openpond 0.2.63 → 0.2.64

package/dist/web/assets/{training-QrdkRQZQ.js → training-DCdoHJ0Y.js}

@@ -1,2 +1,2 @@
-import{C as e,T as t,a as n,b as r,c as i,d as a,g as o,h as s,i as ee,m as c,n as te,o as l,p as u,r as d,s as ne,v as f,x as p}from"./schemas-lO3MoWQr.js";import{a as re,c as m,d as h,i as g,l as _,n as ie,o as v,s as y,t as ae}from"./common-BMjyKlX-.js";import{D as b,O as oe,r as se}from"./settings-D7C-Auyi.js";import{A as ce,D as le,R as ue,U as de,W as fe,_ as pe,c as me,d as x,h as he,i as S,o as C,p as ge,s as w,t as _e,v as ve}from"./chunk-H3JZ3P5T-B0SCoPw4.js";import{A as ye,D as be,E as xe,H as Se,I as Ce,M as we,O as Te,R as Ee,U as De}from"./contracts-BLiqJxBv.js";import{C as Oe,G as ke,H as Ae,O as je,Q as Me,T as Ne,Y as Pe,_ as Fe,b as Ie,d as Le,f as Re,i as ze,k as Be,l as Ve,n as He,r as Ue,s as We,t as Ge,u as Ke,v as qe,y as Je}from"./chunk-ONGOAEY6-C__DQmRT.js";var Ye=f().trim().min(1).max(200),Xe=c({sessionId:Ye,turnId:Ye,source:ge,admissionHash:v,modelConfigurationHash:v,traceHash:v,runtimeEventRefs:n(Ye).max(1e4),startedAt:m,completedAt:m,outputHash:v,partialOutputAvailable:l()}).strict().superRefine((e,t)=>{new Set(e.runtimeEventRefs).size!==e.runtimeEventRefs.length&&t.addIssue({code:`custom`,path:[`runtimeEventRefs`],message:`Native event references must be unique.`}),Date.parse(e.completedAt)<Date.parse(e.startedAt)&&t.addIssue({code:`custom`,path:[`completedAt`],message:`Native completion precedes its start.`})});c({schemaVersion:a(`openpond.nativeHarnessExperimentAttempt.v1`),taskId:f().min(1).max(500),status:d([`completed`,`cancelled`,`timed_out`,`policy_failure`]),output:f().max(8388608).nullable(),error:f().max(2e3).nullable(),messages:n(p()).max(4002),collected:l(),environmentCleanupComplete:a(!0),snapshot:ee(),native:Xe}).strict().superRefine((e,t)=>{let n=e.status===`completed`;(n!==e.collected||n!==(e.output!==null)||n!==(e.error===null))&&t.addIssue({code:`custom`,message:`Native status, output, collection and error must agree.`}),n&&(e.native.partialOutputAvailable||e.native.outputHash!==h(e.output))&&t.addIssue({code:`custom`,path:[`native`,`outputHash`],message:`Native output differs from its retained evidence.`})}).extend({contentHash:v}).strict();var Ze=c({schemaVersion:a(`openpond.humanResultView.v1`),executionId:f().min(1),executionHash:f().regex(/^[a-f0-9]{64}$/),selection:n(ue).max(1e4),coverage:c({pending:u().int().nonnegative(),submitted:u().int().nonnegative(),accepted:u().int().nonnegative(),rejected:u().int().nonnegative(),cancelled:u().int().nonnegative()}).strict(),results:n(c({review:ue,decision:ue.nullable(),attemptIds:n(f().min(1)).min(1).max(2),mode:d([`individual`,`pairwise`]),origin:d([`hosted_retained`,`recorded_evidence`,`owner_attested_local`,`native_local`]),status:d([`pending`,`submitted`,`accepted`,`rejected`,`cancelled`]),grader:ue,answers:n(ce).max(100)}).strict()).max(1e4)}).strict().extend({contentHash:f().regex(/^[a-f0-9]{64}$/)}).strict(),Qe=f().regex(/^[a-f0-9]{64}$/),$e=f().min(1).max(240),et={feedbackKey:$e,criterionId:$e},tt=i(`kind`,[c({...et,kind:a(`normalized_score`),minimum:u().finite(),maximum:u().finite(),passThreshold:u().min(0).max(1).optional()}).strict(),c({...et,kind:a(`boolean`)}).strict(),c({...et,kind:a(`category`)}).strict(),c({...et,kind:a(`preference_score`),tieScore:u().min(0).max(1).nullable()}).strict()]),nt=c({execution:g,resultHash:Qe,reviews:n(c({review:ue,decision:ue}).strict()).min(1).max(1e4)}).strict(),rt=c({rules:n(tt).min(1).max(100),baseline:nt,candidate:nt}).strict().superRefine((e,t)=>{new Set(e.rules.map(e=>e.feedbackKey)).size!