=j(a);if(d.artifactRefs.length!==a.artifactRefs.length||d.artifactRefs.length>z.maxAttachments)throw new Error("Evaluation case has an unauthorized task attachment.");if(d.artifactRefs.length&&!e.storeDir)throw new Error("Evaluation case attachment storage is unavailable.");let k=f?.packageValue??(d.artifactRefs.length||a.requiredOutputs?.length?await e.loadTasksetPackage(s,o.ref.profileId,t.binding.harnessRelease):null);if(k&&(k.contentHash!==t.manifest.packageHash||k.taskset.contentHash!==t.taskset.contentHash))throw new Error("Evaluation case Taskset differs from its frozen package or admitted manifest.");let w=d.artifactRefs.map(u=>{if(u.mediaType!=="application/pdf"||u.sizeBytes>z.maxAttachmentBytes)throw new Error(`Evaluation case attachment ${u.id} is not an admitted PDF.`);let m=k?.files.find(E=>E.asset.id===u.id);if(!m||c(m.asset)!==c(u))throw new Error(`Evaluation case attachment ${u.id} differs from its frozen Taskset file.`);return N(m),V.parse({id:u.id,name:oe.basename(u.path),kind:"file",mediaType:u.mediaType,sizeBytes:u.sizeBytes,contentsBase64:m.base64})}),g=te({manifest:t.manifest,profileRef:o.ref,binding:t.binding,modelRef:t.modelRef,modelConfigurationHash:t.modelConfigurationHash,createSession:e.createSession,sendTurn:e.sendTurn,...e.admitSession?{admitSession:h(u=>e.admitSession(u,t.manifest,t.taskId,t.seed,w),"admitSession")}:{},...e.settleSession?{settleSession:e.settleSession}:{},interruptSessionTurn:e.interruptSessionTurn,runtimeEventsForTurn:h(u=>e.store.runtimeEventsForTurn(u),"runtimeEventsForTurn"),attachments:w,requiredOutputs:a.requiredOutputs??[]});f&&await f.authorize(),p?.throwIfAborted();let v=await g({task:d,seed:t.seed,source:r,...p?{signal:p}:{}}),R=await Q({storeDir:e.storeDir,readManaged:e.readManagedArtifact,attempt:{artifactRefs:v.artifactRefs,profileNative:v.retainedEvidenceRef?.sessionId&&v.retainedEvidenceRef.turnId?{sessionId:v.retainedEvidenceRef.sessionId,turnId:v.retainedEvidenceRef.turnId,traceHash:v.traceHash}:null},events:h(u=>e.store.runtimeEventsForTurn(u),"events"),signal:p,authorize:h(async()=>{p?.throwIfAborted();let u=await e.selectedProfile();if(!u||c(u)!==c(o))throw new Error("The current artifact Profile authority changed.");f&&await f.authorize()},"authorize")});return{...v,evidence:{...v.evidence,artifacts:R,caseOwner:{...v.retainedEvidenceRef,traceHash:v.traceHash}}}}}h(De,"createProfileEvaluationCaseService");var ie=P.object({manifest:C,taskset:x,profileRef:O,binding:P.union([H,b]),modelRef:T,modelConfigurationHash:S}).strict();function Ve(e){let i=new Map,p=h(async(t,o)=>{let l=await e.selectedProfile();if(!l||c(l.ref)!==c(t.profileRef)||l.sourceRevision!==t.binding.sourceRevision)throw new Error("Evaluation Profile differs from the app-server's authorized selection.");let r=await e.loadCatalog({ref:l.ref,sourceRevision:l.sourceRevision,harnessRelease:t.binding.harnessRelease}),f=await _({manifest:t.manifest,taskset:t.taskset,selected:{profileRef:l.ref,sourceRevision:l.sourceRevision,harnessRelease:t.binding.harnessRelease},loadCatalog:e.loadCatalog,resolveExternalDataset:e.resolveExternalDataset}),s=f?.catalog??{schemaVersion:"openpond.profileEvaluations.v1",definitions:r.definitions,suites:r.suites};if(A(t.manifest,t.taskset,s),!e.privateGrading&&(t.taskset.graders.some(m=>m.kind==="model_judge"||m.kind==="custom_verifier")||t.taskset.metrics?.aggregation==="custom"))throw new Error("Profile evaluation cannot run this Taskset's model judge, custom verifier, or custom metric in the current app-server runtime.");let a=t.manifest.profileEvaluation,d=t.manifest.policy;if(a.profileId!