npm - openpond - Versions diffs - 0.2.57 → 0.2.59 - Mend

openpond 0.2.57 → 0.2.59

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (291) hide show

package/dist/chunks/{chunk-QOGJAQDO.js → chunk-TKLBP3MM.js} RENAMED Viewed

@@ -1,3 +1,3 @@
 import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{k as ne,q as ae,t as ie}from"./chunk-EYPCBBMS.js";import{a as re,b as ze,c as Oe}from"./chunk-6MI4KKWF.js";import{A as He,h as Te,i as Ae,l as Ce,p as te,r as we,s as Me,t as Pe,u as Ee,y as V}from"./chunk-PRGBLQVH.js";import{i as O}from"./chunk-7E6T7GP5.js";import{k as je}from"./chunk-FTEOCCPB.js";import{B as Re,k as Z,m as X,ra as Ie,ua as ee}from"./chunk-25MBPJVY.js";import{x as ke}from"./chunk-523TE4BI.js";import{h as _e}from"./chunk-Y4OWBBDD.js";import{f as z}from"./chunk-FFRFSDBM.js";import{b as n,c as o,d as c,f as r,g}from"./chunk-KGNCKIFP.js";import{b as e}from"./chunk-ES2ISFJC.js";import{a as v}from"./chunk-FTQLQJQV.js";var B=e.string().trim().min(1).max(240),L=e.string().trim().min(1),oe=e.string().trim().min(8).max(256),Ve=e.record(e.string(),e.unknown()).default({}),kt=e.enum(["created","inspecting","awaiting_source_review","awaiting_mapping","awaiting_materialization_approval","materializing","validating","ready","cancelling","cancelled","failed"]),Be=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()}),Rt=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)}),It=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()}),jt=e.object({schemaVersion:e.literal("openpond.huggingFaceDatasetInspection.v1"),id:B,locator:Be,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(It).min(1).max(1e4),columns:e.array(Rt).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:oe.nullable()})).max(1e5),inspectedAt:L,metadata:Ve}),Tt=e.enum(["row_id","cluster_id","prompt","messages","demonstration","chosen","rejected","expected_output","privileged_context","reward","feedback","tag","metadata"]),At=e.object({sourcePath:e.string().trim().min(1).max(1e3),target:Tt,transform:e.enum(["identity","string","json","messages","numeric","math_final_answer"]),policy:e.enum(["visible","privileged","metadata"]),required:e.boolean()}),De=e.object({schemaVersion:e.literal("openpond.datasetImportMapping.v1"),sourceSchemaHash:oe,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(At).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(),Z),validationPercent:e.number().min(0).max(100),frozenEvalPercent:e.number().min(0).max(100)}),importerVersion:e.string().trim().min(1).max(200),mappingHash:oe}),Ct=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()}),Le=e.object({schemaVersion:e.literal("openpond.datasetImportJob.v1"),id:B,profileId:B,sourceKind:e.enum(["huggingface","uploaded_file"]),status:kt,locator:Be.nullable(),inspection:jt.nullable(),mapping:De.nullable(),progress:Ct,targetStorageRoot:e.string().trim().min(1).max(4e3).nullable(),tasksetId:B.nullable(),tasksetRevision:e.number().int().positive().nullable(),artifactId:B.nullable(),error:e.string().trim().min(1).max(2e4).nullable(),cancellationRequested:e.boolean(),createdAt:L,updatedAt:L,completedAt:L.nullable(),metadata:Ve}),pa=e.object({profileId:B,url:e.string().trim().min(1).max(2e3)}),ba=e.object({name:e.string().trim().min(1).max(500),objective:e.string().trim().min(1).max(2e4),mapping:De,targetStorageRoot:e.string().trim().min(1).max(4e3).nullable(),licenseApproved:e.boolean().default(!1)});var wt=e.enum(["draft","active","completed","archived"]),Mt=e.enum(["draft","ready","queued","running","candidate","accepted","rejected","no_signal","failed","cancelled"]),qe=e.enum(["seed","daily_residual","rank_candidate","weekly_rollup","full_refresh"]),Pt=e.enum(["base_model","previous_release","seed_release","accepted_daily_head","accepted_seed"]),Et=e.enum(["seed_taskset","nightly_selection","daily_cohort_union","eligible_task_pool"]),Ht=e.enum(["live_client","evaluation_attempt","replay_evidence","manual"]),zt=e.object({id:n,ordinal:e.number().int().nonnegative(),label:e.string().trim().min(1).max(100),role:qe,parentRule:Pt,taskSource:Et,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."})}),Ot=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(),Ue=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:wt,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:Oe.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(zt).min(1).max(1e3),scheduleSealedAt:c.nullable(),advancementPolicy:Ot,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,I)=>u.ordinal-I.ordinal).some((u,I)=>u.ordinal!==I)&&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,I)=>u+I.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."})}),Ne=e.object({source:Ht,taskIds:e.array(n).min(1).max(1e5),observedFrom:c,observedTo:c,reviewedAt:c,reviewedBy:n,sourceTaskset:g}).strict(),Vt=Ne.extend({derivedFamilyKeys:e.array(n).min(1).max(1e5),familyDerivation:e.object({algorithm:n,sourceTasksetHash:o}).strict(),selectionHash:o}).strict(),se=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()]),Bt=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."})}),Dt=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(),Lt=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(),qt=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(),T=e.object({seriesId:n,entryId:n,scheduleEntryId:n,ordinal:e.number().int().nonnegative(),releaseHash:o}).strict(),$e=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:qe,branch:e.enum(["daily","rank_candidate","weekly_rollup","full_refresh"]),status:Mt,parent:se,taskset:g,sourceTasksets:e.array(g).min(1).max(1e3),taskSelection:Vt.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(Bt).min(1).max(1024),attemptOrdinal:e.number().int().positive().default(1),priorRunAttempts:e.array(qt).max(100).default([]),trainingPlanId:n.nullable(),modelRunId:n.nullable(),modelVersionId:n.nullable(),evaluations:e.array(Lt).max(1e5),decision:Dt.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,_)=>x.offsetStart-_.offsetStart);s.some((x,_)=>x.offsetStart!==(_===0?0:s[_-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,_)=>x!==_+1))&&i.addIssue({code:"custom",path:["priorRunAttempts"],message:"Prior Run attempts must be a contiguous history before the active attempt."})}),xa=e.object({seriesId:n,scheduleEntryId:n,taskSelection:Ne.nullable(),expectedSeriesRevision:e.number().int().positive()}).strict();var C=n.nullable(),A=e.number().int().nonnegative(),Ut=e.string().regex(/^[a-f0-9]{40}$/),Nt=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(),Fe=e.object({schemaVersion:e.literal("openpond.correlatedTelemetryReceipt.v1"),stage:e.enum(["training","evaluation","deployment","inference"]),correlation:e.object({modelRunId:C,modelVersionId:C,policyVersion:A.nullable(),taskId:C,rolloutGroupId:C,providerResourceId:C,deploymentId:C,inferenceRequestId:C}).strict(),spans:e.array(Nt).max(1e4),usage:e.object({promptTokens:A.nullable(),generatedTokens:A.nullable(),gpuSeconds:e.number().nonnegative().nullable(),workerActiveSeconds:e.number().nonnegative().nullable(),optimizerSteps:A.nullable(),rolloutGroups:A.nullable(),successfulTrajectories:A.nullable(),failedTrajectories:A.nullable(),peakGpuMemoryBytes:A.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:A.nullable(),baseProfileId:C,baseRepository:e.string().trim().min(1).max(512).nullable(),baseRevision:Ut.