==e.rules.length&&t.addIssue({code:`custom`,message:`Select each Human feedback key once.`})}),it=c({manifest:xe,result:be}).strict(),at=c({id:$e,manifestHash:Qe,resultHash:Qe}).strict(),ot=c({schemaVersion:a(`openpond.humanComparisonProjection.v1`),teamId:$e,projectId:$e,selections:rt,original:c({baseline:at,candidate:at}).strict(),humanResults:c({baseline:Ze,candidate:Ze}).strict(),baseline:it,candidate:it}).strict().extend({contentHash:Qe}).strict();function st(e){throw new De(e,409)}function ct(e,t){let n=ot.parse(e),{contentHash:r,...i}=n;r!==h(i)&&st(`human_comparison_projection_invalid`);for(let e of[`baseline`,`candidate`]){let r=ye(n[e]),i=r.manifest.lineage,a=i?.humanProjection,o=n.original[e],s=n.selections[e];(!i||!a||a.originalManifest.id!==o.id||a.originalManifest.contentHash!==o.manifestHash||a.selectionHash!==h(s)||a.rulesHash!==h(n.selections.rules)||h(i.execution)!==h(s.execution)||r.manifest.teamId!==n.teamId||n.humanResults[e].contentHash!==s.resultHash)&&st(`human_comparison_projection_lineage_changed`);let{contentHash:ee,...c}=n.humanResults[e];if((h(c)!==ee||c.executionId!==s.execution.id||c.executionHash!==s.execution.contentHash||h(c.selection)!==h(s.reviews.map(e=>e.review)))&&st(`human_comparison_human_result_changed`),t){let n=ye(t[e]);(n.manifest.lineage?.humanProjection||n.manifest.id!==o.id||n.manifest.contentHash!==o.manifestHash||n.result.contentHash!==o.resultHash||h(n.manifest.lineage?.execution)!==h(s.execution))&&st(`human_comparison_original_changed`)}}return n}var lt=c({adapterId:y,placement:d([`local`,`remote`,`colocated`,`provider_native`]),runtimeVersion:f().trim().min(1).max(200),capabilityReceipt:v}).strict(),ut=c({maxTurns:u().int().positive().max(1e5),timeoutMs:u().int().positive().max(864e5),maxOutputBytes:u().int().positive().max(25e7),maximumSpendUsd:u().nonnegative().nullable()}).strict();c({schemaVersion:a(`openpond.runManifest.v1`),id:y,harnessRelease:g,tasksetRelease:g,model:oe,runtimeTarget:lt,limits:ut,approval:c({id:y,contentHash:v}).strict().nullable(),createdAt:m,metadata:re}).strict().extend({contentHash:v}).strict();var dt=c({schemaVersion:a(`openpond.attemptReceipt.v1`),id:y,runManifest:g,taskId:y,seed:f().trim().min(1).max(500),terminal:l(),failureClass:ae.nullable(),outputHash:v.nullable(),traceHash:v,artifactRefs:n(ie).max(1e5),graderEvidenceRefs:n(ie).max(1e4),startedAt:m,completedAt:m,latencyMs:u().int().nonnegative(),costUsd:u().nonnegative().nullable(),legacyAttemptRef:y.nullable().default(null),metadata:re}).strict().extend({contentHash:v}).strict(),ft=c({schemaVersion:a(`openpond.evaluationResult.v1`),id:y,runManifest:g,harnessRelease:g,tasksetRelease:g,model:oe,receiptRefs:n(g).min(1).max(1e6),attemptCount:u().int().positive(),rewardEligibleCount:u().int().nonnegative(),terminalCount:u().int().nonnegative(),meanScore:u().min(0).max(1).nullable(),failureCounts:o(ae,u().int().nonnegative()),authoredMetric:fe.optional(),metadata:re}).strict().extend({contentHash:v}).strict();c({schemaVersion:a(`openpond.harnessCompatibility.v1`),id:y,baseHarnessRelease:g,candidateHarnessRelease:g,tasksetRelease:g,environmentHash:v,toolContractHash:v,policyHash:v,graderInterfaceHash:v,metadata:re}).strict().extend({contentHash:v}).strict();var pt=i(`kind`,[c({kind:a(`model`),model:oe,configurationHash:v}).strict(),c({kind:a(`fixture`)}).strict()]),mt=i(`kind`,[c({kind:a(`taskset`),environmentRelease:g,verifierSetRelease:g,policyHash:v}).strict(),c({kind:a(`harness`),harnessRelease:g}).strict()]),ht=c({receiptId:y,taskId:y,seed:f().trim().min(1).max(500),fixtureId:y.nullable()}).strict(),gt=c({schemaVersion:a(`openpond.tasksetRunManifest.v1`),id:y,tasksetRelease:g,packageHash:v,execution:mt,profileEvaluation:he.optional(),policy:pt,gradingRole:a(`evaluation`),metricPolicy:de,population:n(ht).min(1).max(1e5),runtimeTarget:lt,limits:ut,createdAt:m,metadata:re}).strict().superRefine((e,t)=>{new Set(e.population.map(e=>e.receiptId)).size!==e.population.length&&t.addIssue({code:`custom`,path:[`population`],message:`Run receipt identities must be unique.`}),e.population.some(t=>t.fixtureId!==null!=(e.policy.kind===`fixture`))&&t.addIssue({code:`custom`,path:[`population`],message:`Only fixture runs require a fixture identity for every member.`}),e.profileEvaluation&&(e.execution.kind!==`harness`||e.execution.harnessRelease.id!==e.profileEvaluation.harnessRelease.id||e.execution.harnessRelease.contentHash!==e.profileEvaluation.harnessRelease.contentHash)&&t.addIssue({code:`custom`,path:[`profileEvaluation`],message:`Profile evaluation must execute its exact bound Harness release.`})}).safeExtend({contentHash:v}),T=f().trim().min(1).max(240),E=f().trim().min(1).max(5e3),_t=d([`no_training`,`prompting`,`retrieval`,`sft`,`preference`,`grpo_rft`,`sdft_opsd`,`sdpo`,`agentic_rl`]),vt=c({id:T,taskIndex:u().int().nonnegative(),label:Je,output:o(f(),p()),infrastructureError:f().trim().min(1).max(1e4).nullable(),expectedPassed:l(),expectedRewardEligible:l(),metadata:o(f(),p()).default({})}),yt=c({schemaVersion:a(`openpond.capabilityDiagnosis.v1`),summary:f().trim().min(1).max(1e4),stableBehavior:n(E).max(100).default([]),changingKnowledge:n(E).max(100).default([]),requiredContext:n(E).max(100).default([]),requiredTools:n(T).max(100).default([]),intervention:_t,trainingEligible:l(),rationale:n(E).min(1).max(100),confidence:u().min(0).max(1)}),bt=c({id:T,sourceId:T,sourceTurnId:T.nullable(),split:je,origin:d([`extracted`,`corrected`,`synthetic`,`expert_authored`]),inputPrompt:f().trim().min(1).max(1e5),expectedOutputText:f().trim().min(1).max(2e5).nullable(),rationale:E}),xt=c({schemaVersion:a(`openpond.taskDesignProposal.v1`),id:T,name:f().trim().min(1).max(500),objective:f().trim().min(1).max(2e4),diagnosis:yt.default({schemaVersion:`openpond.capabilityDiagnosis.v1`,summary:`Reproduce the selected approved behavior.`,stableBehavior:[],changingKnowledge:[],requiredContext:[],requiredTools:[],intervention:`sft`,trainingEligible:!0,rationale:[`The selected examples were supplied as demonstrations.