==l.ref.profileId||a.sourceRevision!==t.binding.sourceRevision||a.harnessRelease.id!==t.binding.harnessRelease.id||a.harnessRelease.contentHash!==t.binding.harnessRelease.contentHash||(t.binding.schemaVersion==="openpond.profileWorkflowBinding.v1"?a.target.kind!=="workflow"||a.target.workflowId!==t.binding.workflowId:a.target.kind==="workflow"||c(a.target)!==c(t.binding.target))||d.kind!=="model"||d.model.provider!==t.modelRef.providerId||d.model.model!==t.modelRef.modelId||d.configurationHash!==t.modelConfigurationHash)throw new Error("Evaluation request differs from its admitted Profile workflow or model configuration.");let k=s.definitions.find(m=>m.id===a.definitionId),w=e.privateGrading?f?.packageValue??await(e.loadTasksetPackage?e.loadTasksetPackage(k,l.ref.profileId,a.harnessRelease):Promise.reject(new Error("The current private Taskset owner is unavailable."))):null,g=w?await e.privateGrading?.(t.manifest,w,o):void 0;await g?.authorize();let v=await e.store.getProfileEvaluationRun(t.manifest.id);if(v){if(v.manifest.contentHash!==t.manifest.contentHash||c(v.profileRef)!==c(l.ref))throw new Error(`Profile evaluation run ${t.manifest.id} already exists with another manifest.`);return v}f&&await f.authorize(),o?.throwIfAborted();let R=await q({manifest:t.manifest,taskset:t.taskset,catalog:s,...g?{customVerifier:g.customVerifier,modelJudge:g.modelJudge,metricSource:g.metricSource,metricExecutor:g.metricExecutor}:{},execute:h(({task:m,seed:E,signal:n})=>e.executeCase({...t,taskId:m.id,seed:E},n),"execute"),...o?{signal:o}:{},saveGrade:h(async m=>(await e.store.saveProfileEvaluationGrade(m),{id:m.contentHash,contentHash:m.contentHash,mediaType:"application/json",sizeBytes:null}),"saveGrade"),saveReceipt:h(async m=>{await e.store.saveProfileEvaluationReceipt(m)},"saveReceipt"),loadCompletedMember:h(async({receiptId:m})=>{let E=await e.store.getProfileEvaluationReceipt(m);if(!E)return null;let n=E.graderEvidenceRefs[0],y=n&&await e.store.getProfileEvaluationGrade(n.contentHash);if(!y)throw new Error(`Retained Profile evaluation receipt ${m} is missing its grade.`);return{receipt:E,grade:y}},"loadCompletedMember")});await g?.authorize(),f&&await f.authorize();let u={profileRef:l.ref,manifest:t.manifest,metric:R.metric,gradeRefs:R.grades.map(m=>({id:m.contentHash,contentHash:m.contentHash})),receiptRefs:R.receipts.map(m=>({id:m.id,contentHash:m.contentHash})),passRate:R.passRate,passed:R.passed,completedAt:new Date().toISOString()};return e.store.saveProfileEvaluationRun({...u,contentHash:c(u)})},"run");return(t,o)=>{let l=ie.parse(t),r=i.get(l.manifest.id);if(r){if(r.manifestHash!==l.manifest.contentHash)throw new Error(`Profile evaluation run ${l.manifest.id} is already running with another manifest.`);return r.promise}let f=p(l,o);return i.set(l.manifest.id,{manifestHash:l.manifest.contentHash,promise:f}),f.finally(()=>{i.get(l.manifest.id)?.promise===f&&i.delete(l.manifest.id)}).catch(()=>{}),f}}h(Ve,"createProfileEvaluationRunService");var se=P.object({id:I,suiteId:I,createdAt:W,modelRef:T,hostModelConfigurationHash:S.optional()}).strict();function Xe(e){let i=new Map,p=h(async t=>{let o=await e.selectedWorkflows(),l=await e.loadCatalog({ref:o.profileRef,sourceRevision:o.sourceRevision,harnessRelease:o.harnessRelease}),r={schemaVersion:"openpond.profileEvaluations.v1",definitions:l.definitions,suites:l.suites},f=r.suites.find(w=>w.id===t.suiteId);if(!f)throw new Error(`Profile evaluation suite ${t.suiteId} is absent from the selected release.