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 $t=e.enum(["base_reference","lora_adapter"]),le=e.object({baseModel:e.object({source:e.literal("huggingface"),repoId:n,revision:e.string().regex(/^[a-f0-9]{40}$/),configHash:o,tokenizerHash:o,licenseId:n,gated:e.boolean()}).strict(),processor:e.object({repository:n,revision:e.string().regex(/^[a-f0-9]{40}$/),configHash:o}).strict()}).strict().superRefine((t,i)=>{(t.baseModel.repoId!==t.processor.repository||t.baseModel.revision!==t.processor.revision)&&i.addIssue({code:"custom",path:["processor"],message:"Reward Model processor must match the pinned scorer base model."})}),We=e.object({schemaVersion:e.literal("openpond.modelVersion.v1"),id:n,modelId:n,profileId:n,version:e.number().int().nonnegative(),kind:$t,status:e.enum(["available","failed"]),baseModel:V,taskset:g,comparisonSeriesEntry:T.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."})}),Ke=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:V,runtime:le.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._/-]+$/),files:e.array(e.object({path:e.string().trim().min(1).max(512),sizeBytes:e.number().int().positive(),sha256:o}).strict()).min(4).max(128)}).strict(),adapter:r,scalarHead:r,bucketHead:r.nullable(),processorRelease:r}).strict(),qualificationReport:r.nullable(),createdAt:c,contentHash:o}).strict(),Ft=e.object({schemaVersion:e.literal("openpond.rewardModelRunReceipt.v1"),provider:n,providerRunId:n,resolvedBundleHash:o,finalCheckpoint:r.extend({objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/)}).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(),Qe=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:Ft.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."})}),Gt=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:Fe.nullable().default(null),contentHash:o}).strict(),w=e.object({inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),totalTokens:e.number().int().nonnegative(),costUsd:e.number().nonnegative().nullable()}).strict(),Ge=e.string().regex(/^(?:[a-f0-9]{40}|[a-f0-9]{64})$/),me=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:te.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(),Je=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(),Ze=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:Je,heldOut:Je}).strict()}).strict(),ce=e.object({baseline:w,adaptation:w,candidateAdaptation:w,candidate:w,refiner:w,grader:w}).strict(),Jt=e.object({usage:ce,observedSpendUsd:e.number().nonnegative(),attempts:e.array(me).max(1e4)}).strict(),Yt=e.object({id:n,contentHash:o,artifactPath:e.string().trim().min(1).max(2e3)}).strict(),Wt=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:ce,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:Ze.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(me).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:Ge,baseCommit:Ge}).strict().nullable(),contentHash:o}).strict(),Kt=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:ce,budget:e.object({maximumSpendUsd:e.number().nonnegative(),observedSpendUsd:e.number().nonnegative(),enforced:e.boolean()}).strict(),evidenceSnapshot:r,attempts:e.array(me).max(1e4),contentHash:o}).strict(),Ye=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."})}),Qt=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:z.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:Ye.nullable()}).strict(),judge:e.object({judgeModel:z,judgeRelease:r,rubricRelease:r,calibrationRelease:r,score:e.number().min(0).max(100),scoreCi95:Ye,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:te.nullable()}).strict()).max(1e5),usage:e.object({policy:w.nullable(),judge:w.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."})}),Zt=e.object({advancedEvaluation:ne.optional(),benchmarkId:e.literal("harness-refiner"),model:z,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(),Xt=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:z.nullable()}).strict(),grader:r,judge:e.object({model:z,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:se,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(),en=e.union([Zt,Xt]),Xe=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:T.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:en.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:Jt.nullable().default(null),advancedAccounting:ie.optional(),evidenceSnapshot:Yt.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([Gt,Wt,Kt,Qt,ae]).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 Na(t){let i=new Map;for(let p of t.attempts){let j=`${p.phase}\0${p.taskId}`,S=i.get(j);(!S||p.startedAt>S.startedAt||p.startedAt===S.startedAt&&p.attemptId>S.attemptId)&&i.set(j,p)}let s=[...i.values()],l=[["adaptation","adaptation","candidate_adaptation"],["held_out","baseline","candidate"]].flatMap(([p,j,S])=>{let _t=new Map(s.filter(k=>k.phase===S).map(k=>[k.taskId,k]));return s.filter(k=>k.phase===j).flatMap(k=>{let Q=_t.get(k.taskId);return Q?[{cohort:p,taskId:k.taskId,baseline:k,refined:Q,tokenDelta:Q.totalTokens-k.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},_=v((p,j)=>({targetTaskCount:j,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=_(l,h),I=l.reduce((p,j)=>p+j.baseline.totalTokens,0),xe=l.reduce((p,j)=>p+j.refined.totalTokens,0),Se=xe-I,ye=l.length===h;return{summary:Ze.parse({target:"all_tasks_lower",...u,baselineTokens:I,refinedTokens:xe,tokenDelta:Se,tokenDeltaPercent:I>0?Se/I*100:null,complete:ye,passed:ye&&u.passedTaskCount===h,cohorts:{adaptation:_(l.filter(p=>p.cohort==="adaptation"),x.adaptation),heldOut:_(l.filter(p=>p.cohort==="held_out"),x.held_out)}}),pairs:l}}v(Na,"summarizeModelEvaluationTaskEfficiency");var tn=e.enum(["measuring","up_to_date","needs_attention"]),nn=e.enum(["fixed","retained","regressed","unresolved","unavailable"]),et=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(),ue=e.object({level:e.literal(.95),lower:e.number().finite(),upper:e.number().finite()}).strict(),an=e.object({panelId:n,panelRole:re,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:ue.nullable(),judgeScore:e.number().min(0).max(100).nullable(),judgeScoreCi95:ue.nullable()}).strict(),tt=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:re,taskId:n,issueFamilyId:n,seed:e.number().int(),repetition:e.number().int().nonnegative(),classification:nn,parentPassed:e.boolean().nullable(),candidatePassed:e.boolean().nullable(),parentAttempt:et.nullable(),candidateAttempt:et.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(an).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:ue.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:tn,evidenceReasons:e.array(n).max(100),projectedAt:c}).strict();var q=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()]),D=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()]),rn=e.union([e.object({taskId:n,target:q,attempt:D}).strict(),e.object({taskId:n,target:e.enum(["base","current"]).default("current"),attempt:D}).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}}))]),ei=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(rn).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}:${on(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."})}),nt=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:D.nullable(),baselineAttempt:D.nullable().default(null),responses:e.array(e.object({target:q,attempt:D}).strict()).max(100).default([]),stagedAt:c.nullable().default(null),queuedEntry:T.