openpond 0.2.61 → 0.2.62

package/dist/chunks/{chunk-LNRFRAV5.js → chunk-INO67SEN.js}

@@ -1,3 +1,3 @@
 import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{a as le,b as Ve,c as Be}from"./chunk-2ITZW2CN.js";import{k as re,q as oe,t as se}from"./chunk-CQF6QZEL.js";import{d as ie,f as Pe,g as Ee,h as He,i as ze,l as D,n as Oe}from"./chunk-WFVOG5PU.js";import{a as Me}from"./chunk-VWJ5IYQP.js";import{i as B}from"./chunk-E67M5WGY.js";import{k as te,la as Ce,m as ne,qa as we,ta as ae,z as Ae}from"./chunk-M4ZRCRJM.js";import{x as Te}from"./chunk-XK7SPL63.js";import{h as je}from"./chunk-M6JRS5PA.js";import{f as V}from"./chunk-OE22TAXP.js";import{c as n,d as o,e as c,g as r,h as g}from"./chunk-VVNV4NNW.js";import{A as e}from"./chunk-PF56N7ZY.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),De=e.record(e.string(),e.unknown()).default({}),At=e.enum(["created","inspecting","awaiting_source_review","awaiting_mapping","awaiting_materialization_approval","materializing","validating","ready","cancelling","cancelled","failed"]),Le=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()}),Ct=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)}),wt=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()}),Mt=e.object({schemaVersion:e.literal("openpond.huggingFaceDatasetInspection.v1"),id:L,locator:Le,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(wt).min(1).max(1e4),columns:e.array(Ct).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:De}),Pt=e.enum(["row_id","cluster_id","prompt","messages","demonstration","chosen","rejected","expected_output","privileged_context","reward","feedback","tag","metadata"]),Et=e.object({sourcePath:e.string().trim().min(1).max(1e3),target:Pt,transform:e.enum(["identity","string","json","messages","numeric","math_final_answer"]),policy:e.enum(["visible","privileged","metadata"]),required:e.boolean()}),qe=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(Et).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}),Ht=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()}),Ue=e.object({schemaVersion:e.literal("openpond.datasetImportJob.v1"),id:L,profileId:L,sourceKind:e.enum(["huggingface","uploaded_file"]),status:At,locator:Le.nullable(),inspection:Mt.nullable(),mapping:qe.nullable(),progress:Ht,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:De}),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:qe,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),Ne=e.record(e.string(),e.unknown()).default({}),$e=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:Ne}),Ot=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)}),Vt=e.object({tactic:Me,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:$e,fingerprint:e.string().trim().min(1).max(512),title:e.string().trim().min(1).max(500),summary:e.string().trim().min(1).max(1e4),workflowSignature:e.string().trim().min(1).max(2e3),evidence:e.array(zt).min(1).max(1e5),scorecard:Ot,recommendation:Vt,mergedIntoId:y.nullable(),createdAt:w,updatedAt:w,metadata:Ne}),q=e.object({schemaVersion:e.literal("openpond.taskMinerConfig.v1"),enabled:e.boolean().default(!1),localOnly:e.boolean().default(!0),observationWindowDays:e.number().int().min(1).max(365).default(30),minimumRecurrence:e.number().int().min(2).max(100).default(3),clustering:e.literal("hybrid_deterministic_first").default("hybrid_deterministic_first"),consentRequired:e.boolean().default(!0)}),ja=e.object({items:e.array(ce),generatedAt:w,config:q}),Ta=e.object({status:$e.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()}),Fe=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"]),Dt=e.enum(["draft","ready","queued","running","candidate","accepted","rejected","no_signal","failed","cancelled"]),Ge=e.enum(["seed","daily_residual","rank_candidate","weekly_rollup","full_refresh"]),Lt=e.enum(["base_model","previous_release","seed_release","accepted_daily_head","accepted_seed"]),qt=e.enum(["seed_taskset","nightly_selection","daily_cohort_union","eligible_task_pool"]),Ut=e.enum(["live_client","evaluation_attempt","replay_evidence","manual"]),Nt=e.object({id:n,ordinal:e.number().int().nonnegative(),label:e.string().trim().min(1).max(100),role:Ge,parentRule:Lt,taskSource:qt,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 ${f}.`}),t.ordinal===0&&t.role!=="seed"&&i.addIssue({code:"custom",path:["role"],message:"The first series entry must be the seed."})}),$t=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(),Je=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(Nt).min(1).max(1e3),scheduleSealedAt:c.nullable(),advancementPolicy:$t,executionPolicy:e.object({startWhenReady:e.literal(!1)}).strict(),acceptedSeedEntryId:n.nullable(),acceptedDailyHeadEntryId:n.nullable(),promotedBindingId:n.nullable(),createdBy:n,createdAt:c,updatedAt:c}).strict().superRefine((t,i)=>{[...t.schedule].sort((u,j)=>u.ordinal-j.ordinal).some((u,j)=>u.ordinal!==j)&&i.addIssue({code:"custom",path:["schedule"],message:"Schedule ordinals must be unique and contiguous from zero."}),(new Set(t.schedule.map(u=>u.id)).size!==t.schedule.length||new Set(t.schedule.map(u=>u.label)).size!==t.schedule.length)&&i.addIssue({code:"custom",path:["schedule"],message:"Schedule ids and labels must be unique."}),t.residualProfile.maximumEnabledRank>t.residualProfile.serializedEnvelopeRank&&i.addIssue({code:"custom",path:["residualProfile"],message:"Maximum enabled rank cannot exceed the serialized envelope."});let l=Object.values(t.evaluationTasksets);new Set(l.map(u=>`${u.id}:${u.revision}:${u.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(u=>u.role==="seed");if(!f||t.schedule.filter(u=>u.role==="seed").length!==1){i.addIssue({code:"custom",path:["schedule"],message:"A series requires exactly one seed entry."});return}let h=t.schedule.filter(u=>u.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(u=>u.trainableRank)).size!==h.length&&i.addIssue({code:"custom",path:["schedule"],message:"Causal rank candidates must use distinct trainable ranks."}),[f.trainableRank+t.schedule.filter(u=>u.role==="daily_residual").reduce((u,j)=>u+j.trainableRank,0),...t.schedule.filter(u=>u.role==="weekly_rollup").map(u=>f.trainableRank+u.trainableRank),...t.schedule.filter(u=>u.role==="rank_candidate").map(u=>f.trainableRank+u.trainableRank),...t.schedule.filter(u=>u.role==="full_refresh").map(u=>u.trainableRank)].some(u=>u>t.residualProfile.maximumEnabledRank)&&i.addIssue({code:"custom",path:["schedule"],message:"A declared branch exceeds the maximum enabled rank."})}),We=e.object({source:Ut,taskIds:e.array(n).min(1).max(1e5),observedFrom:c,observedTo:c,reviewedAt:c,reviewedBy:n,sourceTaskset:g}).strict(),Ft=We.extend({derivedFamilyKeys:e.array(n).min(1).max(1e5),familyDerivation:e.object({algorithm:n,sourceTasksetHash:o}).strict(),selectionHash:o}).strict(),ue=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()]),Gt=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."})}),Jt=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(),Wt=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(),Yt=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(),Ye=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:Ge,branch:e.enum(["daily","rank_candidate","weekly_rollup","full_refresh"]),status:Dt,parent:ue,taskset:g,sourceTasksets:e.array(g).min(1).max(1e3),taskSelection:Ft.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(Gt).min(1).max(1024),attemptOrdinal:e.number().int().positive().default(1),priorRunAttempts:e.array(Yt).max(100).default([]),trainingPlanId:n.nullable(),modelRunId:n.nullable(),modelVersionId:n.nullable(),evaluations:e.array(Wt).max(1e5),decision:Jt.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:We.nullable(),expectedSeriesRevision:e.number().int().positive()}).strict();var M=n.nullable(),C=e.number().int().nonnegative(),Kt=e.string().regex(/^[a-f0-9]{40}$/),Qt=e.object({name:e.string().trim().min(1).max(160),startedAt:c,completedAt:c,durationMs:e.number().nonnegative(),clock:e.enum(["monotonic","provider","wall"]),outcome:e.enum(["succeeded","failed","cancelled"])}).strict(),Ke=e.object({schemaVersion:e.literal("openpond.correlatedTelemetryReceipt.v1"),stage:e.enum(["training","evaluation","deployment","inference"]),correlation:e.object({modelRunId:M,modelVersionId:M,policyVersion:C.nullable(),taskId:M,rolloutGroupId:M,providerResourceId:M,deploymentId:M,inferenceRequestId:M}).strict(),spans:e.array(Qt).max(1e4),usage:e.object({promptTokens:C.nullable(),generatedTokens:C.nullable(),gpuSeconds:e.number().nonnegative().nullable(),workerActiveSeconds:e.number().nonnegative().nullable(),optimizerSteps:C.nullable(),rolloutGroups:C.nullable(),successfulTrajectories:C.nullable(),failedTrajectories:C.nullable(),peakGpuMemoryBytes:C.nullable(),peakGpuUtilizationPercent:e.number().min(0).max(100).nullable()}).strict(),resource:e.object({provider:n,resourceIds:e.array(n).max(1e3),gpuType:e.string().trim().min(1).max(300).nullable(),gpuCount:C.nullable(),baseProfileId:M,baseRepository:e.string().trim().min(1).max(512).nullable(),baseRevision:Kt.nullable(),adapterContentHash:o.nullable()}).strict(),cost:e.object({currency:e.literal("USD"),providerReportedUsd:e.number().nonnegative().nullable(),quotedHourlyUsd:e.number().nonnegative().nullable(),estimatedUsd:e.number().nonnegative().nullable(),methodologyVersion:e.string().trim().min(1).max(160),pricingInputs:e.record(e.string().trim().min(1).max(160),e.number().finite()),unitEstimates:e.record(e.string().trim().min(1).max(160),e.number().nonnegative())}).strict(),recordedAt:c,contentHash:o}).strict();var Zt=e.enum(["base_reference","lora_adapter"]),de=e.object({baseModel:e.object({source:e.literal("huggingface"),repoId:n,revision:e.string().regex(/^[a-f0-9]{40}$/),configHash:o,tokenizerHash:o,licenseId:n,gated:e.boolean()}).strict(),processor:e.object({repository:n,revision:e.string().regex(/^[a-f0-9]{40}$/),configHash:o}).strict()}).strict().superRefine((t,i)=>{(t.baseModel.repoId!==t.processor.repository||t.baseModel.revision!==t.processor.revision)&&i.addIssue({code:"custom",path:["processor"],message:"Reward Model processor must match the pinned scorer base model."})}),et=e.object({schemaVersion:e.literal("openpond.modelVersion.v1"),id:n,modelId:n,profileId:n,version:e.number().int().nonnegative(),kind:Zt,status:e.enum(["available","failed"]),baseModel:D,taskset:g,comparisonSeriesEntry:A.nullable().optional(),evaluationTasksetRef:g.nullable().optional(),releaseGraph:e.object({resolvedBundleHash:o,profileRelease:g,harnessRelease:r,agentRelease:r.nullable(),grader:e.object({id:n,contentHash:o}).strict()}).strict(),artifactLineageId:n.nullable(),adapterStatus:e.enum(["not_trained","trained"]),createdAt:c,contentHash:o}).strict().superRefine((t,i)=>{t.kind==="base_reference"&&(t.version!==0||t.adapterStatus!=="not_trained"||t.artifactLineageId!==null)&&i.addIssue({code:"custom",message:"A base-reference Model Version must be version 0 without adapter lineage."}),t.kind==="lora_adapter"&&(t.version===0||t.adapterStatus!=="trained"||!t.artifactLineageId)&&i.addIssue({code:"custom",message:"A LoRA Model Version must be version 1+ with adapter lineage."})}),tt=e.object({schemaVersion:e.literal("openpond.rewardModelVersion.v1"),id:n,modelId:n,profileId:n,version:e.number().int().positive(),role:e.literal("reward"),status:e.enum(["available","failed"]),scope:e.enum(["synthetic_smoke","human_preference"]),baseModel:D,runtime:de.nullable().default(null),taskset:g,preferenceDatasetRelease:r,releaseGraph:e.object({resolvedBundleHash:o,profileRelease:g,harnessRelease:r,grader:e.object({id:n,contentHash:o}).strict()}).strict(),artifacts:e.object({checkpoint:r.extend({objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),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(),Xt=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(),nt=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:Xt.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."})}),en=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:Ke.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(),Qe=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(),Ze=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(),at=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:Ze,heldOut:Ze}).strict()}).strict(),be=e.object({baseline:P,adaptation:P,candidateAdaptation:P,candidate:P,refiner:P,grader:P}).strict(),tn=e.object({usage:be,observedSpendUsd:e.number().nonnegative(),attempts:e.array(pe).max(1e4)}).strict(),nn=e.object({id:n,contentHash:o,artifactPath:e.string().trim().min(1).max(2e3)}).strict(),an=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:at.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:Qe,baseCommit:Qe}).strict().nullable(),contentHash:o}).strict(),rn=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(),Xe=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."})}),on=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:Xe.nullable()}).strict(),judge:e.object({judgeModel:V,judgeRelease:r,rubricRelease:r,calibrationRelease:r,score:e.number().min(0).max(100),scoreCi95:Xe,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."})}),sn=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(),ln=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:ue,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(),mn=e.union([sn,ln]),it=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:mn.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:tn.nullable().default(null),advancedAccounting:se.optional(),evidenceSnapshot:nn.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([en,an,rn,on,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 Tt=new Map(s.filter(R=>R.phase===S).map(R=>[R.taskId,R]));return s.filter(R=>R.phase===T).flatMap(R=>{let ee=Tt.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"),u=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:at.parse({target:"all_tasks_lower",...u,baselineTokens:j,refinedTokens:ke,tokenDelta:Re,tokenDeltaPercent:j>0?Re/j*100:null,complete:Ie,passed:Ie&&u.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 cn=e.enum(["measuring","up_to_date","needs_attention"]),un=e.enum(["fixed","retained","regressed","unresolved","unavailable"]),rt=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(),dn=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(),ot=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:rt.nullable(),candidateAttempt:rt.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(dn).