`);let s=[];for(let w of f.definitionIds)s.push(await e.prepareRun({id:`suite-${c([t.id,w]).slice(0,32)}`,createdAt:t.createdAt,definitionId:w,modelRef:t.modelRef,...t.hostModelConfigurationHash?{hostModelConfigurationHash:t.hostModelConfigurationHash}:{}}));let a=await e.store.getProfileEvaluationSuiteRun(t.id);if(a){if(a.suiteId!==f.id||a.catalogHash!==c(r)||a.profileId!==o.profileRef.profileId||a.sourceRevision!==o.sourceRevision||a.harnessRelease.id!==o.harnessRelease.id||a.harnessRelease.contentHash!==o.harnessRelease.contentHash||a.createdAt!==t.createdAt)throw new Error(`Profile evaluation suite run ${t.id} already exists with another request.`);for(let[w,g]of a.members.entries()){let v=await e.store.getProfileEvaluationRun(g.runManifest.id);if(!v||v.contentHash!==g.runHash||v.manifest.contentHash!==g.runManifest.contentHash||v.manifest.contentHash!==s[w]?.manifest.contentHash)throw new Error("Retained Profile evaluation suite is missing member evidence.");let R=v.manifest.policy;if(R.kind!=="model"||R.model.provider!==t.modelRef.providerId||R.model.model!==t.modelRef.modelId)throw new Error(`Profile evaluation suite run ${t.id} already exists with another model.`)}return a}let d=[];for(let[w,g]of f.definitionIds.entries()){let v=s[w],R=await e.executeRun(v);d.push({definitionId:g,manifest:R.manifest,runHash:R.contentHash,passed:R.passed,score:R.metric.value})}let k=F({id:t.id,suiteId:f.id,catalog:r,members:d,createdAt:t.createdAt,completedAt:new Date().toISOString()});return e.store.saveProfileEvaluationSuiteRun(o.profileRef,r,k)},"run");return t=>{let o=se.parse(t),l=c(o),r=i.get(o.id);if(r){if(r.requestHash!==l)throw new Error(`Profile evaluation suite run ${o.id} is already running with another request.`);return r.promise}let f=p(o);return i.set(o.id,{requestHash:l,promise:f}),f.finally(()=>{i.get(o.id)?.promise===f&&i.delete(o.id)}).catch(()=>{}),f}}h(Xe,"createProfileEvaluationSuiteService");export{Q as a,te as b,De as c,Ve as d,Xe as e};

package/dist/chunks/chunk-H6EUFJJK.js

@@ -0,0 +1,2 @@
+import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
+import{Cb as ee,L as q,Q as Y,W as $,ja as Z}from"./chunk-5ZNE43XC.js";import{m as D}from"./chunk-HIQPGYVH.js";import{q as W,r as V,s as I,t as X,w as x}from"./chunk-263WS3TB.js";import{a as U}from"./chunk-SD2XDZQW.js";import{K as A,L as C,M as g,P as Q,o as K}from"./chunk-XHU4BZQY.js";import{Ra as N,Wa as J,bb as _,ga as O,ia as F,v as P,x as B}from"./chunk-ZY32SZ2P.js";import{A as n}from"./chunk-X2J2P5ZT.js";import{a as h}from"./chunk-FTQLQJQV.js";var j=h(t=>t&&typeof t=="object"&&!Array.isArray(t)?t:{},"fields");function te(t){let e=t?.method==="sft"?t:void 0;return{schemaVersion:"openpond.sftRecipe.v1",method:"sft",parameterization:"lora",...e?.baseModel?{baseModel:e.baseModel}:{},dataset:{completionOnly:!0,maxSequenceLength:8192,selectionStrategy:"stable_hash_top_n",selectionSeed:17,...j(e?.dataset)},lora:{rank:16,alpha:32,dropout:0,targetModules:["q_proj","k_proj","v_proj","o_proj","gate_proj","up_proj","down_proj"],...j(e?.lora)},optimizer:{learningRate:1e-4,epochs:3,maxSteps:30,batchSize:1,gradientAccumulationSteps:1,seed:17,...j(e?.optimizer)},resourceLimits:{cpuThreads:4,memoryBytes:48*1024**3,wallTimeMs:3600*1e3,...j(e?.resourceLimits)}}}h(te,"createSupervisedTrainingRecipeDefaults");function S(t){return t&&typeof t=="object"&&!Array.isArray(t)?t:{}}h(S,"fields2");function re(t,e=t?.method==="sft"?"sft":"grpo"){if(e==="sft")return