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(q).max(100).default([]),runs:e.array(e.object({target:q,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:T.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=>it(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."})}),at=e.object({schemaVersion:e.literal("openpond.continualBenchIssueReview.v1"),id:n,seriesId:n,scheduleEntryId:n,passLabel:n,packet:ze,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:T.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 it(t){return`${t.kind}:${t.id}`}v(it,"responseTargetKey");function on(t){return typeof t=="string"?t:it(t)}v(on,"observedTargetKey");var R=e.string().trim().min(1).max(240),y=e.string().trim().min(8).max(256),U=e.enum(["grpo","dapo","gspo-token"]),P=e.object({id:R,revision:e.string().trim().min(1).max(256),tokenizerRevision:e.string().trim().min(1).max(256),chatTemplateHash:y}),rt=e.object({maxRollouts:e.number().int().positive().max(1e6),maxEnvironmentExecutions:e.number().int().positive().max(1e6),maxInputTokens:e.number().int().positive(),maxOutputTokens:e.number().int().positive(),maxOptimizerSteps:e.number().int().positive().max(1e6),wallTimeMs:e.number().int().positive(),maximumCostUsd:e.number().nonnegative().nullable()}),ot=e.object({method:e.literal("grpo"),groupSize:e.number().int().min(2).max(128),normalization:e.literal("group_standardized"),advantageEpsilon:e.number().positive().max(.01).default(1e-8),loss:U,clipRange:e.number().positive().max(1).default(.2),iterations:e.number().int().min(2).max(16).default(2),microbatchSize:e.number().int().positive().max(128).default(1),gradientAccumulationSteps:e.number().int().positive().max(128).default(1),adamw:O}),N=e.object({method:e.literal("ppo"),valueModel:P,gamma:e.number().min(0).max(1),gaeLambda:e.number().min(0).max(1),policyClip:e.number().positive().max(1),valueClip:e.number().positive().max(10),valueLossCoefficient:e.number().nonnegative().max(10),ppoEpochs:e.number().int().positive().max(100),minibatchSize:e.number().int().positive().max(1e5)}),st=e.discriminatedUnion("method",[ot,N]),$=e.object({rewardModelVersion:e.object({id:R,contentHash:y}).strict(),qualificationReport:e.object({id:R,contentHash:y}).strict().nullable().default(null),evaluationReferences:e.array(e.object({id:R,contentHash:y}).strict()).max(1e3).default([]),checkpoint:e.object({id:R,contentHash:y,objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),files:e.array(e.object({path:e.string().trim().min(1).max(512),sizeBytes:e.number().int().positive(),sha256:y}).strict()).min(4).max(128)}).strict(),runtime:le,processorRelease:e.object({id:R,contentHash:y}).strict(),rewardComposerRelease:e.object({id:R,contentHash:y}).strict(),executionReceipt:e.object({id:R,contentHash:y}).strict()}).strict(),F=e.object({schemaVersion:e.literal("openpond.policyOptimization.v1"),policyModel:P,referenceModel:P,dataset:e.object({tasksetId:R,tasksetHash:y,split:e.literal("train"),selectionStrategy:e.enum(["stable_hash_top_n","rft_easy_curriculum_v1"]),selectionSeed:e.number().int(),maxExamples:e.number().int().positive().max(1e5)}),sampler:e.object({temperature:e.number().min(0).max(2),topP:e.number().min(0).max(1),maxOutputTokens:e.number().int().positive().max(32768),maxTurns:e.number().int().positive().max(1e3),concurrency:e.number().int().positive().max(1e3)}),environment:e.object({id:R,version:e.string().trim().min(1).max(256),toolContractHash:y}),reward:e.object({graderId:R,graderHash:y,learnedPreference:$.nullable().default(null)}),kl:e.object({coefficient:e.number().nonnegative().nullable(),referenceConstraint:e.literal("fixed_reference")}),budgets:rt,checkpointEverySteps:e.number().int().positive(),seed:e.number().int(),evaluationSplit:e.literal("frozen_eval"),optimizer:st});var a=e.string().trim().min(1).max(240),b=e.string().trim().min(1),m=e.string().trim().min(8).max(256),E=e.record(e.string(),e.unknown()).default({});var de=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)}),lt=de.parse({}),mt=e.object({schemaVersion:e.literal("openpond.managedAdapterServingProjection.v1"),teamId:a.nullable().default(null),source:e.literal("sandbox_managed_rl"),sourceRef:a,canonicalArtifactId:a.nullable(),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 sn(t){return t.state==="ready"&&t.canonicalArtifactState==="promotable"&&t.canonicalDeploymentState==="ready"&&t.customerBindingAllowed&&!!t.teamId&&!!t.canonicalArtifactId&&!!t.canonicalDeploymentId}v(sn,"managedAdapterProjectionReady");function di(t){let i=t.managedServing;return i?sn(i)&&i.canonicalArtifactId&&i.canonicalDeploymentId?{kind:"sandbox_serving",canonicalArtifactId:i.canonicalArtifactId,canonicalDeploymentId:i.canonicalDeploymentId}:null:{kind:"source_artifact",canonicalArtifactId:null,canonicalDeploymentId:null}}v(di,"resolveModelBindingEligibility");var fe="openpond.crossSystemOperations.v1",hi="1.0.0",ut="openpond.crossSystemTools.v1",ln=["search_crm","query_billing","search_support","run_python"],mn=e.enum(["train","validation","frozen_eval"]),cn=e.enum(["easy","medium","hard"]),un=e.enum(["renewal_risk_v2"]),fi=e.enum(["renewal_exposure","collections_prioritization","invoice_reconciliation","sla_escalation","contract_billing_mismatch"]),vi=e.object({seed:e.number().int().nonnegative(),split:mn,difficulty:cn,scenarioProfile:un.optional()});var pe={anyOf:[{type:"string",minLength:1,maxLength:256},{type:"null"}],default:null},be={type:"integer",minimum:1,maximum:50,default:25},dt=[{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:pe,limit:be}}},{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:pe,limit:be}}},{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:pe,limit:be}}},{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}}}}],W=Sn({schemaVersion:ut,tools:dt}),dn=[`Use only the four registered synthetic Cross-System Operations tools. Contract ${W}.`,"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(" "),pn=15;function bn(t="auto"){return[dn,"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: ${pn}.`,`Registered tool signatures: ${dt.map(gn).join("; ")}`].join(`
+import{k as ne,q as ae,t as ie}from"./chunk-RBZH6A6M.js";import{a as re,b as ze,c as Oe}from"./chunk-6MI4KKWF.js";import{A as He,h as Te,i as Ae,l as Ce,p as te,r as we,s as Me,t as Pe,u as Ee,y as V}from"./chunk-JMBQYBPI.js";import{i as O}from"./chunk-LPWBIGKW.js";import{k as je}from"./chunk-25JDHAZ5.js";import{B as Re,k as Z,m as X,ra as Ie,ua as ee}from"./chunk-XMKCCLGB.js";import{x as ke}from"./chunk-523TE4BI.js";import{h as _e}from"./chunk-Y4OWBBDD.js";import{f as z}from"./chunk-FFRFSDBM.js";import{b as n,c as o,d as c,f as r,g}from"./chunk-KGNCKIFP.js";import{b as e}from"./chunk-ES2ISFJC.js";import{a as v}from"./chunk-FTQLQJQV.js";var B=e.string().trim().min(1).max(240),L=e.string().trim().min(1),oe=e.string().trim().min(8).max(256),Ve=e.record(e.string(),e.unknown()).default({}),kt=e.enum(["created","inspecting","awaiting_source_review","awaiting_mapping","awaiting_materialization_approval","materializing","validating","ready","cancelling","cancelled","failed"]),Be=e.object({schemaVersion:e.literal("openpond.huggingFaceDatasetLocator.v1"),repositoryId:e.string().trim().regex(/^[^/\s]+\/[^/\s]+$/).max(500),repositoryUrl:e.string().url().max(2e3),requestedRevision:e.string().trim().min(1).max(256).nullable()}),Rt=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)}),It=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()}),jt=e.object({schemaVersion:e.literal("openpond.huggingFaceDatasetInspection.v1"),id:B,locator:Be,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(It).min(1).max(1e4),columns:e.array(Rt).