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:cn,evidenceReasons:e.array(n).max(100),projectedAt:c}).strict();var $=e.discriminatedUnion("kind",[e.object({kind:e.literal("base_model"),id:n,revision:e.string().trim().min(1).max(500),label:e.string().trim().min(1).max(300)}).strict(),e.object({kind:e.literal("model_version"),id:n,label:e.string().trim().min(1).max(300)}).strict(),e.object({kind:e.literal("captured_model"),id:n,label:e.string().trim().min(1).max(300)}).strict()]),U=e.union([e.object({source:e.literal("evaluation_run").default("evaluation_run"),evaluationRunId:n,attemptId:n,modelLabel:e.string().trim().min(1).max(300).nullable().default(null),modelVersionId:n.nullable().default(null),reward:e.number().finite().nullable(),components:e.record(e.string(),e.number().finite()).default({})}).strict(),e.object({source:e.literal("imported_response"),modelLabel:e.string().trim().min(1).max(300),modelVersionId:n.nullable().default(null),response:e.unknown().refine(t=>t!==void 0,"An imported model response is required."),reward:e.number().finite().nullable().default(null),components:e.record(e.string(),e.number().finite()).default({})}).strict()]),pn=e.union([e.object({taskId:n,target:$,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(pn).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=>`${l.taskId}:${bn(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."})}),st=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=>mt(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."})}),lt=e.object({schemaVersion:e.literal("openpond.continualBenchIssueReview.v1"),id:n,seriesId:n,scheduleEntryId:n,passLabel:n,packet:Ve,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 mt(t){return`${t.kind}:${t.id}`}v(mt,"responseTargetKey");function bn(t){return typeof t=="string"?t:mt(t)}v(bn,"observedTargetKey");var I=e.string().trim().min(1).max(240),_=e.string().trim().min(8).max(256),F=e.enum(["grpo","dapo","gspo-token"]),H=e.object({id:I,revision:e.string().trim().min(1).max(256),tokenizerRevision:e.string().trim().min(1).max(256),chatTemplateHash:_}),ct=e.object({maxRollouts:e.number().int().positive().max(1e6),maxEnvironmentExecutions:e.number().int().positive().max(1e6),maxInputTokens:e.number().int().positive(),maxOutputTokens:e.number().int().positive(),maxOptimizerSteps:e.number().int().positive().max(1e6),wallTimeMs:e.number().int().positive(),maximumCostUsd:e.number().nonnegative().nullable()}),ut=e.object({method:e.literal("grpo"),groupSize:e.number().int().min(2).max(128),normalization:e.literal("group_standardized"),advantageEpsilon:e.number().positive().max(.01).default(1e-8),loss:F,clipRange:e.number().positive().max(1).default(.2),iterations:e.number().int().min(2).max(16).default(2),microbatchSize:e.number().int().positive().max(128).default(1),gradientAccumulationSteps:e.number().int().positive().max(128).default(1),adamw:B}),G=e.object({method:e.literal("ppo"),valueModel:H,gamma:e.number().min(0).max(1),gaeLambda:e.number().min(0).max(1),policyClip:e.number().positive().max(1),valueClip:e.number().positive().max(10),valueLossCoefficient:e.number().nonnegative().max(10),ppoEpochs:e.number().int().positive().max(100),minibatchSize:e.number().int().positive().max(1e5)}),dt=e.discriminatedUnion("method",[ut,G]),J=e.object({rewardModelVersion:e.object({id:I,contentHash:_}).strict(),qualificationReport:e.object({id:I,contentHash:_}).strict().nullable().default(null),evaluationReferences:e.array(e.object({id:I,contentHash:_}).strict()).max(1e3).default([]),checkpoint:e.object({id:I,contentHash:_,objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),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:de,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:ct,checkpointEverySteps:e.number().int().positive(),seed:e.number().int(),evaluationSplit:e.literal("frozen_eval"),optimizer:dt});var a=e.string().trim().min(1).max(240),b=e.string().trim().min(1),m=e.string().trim().min(8).max(256),z=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)}),pt=he.parse({}),bt=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 gn(t){return t.state==="ready"&&t.canonicalArtifactState==="promotable"&&t.canonicalDeploymentState==="ready"&&t.customerBindingAllowed&&!!t.teamId&&!!t.canonicalArtifactId&&!!t.canonicalDeploymentId}v(gn,"managedAdapterProjectionReady");function ji(t){let i=t.managedServing;return i?gn(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",ht="openpond.crossSystemTools.v1",hn=["search_crm","query_billing","search_support","run_python"],fn=e.enum(["train","validation","frozen_eval"]),vn=e.enum(["easy","medium","hard"]),xn=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:fn,difficulty:vn,scenarioProfile:xn.optional()});var fe={anyOf:[{type:"string",minLength:1,maxLength:256},{type:"null"}],default:null},ve={type:"integer",minimum:1,maximum:50,default:25},ft=[{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=An({schemaVersion:ht,tools:ft}),Sn=[`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(" "),yn=15;function _n(t="auto"){return[Sn,"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: ${yn}.`,`Registered tool signatures: ${ft.map(kn).join("; ")}`].join(`
+import{a as le,b as Ve,c as Be}from"./chunk-2ITZW2CN.js";import{k as re,q as oe,t as se}from"./chunk-BEI4YULC.js";import{d as ie,f as Pe,g as Ee,h as He,i as ze,l as D,n as Oe}from"./chunk-NLQ2UPN2.js";import{a as Me}from"./chunk-547AUVKN.js";import{i as B}from"./chunk-IWK3RTLR.js";import{k as te,la as Ce,m as ne,qa as we,ta as ae,z as Ae}from"./chunk-PGO4FRHU.js";import{x as Te}from"./chunk-TVTFEGTU.js";import{h as je}from"./chunk-ABF3W2MO.js";import{f as V}from"./chunk-OE22TAXP.js";import{c as n,d as o,e as c,g as r,h as g}from"./chunk-VVNV4NNW.js";import{A as e}from"./chunk-PF56N7ZY.js";import{a as v}from"./chunk-FTQLQJQV.js";var L=e.string().trim().min(1).max(240),N=e.string().trim().min(1),me=e.string().trim().min(8).max(256),De=e.record(e.string(),e.unknown()).default({}),At=e.enum(["created","inspecting","awaiting_source_review","awaiting_mapping","awaiting_materialization_approval","materializing","validating","ready","cancelling","cancelled","failed"]),Le=e.object({schemaVersion:e.literal("openpond.huggingFaceDatasetLocator.v1"),repositoryId:e.string().trim().regex(/^[^/\s]+\/[^/\s]+$/).max(500),repositoryUrl:e.string().url().max(2e3),requestedRevision:e.string().trim().min(1).max(256).nullable()}),Ct=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)}),wt=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()}),Mt=e.object({schemaVersion:e.literal("openpond.huggingFaceDatasetInspection.v1"),id:L,locator:Le,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(wt).min(1).max(1e4),columns:e.array(Ct).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:De}),Pt=e.enum(["row_id","cluster_id","prompt","messages","demonstration","chosen","rejected","expected_output","privileged_context","reward","feedback","tag","metadata"]),Et=e.object({sourcePath:e.string().trim().min(1).max(1e3),target:Pt,transform:e.enum(["identity","string","json","messages","numeric","math_final_answer"]),policy:e.enum(["visible","privileged","metadata"]),required:e.boolean()}),qe=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(Et).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}),Ht=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()}),Ue=e.object({schemaVersion:e.literal("openpond.datasetImportJob.v1"),id:L,profileId:L,sourceKind:e.enum(["huggingface","uploaded_file"]),status:At,locator:Le.nullable(),inspection:Mt.nullable(),mapping:qe.nullable(),progress:Ht,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:De}),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:qe,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),Ne=e.record(e.string(),e.unknown()).default({}),$e=e.enum(["nominated","needs_review","approved_for_creation","creating","needs_task_review","baselining","ready_for_export","rejected","dismissed","retired","blocked"]),zt=e.object({id:y,kind:e.enum(["repeated_success","accepted_correction","agent_action_recurrence","outcome_linked_change","frontier_cost","expert_label","runtime_feedback"]),sourceRefIds:e.array(y).min(1).max(1e4),occurredAt:w,signature:e.string().trim().min(1).max(2e3),summary:e.string().trim().min(1).max(5e3),confidence:e.number().min(0).max(1),consented:e.boolean(),metadata:Ne}),Ot=e.object({frequency:e.number().min(0).max(1),businessValue:e.number().min(0).max(1),frontierCost:e.number().min(0).max(1),signalQuality:e.number().min(0).max(1),verifiability:e.number().min(0).max(1),repeatability:e.number().min(0).max(1),privacyRisk:e.number().min(0).max(1),overall:e.number().min(0).max(1)}),Vt=e.object({tactic:Me,eligible:e.boolean(),reasons:e.array(e.string().trim().min(1).max(5e3)).min(1).max(100),blockers:e.array(e.string().trim().min(1).max(5e3)).max(100).default([]),requiredSignals:e.array(e.string().trim().min(1).max(500)).max(100).default([]),generatedBy:e.enum(["decision_table","baseline_reassessment"])}),ce=e.object({schemaVersion:e.literal("openpond.taskCandidate.v1"),id:y,profileId:y,status:$e,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:Ot,recommendation:Vt,mergedIntoId:y.nullable(),createdAt:w,updatedAt:w,metadata:Ne}),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:$e.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()}),Fe=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"]),Dt=e.enum(["draft","ready","queued","running","candidate","accepted","rejected","no_signal","failed","cancelled"]),Ge=e.enum(["seed","daily_residual","rank_candidate","weekly_rollup","full_refresh"]),Lt=e.enum(["base_model","previous_release","seed_release","accepted_daily_head","accepted_seed"]),qt=e.enum(["seed_taskset","nightly_selection","daily_cohort_union","eligible_task_pool"]),Ut=e.enum(["live_client","evaluation_attempt","replay_evidence","manual"]),Nt=e.object({id:n,ordinal:e.number().int().nonnegative(),label:e.string().trim().min(1).max(100),role:Ge,parentRule:Lt,taskSource:qt,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 ${f}.`}),t.ordinal===0&&t.role!=="seed"&&i.addIssue({code:"custom",path:["role"],message:"The first series entry must be the seed."})}),$t=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(),Je=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(Nt).min(1).max(1e3),scheduleSealedAt:c.nullable(),advancementPolicy:$t,executionPolicy:e.object({startWhenReady:e.literal(!1)}).strict(),acceptedSeedEntryId:n.nullable(),acceptedDailyHeadEntryId:n.nullable(),promotedBindingId:n.nullable(),createdBy:n,createdAt:c,updatedAt:c}).strict().superRefine((t,i)=>{[...t.schedule].sort((u,j)=>u.ordinal-j.ordinal).some((u,j)=>u.ordinal!==j)&&i.addIssue({code:"custom",path:["schedule"],message:"Schedule ordinals must be unique and contiguous from zero."}),(new Set(t.schedule.map(u=>u.id)).size!==t.schedule.length||new Set(t.schedule.map(u=>u.label)).size!==t.schedule.length)&&i.addIssue({code:"custom",path:["schedule"],message:"Schedule ids and labels must be unique."}),t.residualProfile.maximumEnabledRank>t.residualProfile.serializedEnvelopeRank&&i.addIssue({code:"custom",path:["residualProfile"],message:"Maximum enabled rank cannot exceed the serialized envelope."});let l=Object.values(t.evaluationTasksets);new Set(l.map(u=>`${u.id}:${u.revision}:${u.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(u=>u.role==="seed");if(!f||t.schedule.filter(u=>u.role==="seed").length!==1){i.addIssue({code:"custom",path:["schedule"],message:"A series requires exactly one seed entry."});return}let h=t.schedule.filter(u=>u.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(u=>u.trainableRank)).size!==h.length&&i.addIssue({code:"custom",path:["schedule"],message:"Causal rank candidates must use distinct trainable ranks."}),[f.trainableRank+t.schedule.filter(u=>u.role==="daily_residual").reduce((u,j)=>u+j.trainableRank,0),...t.schedule.filter(u=>u.role==="weekly_rollup").map(u=>f.trainableRank+u.trainableRank),...t.schedule.filter(u=>u.role==="rank_candidate").map(u=>f.trainableRank+u.trainableRank),...t.schedule.filter(u=>u.role==="full_refresh").map(u=>u.trainableRank)].some(u=>u>t.residualProfile.maximumEnabledRank)&&i.addIssue({code:"custom",path:["schedule"],message:"A declared branch exceeds the maximum enabled rank."})}),We=e.object({source:Ut,taskIds:e.array(n).min(1).max(1e5),observedFrom:c,observedTo:c,reviewedAt:c,reviewedBy:n,sourceTaskset:g}).strict(),Ft=We.extend({derivedFamilyKeys:e.array(n).min(1).max(1e5),familyDerivation:e.object({algorithm:n,sourceTasksetHash:o}).strict(),selectionHash:o}).strict(),ue=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()]),Gt=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."})}),Jt=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(),Wt=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(),Yt=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(),Ye=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:Ge,branch:e.enum(["daily","rank_candidate","weekly_rollup","full_refresh"]),status:Dt,parent:ue,taskset:g,sourceTasksets:e.array(g).min(1).max(1e3),taskSelection:Ft.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(Gt).min(1).max(1024),attemptOrdinal:e.number().int().positive().default(1),priorRunAttempts:e.array(Yt).max(100).default([]),trainingPlanId:n.nullable(),modelRunId:n.nullable(),modelVersionId:n.nullable(),evaluations:e.array(Wt).max(1e5),decision:Jt.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:We.nullable(),expectedSeriesRevision:e.number().int().positive()}).strict();var M=n.nullable(),C=e.number().int().nonnegative(),Kt=e.string().regex(/^[a-f0-9]{40}$/),Qt=e.object({name:e.string().trim().min(1).max(160),startedAt:c,completedAt:c,durationMs:e.number().nonnegative(),clock:e.enum(["monotonic","provider","wall"]),outcome:e.enum(["succeeded","failed","cancelled"])}).strict(),Ke=e.object({schemaVersion:e.literal("openpond.correlatedTelemetryReceipt.v1"),stage:e.enum(["training","evaluation","deployment","inference"]),correlation:e.object({modelRunId:M,modelVersionId:M,policyVersion:C.nullable(),taskId:M,rolloutGroupId:M,providerResourceId:M,deploymentId:M,inferenceRequestId:M}).strict(),spans:e.array(Qt).max(1e4),usage:e.object({promptTokens:C.nullable(),generatedTokens:C.nullable(),gpuSeconds:e.number().nonnegative().nullable(),workerActiveSeconds:e.number().nonnegative().nullable(),optimizerSteps:C.nullable(),rolloutGroups:C.nullable(),successfulTrajectories:C.nullable(),failedTrajectories:C.nullable(),peakGpuMemoryBytes:C.nullable(),peakGpuUtilizationPercent:e.number().min(0).max(100).nullable()}).strict(),resource:e.object({provider:n,resourceIds:e.array(n).max(1e3),gpuType:e.string().trim().min(1).max(300).nullable(),gpuCount:C.nullable(),baseProfileId:M,baseRepository:e.string().trim().min(1).max(512).nullable(),baseRevision:Kt.nullable(),adapterContentHash:o.nullable()}).strict(),cost:e.object({currency:e.literal("USD"),providerReportedUsd:e.number().nonnegative().nullable(),quotedHourlyUsd:e.number().nonnegative().nullable(),estimatedUsd:e.number().nonnegative().nullable(),methodologyVersion:e.string().trim().min(1).max(160),pricingInputs:e.record(e.string().trim().min(1).max(160),e.number().finite()),unitEstimates:e.record(e.string().trim().min(1).max(160),e.number().nonnegative())}).strict(),recordedAt:c,contentHash:o}).strict();var Zt=e.enum(["base_reference","lora_adapter"]),de=e.object({baseModel:e.object({source:e.literal("huggingface"),repoId:n,revision:e.string().regex(/^[a-f0-9]{40}$/),configHash:o,tokenizerHash:o,licenseId:n,gated:e.boolean()}).strict(),processor:e.object({repository:n,revision:e.string().regex(/^[a-f0-9]{40}$/),configHash:o}).strict()}).strict().superRefine((t,i)=>{(t.baseModel.repoId!