te(t);let r=t?.method==="grpo"?t:void 0,a={schemaVersion:"openpond.rftRecipe.v1",method:"grpo",parameterization:"lora",lora:{rank:2},rollout:{groupSize:8,concurrency:4,maxOutputTokens:1024,temperature:.8,topP:.95,seed:17},optimizer:{learningRate:1e-4,maxSteps:8,clipRange:.2,iterations:2,microbatchSize:1,gradientAccumulationSteps:8,advantageEpsilon:1e-8,adamw:{name:"adamw",weightDecay:0,beta1:.9,beta2:.999,epsilon:1e-8}},loss:{method:"grpo",klBeta:.01},resourceLimits:{wallTimeMs:3960*1e3,maxRollouts:64,maxPayloadBytes:1e6},policyOptimization:null};return{...a,...r,lora:{...a.lora,...S(r?.lora)},rollout:{...a.rollout,...S(r?.rollout)},optimizer:{...a.optimizer,...S(r?.optimizer),adamw:{...a.optimizer.adamw,...S(S(r?.optimizer).adamw)}},loss:{...a.loss,...S(r?.loss)},resourceLimits:{...a.resourceLimits,...S(r?.resourceLimits)}}}h(re,"createModelTrainingRecipeDefaults");function pe(t){let e=x.parse(t.setup),r=typeof t.package=="object"&&t.package!==null&&"schemaVersion"in t.package&&t.package.schemaVersion==="openpond.tasksetPackage.v1"?Z(t.package):null,a=r?null:$(t.package),s=r??a,l=e.tasksetRef;if(!(l&&l.revision===s.taskset.revision&&(r?l.id===r.taskset.id&&l.contentHash===r.taskset.contentHash||l.id===r.taskset.metadata.sourceTasksetId&&l.contentHash===r.taskset.metadata.sourceTasksetHash:l.id===s.taskset.id)))return e;!e.evaluationTasksetRef&&s.taskset.tasks.some(p=>p.split==="frozen_eval"||p.split==="validation")&&(e.evaluationTasksetRef=e.tasksetRef);let o=e.baseModel;if(e.method==="sft"){if(e.recipe&&e.recipe.method!=="sft"||!o?.revision||!o.tokenizerRevision||!o.chatTemplateHash)return e;let p=s.taskset.tasks.filter(b=>b.split==="train");if(!p.length)return e;let E=te(e.recipe),k=S(E.dataset),H=e.evaluationTasksetRef?.id===e.tasksetRef?.id&&e.evaluationTasksetRef?.contentHash===e.tasksetRef?.contentHash;if(!k.validationSplit&&!H)throw new Error("Choose the optimizer validation split from the selected evaluation dataset.");return x.parse({...e,recipe:ee.parse({...E,baseModel:{id:o.modelId,revision:o.revision,tokenizerRevision:o.tokenizerRevision,chatTemplateHash:o.chatTemplateHash},dataset:{...k,trainSplit:"train",validationSplit:k.validationSplit??(s.taskset.tasks.some(b=>b.split==="frozen_eval")?"frozen_eval":"validation"),maxExamples:typeof k.maxExamples=="number"?Math.min(k.maxExamples,p.length):Math.min(1e3,p.length)}})})}if(e.method!=="grpo"||e.recipe&&e.recipe.method!=="grpo"||!o?.revision||!o.tokenizerRevision||!o.chatTemplateHash||e.recipe?.baseModel&&e.recipe.reward&&e.recipe.dataset)return e;let m=s.taskset.tasks.filter(p=>p.split==="train"),R=r?r.taskset.graders:_(a.rewardBinding,a.rewards),f=R.find(p=>p.rewardEligible&&p.weight>0),y=r?.environment??a?.executionResources?.environment??a?.execution?.environment;if(!m.length||!f||!y)return e;let M=re(e.recipe);return x.parse({...e,recipe:{...M,baseModel:{id:o.modelId,revision:o.revision,tokenizerRevision:o.tokenizerRevision,chatTemplateHash:o.chatTemplateHash},dataset:{trainSplit:"train",validationSplit:s.taskset.tasks.some(p=>p.split==="frozen_eval")?"frozen_eval":"validation",maxPromptTokens:4096,maxExamples:Math.min(1e3,m.length),selectionStrategy:"stable_hash_top_n",...S(M.dataset)},rollout:{maxTurns:s.taskset.environment.kind==="text"?1:15,...S(M.rollout)},reward:{graderId:f.id,graderHash:D(R),environmentId:y.id,environmentVersion:String(y.revision),toolContractHash:D(s.taskset.tools),learnedPreference:S(M.reward).learnedPreference??null}}})}h(pe,"prepareModelTrainingDefaults");var