`],confidence:.5}),taskKind:Ne.shape.taskKind,sourceIds:n(T).min(1).max(1e5),assumptions:n(E).max(1e3),successCriteria:n(E).min(1).max(1e3),proposedGraders:n(Ie).max(1e3).default([]),graderFixtures:n(vt).max(1e5).default([]),generatedFiles:n(ze).max(1e3).default([]),proposedExamples:n(bt).max(1e5).default([]),proposedMethod:Ve.shape.recommendedMethod,trainingPath:Re.nullable().default(null),policy:Oe,warnings:n(f().trim().min(1)).default([]),createdAt:f().trim().min(1)}),St=f().regex(/^[a-f0-9]{64}$/),Ct=c({providerId:a(`claude-code`),configurationHash:St,executableHash:St,version:f().regex(/^2\.\d+\.\d+$/),capabilityHash:St,maximumRequestCostUsd:u().finite().positive().max(1e4),requestTimeoutMs:u().int().min(1e3).max(3e5),maximumTurns:u().int().min(1).max(100),holdOpen:l()}).strict();c({id:f().min(1).max(200),receiptId:f().min(1).max(200),sessionId:e(),operationId:f().min(1).max(200),action:d([`message`,`seal`]),text:f().max(262144).optional()}).strict().superRefine((e,t)=>{e.action===`message`&&!e.text?.trim()&&t.addIssue({code:`custom`,message:`A human intervention requires a message.`}),e.action===`seal`&&e.text!==void 0&&t.addIssue({code:`custom`,message:`Sealing does not add a prompt.`})});var wt=c({provider:a(`claude-code`),sessionId:e(),requestId:f().min(1).max(200),modelId:f().min(1).max(500),runtime:Ct,sourceHash:St,outputHash:St,traceHash:St,runtimeEventRefs:n(f().min(1).max(200)).min(1).max(2e4),eventCount:u().int().positive().max(2e4),startedAt:t(),completedAt:t(),cleanupComplete:a(!0)}).strict().superRefine((e,t)=>{(e.runtimeEventRefs.length!==e.eventCount||new Set(e.runtimeEventRefs).size!==e.eventCount)&&t.addIssue({code:`custom`,message:`Process event references must match the exact retained trace population.`})}),Tt=c({providerId:a(`codex`),configurationHash:f().regex(/^[a-f0-9]{64}$/),accountHash:f().regex(/^[a-f0-9]{64}$/),reasoningEffort:d([`low`,`medium`,`high`,`xhigh`]),maximumRequestCostUsd:u().finite().positive().max(1e4),requestTimeoutMs:u().int().min(1e3).max(3e5)}).strict(),Et=c({id:f().trim().min(1).max(240),revision:u().int().positive(),contentHash:f().regex(/^[a-f0-9]{64}$/),targetId:f().trim().min(1).max(240).nullable()}).strict(),Dt=n(c({role:d([`system`,`user`,`assistant`]),content:f().max(1e5)}).strict()).max(100).superRefine((e,t)=>{e.reduce((e,t)=>e+t.content.length,0)>1e5&&t.addIssue({code:`custom`,message:`Evaluation prompt exceeds 100,000 characters.`})}),Ot=c({schemaVersion:a(`openpond.experimentAttemptGrade.v1`),selectionHash:f().regex(/^[a-f0-9]{64}$/),grades:n(Se).min(1).max(100),score:u().min(0).max(1).nullable(),passed:l(),rewardEligible:l(),gradingStatus:d([`scored`,`pending`,`unscorable`,`not_configured`]),failureClass:_e.nullable(),contentHash:f().regex(/^[a-f0-9]{64}$/)}).strict().superRefine((e,t)=>{let{contentHash:n,...r}=e;x(r)!==n&&t.addIssue({code:`custom`,message:`Experiment grading integrity failed.`});let i=e.grades.flatMap(e=>e.components);(i.length!==e.grades.length||new Set(i.map(e=>e.graderId)).size!==i.length)&&t.addIssue({code:`custom`,message:`Each selected criterion requires exactly one distinct grader result.