==t.processor.repository||t.baseModel.revision!==t.processor.revision)&&i.addIssue({code:"custom",path:["processor"],message:"Reward Model processor must match the pinned scorer base model."})}),et=e.object({schemaVersion:e.literal("openpond.modelVersion.v1"),id:n,modelId:n,profileId:n,version:e.number().int().nonnegative(),kind:Zt,status:e.enum(["available","failed"]),baseModel:D,taskset:g,comparisonSeriesEntry:A.nullable().optional(),evaluationTasksetRef:g.nullable().optional(),releaseGraph:e.object({resolvedBundleHash:o,profileRelease:g,harnessRelease:r,agentRelease:r.nullable(),grader:e.object({id:n,contentHash:o}).strict()}).strict(),artifactLineageId:n.nullable(),adapterStatus:e.enum(["not_trained","trained"]),createdAt:c,contentHash:o}).strict().superRefine((t,i)=>{t.kind==="base_reference"&&(t.version!==0||t.adapterStatus!=="not_trained"||t.artifactLineageId!==null)&&i.addIssue({code:"custom",message:"A base-reference Model Version must be version 0 without adapter lineage."}),t.kind==="lora_adapter"&&(t.version===0||t.adapterStatus!=="trained"||!t.artifactLineageId)&&i.addIssue({code:"custom",message:"A LoRA Model Version must be version 1+ with adapter lineage."})}),tt=e.object({schemaVersion:e.literal("openpond.rewardModelVersion.v1"),id:n,modelId:n,profileId:n,version:e.number().int().positive(),role:e.literal("reward"),status:e.enum(["available","failed"]),scope:e.enum(["synthetic_smoke","human_preference"]),baseModel:D,runtime:de.nullable().default(null),taskset:g,preferenceDatasetRelease:r,releaseGraph:e.object({resolvedBundleHash:o,profileRelease:g,harnessRelease:r,grader:e.object({id:n,contentHash:o}).strict()}).strict(),artifacts:e.object({checkpoint:r.extend({objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),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(),Xt=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(),nt=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:Xt.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."})}),en=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:Ke.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(),Qe=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(),Ze=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(),at=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:Ze,heldOut:Ze}).strict()}).strict(),be=e.object({baseline:P,adaptation:P,candidateAdaptation:P,candidate:P,refiner:P,grader:P}).strict(),tn=e.object({usage:be,observedSpendUsd:e.number().nonnegative(),attempts:e.array(pe).max(1e4)}).strict(),nn=e.object({id:n,contentHash:o,artifactPath:e.string().trim().min(1).max(2e3)}).strict(),an=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:at.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:Qe,baseCommit:Qe}).strict().nullable(),contentHash:o}).strict(),rn=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(),Xe=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."})}),on=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:Xe.nullable()}).strict(),judge:e.object({judgeModel:V,judgeRelease:r,rubricRelease:r,calibrationRelease:r,score:e.number().min(0).max(100),scoreCi95:Xe,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."})}),sn=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(),ln=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:ue,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(),mn=e.union([sn,ln]),it=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:mn.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:tn.nullable().default(null),advancedAccounting:se.optional(),evidenceSnapshot:nn.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([en,an,rn,on,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 Tt=new Map(s.filter(R=>R.phase===S).map(R=>[R.taskId,R]));return s.filter(R=>R.phase===T).flatMap(R=>{let ee=Tt.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"),u=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:at.parse({target:"all_tasks_lower",...u,baselineTokens:j,refinedTokens:ke,tokenDelta:Re,tokenDeltaPercent:j>0?Re/j*100:null,complete:Ie,passed:Ie&&u.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 cn=e.enum(["measuring","up_to_date","needs_attention"]),un=e.enum(["fixed","retained","regressed","unresolved","unavailable"]),rt=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(),dn=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(),ot=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:rt.nullable(),candidateAttempt:rt.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(dn).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:cn,evidenceReasons:e.array(n).max(100),projectedAt:c}).strict();var $=e.discriminatedUnion("kind",[e.object({kind:e.literal("base_model"),id:n,revision:e.string().trim().min(1).max(500),label:e.string().trim().min(1).max(300)}).strict(),e.object({kind:e.literal("model_version"),id:n,label:e.string().trim().min(1).max(300)}).strict(),e.object({kind:e.literal("captured_model"),id:n,label:e.string().trim().min(1).max(300)}).strict()]),U=e.union([e.object({source:e.literal("evaluation_run").default("evaluation_run"),evaluationRunId:n,attemptId:n,modelLabel:e.string().trim().min(1).max(300).nullable().default(null),modelVersionId:n.nullable().default(null),reward:e.number().finite().nullable(),components:e.record(e.string(),e.number().finite()).default({})}).strict(),e.object({source:e.literal("imported_response"),modelLabel:e.string().trim().min(1).max(300),modelVersionId:n.nullable().default(null),response:e.unknown().refine(t=>t!==void 0,"An imported model response is required."),reward:e.number().finite().nullable().default(null),components:e.record(e.string(),e.number().finite()).default({})}).strict()]),pn=e.union([e.object({taskId:n,target:$,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(pn).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=>`${l.taskId}:${bn(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."})}),st=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=>mt(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."})}),lt=e.object({schemaVersion:e.literal("openpond.continualBenchIssueReview.v1"),id:n,seriesId:n,scheduleEntryId:n,passLabel:n,packet:Ve,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 mt(t){return`${t.kind}:${t.id}`}v(mt,"responseTargetKey");function bn(t){return typeof t=="string"?t:mt(t)}v(bn,"observedTargetKey");var I=e.string().trim().min(1).max(240),_=e.string().trim().min(8).max(256),F=e.enum(["grpo","dapo","gspo-token"]),H=e.object({id:I,revision:e.string().trim().min(1).max(256),tokenizerRevision:e.string().trim().min(1).max(256),chatTemplateHash:_}),ct=e.object({maxRollouts:e.number().int().positive().max(1e6),maxEnvironmentExecutions:e.number().int().positive().max(1e6),maxInputTokens:e.number().int().positive(),maxOutputTokens:e.number().int().positive(),maxOptimizerSteps:e.number().int().positive().max(1e6),wallTimeMs:e.number().int().positive(),maximumCostUsd:e.number().nonnegative().nullable()}),ut=e.object({method:e.literal("grpo"),groupSize:e.number().int().min(2).max(128),normalization:e.literal("group_standardized"),advantageEpsilon:e.number().positive().max(.01).default(1e-8),loss:F,clipRange:e.number().positive().max(1).default(.2),iterations:e.number().int().min(2).max(16).default(2),microbatchSize:e.number().int().positive().max(128).default(1),gradientAccumulationSteps:e.number().int().positive().max(128).default(1),adamw:B}),G=e.object({method:e.literal("ppo"),valueModel:H,gamma:e.number().min(0).max(1),gaeLambda:e.number().min(0).max(1),policyClip:e.number().positive().max(1),valueClip:e.number().positive().max(10),valueLossCoefficient:e.number().nonnegative().max(10),ppoEpochs:e.number().int().positive().max(100),minibatchSize:e.number().int().positive().max(1e5)}),dt=e.discriminatedUnion("method",[ut,G]),J=e.object({rewardModelVersion:e.object({id:I,contentHash:_}).strict(),qualificationReport:e.object({id:I,contentHash:_}).strict().nullable().default(null),evaluationReferences:e.array(e.object({id:I,contentHash:_}).strict()).max(1e3).default([]),checkpoint:e.object({id:I,contentHash:_,objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),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:de,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:ct,checkpointEverySteps:e.number().int().positive(),seed:e.number().int(),evaluationSplit:e.literal("frozen_eval"),optimizer:dt});var a=e.string().trim().min(1).max(240),b=e.string().trim().min(1),m=e.string().trim().min(8).max(256),z=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)}),pt=he.parse({}),bt=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 gn(t){return t.state==="ready"&&t.canonicalArtifactState==="promotable"&&t.canonicalDeploymentState==="ready"&&t.customerBindingAllowed&&!!t.teamId&&!!t.canonicalArtifactId&&!!t.canonicalDeploymentId}v(gn,"managedAdapterProjectionReady");function ji(t){let i=t.managedServing;return i?gn(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",ht="openpond.crossSystemTools.v1",hn=["search_crm","query_billing","search_support","run_python"],fn=e.enum(["train","validation","frozen_eval"]),vn=e.enum(["easy","medium","hard"]),xn=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:fn,difficulty:vn,scenarioProfile:xn.optional()});var fe={anyOf:[{type:"string",minLength:1,maxLength:256},{type:"null"}],default:null},ve={type:"integer",minimum:1,maximum:50,default:25},ft=[{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=An({schemaVersion:ht,tools:ft}),Sn=[`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(" "),yn=15;function _n(t="auto"){return[Sn,"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: ${yn}.`,`Registered tool signatures: ${ft.map(kn).join("; ")}`].join(`
 `)}v(_n,"crossSystemLocalToolSystemPrompt");var Ei=_n();function kn(t){let i=K(t.parameters),s=K(i.properties),l=vt(i.required),f=Object.fromEntries(l.map(h=>[h,Y(s[h])]));return`${t.name}(${JSON.stringify(f)})`}v(kn,"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=vt(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 vt(t){return xe(t).filter(i=>typeof i=="string")}v(vt,"stringValues");var Q=e.enum(hn),Rn=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)})]),xt=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(Rn).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({})}),St=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)}),In=e.object({id:e.string().trim().min(1).max(240),type:e.literal("function"),function:e.object({name:Q,arguments:e.string().max(2e4)})}),yt=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(In).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(yt).min(3).max(1e3)}),jn=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(yt).min(3).max(1e3)}),Tn=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()}),zi=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:Tn.nullable(),tasks:e.array(jn).min(1).max(100)});function An(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`${ht}:${gt(s)}${gt(l)}`}v(An,"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 gt(t){return(t>>>0).toString(16).padStart(8,"0")}v(gt,"hex");var Cn=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:z}),wn=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}),Mn=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)}),_t=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([xt,Cn,wn]).nullable(),verifier:e.union([St,Mn]).nullable(),providerStatus:e.record(e.string(),e.unknown()).default({}),receivedAt:b,startedAt:b.nullable(),completedAt:b.nullable(),updatedAt:b});function d(t){let i;return new e.ZodObject({type:"object",get shape(){return i??=t()}})}v(d,"deferredObject");var E=e.enum(["sft","dpo","grpo","ppo","sdft","opd","opsd","sdpo"]),X=e.enum(["lora","full"]),O=e.enum(["openpond_managed"]),Pn=d(()=>({schemaVersion:e.literal("openpond.sftRecipe.v1"),method:e.literal("sft"),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"]),completionOnly:e.boolean(),maxSequenceLength:e.number().int().positive().max(32768),maxExamples:e.number().int().positive().max(1e5).default(1e3),selectionStrategy:e.literal("stable_hash_top_n").default("stable_hash_top_n"),selectionSeed:e.number().int().default(17)}),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)}),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()}),resourceLimits:e.object({cpuThreads:e.number().int().positive().max(256),memoryBytes:e.number().int().positive(),wallTimeMs:e.number().int().positive()})})),cr=d(()=>({id:a,input:e.record(e.string(),e.unknown()),expectedOutput:e.record(e.string(),e.unknown()),tags:e.array(a).max(100).default([])})),ur=d(()=>({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)})),dr=d(()=>({id:a,input:e.record(e.string(),e.unknown()),tags:e.array(a).max(100).default([])})),pr=d(()=>({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=d(()=>({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:B}),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(10800).