T=n.string().trim().min(1).max(500),v=W,ae=n.object({schemaVersion:n.literal("openpond.modelStarter.v1"),id:T,revision:n.number().int().positive(),name:n.string().trim().min(1).max(200),description:n.string().trim().min(1).max(2e3),category:n.enum(["extraction","support","operations","knowledge","coding","writing","classification"]),taskset:v,taskDefinition:v,rewardBinding:v,rewards:n.array(v).min(1).max(100),assets:n.array(v).max(1e3),previewTaskIds:n.array(T).min(1).max(8),startingModel:V,supportedMethods:n.array(I).min(1).max(8),defaultMethod:I,provenance:n.object({author:n.string().min(1).max(200),license:n.string().min(1).max(200),sourceDescription:n.string().min(1).max(2e3)}).strict(),evidence:n.object({verifierFixtures:v.nullable(),baseline:v.nullable(),training:v.nullable(),evaluation:v.nullable()}).strict()}).strict().superRefine((t,e)=>{t.supportedMethods.includes(t.defaultMethod)||e.addIssue({code:"custom",path:["defaultMethod"],message:"The default method must be supported by the package."});for(let r of["rewards","assets"])new Set(t[r].map(a=>a.id)).size!==t[r].length&&e.addIssue({code:"custom",path:[r],message:"Dependency identities must be unique."});new Set(t.previewTaskIds).size!==t.previewTaskIds.length&&e.addIssue({code:"custom",path:["previewTaskIds"],message:"Preview tasks must be unique."})}),ie=ae.safeExtend({contentHash:n.string().regex(/^[a-f0-9]{64}$/)}),oe=n.object({starter:ie,taskset:O,taskDefinition:K,rewardBinding:J,rewards:n.array(N).min(1).max(100),assets:n.array(Q).max(1e3),execution:q.optional()}).strict(),ne=n.object({schemaVersion:n.literal("openpond.modelStarterPrivacyReview.v1"),disposition:n.literal("synthetic_only"),reviewedBy:n.string().trim().min(1).max(500),reviewedAt:n.iso.datetime(),reviewedContentHash:n.string().regex(/^[a-f0-9]{64}$/),note:n.string().trim().min(1).max(2e3)}).strict();function se(t){let{contentHash:e,...r}=t.starter,{contentHash:a,...s}=t.taskset,{contentHash:l,...c}=r.taskset,o={...s.metadata},m=o.starterAuthoring;if(m&&typeof m=="object"&&!Array.isArray(m)){let{privacyReview:R,...f}=m;o.starterAuthoring=f}return g({...t,starter:{...r,taskset:c},taskset:{...s,metadata:o}}).contentHash}h(se,"modelStarterPrivacyContentHash");var de=n.object({profileId:T,modelId:T,name:n.string().trim().min(1).max(200),starter:v,startingModel:V,method:I,rewardBindingRef:v.nullable().optional(),recipe:X.nullable().optional(),managedGpuSelection:x.shape.managedGpuSelection,continuationAdapterMode:x.shape.continuationAdapterMode,learning:U.optional()}).strict(),ce=de.extend({schemaVersion:n.literal("openpond.modelStarterCreation.v1"),operationId:T}).strict();function le(t){return P(t,64*1024),ce.parse(t)}h(le,"parseModelStarterCreationRequest");function ue(t,e){let r=le(t),a=L(e);if(!C(r.starter,A(a.starter)))throw new Error("Starter creation references a different package revision.");if(!a.starter.supportedMethods.includes(r.method))throw new Error("The selected training method is not supported by this starter.");return{request:r,resolved:a}}h(ue,"validateModelStarterCreation");function he(t){let{starter:e,taskset:r,taskDefinition:a,rewardBinding:s,rewards:l}=L(t);return{starter:e,inputSchema:a.inputSchema,outputSchema:a.outputSchema,reward:{name:s.name,description:s.description,checks:s.sources.map(c=>{let o=l.find(m=>C(c.reward,A(m)));return{name:o.name,description:o.description,kind:o.implementation.kind,role:c.role,weight:c.weight,required:c.required,hardGate:c.hardGate}})},tasks:e.previewTaskIds.map(c=>{let