`});for(let n of e.grades){let{contentHash:e,...r}=n;x(r)!==e&&t.addIssue({code:`custom`,message:`A retained criterion changed.`});for(let e of n.components){let{contentHash:n,...r}=e;x(r)!==n&&t.addIssue({code:`custom`,message:`A retained grader result changed.`})}}}),D=f().trim().min(1).max(200),kt=f().regex(/^[a-f0-9]{64}$/),At=i(`kind`,[c({kind:a(`hosted_chat`),modelId:D,localRuntime:r([c({providerId:a(`custom-openai-compatible`),configurationHash:kt,maximumRequestCostUsd:u().finite().positive().max(1e4),requestTimeoutMs:u().int().min(1e3).max(3e5)}).strict(),Ct,Tt]).optional(),candidate:c({artifactId:D,contentHash:kt,preparationId:D,dispatchId:D,baseProfileId:D,workerImageDigest:f().regex(/^sha256:[a-f0-9]{64}$/)}).strict().optional(),harness:ge.optional(),messages:Dt.optional(),maxOutputTokens:u().int().min(1).max(1e6).default(1e5),temperature:u().min(0).max(2).default(0),topP:u().gt(0).max(1).default(1)}).strict(),c({kind:a(`fixture`),fixtureId:D}).strict()]),jt=c({schemaVersion:a(`openpond.modelStarterAttemptRequest.v1`),operationId:D,teamId:D,modelProjectId:D.nullable(),project:Et.optional(),taskset:ke,taskId:D,environmentSeed:u().int().min(0).max(2147483647).default(0),policy:At}).strict();c({modelProjectId:D.nullable(),taskset:ke,afterTaskId:D.optional(),limit:u().int().min(1).max(100).default(25)}).strict(),c({modelProjectId:D.nullable(),taskset:ke,available:l(),unavailableReason:f().max(1e3).nullable(),models:n(c({id:D,name:f().max(500),maximumOutputTokens:u().int().min(1).max(1e6).optional()}).strict()).max(200),tasks:n(c({id:D,split:le,inputPreview:f().max(500),fixtures:n(c({id:D,label:f().max(200)}).strict()).max(1e3)}).strict()).max(100),nextCursor:D.nullable()}).strict();var Mt=c({schemaVersion:a(`openpond.modelStarterAttemptSummary.v1`),id:D,revision:u().int().positive(),request:jt,status:d([`queued`,`running`,`cancelling`,`completed`,`failed`,`cancelled`]),policySnapshot:c({modelId:D,provider:D,upstreamModelId:f().min(1).max(500),configurationHash:kt}).strict().nullable(),createdAt:t(),startedAt:t().nullable(),completedAt:t().nullable(),cleanupComplete:l(),resultAvailable:l(),score:u().min(0).max(1).nullable(),gradingStatus:d([`not_started`,`pending`,`scored`,`unscorable`,`not_configured`]).default(`not_started`),passed:l().nullable(),outputPreview:f().max(500).nullable(),error:c({code:D,message:f().max(2e3)}).strict().nullable()}).strict().superRefine((e,t)=>{e.request.policy.kind===`fixture`!=(e.policySnapshot===null)&&t.addIssue({code:`custom`,path:[`policySnapshot`],message:`Fixture execution must not claim a model identity.`}),e.status===`completed`&&(!e.cleanupComplete||!e.resultAvailable||!e.completedAt)&&t.addIssue({code:`custom`,message:`Completed attempts require retained results and confirmed cleanup.