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:oe.nullable()})).max(1e5),inspectedAt:L,metadata:Ve}),Tt=e.enum(["row_id","cluster_id","prompt","messages","demonstration","chosen","rejected","expected_output","privileged_context","reward","feedback","tag","metadata"]),At=e.object({sourcePath:e.string().trim().min(1).max(1e3),target:Tt,transform:e.enum(["identity","string","json","messages","numeric","math_final_answer"]),policy:e.enum(["visible","privileged","metadata"]),required:e.boolean()}),De=e.object({schemaVersion:e.literal("openpond.datasetImportMapping.v1"),sourceSchemaHash:oe,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(At).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(),Z),validationPercent:e.number().min(0).max(100),frozenEvalPercent:e.number().min(0).max(100)}),importerVersion:e.string().trim().min(1).max(200),mappingHash:oe}),Ct=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()}),Le=e.object({schemaVersion:e.literal("openpond.datasetImportJob.v1"),id:B,profileId:B,sourceKind:e.enum(["huggingface","uploaded_file"]),status:kt,locator:Be.nullable(),inspection:jt.nullable(),mapping:De.nullable(),progress:Ct,targetStorageRoot:e.string().trim().min(1).max(4e3).nullable(),tasksetId:B.nullable(),tasksetRevision:e.number().int().positive().nullable(),artifactId:B.nullable(),error:e.string().trim().min(1).max(2e4).nullable(),cancellationRequested:e.boolean(),createdAt:L,updatedAt:L,completedAt:L.nullable(),metadata:Ve}),pa=e.object({profileId:B,url:e.string().trim().min(1).max(2e3)}),ba=e.object({name:e.string().trim().min(1).max(500),objective:e.string().trim().min(1).max(2e4),mapping:De,targetStorageRoot:e.string().trim().min(1).max(4e3).nullable(),licenseApproved:e.boolean().default(!1)});var wt=e.enum(["draft","active","completed","archived"]),Mt=e.enum(["draft","ready","queued","running","candidate","accepted","rejected","no_signal","failed","cancelled"]),qe=e.enum(["seed","daily_residual","rank_candidate","weekly_rollup","full_refresh"]),Pt=e.enum(["base_model","previous_release","seed_release","accepted_daily_head","accepted_seed"]),Et=e.enum(["seed_taskset","nightly_selection","daily_cohort_union","eligible_task_pool"]),Ht=e.enum(["live_client","evaluation_attempt","replay_evidence","manual"]),zt=e.object({id:n,ordinal:e.number().int().nonnegative(),label:e.string().trim().min(1).max(100),role:qe,parentRule:Pt,taskSource:Et,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."})}),Ot=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(),Ue=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:wt,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:Oe.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(zt).min(1).max(1e3),scheduleSealedAt:c.nullable(),advancementPolicy:Ot,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,I)=>u.ordinal-I.ordinal).some((u,I)=>u.ordinal!==I)&&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,I)=>u+I.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."})}),Ne=e.object({source:Ht,taskIds:e.array(n).min(1).max(1e5),observedFrom:c,observedTo:c,reviewedAt:c,reviewedBy:n,sourceTaskset:g}).strict(),Vt=Ne.extend({derivedFamilyKeys:e.array(n).min(1).max(1e5),familyDerivation:e.object({algorithm:n,sourceTasksetHash:o}).strict(),selectionHash:o}).strict(),se=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()]),Bt=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."})}),Dt=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(),Lt=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(),qt=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(),T=e.object({seriesId:n,entryId:n,scheduleEntryId:n,ordinal:e.number().int().nonnegative(),releaseHash:o}).strict(),$e=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:qe,branch:e.enum(["daily","rank_candidate","weekly_rollup","full_refresh"]),status:Mt,parent:se,taskset:g,sourceTasksets:e.array(g).min(1).max(1e3),taskSelection:Vt.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(Bt).min(1).max(1024),attemptOrdinal:e.number().int().positive().default(1),priorRunAttempts:e.array(qt).max(100).default([]),trainingPlanId:n.nullable(),modelRunId:n.nullable(),modelVersionId:n.nullable(),evaluations:e.array(Lt).max(1e5),decision:Dt.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,_)=>x.offsetStart-_.offsetStart);s.some((x,_)=>x.offsetStart!==(_===0?0:s[_-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,_)=>x!==_+1))&&i.addIssue({code:"custom",path:["priorRunAttempts"],message:"Prior Run attempts must be a contiguous history before the active attempt."})}),xa=e.object({seriesId:n,scheduleEntryId:n,taskSelection:Ne.nullable(),expectedSeriesRevision:e.number().int().positive()}).strict();var C=n.nullable(),A=e.number().int().nonnegative(),Ut=e.string().regex(/^[a-f0-9]{40}$/),Nt=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(),Fe=e.object({schemaVersion:e.literal("openpond.correlatedTelemetryReceipt.v1"),stage:e.enum(["training","evaluation","deployment","inference"]),correlation:e.object({modelRunId:C,modelVersionId:C,policyVersion:A.nullable(),taskId:C,rolloutGroupId:C,providerResourceId:C,deploymentId:C,inferenceRequestId:C}).strict(),spans:e.array(Nt).max(1e4),usage:e.object({promptTokens:A.nullable(),generatedTokens:A.nullable(),gpuSeconds:e.number().nonnegative().nullable(),workerActiveSeconds:e.number().nonnegative().nullable(),optimizerSteps:A.nullable(),rolloutGroups:A.nullable(),successfulTrajectories:A.nullable(),failedTrajectories:A.nullable(),peakGpuMemoryBytes:A.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:A.nullable(),baseProfileId:C,baseRepository:e.string().trim().min(1).max(512).nullable(),baseRevision:Ut.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 $t=e.enum(["base_reference","lora_adapter"]),le=e.object({baseModel:e.object({source:e.literal("huggingface"),repoId:n,revision:e.string().regex(/^[a-f0-9]{40}$/),configHash:o,tokenizerHash:o,licenseId:n,gated:e.boolean()}).strict(),processor:e.object({repository:n,revision:e.string().regex(/^[a-f0-9]{40}$/),configHash:o}).strict()}).strict().superRefine((t,i)=>{(t.baseModel.repoId!==t.processor.repository||t.baseModel.revision!==t.processor.revision)&&i.addIssue({code:"custom",path:["processor"],message:"Reward Model processor must match the pinned scorer base model."})}),We=e.object({schemaVersion:e.literal("openpond.modelVersion.v1"),id:n,modelId:n,profileId:n,version:e.number().int().nonnegative(),kind:$t,status:e.enum(["available","failed"]),baseModel:V,taskset:g,comparisonSeriesEntry:T.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."})}),Ke=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:V,runtime:le.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._/-]+$/),files:e.array(e.object({path:e.string().trim().min(1).max(512),sizeBytes:e.number().int().positive(),sha256:o}).strict()).min(4).max(128)}).strict(),adapter:r,scalarHead:r,bucketHead:r.nullable(),processorRelease:r}).strict(),qualificationReport:r.nullable(),createdAt:c,contentHash:o}).strict(),Ft=e.object({schemaVersion:e.literal("openpond.rewardModelRunReceipt.v1"),provider:n,providerRunId:n,resolvedBundleHash:o,finalCheckpoint:r.extend({objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/)}).