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"])}).strict().optional()})),Hn=d(()=>({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()})})),zn=d(()=>({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()})})),On=d(()=>({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([Pn,Hn,En,zn,On]),Bn=d(()=>({schemaVersion:e.literal("openpond.trainingDestinationCapabilities.v1"),destinationId:O,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})),Dn=d(()=>({destinationId:O,available:e.boolean(),methods:e.array(E),parameterizations:e.array(X),nonProduction:e.boolean(),unavailableReason:e.string().trim().min(1).max(5e3).nullable()})),Ln=d(()=>({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:D,available:e.boolean(),nonProduction:e.boolean(),unavailableReason:e.string().trim().min(1).max(5e3).nullable(),methods:e.array(E),executionOptions:e.array(Dn).min(1)})),qn=d(()=>({code:a,severity:e.enum(["warning","error"]),path:e.string().trim().max(2e3).nullable(),message:e.string().trim().min(1).max(5e3)})),Un=d(()=>({schemaVersion:e.literal("openpond.trainingCompatibility.v1"),compatible:e.boolean(),destinationId:O,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"])]),br=d(()=>({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:O.nullable().default(null)})),gr=e.enum(["small","standard","custom","small_experiment"]);var $n=Te,It=d(()=>({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:O,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=d(()=>({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=d(()=>({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})),hr=d(()=>({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})),fr=d(()=>({schemaVersion:e.literal("openpond.trainingPreparedStart.v1"),plan:It,bundle:_e,approvalActor:a.nullable(),preparedAt:b})),vr=d(()=>({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:O,modelId:a,method:E,parameterization:X,maximumCostUsd:e.number().nonnegative().nullable(),approvedBy:a,approvedAt:b})),xr=d(()=>({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:D,method:E})).strict(),Gn=e.enum(["queued","starting","running","cancelling","cancelled","succeeded","failed","reconciling"]),jt=d(()=>({schemaVersion:e.literal("openpond.trainingJob.v1"),id:a,planId:a,bundleHash:m,approvalId:a,destinationId:O,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:z})),Jn=d(()=>({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:z})),Wn=d(()=>({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=d(()=>({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=d(()=>({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})),kt=d(()=>({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()})),Rt=d(()=>({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=d(()=>({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:Rt.nullable(),trainedGrade:Rt.nullable()})),Zn=d(()=>({schemaVersion:e.literal("openpond.trainingEvaluationSummary.v1"),jobId:a,tasksetId:a,base:kt,trained:kt,meanScoreDelta:e.number().min(-1).max(1).nullable(),examples:e.array(Qn).max(1e6)})),Sr=d(()=>({schemaVersion:e.literal("openpond.trainingRunDetail.v1"),job:jt,events:e.array(Jn),stepMetrics:e.array(Wn),policyMetrics:e.array(Yn).default([]),managedEvidence:Kn.nullable().default(null),evaluation:Zn.nullable(),generatedAt:b})),Xn=d(()=>({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:z})),ea=d(()=>({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(pt),managedServing:bt.nullable().default(null)})),ta=e.enum(["chat_manual","agent","extension","authoring_optimizer"]),na=d(()=>({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:z})),yr=d(()=>({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})),_r=d(()=>({schemaVersion:e.literal("openpond.tasksetOperationalState.v1"),tasksetId:a,attempts:e.array(Pe),artifacts:e.array(Ee),grades:e.array(He),evaluationResults:e.array(je).default([]),generatedAt:b})),kr=d(()=>({schemaVersion:e.literal("openpond.trainingState.v1"),profileId:a,sources:e.array(Ae),creations:e.array(Oe),tasksetDrafts:e.array(Ce).default([]),tasksets:e.array(ae),benchmarkRuns:e.array(e.custom()).default([]),benchmarkComparisons:e.array(e.custom()).default([]),datasetImports:e.array(Ue).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(Fe).default([]),modelProjects:e.array($n).default([]),modelVersions:e.array(et).default([]),modelRuns:e.array(it).default([]),comparisonSeries:e.array(Je).default([]),comparisonSeriesEntries:e.array(Ye).default([]),modelCurrencySnapshots:e.array(ot).default([]),continualBenchIssueReviews:e.array(lt).default([]),continualLearningDailyBatches:e.array(st).default([]),rewardModelVersions:e.array(tt).default([]),rewardModelRuns:e.array(nt).default([]),modelTasksets:e.array(ae).default([]),plans:e.array(It),bundles:e.array(_e),jobs:e.array(jt),artifacts:e.array(Xn),models:e.array(ea),rolloutReceipts:e.array(_t).default([]),modelBindings:e.array(na).default([]),destinations:e.array(Bn),baseModelCandidates:e.array(Ln).default([]),activityRevision:m.optional(),generatedAt:b}));export{Le as a,Mt as b,qe as c,Ue as d,ya as e,_a as f,Vt as g,ce as h,q as i,Ta as j,Aa as k,Fe as l,Dt as m,Je as n,Ft as o,Jt as p,Wt as q,A as r,Ye as s,Ea as t,et as u,tt as v,Xt as w,nt as x,an as y,on as z,it as A,ai as B,ot as C,$ as D,bi as E,st as F,lt as G,J as H,he as I,bt as J,gn as K,ji as L,ye as M,wi as N,ht as O,hn as P,Pi as Q,ft as R,Z as S,Ei as T,xt as U,St as V,yt as W,Hi as X,Tn as Y,zi as Z,_t as _,O as $,pr as aa,Vn as ba,Bn as ca,Ln as da,Un as ea,$n as fa,It as ga,_e as ha,hr as ia,fr as ja,vr as ka,xr as la,jt as ma,Jn as na,Wn as oa,Yn as pa,Kn as qa,Zn as ra,Sr as sa,Xn as ta,ea as ua,ta as va,na as wa,_r as xa};

package/dist/chunks/{chunk-E67M5WGY.js → chunk-IWK3RTLR.js}

@@ -1,2 +1,2 @@
 import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{Ab as me,Ib as ue,Nb as fe,Ob as he,Y as ee,ha as te,hb as ae,ib as se,kb as ie,lb as ne,mb as re,ta as K,tb as F,ub as J,v as X,vb as oe,wb as ce,xb as de,yb as le,za as U,zb as pe}from"./chunk-M4ZRCRJM.js";import{b as A,c as E,d as $,f as j,k as L,l as z,m as d}from"./chunk-VNJAAYNQ.js";import{N as O,aa as Z,ba as Y,d as B,e as Q}from"./chunk-4M3YW3EQ.js";import{Ya as N}from"./chunk-PBZCXPPC.js";import{A as n}from"./chunk-PF56N7ZY.js";import{a as S}from"./chunk-FTQLQJQV.js";var xe=n.object({id:A,purpose:n.string().trim().min(1).max(500),audience:n.enum(["orchestrator","environment","privileged_scorer","trainer","infrastructure"]),required:n.boolean(),ttlSeconds:n.number().int().positive().max(86400),scopes:n.array(n.string().trim().min(1).max(500)).max(100).default([])}).strict(),Te=n.object({declarationId:A,leaseRef:n.string().trim().min(8).max(1e3),audience:xe.shape.audience,expiresAt:$}).strict(),He=n.object({adapterId:A,placement:n.enum(["local","remote","colocated","provider_native"]),capabilityReceipt:E,runtimeVersion:n.string().trim().min(1).max(200),dataPlane:n.object({provider:A,dataPlaneId:A,cellId:A,runnerPoolId:A,runtimeImageDigest:n.string().regex(/^sha256:[a-f0-9]{64}$/),capabilityReceipt:E}).strict().nullable()}).strict(),Ee=n.object({adapterId:A,kind:n.enum(["local","ssh","managed","custom"]),deviceOrPool:n.string().trim().min(1).max(1e3),capabilityReceipt:E,provider:A.nullable()}).strict(),Ae=n.object({adapterId:A,workerVersion:n.string().trim().min(1).max(200),workerImageDigest:n.string().regex(/^sha256:[a-f0-9]{64}$/).nullable(),upstreamRevision:n.string().trim().min(1).max(500),capabilityReceipt:E}).strict(),be=n.object({schemaVersion:n.literal("openpond.harnessRunManifest.v1"),executionMode:n.literal("qualification").optional(),qualificationSamples:n.number().int().min(1).max(32).optional(),id:A,harnessRelease:j,datasetRelease:j,evidenceSets:n.array(j).max(1e4),model:n.object({source:n.string().trim().min(1).max(200),revision:n.string().trim().min(1).max(500),artifactHash:E.nullable(),tokenizerRevision:n.string().trim().min(1).max(500),chatTemplateHash:E}).strict(),recipe:n.object({method:n.string().trim().min(1).max(100),version:n.string().trim().min(1).max(200),configHash:E}).strict(),runtimeTarget:He,computeTarget:Ee,engine:Ae,resolvedBundleHash:E,secretLeaseRefs:n.array(Te).max(1e3),approval:n.object({approvalHash:E,approvedAt:$,maximumSpendUsd:n.number().nonnegative().nullable()}).strict(),createdAt:$}).strict(),we=be.extend({contentHash:E}).strict(),ke=n.object({schemaVersion:n.literal("openpond.resolvedTrainingBundle.v1"),executionMode:n.literal("qualification").optional(),qualificationSamples:n.number().int().min(1).max(32).optional(),projection:n.literal("trainer"),harnessRelease:j,datasetRelease:j,evidenceSetRelease:j.nullable(),files:n.array(n.object({path:n.string().trim().min(1).max(2e3),sha256:E,sizeBytes:n.number().int().nonnegative()}).strict()).max(1e5)}).strict(),Se=ke.extend({contentHash:E}).strict(),Ie=[{label:"OpenAI-style API key",pattern:/\bsk-[A-Za-z0-9_-]{16,}\b/g},{label:"OpenPond API key",pattern:/\bopk_[A-Za-z0-9_-]{12,}\b/g},{label:"private key",pattern:/-----BEGIN (?:RSA |EC |OPENSSH )?PRIVATE KEY-----/g},{label:"credential assignment",pattern:/\b(?:api[_-]?key|password|secret|token)["']?\s*[:=]\s*["']?[^\s"']{8,}/gi}],V=/\b[A-Z0-9._%+-]+@[A-Z0-9.-]+\.[A-Z]{2,}\b/gi,D=/\b(?:\+?1[-.\s]?)?\(?\d{3}\)?[-.\s]\d{3}[-.\s]\d{4}\b/g;function Me(e){let t=[],a=e;for(let s of Ie)s.pattern.lastIndex=0,s.pattern.test(a)&&t.push(s.label),s.pattern.lastIndex=0,a=a.replace(s.pattern,`[REDACTED ${s.label}]`);V.lastIndex=0;let r=V.test(a);V.lastIndex=0,r&&t.push("email address"),a=a.replace(V,"[REDACTED email]"),D.lastIndex=0;let o=D.test(a);return D.lastIndex=0,o&&t.push("phone number"),a=a.replace(D,"[REDACTED phone]"),{secretStatus:t.some(s=>!s.includes("email")&&!s.includes("phone"))?"blocked":"passed",piiStatus:r||o?"review":"passed",findings:t,redacted:a}}S(Me,"scanAndRedactEvidence");function qe(e){let t=new Map;return{tasks:e.tasks.map(({artifactRefs:r,...o})=>({...o,assets:r.map(s=>{if(s.visibility!=="policy")throw new Error("Learning task inputs must be policy-visible assets.");let i=e.assets?.find(m=>m.id===s.id),l=e.assetBytes?.get(s.id)??(i?new TextEncoder().encode(B(i,s)):void 0);if(!l||l.byteLength!==s.sizeBytes||z(l)!==s.contentHash)throw new Error(`Learning task input bytes are missing or changed: ${s.id}.`);let p=`learning/assets/${d({task:o.id,asset:s.id}).slice(0,40)}/${s.path}`;return t.set(p,new Uint8Array(l)),{id:s.id,sourceRefId:e.sourceId,artifactRef:p,fileName:s.path.split("/").at(-1),stagingPath:s.path,mediaType:s.mediaType,sha256:s.contentHash,sizeBytes:s.sizeBytes,split:o.split,metadata:{portableAsset:s}}}),...o.requiredOutputs?{requiredOutputs:o.requiredOutputs.map(s=>({...s,schemaRef:s.schemaRef?.id??null,maxBytes:s.maxBytes??void 0}))}:{}})),tasksetAssetBytes:t}}S(qe,"materializeLearningBatchAssets");function je(e){let t=Z(e),a=Y(t),r=t.environment.kind==="work";if(!r&&(t.environment.kind!=="text"||t.tools.length||t.tasks.some(c=>c.artifactRefs.length))||r&&t.tasks.some(c=>!c.requiredOutputs?.length))throw new Error("This learning batch requires a supported text or Work environment with declared outputs.");let o=e.source.id,s=qe({tasks:t.tasks,sourceId:o,assets:e.assets,assetBytes:e.assetBytes}),i={demonstrations:[],preferences:[],corrections:[],feedback:[],rewards:[],labels:[]},l=[];for(let c of a.admissions){let v=e.evidence.find(h=>h.id===c.evidence.id),f=e.decisions.find(h=>h.id===c.decision.id);e.batch.purpose==="reward_training"&&f.grade?.training.status==="scored"&&i.rewards.push({id:`reward-${c.taskId}`,taskId:c.taskId,kind:"reward",task:e.definition.instructions,rules:[{id:e.binding.id,points:1,condition:"Execute the released Reward binding with its declared normalization, weights and gates."}],otherwisePoints:0,executable:!0,approved:!0,confidence:1,sourceRefs:[o],artifactRef:e.binding.id,metadata:{rewardBinding:O(e.binding),evidence:c.evidence,decision:c.decision}}),c.supervisedTarget!==null&&i.demonstrations.push({kind:"demonstration",id:`demo-${c.taskId}`,taskId:c.taskId,sourceRefs:[o],artifactRef:c.decision.id,approved:!0,confidence:1,prompt:null,response:JSON.stringify(c.supervisedTarget),metadata:{evidence:c.evidence,decision:c.decision,targetGradeHash:f.targetGrade?.contentHash}});for(let[h,T,y]of[["observed",v.submission.observedOutput,f.grade],["target",f.approvedTarget,f.targetGrade]]){let k=_e(y);T===null||k===null||l.push({id:`fixture-${h}-${c.taskId}`,taskId:c.taskId,label:k.passed?"positive":"negative",output:T,infrastructureError:null,expectedPassed:k.passed,expectedRewardEligible:k.rewardEligible,metadata:{gradeReceiptHash:y.contentHash,evidence:c.evidence,decision:c.decision}})}}if(!l.length&&e.batch.purpose!=="reward_training")throw new Error("Grade at least one reviewed response before preparing this batch. Grader fixtures must come from actual receipts.");let p=he(e.binding,e.rewards,e.assets),m=[];for(let c of t.graders){if(c.kind!=="custom_verifier")continue;let v=e.assets?.find(h=>h.id===c.verifierRef.id);if(!v)throw new Error(`Reward ${c.id} is missing its immutable verifier source.`);let f=c.verifierRef.path;m.some(h=>h.path===f)||m.push({path:f,role:"verifier",content:B(v,c.verifierRef)})}let w=K.parse({schemaVersion:"openpond.taskset.v1",id:`learning-${d([e.profileId,e.batch.contentHash]).slice(0,40)}`,revision:1,profileId:e.profileId,profileRelease:null,createImproveRunId:null,name:`${e.definition.name} \xB7 ${e.batch.examples.length} approved examples`,objective:e.definition.instructions,purpose:"general",benchmark:null,status:"needs_review",sourceRefs:[e.source],datasetArtifact:null,policy:t.policy,environment:{protocolVersion:"openpond.taskEnvironment.v1",kind:r?"work":"chat",entrypoint:t.environment.entrypoint,stateful:t.environment.stateful,deterministicSeeds:t.environment.deterministicSeeds,toolNames:t.tools.map(c=>c.name),lifecycle:["create","reset","step","grade","cleanup"],defaultTimeoutMs:t.environment.defaultTimeoutMs,networkPolicy:t.environment.networkPolicy,metadata:{portableEnvironment:t.environment,portableTools:t.tools}},capabilities:{schemaVersion:"openpond.tasksetCapabilities.v1",taskKind:r?"single_agent":"chat",supportedSignals:i.demonstrations.length?["demonstration","reward"]:["reward"],compatibleMethods:e.batch.purpose==="supervised_training"?["sft"]:e.batch.purpose==="reward_training"?["grpo","ppo"]:["none"],rewardKinds:[...new Set(t.graders.map(c=>c.kind==="human"?"human":c.kind==="model_judge"?"model_judge":"deterministic"))],requiresTools:t.tools.length>0,requiresState:t.environment.stateful,requiresPrivilegedGrading:!0,environmentPlacements:["local","remote"],exportable:!0,portabilityBlockers:[]},tasks:s.tasks.map(c=>({...c,schemaVersion:"openpond.taskData.v1",sourceRefs:[o],metadata:{admission:a.admissions.find(v=>v.taskId===c.id),exampleOrigin:e.evidence.find(v=>v.id===c.id)?.correctionFeedbackId?"corrected":"extracted"}})),graders:p,graderFixtures:l,learningSignals:i,authoringProvenance:{schemaVersion:"openpond.taskAuthoringProvenance.v1",model:null,modelConfig:{},skillHash:d("openpond-learning-batch-v1"),promptTemplateVersion:"learning-batch-v1",buildIntent:e.batch.purpose==="supervised_training"?"demonstrations":"verifiable_reward",buildSpecification:null,evidenceHashes:e.batch.examples.map(c=>c.evidence.contentHash),tasksetSdkVersion:"learning-batch-v1",sourceCommit:null,repairHistory:[],createdAt:e.batch.sealedAt},readiness:null,contentHash:"00000000",createdAt:e.batch.sealedAt,updatedAt:e.batch.sealedAt,metadata:{learning:a,learningRelease:O(t),trainingMethod:e.batch.purpose==="supervised_training"?"sft":e.batch.purpose==="reward_training"?"grpo":"none",tasksetOutputContract:r?{mode:"artifacts",requiredOutputSource:"task.requiredOutputs"}:{mode:"structured_json",jsonSchema:e.definition.outputSchema,renderer:null}}});return{taskset:K.parse({...w,contentHash:U(w)}),release:t,generatedFiles:m,tasksetAssetBytes:s.tasksetAssetBytes}}S(je,"materializeLearningBatchTaskset");function _e(e){if(!e)return null;let t=e.training.status==="not_configured"?e.evaluation:e.training;return t.status!