`})});c({modelProjectId:D,afterId:D.optional(),limit:u().int().min(1).max(100).default(25)}).strict(),c({items:n(Mt).max(100),nextCursor:D.nullable()}).strict(),c({schemaVersion:a(`openpond.modelStarterAttemptResult.v1`),attempt:Mt,output:f().max(1048576).nullable(),messages:n(o(f(),p())).max(4002),composition:Ce.nullable(),grade:Se.nullable().optional(),experimentGrade:Ot.nullable().optional(),environment:c({status:d([`completed`,`budget_exhausted`,`cancelled`,`timed_out`,`policy_failure`,`environment_failure`]),collected:l(),definition:ke,initialStateHash:kt.nullable(),finalStateHash:kt.nullable(),attemptHash:kt}).strict(),native:Xe.optional(),providerRequestIds:n(D).max(1001),contentHash:kt}).strict().superRefine((e,t)=>{let n=e.attempt.request.policy.kind===`hosted_chat`?e.attempt.request.policy.harness:void 0;!!n!=!!e.native&&t.addIssue({code:`custom`,path:[`native`],message:`Standalone Harness results require their exact native evidence.`}),e.native&&(!n||x(n)!==x(e.native.source)||e.native.modelConfigurationHash!==e.attempt.policySnapshot?.configurationHash||e.environment.initialStateHash!==null||e.environment.finalStateHash!==null||e.environment.status===`completed`&&(!e.environment.collected||e.output===null||e.native.partialOutputAvailable||e.native.outputHash!==x(e.output)))&&t.addIssue({code:`custom`,path:[`native`],message:`Native evidence differs from the admitted source, model or output.`}),e.experimentGrade&&((e.grade||e.composition)&&t.addIssue({code:`custom`,message:`Selected Experiment grading cannot claim another grade population.`}),(![`completed`,`failed`].includes(e.attempt.status)||!e.attempt.resultAvailable||e.attempt.score!==e.experimentGrade.score||e.attempt.gradingStatus!==e.experimentGrade.gradingStatus||e.attempt.passed!==(e.experimentGrade.score===null?null:e.experimentGrade.passed))&&t.addIssue({code:`custom`,message:`Experiment grading differs from its retained attempt summary.`})),e.grade&&(e.composition&&t.addIssue({code:`custom`,path:[`grade`],message:`An ordinary grade cannot also claim a Reward composition.`}),(![`completed`,`failed`].includes(e.attempt.status)||!e.attempt.resultAvailable||e.attempt.score!==e.grade.score||e.attempt.gradingStatus!==e.grade.gradingStatus||e.attempt.passed!==(e.grade.score===null?null:e.grade.passed))&&t.addIssue({code:`custom`,path:[`grade`],message:`Ordinary grade differs from its terminal attempt summary.`}))});var Nt=f().trim().min(1).max(240),Pt=f().regex(/^[a-f0-9]{64}$/),Ft=c({maxOutputTokens:u().int().min(1).max(1e6).default(4096),temperature:u().min(0).max(2).default(0),topP:u().gt(0).max(1).default(1)}).strict(),It=c({kind:a(`hosted_harness`),modelId:Nt,maxOutputTokens:Ft.shape.maxOutputTokens.removeDefault().optional(),temperature:Ft.shape.temperature.removeDefault().optional(),topP:Ft.shape.topP.removeDefault().optional(),localRuntime:Tt.optional(),candidate:c({artifactId:Nt,contentHash:Pt,preparationId:Nt,dispatchId:Nt,baseProfileId:Nt,workerImageDigest:f().regex(/^sha256:[a-f0-9]{64}$/)}).strict().optional(),sourceCandidate:he.shape.sourceCandidate,reasoningEffort:d([`off`,`low`,`medium`,`high`,`xhigh`,`max`]).optional(),profileRepositoryId:Nt,source:he,modelConfigurationHash:Pt,packageHash:Pt,externalDatasetBinding:pe.optional()}).strict().superRefine((e,t)=>{JSON.stringify(e.externalDatasetBinding??null)!==JSON.stringify(e.source.externalDatasetBinding??null)&&t.addIssue({code:`custom`,message:`The policy and source must bind the same external Dataset recipe.`}),JSON.stringify(e.sourceCandidate??null)!