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(),Qe=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:Ft.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."})}),Gt=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:Fe.nullable().default(null),contentHash:o}).strict(),w=e.object({inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),totalTokens:e.number().int().nonnegative(),costUsd:e.number().nonnegative().nullable()}).strict(),Ge=e.string().regex(/^(?:[a-f0-9]{40}|[a-f0-9]{64})$/),me=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:te.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(),Je=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(),Ze=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:Je,heldOut:Je}).strict()}).strict(),ce=e.object({baseline:w,adaptation:w,candidateAdaptation:w,candidate:w,refiner:w,grader:w}).strict(),Jt=e.object({usage:ce,observedSpendUsd:e.number().nonnegative(),attempts:e.array(me).max(1e4)}).strict(),Yt=e.object({id:n,contentHash:o,artifactPath:e.string().trim().min(1).max(2e3)}).strict(),Wt=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:ce,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:Ze.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(me).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:Ge,baseCommit:Ge}).strict().nullable(),contentHash:o}).strict(),Kt=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:ce,budget:e.object({maximumSpendUsd:e.number().nonnegative(),observedSpendUsd:e.number().nonnegative(),enforced:e.boolean()}).strict(),evidenceSnapshot:r,attempts:e.array(me).max(1e4),contentHash:o}).strict(),Ye=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."})}),Qt=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:z.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:Ye.nullable()}).strict(),judge:e.object({judgeModel:z,judgeRelease:r,rubricRelease:r,calibrationRelease:r,score:e.number().min(0).max(100),scoreCi95:Ye,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:te.nullable()}).strict()).max(1e5),usage:e.object({policy:w.nullable(),judge:w.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."})}),Zt=e.object({advancedEvaluation:ne.optional(),benchmarkId:e.literal("harness-refiner"),model:z,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(),Xt=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:z.nullable()}).strict(),grader:r,judge:e.object({model:z,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:se,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(),en=e.union([Zt,Xt]),Xe=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:T.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:en.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:Jt.nullable().default(null),advancedAccounting:ie.optional(),evidenceSnapshot:Yt.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([Gt,Wt,Kt,Qt,ae]).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 Na(t){let i=new Map;for(let p of t.attempts){let j=`${p.phase}\0${p.taskId}`,S=i.get(j);(!S||p.startedAt>S.startedAt||p.startedAt===S.startedAt&&p.attemptId>S.attemptId)&&i.set(j,p)}let s=[...i.values()],l=[["adaptation","adaptation","candidate_adaptation"],["held_out","baseline","candidate"]].flatMap(([p,j,S])=>{let _t=new Map(s.filter(k=>k.phase===S).map(k=>[k.taskId,k]));return s.filter(k=>k.phase===j).flatMap(k=>{let Q=_t.get(k.taskId);return Q?[{cohort:p,taskId:k.taskId,baseline:k,refined:Q,tokenDelta:Q.totalTokens-k.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},_=v((p,j)=>({targetTaskCount:j,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=_(l,h),I=l.reduce((p,j)=>p+j.baseline.totalTokens,0),xe=l.reduce((p,j)=>p+j.refined.totalTokens,0),Se=xe-I,ye=l.length===h;return{summary:Ze.parse({target:"all_tasks_lower",...u,baselineTokens:I,refinedTokens:xe,tokenDelta:Se,tokenDeltaPercent:I>0?Se/I*100:null,complete:ye,passed:ye&&u.passedTaskCount===h,cohorts:{adaptation:_(l.filter(p=>p.cohort==="adaptation"),x.adaptation),heldOut:_(l.filter(p=>p.cohort==="held_out"),x.held_out)}}),pairs:l}}v(Na,"summarizeModelEvaluationTaskEfficiency");var tn=e.enum(["measuring","up_to_date","needs_attention"]),nn=e.enum(["fixed","retained","regressed","unresolved","unavailable"]),et=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(),ue=e.object({level:e.literal(.95),lower:e.number().finite(),upper:e.number().finite()}).strict(),an=e.object({panelId:n,panelRole:re,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:ue.nullable(),judgeScore:e.number().min(0).max(100).nullable(),judgeScoreCi95:ue.nullable()}).strict(),tt=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:re,taskId:n,issueFamilyId:n,seed:e.number().int(),repetition:e.number().int().nonnegative(),classification:nn,parentPassed:e.boolean().nullable(),candidatePassed:e.boolean().nullable(),parentAttempt:et.nullable(),candidateAttempt:et.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(an).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:ue.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:tn,evidenceReasons:e.array(n).max(100),projectedAt:c}).strict();var q=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()]),D=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()]),rn=e.union([e.object({taskId:n,target:q,attempt:D}).strict(),e.object({taskId:n,target:e.enum(["base","current"]).default("current"),attempt:D}).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}}))]),ei=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(rn).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}:${on(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."})}),nt=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:D.nullable(),baselineAttempt:D.nullable().default(null),responses:e.array(e.object({target:q,attempt:D}).strict()).max(100).default([]),stagedAt:c.nullable().default(null),queuedEntry:T.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(q).max(100).default([]),runs:e.array(e.object({target:q,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:T.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=>it(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."})}),at=e.object({schemaVersion:e.literal("openpond.continualBenchIssueReview.v1"),id:n,seriesId:n,scheduleEntryId:n,passLabel:n,packet:ze,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:T.