o=r.tasks.find(m=>m.id===c);return{id:o.id,input:o.input,policyVisibleContext:o.policyVisibleContext}}),counts:{train:r.tasks.filter(c=>c.split==="train").length,validation:r.tasks.filter(c=>c.split==="validation").length,frozenEvaluation:r.tasks.filter(c=>c.split==="frozen_eval").length}}}h(he,"previewModelStarter");function L(t){P(t,16*1024*1024);let e=oe.parse(t),{starter:r,taskset:a,taskDefinition:s,rewardBinding:l,rewards:c,assets:o}=e,m=a.metadata.starterAuthoring;if(m&&typeof m=="object"&&!Array.isArray(m)&&"privacyReview"in m&&ne.parse(m.privacyReview).reviewedContentHash!==se(e))throw new Error("Starter privacy review differs from the published contents.");let R=[r,a,s,l,...c,...o];for(let i of R){let{contentHash:d,...u}=i;if(g(u).contentHash!==d)throw new Error(`Starter resource integrity failed: ${i.id}.`)}F(a),Y(e.execution,a,o);function f(i,d){if(!C(i,A(d)))throw new Error(`Starter dependency mismatch: ${i.id}.`)}if(h(f,"exact"),f(r.taskset,a),f(r.taskDefinition,s),f(r.rewardBinding,l),f(s.rewardBinding,l),l.recipeRef)throw new Error("Starter binding recipes must be resolved into a standalone binding before publication.");let{policy:y,environment:M,environmentRelease:p,tools:E,capabilities:k,verifierSetRelease:H}=a,b={policy:y,environment:M,...p?{environmentRelease:p}:{},tools:E,capabilities:k,...H?{verifierSetRelease:H}:{}};if(g(b).contentHash!==g(s.execution).contentHash)throw new Error("Starter execution context differs from its task format.");if(g({graders:a.graders}).contentHash!==g({graders:_(l,c)}).contentHash)throw new Error("Starter executable graders differ from its Reward binding.");for(let[i,d]of[[r.rewards,c],[r.assets,o]]){if(i.length!==d.length||new Set(d.map(u=>u.id)).size!==d.length)throw new Error("Starter dependency inventory mismatch.");for(let u of i){let w=d.find(z=>z.id===u.id);if(!w)throw new Error(`Starter dependency is missing: ${u.id}.`);f(u,w)}}for(let i of l.sources){let d=c.find(u=>u.id===i.reward.id);if(!d)throw new Error(`Starter Reward is missing: ${i.reward.id}.`);f(i.reward,d)}for(let i of c){let d=i.implementation,u=[...i.assets,..."verifierRef"in d?[d.verifierRef]:[],..."rubricRef"in d?[d.rubricRef]:[],..."inputContract"in d?[d.inputContract]:[]];for(let w of u){let z=o.find(me=>me.id===w.id);if(!z||g(z.asset).contentHash!==g(w).contentHash)throw new Error(`Starter Reward source is missing or changed: ${w.id}.`);if(w.visibility==="policy")throw new Error(`Starter Reward source must remain private to its evaluator: ${w.id}.`)}}if(new Set(a.tasks.map(i=>i.id)).size!==a.tasks.length)throw new Error("Starter task identities must be unique.");let G=new Map;for(let i of a.tasks){if(i.policyVisibleContext.instructions!==s.instructions)throw new Error(`Starter task does not carry its declared model instructions: ${i.id}.`);let d=g({input:i.input,context:i.policyVisibleContext}).contentHash,u=G.get(d);if(u&&u!==i.split)throw new Error("Starter policy input appears in multiple splits.");if(G.set(d,i.split),!B(s.inputSchema,i.input).valid)throw new Error(`Starter task input does not match its format: ${i.id}.`);if(i.expectedOutput!==null&&!B(s.outputSchema,i.expectedOutput).valid)throw new Error(`Starter expected output does not match its format: ${i.id}.`)}for(let i of r.previewTaskIds){let d=a.tasks.find(u=>u.id===i);if(!d||d.split==="frozen_eval")throw new Error(`Starter preview task is unavailable: ${i}.`)}return e}h(L,"validateResolvedModelStarter");export{pe as a,ie as b,ne as c,le as d,ue as e,he as f,L as g};