=="scored"?null:{passed:t.passed===!0,rewardEligible:e.training.status==="scored"}}S(_e,"fixtureOutcome");function tt(e){let t=[],a=0;for(let s of new Map(e.evidence.flatMap(i=>i.submission.assets.map(l=>[l.id,l]))).values()){let i=e.assetBytes?.get(s.id),l=e.assets?.find(m=>m.id===s.id),p=i?.byteLength??(l?new TextEncoder().encode(l.text).byteLength:0);if(a+=p,a>ee)throw new Error("Learning input assets exceed the training package limit.");i?t.push(new TextDecoder().decode(i)):l&&t.push(l.text)}let r=Me(JSON.stringify({tasks:e.evidence.map(s=>s.submission),targets:e.decisions.map(s=>s.approvedTarget),definition:e.definition,inputAssets:t}));if(r.secretStatus!=="passed"||r.piiStatus!=="passed")throw new Error(`Learning batch contains unresolved data findings (${r.findings.join(", ")}). Correct the source examples and seal a revised batch before training.`);let o=X.parse({schemaVersion:"openpond.learningBatchDatasetSource.v1",kind:"learning_batch",id:`batch-source-${e.batch.contentHash.slice(0,40)}`,profileId:e.profileId,title:e.definition.name,sourceHash:e.batch.contentHash,occurredAt:e.batch.sealedAt,batch:O(e.batch),taskDefinition:e.batch.taskDefinition,admittedBy:e.batch.sealedBy,licensingStatus:"approved",secretScanStatus:r.secretStatus,piiScanStatus:r.piiStatus,metadata:{admission:e.admissionContext??"explicit_local_batch_review",privacyScanner:"openpond-evidence-v1",findings:r.findings}});return je({...e,source:o})}S(tt,"prepareReviewedLearningBatch");async function st(e){let t=structuredClone(e),{project:a,recipe:r,taskset:o,approval:s}=t,i=we.parse(t.manifest),l=Se.parse(t.bundleManifest),p=t.executionMode==="qualification",m=i.qualificationSamples===void 0?null:n.object({rollout:n.object({groupSize:n.number().int().min(2).max(32)}).passthrough()}).passthrough().parse(r).rollout.groupSize;if(i.qualificationSamples!==l.qualificationSamples||i.qualificationSamples!==void 0&&(!p||i.qualificationSamples>m))throw new Error("Qualification samples must be within the pinned rollout group and cannot be used for training.");if(i.executionMode!==t.executionMode||l.executionMode!==t.executionMode)throw new Error("Managed submission execution mode differs from its immutable bundle.");if(p&&(t.evaluationSource||t.resumeFrom||t.rewardSource.kind!=="deterministic"))throw new Error("Qualification requires the original learner, deterministic rewards, and no evaluation source.");let{contentHash:w,...c}=i,{contentHash:v,...f}=l;if(d(c)!==w||d(f)!==v||i.resolvedBundleHash!==v||d(i.harnessRelease)!==d(l.harnessRelease)||d(i.datasetRelease)!==d(l.datasetRelease)||d(i.evidenceSets)!==d(l.evidenceSetRelease?[l.evidenceSetRelease]:[]))throw new Error("The managed training manifest or resolved bundle changed.");if(!a.hosted||a.hosted.syncedSourceRevision!==a.revision||a.hosted.portableProjectId!==a.id)throw new Error("Sync the exact Model Project revision before preparing managed training.");let h=a.trainingSetup.baseModel;if(!h||h.modelId!==i.model.source||h.revision!==i.model.revision||h.tokenizerRevision!==i.model.tokenizerRevision||h.chatTemplateHash!==i.model.chatTemplateHash)throw new Error("The synced Model Project base Model does not match the Run manifest.");if(r.method!=="grpo"||d(r)!==i.recipe.configHash||s.approvalHash!==i.approval.approvalHash||s.maximumSpendUsd!==i.approval.maximumSpendUsd)throw new Error("Managed training requires the exact approved GRPO recipe and spend ceiling.");let T=n.object({wallTimeMs:n.number().positive()}).parse(r.resourceLimits);t.idempotencyKey!==void 0&&n.string().trim().min(16).max(191).parse(t.idempotencyKey);let y=Pe(l,t.files),k=p?null:J.parse(t.evaluationSource),R=y.find(g=>g.path===F);if(k&&(!R||d(JSON.parse(Buffer.from(R.content,"base64").toString("utf8")))!==d(k)))throw new Error("The evaluation source differs from the immutable training bundle.");if(p&&(R||y.some(g=>g.path==="dataset/train.json")||!y.some(g=>g.path==="dataset/qualification.json")||!y.some(g=>g.path==="qualification-source.json")))throw new Error("Qualification requires its dedicated dataset and no training/evaluation datasets.");if(!p&&y.some(g=>g.path==="dataset/qualification.json"||g.path==="qualification-source.json"))throw new Error("Qualification datasets cannot be prepared as training.");let M=k?await ce(k):null,P={schemaVersion:"openpond.managedRlPortableSubmission.v1",...p?{executionMode:"qualification"}:{},...t.origin?{origin:Q.parse(t.origin)}:{},sourceRunRef:`openpond:model-run:${i.id}`,name:`OpenPond Managed \xB7 ${a.id}`.slice(0,191),idempotencyKey:t.idempotencyKey??`openpond-managed:${w}`.slice(0,191),modelProject:{id:a.hosted.projectId,portableProjectId:a.hosted.portableProjectId},manifest:i,sourceTaskset:o,modelImprovementQualification:t.modelImprovementQualification??null,recipe:r,resolvedBundle:{manifest:l,files:y},validationTasks:k?.tasks??[],validationAssets:k?.assets??[]},q={schemaVersion:"openpond.trainingInputArtifactUpload.v2",kind:"portable_training_bundle",idempotencyKey:`stage:${t.idempotencyKey??w}`,sourceManifest:{id:i.id,contentHash:w},payload:P},H=await le({...q,contentHash:await de(q)}),C={schemaVersion:"openpond.trainingJobSubmission.v2",idempotencyKey:t.idempotencyKey??`openpond-training-v2:${w}`,name:`OpenPond Managed \xB7 ${a.id}`.slice(0,200),source:{...t.origin?{origin:Q.parse(t.origin)}:{},modelProject:{id:a.hosted.projectId,portableProjectId:a.hosted.portableProjectId,revision:a.revision,contentHash:a.hosted.etag},harnessRunManifest:q.sourceManifest,harnessRelease:i.harnessRelease,taskset:o,tasksetRelease:{id:o.id,contentHash:o.contentHash},dataset:l.datasetRelease,...M?{evaluation:M}:{},evidenceSets:l.evidenceSetRelease?[l.evidenceSetRelease]:[]},job:{kind:"policy_optimize",...p?{executionMode:"qualification"}:{},baseModel:h,recipe:r,rewardSource:t.rewardSource,resumeFrom:t.resumeFrom},requestedCapabilities:[{id:p?"managed_rl.policy.qualification":"managed_rl.policy.grpo",version:"1",required:!0},{id:`managed_rl.rollouts.${i.runtimeTarget.placement}`,version:"1",required:!0},...a.trainingSetup.managedGpuRequirement==="h100_hbm3"?[{id:"managed_rl.gpu.h100_hbm3",version:"1",required:!0}]:[]],placementObjective:a.trainingSetup.managedGpuPlacementObjective,budget:{maximumSpendUsd:s.maximumSpendUsd,maximumWallSeconds:Math.ceil(T.wallTimeMs/1e3)},approval:{...s,approvedAt:i.approval.approvedAt,exportApproved:!0}},u=await me({...C,contentHash:await pe(C)});return{artifact:H,submission:u}}S(st,"prepareManagedTrainingSubmission");function Pe(e,t){if(t.length!==e.files.length||new Set(t.map(r=>r.path)).size!==t.length)throw new Error("The managed bundle file inventory changed.");let a=new Map(t.map(r=>[r.path,r]));return e.files.map(r=>{let o=a.get(r.path);if(!o||o.path.includes("\\")||o.path.includes("\0")||o.path.split("/").some(i=>!i||i==="."||i==="..")||o.path==="bundle-manifest.json"||o.encoding!=="base64")throw new Error("The managed bundle contains an invalid file.");let s=Buffer.from(o.content,"base64");if(s.toString("base64")!==o.content||o.sha256!==r.sha256||o.sizeBytes!==r.sizeBytes||s.byteLength!==r.sizeBytes||z(s)!==r.sha256)throw new Error(`Managed training bundle file ${o.path} changed before upload.`);return o})}S(Pe,"verifyFiles");var Be=n.object({name:n.literal("adamw").default("adamw"),weightDecay:n.number().nonnegative().max(1).default(0),beta1:n.number().positive().lt(1).default(.9),beta2:n.number().positive().lt(1).default(.999),epsilon:n.number().positive().max(.01).default(1e-8)}).strict().default({name:"adamw",weightDecay:0,beta1:.9,beta2:.999,epsilon:1e-8});function nt(e,t){if(!t||typeof t!="object"||Array.isArray(t))return t;let a=t;if(a.method==="dpo"){let R=x(a.policyModel),M=x(a.referenceModel),P=x(a.dataset),q=d({tasksetHash:e.contentHash,preferenceSignals:e.learningSignals.preferences.map(H=>({id:H.id,artifactRef:H.artifactRef,prompt:H.prompt,chosen:H.chosen,rejected:H.rejected,approved:H.approved})),policyModel:R,referenceModel:M,dataset:P});return{...a,referenceLogprobs:{cacheSchemaVersion:"openpond.dpoReferenceLogprobs.v1",cacheKey:d(["dpo-reference-logprobs",q]),invalidationHash:q}}}if(a.method==="ppo"){let R=x(a.policyOptimization),M=x(R.policyModel),P=x(R.referenceModel),q=x(R.optimizer),H=x(q.valueModel),C=x(R.dataset),u=x(R.reward),g=d(M),G=d(P),b=d(H);return{...a,policyOptimization:{...R,dataset:{...C,tasksetId:e.id,tasksetHash:e.contentHash},reward:{...u,graderHash:d(e.graders)}},resume:{...x(a.resume),policyHash:g,referenceHash:G,valueModelHash:b}}}if(a.method!=="grpo")return t;let r=a.reward&&typeof a.reward=="object"&&!Array.isArray(a.reward)?a.reward:{},o=e.environment.metadata.toolContractHash,s=typeof o=="string"&&o.trim()?o:r.toolContractHash,i=x(a.baseModel),l=x(a.dataset),p=x(a.rollout),m=x(a.optimizer),w=x(a.loss),c=x(a.resourceLimits),v=I(l.maxExamples,1),f=I(p.groupSize,2),h=I(p.maxOutputTokens,1),T=I(l.maxPromptTokens,1),y=I(m.maxSteps,1),k=I(m.iterations,2);return{...a,resourceLimits:{...c,maxGpuSeconds:c.maxGpuSeconds??Math.min(10800,Math.ceil(I(c.wallTimeMs,18e4)/1e3))},reward:{...r,graderHash:d(e.graders),toolContractHash:s},policyOptimization:{schemaVersion:"openpond.policyOptimization.v1",policyModel:i,referenceModel:i,dataset:{tasksetId:e.id,tasksetHash:e.contentHash,split:"train",selectionStrategy:l.selectionStrategy,selectionSeed:p.seed,maxExamples:v},sampler:{temperature:p.temperature,topP:p.topP,maxOutputTokens:h,maxTurns:p.maxTurns,concurrency:p.concurrency},environment:{id:r.environmentId,version:r.environmentVersion,toolContractHash:s},reward:{graderId:r.graderId,graderHash:d(e.graders),learnedPreference:r.learnedPreference??null},kl:{coefficient:w.klBeta??null,referenceConstraint:"fixed_reference"},budgets:{maxRollouts:I(c.maxRollouts,v*f),maxEnvironmentExecutions:I(c.maxRollouts,v*f),maxInputTokens:v*f*T,maxOutputTokens:v*f*h,maxOptimizerSteps:y*k,wallTimeMs:I(c.wallTimeMs,18e4),maximumCostUsd:null},checkpointEverySteps:1,seed:p.seed,evaluationSplit:"frozen_eval",optimizer:{method:"grpo",groupSize:f,normalization:"group_standardized",advantageEpsilon:typeof m.advantageEpsilon=="number"?m.advantageEpsilon:1e-8,loss:w.method??"grpo",clipRange:typeof m.clipRange=="number"?m.clipRange:.2,iterations:k,microbatchSize:I(m.microbatchSize,1),gradientAccumulationSteps:I(m.gradientAccumulationSteps,1),adamw:Be.parse(m.adamw)}}}}S(nt,"withAuthoritativeRecipeHashes");function x(e){return e&&typeof e=="object"&&!Array.isArray(e)?e:{}}S(x,"record");function I(e,t){return typeof e=="number"&&Number.isInteger(e)&&e>0?e:t}S(I,"positiveInteger");function rt(e){let t=e.sourceRelease,a=[],r=e.skills??[],o=e.agents??[];if([...r,...o].some(m=>m.visibility!=="policy"))throw new Error("The portable policy Harness requires policy-visible dependencies.");let s=ge({id:"desktop-dependency-lock",path:".openpond/harness/dependency-lock.json",hashInput:{profileHead:e.profileHead??null,sourceRelease:t,tools:a,skills:r.map(({id:m,contentHash:w})=>({id:m,contentHash:w})),agents:o.map(({id:m,contentHash:w})=>({id:m,contentHash:w}))},mediaType:"application/json",visibility:"policy"}),i=ae({schemaVersion:"openpond.agentSnapshot.v2",id:`agent-snapshot-${d([t,a,s.contentHash]).slice(0,24)}`,sourceRelease:t,instructions:[],skills:r,agents:o,toolDeclarations:a,capabilityRequirements:[],dependencyLock:s,portability:{portable:!0,blockers:[],localOnlyAssetRefs:[],hostPrivateAssetRefs:[]},metadata:{sourceReleaseId:t?.id??null}}),l=ge({id:"desktop-harness-program",path:".openpond/harness/program.json",hashInput:{program:"openpond.desktop-agent-loop.v1"},mediaType:"application/json",visibility:"policy"}),p=se({schemaVersion:"openpond.harnessRelease.v2",id:`harness-${d([i.contentHash,l.contentHash,a]).slice(0,24)}`,agentSnapshot:{id:i.id,contentHash:i.contentHash},program:l,tools:a,lifecycle:{create:!0,reset:!0,step:!0,collect:!0,destroy:!0,resetScope:"attempt"},graderInterface:{visibleEvidence:["output","runtime_events","artifacts"],privilegedEvidence:["expected_output","private_verifier"],privateVerifierIsolation:!0},files:[...r,...o],metadata:{runtimeProtocol:"openpond.desktop-agent-loop.v1"}});return{agentSnapshot:i,harnessRelease:p}}S(rt,"createPolicyHarnessContext");function ge(e){let t=L(e.hashInput);return{id:e.id,path:e.path,contentHash:d(e.hashInput),sizeBytes:new TextEncoder().encode(t).byteLength,mediaType:e.mediaType,visibility:e.visibility}}S(ge,"asset");function ot(e){let{taskset:t,modelProject:a,harnessRelease:r,tasksetRelease:o}=e,s=a.trainingSetup,i=e.executionMode==="qualification",l=e.qualificationSamples===void 0?null:n.object({rollout:n.object({groupSize:n.number().int().min(2).max(32)}).passthrough()}).passthrough().parse(e.modelProject.trainingSetup.recipe).rollout.groupSize;if(e.qualificationSamples!==void 0&&(!i||!Number.isInteger(e.qualificationSamples)||e.qualificationSamples<1||e.qualificationSamples>l))throw new Error("Qualification samples must be within the pinned rollout group and cannot be used for training.");if(i&&e.evaluationSource)throw new Error("Qualification does not consume an independent evaluation source.");let p=i?te(e.qualificationSource):null;if(!i&&e.qualificationSource)throw new Error("Qualification sources cannot be used for training.");if(!i&&e.qualificationTaskId!==void 0)throw new Error("Qualification selection cannot be used for training.");if(i&&(!e.qualificationTaskId||!t.tasks.some(u=>u.id===e.qualificationTaskId)))throw new Error("Qualification requires one explicit task ID from its published source.");if(p&&(p.modelResources||p.learningResources||d(t.tasks.map(fe))!==d(p.taskset.tasks)||ve(e.tasksetRelease)!==ve(p.taskset)))throw new Error("Qualification must preserve its ordinary source release and complete task population.");let m=j.parse(r),w=j.parse(o);if(e.openpondRelease,e.workerProtocol,U(t)!==t.contentHash)throw new Error("Taskset authoring state changed after its release was selected.");if(!s.baseModel?.revision||!s.baseModel.tokenizerRevision||!s.baseModel.chatTemplateHash||!s.recipe||!s.tasksetRef||s.tasksetRef.id!==(i?o.id:t.id)||s.tasksetRef.contentHash!==(i?o.contentHash:t.contentHash))throw new Error("Model Run must bind an exact Taskset, Model revision, tokenizer, chat template, and Recipe.");let v=t.profileRelease??{id:`profile_${t.profileId}`,revision:t.revision,contentHash:d({profileId:t.profileId,sourceCommit:t.authoringProvenance.sourceCommit,skillHash:t.authoringProvenance.skillHash})},f=new Map;if(p&&_(f,"qualification-source.json",p),e.evaluationSource){let u=J.parse(e.evaluationSource);if(d(u.taskset)!==d(s.evaluationTasksetRef))throw new Error("The evaluation source differs from the prepared Model configuration.");oe(t.tasks.filter(g=>g.split==="train"),u.tasks),f.set(F,new TextEncoder().encode(L(u)))}else if(!i&&s.evaluationTasksetRef)throw new Error("The prepared Model requires its pinned evaluation source bytes.");let h=e.harnessSource?e.harnessSource.schemaVersion==="openpond.harnessPolicySourcePackage.v1"?ne(e.harnessSource,m):re(e.harnessSource,m):null;if(s.harnessRelease&&!h)throw new Error("A selected Harness requires its complete immutable source package.");if(h&&(!s.harnessRelease||s.harnessRelease.id!