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 it(t){return`${t.kind}:${t.id}`}v(it,"responseTargetKey");function on(t){return typeof t=="string"?t:it(t)}v(on,"observedTargetKey");var R=e.string().trim().min(1).max(240),y=e.string().trim().min(8).max(256),U=e.enum(["grpo","dapo","gspo-token"]),P=e.object({id:R,revision:e.string().trim().min(1).max(256),tokenizerRevision:e.string().trim().min(1).max(256),chatTemplateHash:y}),rt=e.object({maxRollouts:e.number().int().positive().max(1e6),maxEnvironmentExecutions:e.number().int().positive().max(1e6),maxInputTokens:e.number().int().positive(),maxOutputTokens:e.number().int().positive(),maxOptimizerSteps:e.number().int().positive().max(1e6),wallTimeMs:e.number().int().positive(),maximumCostUsd:e.number().nonnegative().nullable()}),ot=e.object({method:e.literal("grpo"),groupSize:e.number().int().min(2).max(128),normalization:e.literal("group_standardized"),advantageEpsilon:e.number().positive().max(.01).default(1e-8),loss:U,clipRange:e.number().positive().max(1).default(.2),iterations:e.number().int().min(2).max(16).default(2),microbatchSize:e.number().int().positive().max(128).default(1),gradientAccumulationSteps:e.number().int().positive().max(128).default(1),adamw:O}),N=e.object({method:e.literal("ppo"),valueModel:P,gamma:e.number().min(0).max(1),gaeLambda:e.number().min(0).max(1),policyClip:e.number().positive().max(1),valueClip:e.number().positive().max(10),valueLossCoefficient:e.number().nonnegative().max(10),ppoEpochs:e.number().int().positive().max(100),minibatchSize:e.number().int().positive().max(1e5)}),st=e.discriminatedUnion("method",[ot,N]),$=e.object({rewardModelVersion:e.object({id:R,contentHash:y}).strict(),qualificationReport:e.object({id:R,contentHash:y}).strict().nullable().default(null),evaluationReferences:e.array(e.object({id:R,contentHash:y}).strict()).max(1e3).default([]),checkpoint:e.object({id:R,contentHash:y,objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),files:e.array(e.object({path:e.string().trim().min(1).max(512),sizeBytes:e.number().int().positive(),sha256:y}).strict()).min(4).max(128)}).strict(),runtime:le,processorRelease:e.object({id:R,contentHash:y}).strict(),rewardComposerRelease:e.object({id:R,contentHash:y}).strict(),executionReceipt:e.object({id:R,contentHash:y}).strict()}).strict(),F=e.object({schemaVersion:e.literal("openpond.policyOptimization.v1"),policyModel:P,referenceModel:P,dataset:e.object({tasksetId:R,tasksetHash:y,split:e.literal("train"),selectionStrategy:e.enum(["stable_hash_top_n","rft_easy_curriculum_v1"]),selectionSeed:e.number().int(),maxExamples:e.number().int().positive().max(1e5)}),sampler:e.object({temperature:e.number().min(0).max(2),topP:e.number().min(0).max(1),maxOutputTokens:e.number().int().positive().max(32768),maxTurns:e.number().int().positive().max(1e3),concurrency:e.number().int().positive().max(1e3)}),environment:e.object({id:R,version:e.string().trim().min(1).max(256),toolContractHash:y}),reward:e.object({graderId:R,graderHash:y,learnedPreference:$.nullable().default(null)}),kl:e.object({coefficient:e.number().nonnegative().nullable(),referenceConstraint:e.literal("fixed_reference")}),budgets:rt,checkpointEverySteps:e.number().int().positive(),seed:e.number().int(),evaluationSplit:e.literal("frozen_eval"),optimizer:st});var a=e.string().trim().min(1).max(240),b=e.string().trim().min(1),m=e.string().trim().min(8).max(256),E=e.record(e.string(),e.unknown()).default({});var de=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)}),lt=de.parse({}),mt=e.object({schemaVersion:e.literal("openpond.managedAdapterServingProjection.v1"),teamId:a.nullable().default(null),source:e.literal("sandbox_managed_rl"),sourceRef:a,canonicalArtifactId:a.nullable(),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 sn(t){return t.state==="ready"&&t.canonicalArtifactState==="promotable"&&t.canonicalDeploymentState==="ready"&&t.customerBindingAllowed&&!!t.teamId&&!!t.canonicalArtifactId&&!!t.canonicalDeploymentId}v(sn,"managedAdapterProjectionReady");function di(t){let i=t.managedServing;return i?sn(i)&&i.canonicalArtifactId&&i.canonicalDeploymentId?{kind:"sandbox_serving",canonicalArtifactId:i.canonicalArtifactId,canonicalDeploymentId:i.canonicalDeploymentId}:null:{kind:"source_artifact",canonicalArtifactId:null,canonicalDeploymentId:null}}v(di,"resolveModelBindingEligibility");var fe="openpond.crossSystemOperations.v1",hi="1.0.0",ut="openpond.crossSystemTools.v1",ln=["search_crm","query_billing","search_support","run_python"],mn=e.enum(["train","validation","frozen_eval"]),cn=e.enum(["easy","medium","hard"]),un=e.enum(["renewal_risk_v2"]),fi=e.enum(["renewal_exposure","collections_prioritization","invoice_reconciliation","sla_escalation","contract_billing_mismatch"]),vi=e.object({seed:e.number().int().nonnegative(),split:mn,difficulty:cn,scenarioProfile:un.optional()});var pe={anyOf:[{type:"string",minLength:1,maxLength:256},{type:"null"}],default:null},be={type:"integer",minimum:1,maximum:50,default:25},dt=[{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:pe,limit:be}}},{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:pe,limit:be}}},{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:pe,limit:be}}},{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}}}}],W=Sn({schemaVersion:ut,tools:dt}),dn=[`Use only the four registered synthetic Cross-System Operations tools. Contract ${W}.`,"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(" "),pn=15;function bn(t="auto"){return[dn,"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: ${pn}.`,`Registered tool signatures: ${dt.map(gn).join("; ")}`].join(`
 `)}v(bn,"crossSystemLocalToolSystemPrompt");var xi=bn();function gn(t){let i=J(t.parameters),s=J(i.properties),l=pt(i.required),f=Object.fromEntries(l.map(h=>[h,G(s[h])]));return`${t.name}(${JSON.stringify(f)})`}v(gn,"compactToolSignature");function G(t){let i=J(t),s=ge(i.anyOf);if(s.length)return s.map(G).join("|");let l=ge(i.enum);if(l.length)return l.join("|");if(i.type==="array")return[G(i.items)];if(i.type==="object"){let f=J(i.properties),h=pt(i.required);return Object.fromEntries(h.map(x=>[x,G(f[x])]))}return typeof i.type=="string"?i.type:"unknown"}v(G,"compactSchema");function J(t){return t&&typeof t=="object"&&!Array.isArray(t)?t:{}}v(J,"recordValue");function ge(t){return Array.isArray(t)?t:[]}v(ge,"arrayValue");function pt(t){return ge(t).filter(i=>typeof i=="string")}v(pt,"stringValues");var Y=e.enum(ln),hn=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:Y,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:Y,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)})]),bt=e.object({schemaVersion:e.literal(fe),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(W),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(hn).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({})}),gt=e.object({schemaVersion:e.literal(fe),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)}),fn=e.object({id:e.string().trim().min(1).max(240),type:e.literal("function"),function:e.object({name:Y,arguments:e.string().max(2e4)})}),ht=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(fn).max(4).optional()}),Si=e.object({schemaVersion:e.literal(fe),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(W),approval:e.object({status:e.literal("approved"),approvedBy:e.string().trim().min(1).max(240),approvedAt:e.string().trim().min(1)}),messages:e.array(ht).min(3).max(1e3)}),vn=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(Y).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(ht).min(3).max(1e3)}),xn=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()}),yi=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(W),status:e.enum(["ready_for_review","approved"]),approval:xn.nullable(),tasks:e.array(vn).min(1).max(100)});function Sn(t){let i=he(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`${ut}:${ct(s)}${ct(l)}`}v(Sn,"contractHash");function he(t){return Array.isArray(t)?`[${t.map(he).join(",")}]`:t&&typeof t=="object"?