==JSON.stringify(e.source.sourceCandidate??null)&&t.addIssue({code:`custom`,message:`The policy and prepared source must bind the same actual source candidate.`})}),O=f().trim().min(1).max(200),Lt=f().regex(/^[a-f0-9]{64}$/),Rt=i(`kind`,[At.options[0],c({kind:a(`fixture`)}).strict(),It]),zt=c({schemaVersion:a(`openpond.modelTasksetRunRequest.v1`),operationId:O,teamId:O,modelProjectId:O.nullable(),name:f().trim().min(1).max(200).optional(),project:Et.optional(),taskset:ke,policy:Rt,population:n(ht).min(1).max(1e4)}).strict().superRefine((e,t)=>{new Set(e.population.map(e=>e.receiptId)).size!==e.population.length&&t.addIssue({code:`custom`,path:[`population`],message:`Run receipt identities must be unique.`}),e.population.some(t=>t.fixtureId!==null!=(e.policy.kind===`fixture`))&&t.addIssue({code:`custom`,path:[`population`],message:`Only fixture runs require a fixture identity for every member.`}),e.population.some(e=>!/^(0|[1-9][0-9]*)$/.test(e.seed)||Number(e.seed)>2147483647)&&t.addIssue({code:`custom`,path:[`population`],message:`Hosted environment seeds must be integers from 0 to 2147483647.`}),e.policy.kind===`hosted_harness`&&e.modelProjectId!==null&&t.addIssue({code:`custom`,path:[`modelProjectId`],message:`A Harness target retains its released Profile binding separately from Model artifacts.`}),e.policy.kind===`hosted_harness`&&new Set(e.population.map(e=>Pe([e.taskId,e.seed]))).size!==e.population.length&&t.addIssue({code:`custom`,path:[`population`],message:`Harness case identities must be unique for each task and seed.`})}),Bt=c({schemaVersion:a(`openpond.modelTasksetRunSummary.v1`),id:O,revision:u().int().positive(),teamId:O,modelProjectId:O.nullable(),name:f().trim().min(1).max(200).optional(),project:Et.optional(),operationId:O,taskset:ke,policyKind:d([`hosted_chat`,`fixture`,`hosted_harness`]),manifestHash:Lt,status:d([`queued`,`running`,`cancelling`,`completed`,`failed`,`cancelled`]),totalCount:u().int().min(1).max(1e4),counts:c({pending:u().int().nonnegative(),running:u().int().nonnegative(),completed:u().int().nonnegative(),failed:u().int().nonnegative(),cancelled:u().int().nonnegative()}).strict(),createdAt:t(),startedAt:t().nullable(),completedAt:t().nullable(),cleanupComplete:l(),resultAvailable:l(),score:u().min(0).max(1).nullable(),metricName:f().min(1).max(240),error:c({code:O,message:f().max(2e3)}).strict().nullable()}).strict().superRefine((e,t)=>{Object.values(e.counts).reduce((e,t)=>e+t,0)!==e.totalCount&&t.addIssue({code:`custom`,path:[`counts`],message:`Run counts must account for the complete population.`}),e.status!==`completed`&&e.score!==null&&t.addIssue({code:`custom`,path:[`score`],message:`Only completed evaluations publish a metric value.`}),[`completed`,`failed`,`cancelled`].includes(e.status)&&(!e.completedAt||e.counts.pending||e.counts.running)&&t.addIssue({code:`custom`,message:`Terminal runs require complete population accounting.`}),e.status===`cancelled`&&!e.cleanupComplete&&t.addIssue({code:`custom`,message:`Cancelled runs require confirmed cleanup.`}),e.resultAvailable&&![`completed`,`failed`,`cancelled`].includes(e.status)&&t.addIssue({code:`custom`,message:`Run results require terminal population accounting.