==m.id||s.harnessRelease.contentHash!==m.contentHash))throw new Error("Harness source differs from the Model's explicit selection.");if(h){if((e.runtime.placement!=="local"||e.compute.kind!=="local")&&(!h.agentSnapshot.portability.portable||h.schemaVersion==="openpond.harnessSourcePackage.v1"&&h.harnessRelease.files.some(g=>g.visibility!=="policy")))throw new Error("The selected Harness contains source that cannot leave its local host.");_(f,"harness/source-package.json",h)}_(f,"harness/execution.json",ie.parse({schemaVersion:"openpond.harnessSourceSelection.v1",mode:h?"selected_release":"taskset_owned",harnessRelease:m,sourcePackageHash:h?.contentHash??null}));let T=ue(t,e.rewardExecution);if(p&&(!T||d(N(T.binding,T.rewards))!==d(p.taskset.graders)||p.taskset.graders.some(u=>u.kind==="model_judge"||u.kind==="human")))throw new Error("Qualification requires the source package's exact deterministic graders.");if(T){let u=e.verifierAssets??[],g=[...T.rewards.flatMap(b=>b.assets),...N(T.binding,T.rewards).flatMap(b=>b.kind==="custom_verifier"?[b.verifierRef]:b.kind==="model_judge"?[b.rubricRef]:[])],G=new Set(g.map(b=>b.id));if(new Set(u.map(b=>b.id)).size!==u.length||new Set(u.map(b=>b.asset.path)).size!==u.length||u.some(b=>!G.has(b.id)))throw new Error("Unexpected or duplicate private verifier asset.");for(let b of g){let W=u.find(Re=>Re.id===b.id);if(!W||b.visibility!=="verifier")throw new Error("Training bundle requires its private verifier asset.");B(W,b)}_(f,"reward-binding.json",{kind:"reward_binding_v1",binding:T.binding,rewards:T.rewards,assets:u})}else if(e.verifierAssets?.length)throw new Error("Private verifier assets require a published Reward binding.");_(f,"environment.json",{schemaVersion:"openpond.harnessEnvironment.v1",environment:i?{...t.environment,metadata:{...t.environment.metadata,policySystemPrompt:t.objective}}:t.environment,capabilities:t.capabilities,policy:t.policy,outputContract:t.metadata.tasksetOutputContract?.mode==="text"?null:t.metadata.tasksetOutputContract??null}),_(f,"graders.json",{schemaVersion:"openpond.harnessGraders.v1",graders:t.graders,fixtures:t.graderFixtures}),_(f,"tool-contract.json",{schemaVersion:"openpond.harnessToolContract.v1",toolNames:t.environment.toolNames,actionBindings:t.environment.actionBindings??[],capabilities:t.capabilities,connectedAppScopes:t.policy.connectedAppScopes});let y=i?t.tasks.filter(u=>u.id===e.qualificationTaskId):t.tasks.filter(u=>u.split==="train");if(y.length>0)_(f,i?"dataset/qualification.json":"dataset/train.json",{schemaVersion:i?"openpond.qualificationDataset.v1":"openpond.datasetSplit.v1",...i?{selectedTaskId:e.qualificationTaskId}:{split:"train"},tasks:y}),ze({assets:f,tasks:y,tasksetAssetBytes:e.tasksetAssetBytes??new Map});else if(t.datasetArtifact)_(f,"dataset/artifact.json",t.datasetArtifact);else throw new Error("Taskset has no training dataset.");let k=Object.values(t.learningSignals).flat().filter(u=>u.approved);for(let u of k){let g=ye(u);f.set(`evidence/signals/${z(g)}.json`,g)}let R={id:`dataset_${t.id}_r${t.revision}`,contentHash:d({taskset:w,sourceRefs:t.sourceRefs,datasetArtifact:t.datasetArtifact??null,...i?{executionMode:"qualification",tasks:y}:{trainTasks:y}})},M=k.length>0?{id:`evidence_${t.id}_r${t.revision}`,contentHash:d({taskset:w,harnessRelease:m,datasetRelease:R,profileRelease:v,signals:k})}:null,P=ke.parse({schemaVersion:"openpond.resolvedTrainingBundle.v1",...i?{executionMode:"qualification"}:{},...e.qualificationSamples!==void 0?{qualificationSamples:e.qualificationSamples}:{},projection:"trainer",harnessRelease:m,datasetRelease:R,evidenceSetRelease:M,files:[...f.entries()].sort(([u],[g])=>u.localeCompare(g)).map(([u,g])=>({path:u,sha256:z(g),sizeBytes:g.byteLength}))}),q=Se.parse({...P,contentHash:d(P)}),H=be.parse({schemaVersion:"openpond.harnessRunManifest.v1",...i?{executionMode:"qualification"}:{},...e.qualificationSamples!==void 0?{qualificationSamples:e.qualificationSamples}:{},id:`manifest_${e.modelRunId}_${e.approval.approvalHash.slice(0,16)}`,harnessRelease:r,datasetRelease:R,evidenceSets:M?[M]:[],model:{source:s.baseModel.modelId,revision:s.baseModel.revision,artifactHash:null,tokenizerRevision:s.baseModel.tokenizerRevision,chatTemplateHash:s.baseModel.chatTemplateHash},recipe:{method:s.recipe.method,version:"openpond.trainingRecipe.v1",configHash:d(s.recipe)},runtimeTarget:e.runtime,computeTarget:e.compute,engine:e.engine,resolvedBundleHash:q.contentHash,secretLeaseRefs:e.secretLeaseRefs??[],approval:e.approval,createdAt:a.updatedAt});return{manifest:we.parse({...H,contentHash:d(H)}),resolvedBundleManifest:q,assets:f,profileRelease:v,harnessRelease:m,tasksetRelease:w,datasetRelease:R,evidenceSetRelease:M}}S(ot,"buildTasksetTrainingBundle");function ze(e){let t=new Set;for(let a of e.tasks)for(let r of a.assets??[]){t.add(r.artifactRef);let o=e.tasksetAssetBytes.get(r.artifactRef);if(!o)throw new Error(`Resolved Training Bundle is missing Work asset ${r.artifactRef}.`);if(o.byteLength!==r.sizeBytes||z(o)!==r.sha256)throw new Error(`Resolved Training Bundle Work asset ${r.artifactRef} failed immutable verification.`);e.assets.set(r.artifactRef,o)}for(let a of e.tasksetAssetBytes.keys())if(!t.has(a))throw new Error(`Resolved Training Bundle received unexpected Work asset ${a}.`)}S(ze,"addTasksetAssets");function _(e,t,a){e.set(t,ye(a))}S(_,"addJsonAsset");function ye(e){return Buffer.from(L(e),"utf8")}S(ye,"bytes");function ve(e){return d({id:e.id,contentHash:e.contentHash})}S(ve,"releasedRefHash");export{He as a,Ee as b,Ae as c,we as d,Se as e,Me as f,tt as g,st as h,Be as i,nt as j,rt as k,ot as l};
+import{Ab as me,Ib as ue,Nb as fe,Ob as he,Y as ee,ha as te,hb as ae,ib as se,kb as ie,lb as ne,mb as re,ta as K,tb as F,ub as J,v as X,vb as oe,wb as ce,xb as de,yb as le,za as U,zb as pe}from"./chunk-PGO4FRHU.js";import{b as A,c as E,d as $,f as j,k as L,l as z,m as d}from"./chunk-VNJAAYNQ.js";import{N as O,aa as Z,ba as Y,d as B,e as Q}from"./chunk-4M3YW3EQ.js";import{Ya as N}from"./chunk-PBZCXPPC.js";import{A as n}from"./chunk-PF56N7ZY.js";import{a as S}from"./chunk-FTQLQJQV.js";var xe=n.object({id:A,purpose:n.string().trim().min(1).max(500),audience:n.enum(["orchestrator","environment","privileged_scorer","trainer","infrastructure"]),required:n.boolean(),ttlSeconds:n.number().int().positive().max(86400),scopes:n.array(n.string().trim().min(1).max(500)).max(100).default([])}).strict(),Te=n.object({declarationId:A,leaseRef:n.string().trim().min(8).max(1e3),audience:xe.shape.audience,expiresAt:$}).strict(),He=n.object({adapterId:A,placement:n.enum(["local","remote","colocated","provider_native"]),capabilityReceipt:E,runtimeVersion:n.string().trim().min(1).max(200),dataPlane:n.object({provider:A,dataPlaneId:A,cellId:A,runnerPoolId:A,runtimeImageDigest:n.string().regex(/^sha256:[a-f0-9]{64}$/),capabilityReceipt:E}).strict().nullable()}).strict(),Ee=n.object({adapterId:A,kind:n.enum(["local","ssh","managed","custom"]),deviceOrPool:n.string().trim().min(1).max(1e3),capabilityReceipt:E,provider:A.nullable()}).strict(),Ae=n.object({adapterId:A,workerVersion:n.string().trim().min(1).max(200),workerImageDigest:n.string().regex(/^sha256:[a-f0-9]{64}$/).nullable(),upstreamRevision:n.string().trim().min(1).max(500),capabilityReceipt:E}).strict(),be=n.object({schemaVersion:n.literal("openpond.harnessRunManifest.v1"),executionMode:n.literal("qualification").optional(),qualificationSamples:n.number().int().min(1).max(32).optional(),id:A,harnessRelease:j,datasetRelease:j,evidenceSets:n.array(j).max(1e4),model:n.object({source:n.string().trim().min(1).max(200),revision:n.string().trim().min(1).max(500),artifactHash:E.nullable(),tokenizerRevision:n.string().trim().min(1).max(500),chatTemplateHash:E}).strict(),recipe:n.object({method:n.string().trim().min(1).max(100),version:n.string().trim().min(1).max(200),configHash:E}).strict(),runtimeTarget:He,computeTarget:Ee,engine:Ae,resolvedBundleHash:E,secretLeaseRefs:n.array(Te).max(1e3),approval:n.object({approvalHash:E,approvedAt:$,maximumSpendUsd:n.number().nonnegative().nullable()}).strict(),createdAt:$}).strict(),we=be.extend({contentHash:E}).strict(),ke=n.object({schemaVersion:n.literal("openpond.resolvedTrainingBundle.v1"),executionMode:n.literal("qualification").optional(),qualificationSamples:n.number().int().min(1).max(32).optional(),projection:n.literal("trainer"),harnessRelease:j,datasetRelease:j,evidenceSetRelease:j.nullable(),files:n.array(n.object({path:n.string().trim().min(1).max(2e3),sha256:E,sizeBytes:n.number().int().nonnegative()}).strict()).max(1e5)}).strict(),Se=ke.extend({contentHash:E}).strict(),Ie=[{label:"OpenAI-style API key",pattern:/\bsk-[A-Za-z0-9_-]{16,}\b/g},{label:"OpenPond API key",pattern:/\bopk_[A-Za-z0-9_-]{12,}\b/g},{label:"private key",pattern:/-----BEGIN (?:RSA |EC |OPENSSH )?PRIVATE KEY-----/g},{label:"credential assignment",pattern:/\b(?:api[_-]?key|password|secret|token)["']?\s*[:=]\s*["']?[^\s"']{8,}/gi}],V=/\b[A-Z0-9._%+-]+@[A-Z0-9.-]+\.[A-Z]{2,}\b/gi,D=/\b(?:\+?1[-.\s]?)?\(?\d{3}\)?[-.\s]\d{3}[-.\s]\d{4}\b/g;function Me(e){let t=[],a=e;for(let s of Ie)s.pattern.lastIndex=0,s.pattern.test(a)&&t.push(s.label),s.pattern.lastIndex=0,a=a.replace(s.pattern,`[REDACTED ${s.label}]`);V.lastIndex=0;let r=V.test(a);V.lastIndex=0,r&&t.push("email address"),a=a.replace(V,"[REDACTED email]"),D.lastIndex=0;let o=D.test(a);return D.lastIndex=0,o&&t.push("phone number"),a=a.replace(D,"[REDACTED phone]"),{secretStatus:t.some(s=>!s.includes("email")&&!s.includes("phone"))?"blocked":"passed",piiStatus:r||o?"review":"passed",findings:t,redacted:a}}S(Me,"scanAndRedactEvidence");function qe(e){let t=new Map;return{tasks:e.tasks.map(({artifactRefs:r,...o})=>({...o,assets:r.map(s=>{if(s.visibility!=="policy")throw new Error("Learning task inputs must be policy-visible assets.");let i=e.assets?.find(m=>m.id===s.id),l=e.assetBytes?.get(s.id)??(i?new TextEncoder().encode(B(i,s)):void 0);if(!l||l.byteLength!==s.sizeBytes||z(l)!==s.contentHash)throw new Error(`Learning task input bytes are missing or changed: ${s.id}.`);let p=`learning/assets/${d({task:o.id,asset:s.id}).slice(0,40)}/${s.path}`;return t.set(p,new Uint8Array(l)),{id:s.id,sourceRefId:e.sourceId,artifactRef:p,fileName:s.path.split("/").at(-1),stagingPath:s.path,mediaType:s.mediaType,sha256:s.contentHash,sizeBytes:s.sizeBytes,split:o.split,metadata:{portableAsset:s}}}),...o.requiredOutputs?{requiredOutputs:o.requiredOutputs.map(s=>({...s,schemaRef:s.schemaRef?.id??null,maxBytes:s.maxBytes??void 0}))}:{}})),tasksetAssetBytes:t}}S(qe,"materializeLearningBatchAssets");function je(e){let t=Z(e),a=Y(t),r=t.environment.kind==="work";if(!r&&(t.environment.kind!=="text"||t.tools.length||t.tasks.some(c=>c.artifactRefs.length))||r&&t.tasks.some(c=>!c.requiredOutputs?.length))throw new Error("This learning batch requires a supported text or Work environment with declared outputs.");let o=e.source.id,s=qe({tasks:t.tasks,sourceId:o,assets:e.assets,assetBytes:e.assetBytes}),i={demonstrations:[],preferences:[],corrections:[],feedback:[],rewards:[],labels:[]},l=[];for(let c of a.admissions){let v=e.evidence.find(h=>h.id===c.evidence.id),f=e.decisions.find(h=>h.id===c.decision.id);e.batch.purpose==="reward_training"&&f.grade?.training.status==="scored"&&i.rewards.push({id:`reward-${c.taskId}`,taskId:c.taskId,kind:"reward",task:e.definition.instructions,rules:[{id:e.binding.id,points:1,condition:"Execute the released Reward binding with its declared normalization, weights and gates."}],otherwisePoints:0,executable:!0,approved:!0,confidence:1,sourceRefs:[o],artifactRef:e.binding.id,metadata:{rewardBinding:O(e.binding),evidence:c.evidence,decision:c.decision}}),c.supervisedTarget!==null&&i.demonstrations.push({kind:"demonstration",id:`demo-${c.taskId}`,taskId:c.taskId,sourceRefs:[o],artifactRef:c.decision.id,approved:!0,confidence:1,prompt:null,response:JSON.stringify(c.supervisedTarget),metadata:{evidence:c.evidence,decision:c.decision,targetGradeHash:f.targetGrade?.contentHash}});for(let[h,T,y]of[["observed",v.submission.observedOutput,f.grade],["target",f.approvedTarget,f.targetGrade]]){let k=_e(y);T===null||k===null||l.push({id:`fixture-${h}-${c.taskId}`,taskId:c.taskId,label:k.passed?"positive":"negative",output:T,infrastructureError:null,expectedPassed:k.passed,expectedRewardEligible:k.rewardEligible,metadata:{gradeReceiptHash:y.contentHash,evidence:c.evidence,decision:c.decision}})}}if(!l.length&&e.batch.purpose!=="reward_training")throw new Error("Grade at least one reviewed response before preparing this batch. Grader fixtures must come from actual receipts.");let p=he(e.binding,e.rewards,e.assets),m=[];for(let c of t.graders){if(c.kind!=="custom_verifier")continue;let v=e.assets?.find(h=>h.id===c.verifierRef.id);if(!v)throw new Error(`Reward ${c.id} is missing its immutable verifier source.`);let f=c.verifierRef.path;m.some(h=>h.path===f)||m.push({path:f,role:"verifier",content:B(v,c.verifierRef)})}let w=K.parse({schemaVersion:"openpond.taskset.v1",id:`learning-${d([e.profileId,e.batch.contentHash]).slice(0,40)}`,revision:1,profileId:e.profileId,profileRelease:null,createImproveRunId:null,name:`${e.definition.name} \xB7 ${e.batch.examples.length} approved examples`,objective:e.definition.instructions,purpose:"general",benchmark:null,status:"needs_review",sourceRefs:[e.source],datasetArtifact:null,policy:t.policy,environment:{protocolVersion:"openpond.taskEnvironment.v1",kind:r?"work":"chat",entrypoint:t.environment.entrypoint,stateful:t.environment.stateful,deterministicSeeds:t.environment.deterministicSeeds,toolNames:t.tools.map(c=>c.name),lifecycle:["create","reset","step","grade","cleanup"],defaultTimeoutMs:t.environment.defaultTimeoutMs,networkPolicy:t.environment.networkPolicy,metadata:{portableEnvironment:t.environment,portableTools:t.tools}},capabilities:{schemaVersion:"openpond.tasksetCapabilities.v1",taskKind:r?"single_agent":"chat",supportedSignals:i.demonstrations.length?["demonstration","reward"]:["reward"],compatibleMethods:e.batch.purpose==="supervised_training"?["sft"]:e.batch.purpose==="reward_training"?