`{${Object.entries(t).sort(([i],[s])=>i.localeCompare(s)).map(([i,s])=>`${JSON.stringify(i)}:${he(s)}`).join(",")}}`:JSON.stringify(t)}v(he,"stableJson");function ct(t){return(t>>>0).toString(16).padStart(8,"0")}v(ct,"hex");var yn=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:E}),_n=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}),kn=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)}),ft=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([bt,yn,_n]).nullable(),verifier:e.union([gt,kn]).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 M=e.enum(["sft","dpo","grpo","ppo","sdft","opd","opsd","sdpo"]),K=e.enum(["lora","full"]),H=e.enum(["openpond_managed"]),Rn=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()})})),Ki=d(()=>({id:a,input:e.record(e.string(),e.unknown()),expectedOutput:e.record(e.string(),e.unknown()),tags:e.array(a).max(100).default([])})),Qi=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)})),Zi=d(()=>({id:a,input:e.record(e.string(),e.unknown()),tags:e.array(a).max(100).default([])})),Xi=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."})}),In=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:O}),loss:e.object({method:U.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:$.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:F.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()})),jn=d(()=>({schemaVersion:e.literal("openpond.dpoRecipe.v1"),method:e.literal("dpo"),parameterization:e.literal("lora"),policyModel:P,referenceModel:P,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()})})),Tn=d(()=>({schemaVersion:e.literal("openpond.ppoRecipe.v1"),method:e.literal("ppo"),parameterization:e.literal("lora"),policyOptimization:F.extend({optimizer:N}),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()})})),An=d(()=>({schemaVersion:e.literal("openpond.unsupportedRecipe.v1"),method:M.exclude(["sft","dpo","grpo","ppo"]),parameterization:K,unsupportedReason:e.string().trim().min(1).max(5e3)})),Cn=e.union([Rn,jn,In,Tn,An]),wn=d(()=>({schemaVersion:e.literal("openpond.trainingDestinationCapabilities.v1"),destinationId:H,available:e.boolean(),methods:e.array(M),parameterizations:e.array(K),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})),Mn=d(()=>({destinationId:H,available:e.boolean(),methods:e.array(M),parameterizations:e.array(K),nonProduction:e.boolean(),unavailableReason:e.string().trim().min(1).max(5e3).nullable()})),Pn=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:V,available:e.boolean(),nonProduction:e.boolean(),unavailableReason:e.string().trim().min(1).max(5e3).nullable(),methods:e.array(M),executionOptions:e.array(Mn).min(1)})),En=d(()=>({code:a,severity:e.enum(["warning","error"]),path:e.string().trim().max(2e3).nullable(),message:e.string().trim().min(1).max(5e3)})),Hn=d(()=>({schemaVersion:e.literal("openpond.trainingCompatibility.v1"),compatible:e.boolean(),destinationId:H,tasksetId:a,recipeMethod:M,issues:e.array(En),checkedAt:b})),zn=e.union([Ie,e.enum(["destination_unavailable","destination_method_unsupported","destination_model_unsupported","destination_parameterization_unsupported","experimental_destination"])]),er=d(()=>({method:M,state:e.enum(["recommended","compatible","needs_dataset_work","destination_unavailable","experimental_destination"]),reasonCodes:e.array(zn).default([]),reasons:e.array(e.string().trim().min(1).max(5e3)).default([]),destinationId:H.nullable().default(null)})),tr=e.enum(["small","standard","custom","small_experiment"]);var On=ke,St=d(()=>({schemaVersion:e.literal("openpond.trainingPlan.v1"),id:a,modelId:a,tasksetId:a,tasksetHash:m,comparisonSeriesEntry:T.nullable().optional(),evaluationTasksetRef:g.nullable().optional(),harnessRelease:r.nullable().optional(),modelImprovementQualification:r.nullable().optional(),destinationId:H,recipe:Cn,environmentPlacement:e.enum(["none","local","remote","colocated","provider_native"]),compatibility:Hn,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})),Vn=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"])})),ve=d(()=>({schemaVersion:e.literal("openpond.trainingBundle.v1"),id:a,planId:a,tasksetId:a,tasksetHash:m,recipeHash:m,files:e.array(Vn).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})),nr=d(()=>({schemaVersion:e.literal("openpond.trainingBundleExport.v1"),manifest:ve,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})),ar=d(()=>({schemaVersion:e.literal("openpond.trainingPreparedStart.v1"),plan:St,bundle:ve,approvalActor:a.nullable(),preparedAt:b})),ir=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:H,modelId:a,method:M,parameterization:K,maximumCostUsd:e.number().nonnegative().nullable(),approvedBy:a,approvedAt:b})),rr=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:V,method:M})).strict(),Bn=e.enum(["queued","starting","running","cancelling","cancelled","succeeded","failed","reconciling"]),yt=d(()=>({schemaVersion:e.literal("openpond.trainingJob.v1"),id:a,planId:a,bundleHash:m,approvalId:a,destinationId:H,status:Bn,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:E})),Dn=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:E})),Ln=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()})),qn=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()})),Un=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})),vt=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()})),xt=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)})),Nn=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:xt.nullable(),trainedGrade:xt.nullable()})),$n=d(()=>({schemaVersion:e.literal("openpond.trainingEvaluationSummary.v1"),jobId:a,tasksetId:a,base:vt,trained:vt,meanScoreDelta:e.number().min(-1).max(1).nullable(),examples:e.array(Nn).max(1e6)})),or=d(()=>({schemaVersion:e.literal("openpond.trainingRunDetail.v1"),job:yt,events:e.array(Dn),stepMetrics:e.array(Ln),policyMetrics:e.array(qn).default([]),managedEvidence:Un.nullable().default(null),evaluation:$n.nullable(),generatedAt:b})),Fn=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:E})),Gn=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:de.default(lt),managedServing:mt.nullable().default(null)})),Jn=e.enum(["chat_manual","agent","extension","authoring_optimizer"]),Yn=d(()=>({schemaVersion:e.literal("openpond.modelBinding.v1"),id:a,profileId:a,role:Jn,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:E})),sr=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})),lr=d(()=>({schemaVersion:e.literal("openpond.tasksetOperationalState.v1"),tasksetId:a,attempts:e.array(we),artifacts:e.array(Me),grades:e.array(Pe),evaluationResults:e.array(_e).default([]),generatedAt:b})),mr=d(()=>({schemaVersion:e.literal("openpond.trainingState.v1"),profileId:a,sources:e.array(Re),creations:e.array(He),tasksetDrafts:e.array(je).default([]),tasksets:e.array(ee),benchmarkRuns:e.array(e.custom()).default([]),benchmarkComparisons:e.array(e.custom()).default([]),datasetImports:e.array(Le).default([]),datasetArtifacts:e.array(X).default([]),graderAuditReports:e.array(Ee),evaluationResults:e.array(e.custom()).default([]),candidates:e.array(Te),minerConfig:Ae,minerRuns:e.array(Ce).default([]),modelProjects:e.array(On).default([]),modelVersions:e.array(We).default([]),modelRuns:e.array(Xe).default([]),comparisonSeries:e.array(Ue).default([]),comparisonSeriesEntries:e.array($e).default([]),modelCurrencySnapshots:e.array(tt).default([]),continualBenchIssueReviews:e.array(at).default([]),continualLearningDailyBatches:e.array(nt).default([]),rewardModelVersions:e.array(Ke).default([]),rewardModelRuns:e.array(Qe).default([]),modelTasksets:e.array(ee).default([]),plans:e.array(St),bundles:e.array(ve),jobs:e.array(yt),artifacts:e.array(Fn),models:e.array(Gn),rolloutReceipts:e.array(ft).default([]),modelBindings:e.array(Yn).default([]),destinations:e.array(wn),baseModelCandidates:e.array(Pn).default([]),activityRevision:m.optional(),generatedAt:b}));export{Be