["grpo","ppo"]:["none"],rewardKinds:[...new Set(t.graders.map(c=>c.kind==="human"?"human":c.kind==="model_judge"?"model_judge":"deterministic"))],requiresTools:t.tools.length>0,requiresState:t.environment.stateful,requiresPrivilegedGrading:!0,environmentPlacements:["local","remote"],exportable:!0,portabilityBlockers:[]},tasks:s.tasks.map(c=>({...c,schemaVersion:"openpond.taskData.v1",sourceRefs:[o],metadata:{admission:a.admissions.find(v=>v.taskId===c.id),exampleOrigin:e.evidence.find(v=>v.id===c.id)?.correctionFeedbackId?"corrected":"extracted"}})),graders:p,graderFixtures:l,learningSignals:i,authoringProvenance:{schemaVersion:"openpond.taskAuthoringProvenance.v1",model:null,modelConfig:{},skillHash:d("openpond-learning-batch-v1"),promptTemplateVersion:"learning-batch-v1",buildIntent:e.batch.purpose==="supervised_training"?"demonstrations":"verifiable_reward",buildSpecification:null,evidenceHashes:e.batch.examples.map(c=>c.evidence.contentHash),tasksetSdkVersion:"learning-batch-v1",sourceCommit:null,repairHistory:[],createdAt:e.batch.sealedAt},readiness:null,contentHash:"00000000",createdAt:e.batch.sealedAt,updatedAt:e.batch.sealedAt,metadata:{learning:a,learningRelease:O(t),trainingMethod:e.batch.purpose==="supervised_training"?"sft":e.batch.purpose==="reward_training"?"grpo":"none",tasksetOutputContract:r?{mode:"artifacts",requiredOutputSource:"task.requiredOutputs"}:{mode:"structured_json",jsonSchema:e.definition.outputSchema,renderer:null}}});return{taskset:K.parse({...w,contentHash:U(w)}),release:t,generatedFiles:m,tasksetAssetBytes:s.tasksetAssetBytes}}S(je,"materializeLearningBatchTaskset");function _e(e){if(!e)return null;let t=e.training.status==="not_configured"?e.evaluation:e.training;return t.status!=="scored"?null:{passed:t.passed===!0,rewardEligible:e.training.status==="scored"}}S(_e,"fixtureOutcome");function tt(e){let t=[],a=0;for(let s of new Map(e.evidence.flatMap(i=>i.submission.assets.map(l=>[l.id,l]))).values()){let i=e.assetBytes?.get(s.id),l=e.assets?.find(m=>m.id===s.id),p=i?.byteLength??(l?new TextEncoder().encode(l.text).byteLength:0);if(a+=p,a>ee)throw new Error("Learning input assets exceed the training package limit.");i?t.push(new TextDecoder().decode(i)):l&&t.push(l.text)}let r=Me(JSON.stringify({tasks:e.evidence.map(s=>s.submission),targets:e.decisions.map(s=>s.approvedTarget),definition:e.definition,inputAssets:t}));if(r.secretStatus!=="passed"||r.piiStatus!=="passed")throw new Error(`Learning batch contains unresolved data findings (${r.findings.join(", ")}). Correct the source examples and seal a revised batch before training.`);let o=X.parse({schemaVersion:"openpond.learningBatchDatasetSource.v1",kind:"learning_batch",id:`batch-source-${e.batch.contentHash.slice(0,40)}`,profileId:e.profileId,title:e.definition.name,sourceHash:e.batch.contentHash,occurredAt:e.batch.sealedAt,batch:O(e.batch),taskDefinition:e.batch.taskDefinition,admittedBy:e.batch.sealedBy,licensingStatus:"approved",secretScanStatus:r.secretStatus,piiScanStatus:r.piiStatus,metadata:{admission:e.admissionContext??"explicit_local_batch_review",privacyScanner:"openpond-evidence-v1",findings:r.findings}});return je({...e,source:o})}S(tt,"prepareReviewedLearningBatch");async function st(e){let t=structuredClone(e),{project:a,recipe:r,taskset:o,approval:s}=t,i=we.parse(t.manifest),l=Se.parse(t.bundleManifest),p=t.executionMode==="qualification",m=i.qualificationSamples===void 0?null:n.object({rollout:n.object({groupSize:n.number().int().min(2).max(32)}).passthrough()}).passthrough().parse(r).rollout.groupSize;if(i.qualificationSamples!==l.qualificationSamples||i.qualificationSamples!==void 0&&(!p||i.qualificationSamples>m))throw new Error("Qualification samples must be within the pinned rollout group and cannot be used for training.");if(i.executionMode!==t.executionMode||l.executionMode!==t.executionMode)throw new Error("Managed submission execution mode differs from its immutable bundle.");if(p&&(t.evaluationSource||t.resumeFrom||t.rewardSource.kind!=="deterministic"))throw new Error("Qualification requires the original learner, deterministic rewards, and no evaluation source.");let{contentHash:w,...c}=i,{contentHash:v,...f}=l;if(d(c)!==w||d(f)!==v||i.resolvedBundleHash!==v||d(i.harnessRelease)!==d(l.harnessRelease)||d(i.datasetRelease)!==d(l.datasetRelease)||d(i.evidenceSets)!==d(l.evidenceSetRelease?[l.evidenceSetRelease]:[]))throw new Error("The managed training manifest or resolved bundle changed.");if(!a.hosted||a.hosted.syncedSourceRevision!==a.revision||a.hosted.portableProjectId!==a.id)throw new Error("Sync the exact Model Project revision before preparing managed training.");let h=a.trainingSetup.baseModel;if(!h||h.modelId!==i.model.source||h.revision!==i.model.revision||h.tokenizerRevision!==i.model.tokenizerRevision||h.chatTemplateHash!==i.model.chatTemplateHash)throw new Error("The synced Model Project base Model does not match the Run manifest.");if(r.method!=="grpo"||d(r)!==i.recipe.configHash||s.approvalHash!==i.approval.approvalHash||s.maximumSpendUsd!==i.approval.maximumSpendUsd)throw new Error("Managed training requires the exact approved GRPO recipe and spend ceiling.");let T=n.object({wallTimeMs:n.number().positive()}).parse(r.resourceLimits);t.idempotencyKey!==void 0&&n.string().trim().min(16).max(191).parse(t.idempotencyKey);let y=Pe(l,t.files),k=p?null:J.parse(t.evaluationSource),R=y.find(g=>g.path===F);if(k&&(!R||d(JSON.parse(Buffer.from(R.content,"base64").toString("utf8")))!==d(k)))throw new Error("The evaluation source differs from the immutable training bundle.");if(p&&(R||y.some(g=>g.path==="dataset/train.json")||!y.some(g=>g.path==="dataset/qualification.json")||!y.some(g=>g.path==="qualification-source.json")))throw new Error("Qualification requires its dedicated dataset and no training/evaluation datasets.");if(!p&&y.some(g=>g.path==="dataset/qualification.json"||g.path==="qualification-source.json"))throw new Error("Qualification datasets cannot be prepared as training.");let M=k?await ce(k):null,P={schemaVersion:"openpond.managedRlPortableSubmission.v1",...p?{executionMode:"qualification"}:{},...t.origin?{origin:Q.parse(t.origin)}:{},sourceRunRef:`openpond:model-run:${i.id}`,name:`OpenPond Managed \xB7 ${a.id}`.slice(0,191),idempotencyKey:t.idempotencyKey??`openpond-managed:${w}`.slice(0,191),modelProject:{id:a.hosted.projectId,portableProjectId:a.hosted.portableProjectId},manifest:i,sourceTaskset:o,modelImprovementQualification:t.modelImprovementQualification??null,recipe:r,resolvedBundle:{manifest:l,files:y},validationTasks:k?.tasks??[],validationAssets:k?.assets??[]},q={schemaVersion:"openpond.trainingInputArtifactUpload.v2",kind:"portable_training_bundle",idempotencyKey:`stage:${t.idempotencyKey??w}`,sourceManifest:{id:i.id,contentHash:w},payload:P},H=await le({...q,contentHash:await de(q)}),C={schemaVersion:"openpond.trainingJobSubmission.v2",idempotencyKey:t.idempotencyKey??`openpond-training-v2:${w}`,name:`OpenPond Managed \xB7 ${a.id}`.slice(0,200),source:{...t.origin?{origin:Q.parse(t.origin)}:{},modelProject:{id:a.hosted.projectId,portableProjectId:a.hosted.portableProjectId,revision:a.revision,contentHash:a.hosted.etag},harnessRunManifest:q.sourceManifest,harnessRelease:i.harnessRelease,taskset:o,tasksetRelease:{id:o.id,contentHash:o.contentHash},dataset:l.datasetRelease,...M?{evaluation:M}:{},evidenceSets:l.evidenceSetRelease?[l.evidenceSetRelease]:[]},job:{kind:"policy_optimize",...p?{executionMode:"qualification"}:{},baseModel:h,recipe:r,rewardSource:t.rewardSource,resumeFrom:t.resumeFrom},requestedCapabilities:[{id:p?"managed_rl.policy.qualification":"managed_rl.policy.grpo",version:"1",required:!0},{id:`managed_rl.rollouts.${i.runtimeTarget.placement}`,version:"1",required:!0},...a.trainingSetup.managedGpuRequirement==="h100_hbm3"?[{id:"managed_rl.gpu.h100_hbm3",version:"1",required:!0}]:[]],placementObjective:a.trainingSetup.managedGpuPlacementObjective,budget:{maximumSpendUsd:s.maximumSpendUsd,maximumWallSeconds:Math.ceil(T.wallTimeMs/1e3)},approval:{...s,approvedAt:i.approval.approvedAt,exportApproved:!0}},u=await me({...C,contentHash:await pe(C)});return{artifact:H,submission:u}}S(st,"prepareManagedTrainingSubmission");function Pe(e,t){if(t.length!==e.files.length||new Set(t.map(r=>r.path)).size!==t.length)throw new Error("The managed bundle file inventory changed.");let a=new Map(t.map(r=>[r.path,r]));return e.files.map(r=>{let o=a.get(r.path);if(!o||o.path.includes("\\")||o.path.includes("\0")||o.path.split("/").some(i=>!i||i==="."||i==="..")||o.path==="bundle-manifest.json"||o.encoding!=="base64")throw new Error("The managed bundle contains an invalid file.");let s=Buffer.from(o.content,"base64");if(s.toString("base64")!==o.content||o.sha256!==r.sha256||o.sizeBytes!==r.sizeBytes||s.byteLength!==r.sizeBytes||z(s)!==r.sha256)throw new Error(`Managed training bundle file ${o.path} changed before upload.`);return o})}S(Pe,"verifyFiles");var Be=n.object({name:n.literal("adamw").default("adamw"),weightDecay:n.number().nonnegative().max(1).default(0),beta1:n.number().positive().lt(1).default(.9),beta2:n.number().positive().lt(1).default(.999),epsilon:n.number().positive().max(.01).default(1e-8)}).strict().default({name:"adamw",weightDecay:0,beta1:.9,beta2:.999,epsilon:1e-8});function nt(e,t){if(!t||typeof t!="object"||Array.isArray(t))return t;let a=t;if(a.method==="dpo"){let R=x(a.policyModel),M=x(a.referenceModel),P=x(a.dataset),q=d({tasksetHash:e.contentHash,preferenceSignals:e.learningSignals.preferences.map(H=>({id:H.id,artifactRef:H.artifactRef,prompt:H.prompt,chosen:H.chosen,rejected:H.rejected,approved:H.approved})),policyModel:R,referenceModel:M,dataset:P});return{...a,referenceLogprobs:{cacheSchemaVersion:"openpond.dpoReferenceLogprobs.v1",cacheKey:d(["dpo-reference-logprobs",q]),invalidationHash:q}}}if(a.method==="ppo"){let R=x(a.policyOptimization),M=x(R.policyModel),P=x(R.referenceModel),q=x(R.optimizer),H=x(q.valueModel),C=x(R.dataset),u=x(R.reward),g=d(M),G=d(P),b=d(H);return{...a,policyOptimization:{...R,dataset:{...C,tasksetId:e.id,tasksetHash:e.contentHash},reward:{...u,graderHash:d(e.graders)}},resume:{...x(a.resume),policyHash:g,referenceHash:G,valueModelHash:b}}}if(a.method!=="grpo")return t;let r=a.reward&&typeof a.reward=="object"&&!Array.isArray(a.reward)?a.reward:{},o=e.environment.metadata.toolContractHash,s=typeof o=="string"&&o.trim()?o:r.toolContractHash,i=x(a.baseModel),l=x(a.dataset),p=x(a.rollout),m=x(a.optimizer),w=x(a.loss),c=x(a.resourceLimits),v=I(l.maxExamples,1),f=I(p.groupSize,2),h=I(p.maxOutputTokens,1),T=I(l.maxPromptTokens,1),y=I(m.maxSteps,1),k=I(m.iterations,2);return{...a,resourceLimits:{...c,maxGpuSeconds:c.maxGpuSeconds??Math.min(10800,Math.ceil(I(c.wallTimeMs,18e4)/1e3))},reward:{...r,graderHash:d(e.graders),toolContractHash:s},policyOptimization:{schemaVersion:"openpond.policyOptimization.v1",policyModel:i,referenceModel:i,dataset:{tasksetId:e.id,tasksetHash:e.contentHash,split:"train",selectionStrategy:l.selectionStrategy,selectionSeed:p.seed,maxExamples:v},sampler:{temperature:p.temperature,topP:p.topP,maxOutputTokens:h,maxTurns:p.maxTurns,concurrency:p.concurrency},environment:{id:r.environmentId,version:r.environmentVersion,toolContractHash:s},reward:{graderId:r.graderId,graderHash:d(e.graders),learnedPreference:r.learnedPreference??null},kl:{coefficient:w.klBeta??null,referenceConstraint:"fixed_reference"},budgets:{maxRollouts:I(c.maxRollouts,v*f),maxEnvironmentExecutions:I(c.maxRollouts,v*f),maxInputTokens:v*f*T,maxOutputTokens:v*f*h,maxOptimizerSteps:y*k,wallTimeMs:I(c.wallTimeMs,18e4),maximumCostUsd:null},checkpointEverySteps:1,seed:p.seed,evaluationSplit:"frozen_eval",optimizer:{method:"grpo",groupSize:f,normalization:"group_standardized",advantageEpsilon:typeof m.advantageEpsilon=="number"?m.advantageEpsilon:1e-8,loss:w.method??"grpo",clipRange:typeof m.clipRange=="number"?m.clipRange:.2,iterations:k,microbatchSize:I(m.microbatchSize,1),gradientAccumulationSteps:I(m.gradientAccumulationSteps,1),adamw:Be.parse(m.adamw)}}}}S(nt,"withAuthoritativeRecipeHashes");function x(e){return e&&typeof e=="object"&&!Array.isArray(e)?e:{}}S(x,"record");function I(e,t){return typeof e=="number"&&Number.isInteger(e)&&e>0?e:t}S(I,"positiveInteger");function rt(e){let t=e.sourceRelease,a=[],r=e.skills??[],o=e.agents??[];if([...r,...o].some(m=>m.visibility!=="policy"))throw new Error("The portable policy Harness requires policy-visible dependencies.");let s=ge({id:"desktop-dependency-lock",path:".openpond/harness/dependency-lock.json",hashInput:{profileHead:e.profileHead??null,sourceRelease:t,tools:a,skills:r.map(({id:m,contentHash:w})=>({id:m,contentHash:w})),agents:o.map(({id:m,contentHash:w})=>({id:m,contentHash:w}))},mediaType:"application/json",visibility:"policy"}),i=ae({schemaVersion:"openpond.agentSnapshot.v2",id:`agent-snapshot-${d([t,a,s.contentHash]).slice(0,24)}`,sourceRelease:t,instructions:[],skills:r,agents:o,toolDeclarations:a,capabilityRequirements:[],dependencyLock:s,portability:{portable:!0,blockers:[],localOnlyAssetRefs:[],hostPrivateAssetRefs:[]},metadata:{sourceReleaseId:t?.id??null}}),l=ge({id:"desktop-harness-program",path:".openpond/harness/program.json",hashInput:{program:"openpond.desktop-agent-loop.v1"},mediaType:"application/json",visibility:"policy"}),p=se({schemaVersion:"openpond.harnessRelease.v2",id:`harness-${d([i.contentHash,l.contentHash,a]).slice(0,24)}`,agentSnapshot:{id:i.id,contentHash:i.contentHash},program:l,tools:a,lifecycle:{create:!0,reset:!0,step:!0,collect:!0,destroy:!0,resetScope:"attempt"},graderInterface:{visibleEvidence:["output","runtime_events","artifacts"],privilegedEvidence:["expected_output","private_verifier"],privateVerifierIsolation:!0},files:[...r,...o],metadata:{runtimeProtocol:"openpond.desktop-agent-loop.v1"}});return{agentSnapshot:i,harnessRelease:p}}S(rt,"createPolicyHarnessContext");function ge(e){let t=L(e.hashInput);return{id:e.id,path:e.path,contentHash:d(e.hashInput),sizeBytes:new TextEncoder().encode(t).byteLength,mediaType:e.mediaType,visibility:e.visibility}}S(ge,"asset");function ot(e){let{taskset:t,modelProject:a,harnessRelease:r,tasksetRelease:o}=e,s=a.trainingSetup,i=e.executionMode==="qualification",l=e.qualificationSamples===void 0?null:n.object({rollout:n.object({groupSize:n.number().int().min(2).max(32)}).passthrough()}).passthrough().parse(e.modelProject.trainingSetup.recipe).rollout.groupSize;if(e.qualificationSamples!==void 0&&(!i||!Number.isInteger(e.qualificationSamples)||e.qualificationSamples<1||e.qualificationSamples>l))throw new Error("Qualification samples must be within the pinned rollout group and cannot be used for training.");if(i&&e.evaluationSource)throw new Error("Qualification does not consume an independent evaluation source.");let p=i?te(e.qualificationSource):null;if(!i&&e.qualificationSource)throw new Error("Qualification sources cannot be used for training.");if(!i&&e.qualificationTaskId!==void 0)throw new Error("Qualification selection cannot be used for training.");if(i&&(!e.qualificationTaskId||!t.tasks.some(u=>u.id===e.qualificationTaskId)))throw new Error("Qualification requires one explicit task ID from its published source.");if(p&&(p.modelResources||p.learningResources||d(t.tasks.map(fe))!==d(p.taskset.tasks)||ve(e.tasksetRelease)!==ve(p.taskset)))throw new Error("Qualification must preserve its ordinary source release and complete task population.");let m=j.parse(r),w=j.parse(o);if(e.openpondRelease,e.workerProtocol,U(t)!==t.contentHash)throw new Error("Taskset authoring state changed after its release was selected.");if(!s.baseModel?.revision||!s.baseModel.tokenizerRevision||!s.baseModel.chatTemplateHash||!s.recipe||!s.tasksetRef||s.tasksetRef.id!==(i?o.id:t.id)||s.tasksetRef.contentHash!==(i?o.contentHash:t.contentHash))throw new Error("Model Run must bind an exact Taskset, Model revision, tokenizer, chat template, and Recipe.");let v=t.profileRelease??{id:`profile_${t.profileId}`,revision:t.revision,contentHash:d({profileId:t.profileId,sourceCommit:t.authoringProvenance.sourceCommit,skillHash:t.authoringProvenance.skillHash})},f=new Map;if(p&&_(f,"qualification-source.json",p),e.evaluationSource){let u=J.parse(e.evaluationSource);if(d(u.taskset)!==d(s.evaluationTasksetRef))throw new Error("The evaluation source differs from the prepared Model configuration.");oe(t.tasks.filter(g=>g.split==="train"),u.tasks),f.set(F,new TextEncoder().encode(L(u)))}else if(!i&&s.evaluationTasksetRef)throw new Error("The prepared Model requires its pinned evaluation source bytes.");let h=e.harnessSource?e.harnessSource.schemaVersion==="openpond.harnessPolicySourcePackage.v1"?ne(e.harnessSource,m):re(e.harnessSource,m):null;if(s.harnessRelease&&!h)throw new Error("A selected Harness requires its complete immutable source package.");if(h&&(!s.harnessRelease||s.harnessRelease.id!==m.id||s.harnessRelease.contentHash!==m.contentHash))throw new Error("Harness source differs from the Model's explicit selection.");if(h){if((e.runtime.placement!=="local"||e.compute.kind!=="local")&&(!h.agentSnapshot.portability.portable||h.schemaVersion==="openpond.harnessSourcePackage.v1"&&h.harnessRelease.files.some(g=>g.visibility!=="policy")))throw new Error("The selected Harness contains source that cannot leave its local host.");_(f,"harness/source-package.json",h)}_(f,"harness/execution.json",ie.parse({schemaVersion:"openpond.harnessSourceSelection.v1",mode:h?"selected_release":"taskset_owned",harnessRelease:m,sourcePackageHash:h?.contentHash??null}));let T=ue(t,e.rewardExecution);if(p&&(!T||d(N(T.binding,T.rewards))!==d(p.taskset.graders)||p.taskset.graders.some(u=>u.kind==="model_judge"||u.kind==="human")))throw new Error("Qualification requires the source package's exact deterministic graders.");if(T){let u=e.verifierAssets??[],g=[...T.rewards.flatMap(b=>b.assets),...N(T.binding,T.rewards).flatMap(b=>b.kind==="custom_verifier"?[b.verifierRef]:b.kind==="model_judge"?[b.rubricRef]:[])],G=new Set(g.map(b=>b.id));if(new Set(u.map(b=>b.id)).size!==u.length||new Set(u.map(b=>b.asset.path)).size!==u.length||u.some(b=>!G.has(b.id)))throw new Error("Unexpected or duplicate private verifier asset.");for(let b of g){let W=u.find(Re=>Re.id===b.id);if(!W||b.visibility!=="verifier")throw new Error("Training bundle requires its private verifier asset.");B(W,b)}_(f,"reward-binding.json",{kind:"reward_binding_v1",binding:T.binding,rewards:T.rewards,assets:u})}else if(e.verifierAssets?.length)throw new Error("Private verifier assets require a published Reward binding.");_(f,"environment.json",{schemaVersion:"openpond.harnessEnvironment.v1",environment:i?{...t.environment,metadata:{...t.environment.metadata,policySystemPrompt:t.objective}}:t.environment,capabilities:t.capabilities,policy:t.policy,outputContract:t.metadata.tasksetOutputContract?.mode==="text"?null:t.metadata.tasksetOutputContract??null}),_(f,"graders.json",{schemaVersion:"openpond.harnessGraders.v1",graders:t.graders,fixtures:t.graderFixtures}),_(f,"tool-contract.json",{schemaVersion:"openpond.harnessToolContract.v1",toolNames:t.environment.toolNames,actionBindings:t.environment.actionBindings??[],capabilities:t.capabilities,connectedAppScopes:t.policy.connectedAppScopes});let y=i?t.tasks.filter(u=>u.id===e.qualificationTaskId):t.tasks.filter(u=>u.split==="train");if(y.length>0)_(f,i?"dataset/qualification.json":"dataset/train.json",{schemaVersion:i?"openpond.qualificationDataset.v1":"openpond.datasetSplit.v1",...i?{selectedTaskId:e.qualificationTaskId}:{split:"train"},tasks:y}),ze({assets:f,tasks:y,tasksetAssetBytes:e.tasksetAssetBytes??new Map});else if(t.datasetArtifact)_(f,"dataset/artifact.json",t.datasetArtifact);else throw new Error("Taskset has no training dataset.");let k=Object.values(t.learningSignals).flat().filter(u=>u.approved);for(let u of k){let g=ye(u);f.set(`evidence/signals/${z(g)}.json`,g)}let R={id:`dataset_${t.id}_r${t.revision}`,contentHash:d({taskset:w,sourceRefs:t.sourceRefs,datasetArtifact:t.datasetArtifact??null,...i?{executionMode:"qualification",tasks:y}:{trainTasks:y}})},M=k.length>0?{id:`evidence_${t.id}_r${t.revision}`,contentHash:d({taskset:w,harnessRelease:m,datasetRelease:R,profileRelease:v,signals:k})}:null,P=ke.parse({schemaVersion:"openpond.resolvedTrainingBundle.v1",...i?{executionMode:"qualification"}:{},...e.qualificationSamples!==void 0?{qualificationSamples:e.qualificationSamples}:{},projection:"trainer",harnessRelease:m,datasetRelease:R,evidenceSetRelease:M,files:[...f.entries()].sort(([u],[g])=>u.localeCompare(g)).map(([u,g])=>({path:u,sha256:z(g),sizeBytes:g.byteLength}))}),q=Se.parse({...P,contentHash:d(P)}),H=be.parse({schemaVersion:"openpond.harnessRunManifest.v1",...i?{executionMode:"qualification"}:{},...e.qualificationSamples!==void 0?{qualificationSamples:e.qualificationSamples}:{},id:`manifest_${e.modelRunId}_${e.approval.approvalHash.slice(0,16)}`,harnessRelease:r,datasetRelease:R,evidenceSets:M?[M]:[],model:{source:s.baseModel.modelId,revision:s.baseModel.revision,artifactHash:null,tokenizerRevision:s.baseModel.tokenizerRevision,chatTemplateHash:s.baseModel.chatTemplateHash},recipe:{method:s.recipe.method,version:"openpond.trainingRecipe.v1",configHash:d(s.recipe)},runtimeTarget:e.runtime,computeTarget:e.compute,engine:e.engine,resolvedBundleHash:q.contentHash,secretLeaseRefs:e.secretLeaseRefs??[],approval:e.approval,createdAt:a.updatedAt});return{manifest:we.parse({...H,contentHash:d(H)}),resolvedBundleManifest:q,assets:f,profileRelease:v,harnessRelease:m,tasksetRelease:w,datasetRelease:R,evidenceSetRelease:M}}S(ot,"buildTasksetTrainingBundle");function ze(e){let t=new Set;for(let a of e.tasks)for(let r of a.assets??[]){t.add(r.artifactRef);let o=e.tasksetAssetBytes.get(r.artifactRef);if(!o)throw new Error(`Resolved Training Bundle is missing Work asset ${r.artifactRef}.`);if(o.byteLength!==r.sizeBytes||z(o)!==r.sha256)throw new Error(`Resolved Training Bundle Work asset ${r.artifactRef} failed immutable verification.`);e.assets.set(r.artifactRef,o)}for(let a of e.tasksetAssetBytes.keys())if(!t.has(a))throw new Error(`Resolved Training Bundle received unexpected Work asset ${a}.`)}S(ze,"addTasksetAssets");function _(e,t,a){e.set(t,ye(a))}S(_,"addJsonAsset");function ye(e){return Buffer.from(L(e),"utf8")}S(ye,"bytes");function ve(e){return d({id:e.id,contentHash:e.contentHash})}S(ve,"releasedRefHash");export{He as a,Ee as b,Ae as c,we as d,Se as e,Me as f,tt as g,st as h,Be as i,nt as j,rt as k,ot as l};

package/dist/chunks/{chunk-U447MS3Y.js → chunk-J3B2GM4A.js}

@@ -1,5 +1,5 @@
 import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{j as sr,v as Bt}from"./chunk-V67IFV26.js";import{a as et}from"./chunk-HIYQPZ7C.js";import{a as tt}from"./chunk-CLFYRN46.js";import{b as at}from"./chunk-ELYRVTTG.js";import{a as or}from"./chunk-AKGCM53D.js";import{a as nr}from"./chunk-FEAKTSRS.js";import{d as Ze}from"./chunk-VZXZ6CHS.js";import{b as Ye}from"./chunk-PAZDQRYU.js";import{a as tr,e as ar,f as rr}from"./chunk-MLTKB5G3.js";import{c as Za,e as He}from"./chunk-TTN63EAM.js";import{a as Mt}from"./chunk-IZ6DLEVY.js";import{a as Ka,b as _e,c as Xa,d as Oe,g as Xe,l as Qe,m as se,n as Ya}from"./chunk-MHYI7ATG.js";import{a as Ga}from"./chunk-RPOD6I64.js";import{a as er}from"./chunk-7KWD2I6W.js";import{c as Fa,y as $a}from"./chunk-3DVWJCOA.js";import{g as he}from"./chunk-FJ3HMHT6.js";import{e as Ra}from"./chunk-MWXW2CA6.js";import{b as Da}from"./chunk-L3KS6MOV.js";import{a as Qa}from"./chunk-IZ56L2ZK.js";import{a as Ct}from"./chunk-WBUK73ST.js";import{A as F,B as Ke,u as Ja,y as Ge}from"./chunk-LNRFRAV5.js";import{a as za,c as Se,d as Je,f as Ua,l as La,p as Va,r as Wa}from"./chunk-CQF6QZEL.js";import{r as Be}from"./chunk-D5ADDBS5.js";import{b as pa,d as fa,e as ha}from"./chunk-CN5OXQ7A.js";import{a as Le,e as ca,f as la,g as ma,h as ua}from"./chunk-OZ6LRNRE.js";import{f as Na}from"./chunk-WFVOG5PU.js";import{Qb as qa,_ as Oa,ha as ja,ta as Me,za as Tt}from"./chunk-M4ZRCRJM.js";import{f as Ha}from"./chunk-LAOLJDVB.js";import{k as ga}from"./chunk-UWHIU24L.js";import{G as We,H as Ea,L as Aa,R as xa,S as Pa,T as Ta,ba as Ca,e as Ia,f as Ve,ga as Ma,n as Sa,pa as Ba,ra as _a}from"./chunk-M6JRS5PA.js";import{Z as da}from"./chunk-KK626T3Z.js";import{M as sa}from"./chunk-6DCXAXVD.js";import{m as ia}from"./chunk-UGOK7VTC.js";import{Da as ba,Pb as Ie,Qb as ya,b as ka}from"./chunk-PBZCXPPC.js";import{r as wa,s as va}from"./chunk-NJLULBTM.js";import{i as oa,l as re,m as fe,n as v}from"./chunk-VVNV4NNW.js";import{A as M}from"./chunk-PF56N7ZY.js";import{a as h}from"./chunk-FTQLQJQV.js";import{execFile as Wr}from"node:child_process";import{readFile as Jr}from"node:fs/promises";import{promisify as Gr}from"node:util";var Kr=5,Xr=20*1024*1024,ir=4e4,Ot=Gr(Wr);function Zn(e){return async({grader:t,task:a,attempt:n})=>{let s=[],o=0,i=await Qr({store:e.store,attempt:n}),d=n.modelRef?.providerId===t.judge.providerId&&n.modelRef.modelId===t.judge.modelId?sr(n.metadata.hostedTokenPricing)??void 0:void 0,g=await e.modelText({model:t.judge,hostedTokenPricing:d,signal:new AbortController().signal,requestId:`task-judge:${n.id}:${t.id}`,onUsage:h((m,r)=>{s.push(m),typeof r=="number"&&Number.isFinite(r)&&r>=0&&(o+=r)},"onUsage"),messages:[{role:"system",content:`Apply this rubric and return JSON only with score (0..1), passed, and feedback.
+import{j as sr,v as Bt}from"./chunk-4PZWGNX4.js";import{a as or}from"./chunk-AKGCM53D.js";import{a as nr}from"./chunk-FEAKTSRS.js";import{a as et}from"./chunk-HIYQPZ7C.js";import{a as tt}from"./chunk-CLFYRN46.js";import{b as at}from"./chunk-CJUVHE4U.js";import{d as Ze}from"./chunk-NE4ESZPT.js";import{b as Ye}from"./chunk-PAE3PQEC.js";import{a as tr,e as ar,f as rr}from"./chunk-SZKFNKWO.js";import{a as Ga}from"./chunk-RTJCELRR.js";import{c as Za,e as He}from"./chunk-NWOK6NLE.js";import{a as Mt}from"./chunk-IZ6DLEVY.js";import{a as Ka,b as _e,c as Xa,d as Oe,g as Xe,l as Qe,m as se,n as Ya}from"./chunk-LRRBYIGH.js";import{a as Qa}from"./chunk-IZ56L2ZK.js";import{a as er}from"./chunk-7KWD2I6W.js";import{c as Fa,y as $a}from"./chunk-B4RKHHF7.js";import{g as he}from"./chunk-557XSPVS.js";import{a as Ct}from"./chunk-WBUK73ST.js";import{A as F,B as Ke,u as Ja,y as Ge}from"./chunk-INO67SEN.js";import{a as za,c as Se,d as Je,f as Ua,l as La,p as Va,r as Wa}from"./chunk-BEI4YULC.js";import{b as pa,d as fa,e as ha}from"./chunk-CN5OXQ7A.js";import{r as Be}from"./chunk-JY3YSI54.js";import{a as Le,e as ca,f as la,g as ma,h as ua}from"./chunk-OZ6LRNRE.js";import{e as Ra}from"./chunk-MWXW2CA6.js";import{b as Da}from"./chunk-L3KS6MOV.js";import{f as Na}from"./chunk-NLQ2UPN2.js";import{Qb as qa,_ as Oa,ha as ja,ta as Me,za as Tt}from"./chunk-PGO4FRHU.js";import{f as Ha}from"./chunk-LAOLJDVB.js";import{k as ga}from"./chunk-UWHIU24L.js";import{G as We,H as Ea,L as Aa,R as xa,S as Pa,T as Ta,ba as Ca,e as Ia,f as Ve,ga as Ma,n as Sa,pa as Ba,ra as _a}from"./chunk-ABF3W2MO.js";import{Z as da}from"./chunk-2EPR4HLF.js";import{M as sa}from"./chunk-BKEDG3A4.js";import{m as ia}from"./chunk-UGOK7VTC.js";import{Da as ba,Pb as Ie,Qb as ya,b as ka}from"./chunk-PBZCXPPC.js";import{r as wa,s as va}from"./chunk-NJLULBTM.js";import{i as oa,l as re,m as fe,n as v}from"./chunk-VVNV4NNW.js";import{A as M}from"./chunk-PF56N7ZY.js";import{a as h}from"./chunk-FTQLQJQV.js";import{execFile as Wr}from"node:child_process";import{readFile as Jr}from"node:fs/promises";import{promisify as Gr}from"node:util";var Kr=5,Xr=20*1024*1024,ir=4e4,Ot=Gr(Wr);function Zn(e){return async({grader:t,task:a,attempt:n})=>{let s=[],o=0,i=await Qr({store:e.store,attempt:n}),d=n.modelRef?.providerId===t.judge.providerId&&n.modelRef.modelId===t.judge.modelId?sr(n.metadata.hostedTokenPricing)??void 0:void 0,g=await e.modelText({model:t.judge,hostedTokenPricing:d,signal:new AbortController().signal,requestId:`task-judge:${n.id}:${t.id}`,onUsage:h((m,r)=>{s.push(m),typeof r=="number"&&Number.isFinite(r)&&r>=0&&(o+=r)},"onUsage"),messages:[{role:"system",content:`Apply this rubric and return JSON only with score (0..1), passed, and feedback.
 
 ${t.rubric}`},{role:"user",content:JSON.stringify({input:a.input,expectedOutput:a.expectedOutput,output:n.output,artifactEvidence:i})}]}),f=rr(g);if(!f)throw new Error("Model judge returned invalid structured output.");return{...f,usage:s,...o>0?{costUsd:o}:{}}}}h(Zn,"createTaskAttemptModelJudge");async function Qr(e){let t=(await e.store.listTaskAttemptArtifacts({attemptId:e.attempt.id})).filter(a=>a.kind==="output_artifact").slice(0,Kr);return Promise.all(t.map(async a=>{let n={artifactId:a.id,mediaType:a.mediaType??null,sha256:a.sha256,sizeBytes:a.sizeBytes,requiredOutputPath:typeof a.metadata.requiredOutputPath=="string"?a.metadata.requiredOutputPath:null};if(a.sizeBytes>Xr)return{...n,extraction:"omitted",reason:"artifact exceeds inspection bound"};try{if(a.mediaType==="application/pdf"){let{stdout:s}=await Ot("pdftotext",[a.path,"-"],{env:{...process.env,PATH:Mt()},maxBuffer:8388608,timeout:15e3});return{...n,extraction:"pdf_text",content:_t(s)}}if(Zr(a.mediaType)){let s=await en(a.path,a.mediaType);return{...n,extraction:"open_xml_text",content:_t(s.content),structure:s.structure}}return Yr(a.mediaType)?{...n,extraction:"text",content:_t(await Jr(a.path,"utf8"))}:{...n,extraction:"metadata_only"}}catch(s){return{...n,extraction:"unavailable",reason:s instanceof Error?s.message.slice(0,500):String(s).slice(0,500)}}}))}h(Qr,"loadTaskAttemptGraderEvidence");function Yr(e){return!!(e?.startsWith("text/")||e==="application/json"||e==="application/javascript"||e==="application/sql"||e==="application/xml"||e==="application/yaml")}h(Yr,"isTextArtifact");function Zr(e){return e==="application/vnd.openxmlformats-officedocument.wordprocessingml.document"||e==="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet"||e==="application/vnd.openxmlformats-officedocument.presentationml.presentation"}h(Zr,"isOpenXmlArtifact");async function en(e,t){let a={env:{...process.env,PATH:Mt()},maxBuffer:16777216,timeout:15e3},{stdout:n}=await Ot("unzip",["-Z1",e],a),s=n.split(/\r?\n/).filter(Boolean),o=tn(s,t).slice(0,250),i=[];for(let d of o){let{stdout:g}=await Ot("unzip",["-p",e,d],a),f=rn(g);f&&i.push(`[${d}]
 ${f}`)}return{content:i.join(`