as a,jt as b,De as c,Le as d,pa as e,ba as f,Mt as g,Ue as h,Vt as i,Dt as j,Lt as k,T as l,$e as m,xa as n,We as o,Ke as p,Ft as q,Qe as r,Wt as s,Qt as t,Xe as u,Na as v,tt as w,q as x,ei as y,nt as z,at as A,$ as B,de as C,mt as D,sn as E,di as F,fe as G,hi as H,ut as I,ln as J,vi as K,dt as L,W as M,xi as N,bt as O,gt as P,ht as Q,Si as R,xn as S,yi as T,ft as U,H as V,Xi as W,Cn as X,wn as Y,Pn as Z,Hn as _,On as $,St as aa,ve as ba,nr as ca,ar as da,ir as ea,rr as fa,yt as ga,Dn as ha,Ln as ia,qn as ja,Un as ka,$n as la,or as ma,Fn as na,Gn as oa,Jn as pa,Yn as qa,lr as ra};

package/dist/chunks/{chunk-M4JWQSV7.js → chunk-U2MAPON2.js} RENAMED Viewed

@@ -1,2 +1,2 @@
 import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{e as g,g as $,h as O,i as C}from"./chunk-FV2R5RUN.js";import{a as w}from"./chunk-3M53YXEI.js";import{a as r}from"./chunk-FTQLQJQV.js";var L=Symbol("openpondCliOptionValues"),a=class extends Error{static{r(this,"CliUsageError")}exitCode=2;constructor(t){super(t),this.name="CliUsageError"}},K=new Set(["maxOutputBytes","timeoutSec"]);function k(n){let t=[...n],o="",i={},e={},u=[];for(;t.length>0;){let s=t.shift();if(s==="--"){u.push(...t);break}if(s.startsWith("--")){let f=s.indexOf("="),c=f>=0?s.slice(2,f):s.slice(2),p=N(c),d=f>=0?s.slice(f+1):P(t,p,o);h(i,e,p,d,o)}else _(s)?y(s,t,i,e):o?u.push(s):o=s}return Object.defineProperty(i,L,{value:e,enumerable:!1,configurable:!1,writable:!1}),V(o,i),{command:o,options:i,rest:u}}r(k,"parseArgs");function N(n){return n.replace(/-([a-z])/g,(t,o)=>o.toUpperCase())}r(N,"normalizeCliOptionKey");function _(n){return/^-[^-].*/.test(n)&&!S(n)}r(_,"isShortOptionToken");function y(n,t,o,i){let e=n.slice(1);if(!e)return;let u=e.indexOf("="),s=u>=0?e.slice(0,u):e,f=u>=0?e.slice(u+1):null;if(s.length!==1){for(let d of s){let m=g[d];if(!m)throw new a(`unknown short option -${d}`);if(C(m)!=="boolean")throw new a(`short option -${d} requires its own value`);h(o,i,m,"true")}return}let c=g[s];if(!c)throw new a(`unknown short option -${s}`);let p=f??P(t,c);h(o,i,c,p)}r(y,"parseShortOptionToken");function P(n,t,o){let i=I(t,o);if(i==="boolean")return n[0]&&x(n[0])?n.shift():"true";if(i){let e=n.shift();if(e===void 0||R(e)&&!S(e))throw e!==void 0&&n.unshift(e),new a(`${l(t)} must be ${b(i)}`);return e}return n[0]&&!n[0].startsWith("--")?n.shift():"true"}r(P,"readOptionValue");function h(n,t,o,i,e){T(o,i,e),n[o]=i,(t[o]??=[]).push(i)}r(h,"setParsedOption");function T(n,t,o){let i=I(n,o);if(i){if(i==="boolean"){if(!x(t))throw new a(`${l(n)} must be a boolean`);return}if(t.trim().length===0||t==="true")throw new a(`${l(n)} must be ${b(i)}`);if(i==="number"||i==="integer"){let e=Number(t);if(!Number.isFinite(e))throw new a(`${l(n)} must be ${b(i)}`);if(i==="integer"&&!Number.isInteger(e))throw new a(`${l(n)} must be an integer`);if(i==="integer"&&K.has(n)&&e<=0)throw new a(`${l(n)} must be a positive integer`);return}if(i==="json")try{w(t,l(n))}catch(e){throw new a(e instanceof Error?e.message:String(e))}}}r(T,"validateParsedOptionValue");function I(n,t){return t?O(t,n):C(n)}r(I,"typedOptionKind");function l(n){return n.replace(/[A-Z]/g,t=>`-${t.toLowerCase()}`)}r(l,"cliOptionLabel");function E(n){return n==="json"?"JSON value":n}r(E,"optionKindLabel");function b(n){let t=E(n);return/^[aeiou]/i.test(t)?`an ${t}`:`a ${t}`}r(b,"optionKindPhrase");function x(n){return/^(true|false|1|0|yes|no)$/i.test(n)}r(x,"isBooleanLiteral");function R(n){return n.startsWith("-")&&n!=="-"}r(R,"isOptionLike");function S(n){return/^-\d+(?:\.\d+)?$/.test(n)}r(S,"isNegativeNumber");function V(n,t){let o=n?$(n):null;for(let i of Object.keys(t))if(!O(n,i)){let e=l(i),u=o?` for ${o.name}`:"";throw new a(`unknown option --${e}${u}`)}}r(V,"validateCommandOptions");export{a,k as b};
+import{e as g,g as $,h as O,i as C}from"./chunk-7UH4YDA7.js";import{a as w}from"./chunk-3M53YXEI.js";import{a as r}from"./chunk-FTQLQJQV.js";var L=Symbol("openpondCliOptionValues"),a=class extends Error{static{r(this,"CliUsageError")}exitCode=2;constructor(t){super(t),this.name="CliUsageError"}},K=new Set(["maxOutputBytes","timeoutSec"]);function k(n){let t=[...n],o="",i={},e={},u=[];for(;t.length>0;){let s=t.shift();if(s==="--"){u.push(...t);break}if(s.startsWith("--")){let f=s.indexOf("="),c=f>=0?s.slice(2,f):s.slice(2),p=N(c),d=f>=0?s.slice(f+1):P(t,p,o);h(i,e,p,d,o)}else _(s)?y(s,t,i,e):o?u.push(s):o=s}return Object.defineProperty(i,L,{value:e,enumerable:!1,configurable:!1,writable:!1}),V(o,i),{command:o,options:i,rest:u}}r(k,"parseArgs");function N(n){return n.replace(/-([a-z])/g,(t,o)=>o.toUpperCase())}r(N,"normalizeCliOptionKey");function _(n){return/^-[^-].*/.test(n)&&!S(n)}r(_,"isShortOptionToken");function y(n,t,o,i){let e=n.slice(1);if(!e)return;let u=e.indexOf("="),s=u>=0?e.slice(0,u):e,f=u>=0?e.slice(u+1):null;if(s.length!==1){for(let d of s){let m=g[d];if(!m)throw new a(`unknown short option -${d}`);if(C(m)!=="boolean")throw new a(`short option -${d} requires its own value`);h(o,i,m,"true")}return}let c=g[s];if(!c)throw new a(`unknown short option -${s}`);let p=f??P(t,c);h(o,i,c,p)}r(y,"parseShortOptionToken");function P(n,t,o){let i=I(t,o);if(i==="boolean")return n[0]&&x(n[0])?n.shift():"true";if(i){let e=n.shift();if(e===void 0||R(e)&&!S(e))throw e!==void 0&&n.unshift(e),new a(`${l(t)} must be ${b(i)}`);return e}return n[0]&&!n[0].startsWith("--")?n.shift():"true"}r(P,"readOptionValue");function h(n,t,o,i,e){T(o,i,e),n[o]=i,(t[o]??=[]).push(i)}r(h,"setParsedOption");function T(n,t,o){let i=I(n,o);if(i){if(i==="boolean"){if(!x(t))throw new a(`${l(n)} must be a boolean`);return}if(t.trim().length===0||t==="true")throw new a(`${l(n)} must be ${b(i)}`);if(i==="number"||i==="integer"){let e=Number(t);if(!Number.isFinite(e))throw new a(`${l(n)} must be ${b(i)}`);if(i==="integer"&&!Number.isInteger(e))throw new a(`${l(n)} must be an integer`);if(i==="integer"&&K.has(n)&&e<=0)throw new a(`${l(n)} must be a positive integer`);return}if(i==="json")try{w(t,l(n))}catch(e){throw new a(e instanceof Error?e.message:String(e))}}}r(T,"validateParsedOptionValue");function I(n,t){return t?O(t,n):C(n)}r(I,"typedOptionKind");function l(n){return n.replace(/[A-Z]/g,t=>`-${t.toLowerCase()}`)}r(l,"cliOptionLabel");function E(n){return n==="json"?"JSON value":n}r(E,"optionKindLabel");function b(n){let t=E(n);return/^[aeiou]/i.test(t)?`an ${t}`:`a ${t}`}r(b,"optionKindPhrase");function x(n){return/^(true|false|1|0|yes|no)$/i.test(n)}r(x,"isBooleanLiteral");function R(n){return n.startsWith("-")&&n!=="-"}r(R,"isOptionLike");function S(n){return/^-\d+(?:\.\d+)?$/.test(n)}r(S,"isNegativeNumber");function V(n,t){let o=n?$(n):null;for(let i of Object.keys(t))if(!O(n,i)){let e=l(i),u=o?` for ${o.name}`:"";throw new a(`unknown option --${e}${u}`)}}r(V,"validateCommandOptions");export{a,k as b};