`}),e.status===`completed`&&(!e.cleanupComplete||!e.resultAvailable)&&t.addIssue({code:`custom`,message:`Completed evaluations require retained results and confirmed cleanup.`})}),Vt=c({summary:Bt,request:zt,manifest:gt,policySnapshot:Mt.shape.policySnapshot}).strict();c({modelProjectId:O.optional(),projectId:Et.shape.id.optional(),afterId:O.optional(),limit:u().int().min(1).max(100).default(25)}).strict(),c({items:n(Bt).max(100),nextCursor:O.nullable()}).strict(),c({schemaVersion:a(`openpond.modelTasksetRunResult.v1`),run:Vt,receipts:n(dt).min(1).max(1e4),metric:fe.nullable(),contentHash:Lt}).strict();var Ht=f().min(1).max(200).refine(e=>![`__proto__`,`prototype`,`constructor`].includes(e),`Use an ordinary field name.`),Ut=n(Ht).min(1).max(20),Wt=d([`input`,`output`,`expectedOutput`,`evaluatorContext`]),Gt=c({destination:Wt,path:Ut,source:Wt,sourcePath:n(Ht).max(20)}).strict(),Kt=n(Gt).max(100).superRefine((e,t)=>{for(let[n,r]of e.entries())e.slice(0,n).some(e=>e.destination===r.destination&&(qt(e.path,r.path)||qt(r.path,e.path)))&&t.addIssue({code:`custom`,path:[n,`path`],message:`Mapped destination paths must not overlap.`})});function qt(e,t){return e.length<=t.length&&e.every((e,n)=>e===t[n])}var k=f().trim().min(1).max(200),Jt=f().regex(/^[a-f0-9]{64}$/),Yt=c({id:k,revision:u().int().positive(),contentHash:Jt}).strict(),Xt=c({id:k,version:f().min(1).max(200),contentHash:Jt,name:f().trim().min(1).max(500).optional(),feedbackKey:we,release:Yt.nullable(),mappings:Kt.optional()}).strict(),Zt=c({id:k,version:f().min(1).max(200),contentHash:Jt,mappings:Kt}).strict(),Qt=c({operationId:k,id:k.optional(),expectedRevision:u().int().nonnegative(),request:zt,maximumCostUsd:u().finite().positive().max(1e4),graders:n(Zt).min(1).max(100).optional()}).strict().superRefine((e,t)=>{e.request.name||t.addIssue({code:`custom`,path:[`request`,`name`],message:`Name this Experiment before saving it.`}),!!e.id!=e.expectedRevision>0&&t.addIssue({code:`custom`,path:[`expectedRevision`],message:`An update requires its saved definition revision.`}),Math.round(e.maximumCostUsd*1e6)<1&&t.addIssue({code:`custom`,path:[`maximumCostUsd`],message:`The budget must be at least $0.000001.`}),e.graders&&new Set(e.graders.map(e=>e.id)).size!==e.graders.length&&t.addIssue({code:`custom`,path:[`graders`],message:`Select each grader once.`})});c({schemaVersion:a(`openpond.experimentDefinition.v1`),id:k,teamId:k,revision:u().int().positive(),request:zt,maximumCostUsd:u().finite().positive().max(1e4),graders:n(Xt).min(1).max(100),createdAt:t(),updatedAt:t()}).strict().extend({contentHash:Jt}),c({operationId:k,definition:Yt}).strict(),c({projectId:k.optional(),datasetHash:Jt.optional(),search:f().trim().max(200).optional(),status:d([`queued`,`running`,`cancelling`,`completed`,`failed`,`cancelled`]).optional(),afterId:k.optional(),limit:u().int().min(1).max(100).default(30)}).strict(),c({definition:Yt.nullable(),configurationHash:Jt.optional(),admissionRequestHash:Jt.optional(),sourceExperimentId:k.optional(),maximumCostUsd:u().finite().positive().max(1e4),graders:n(Xt).min(1).max(100)}).strict().superRefine((e,t)=>{e.definition===null&&(!e.configurationHash||!e.admissionRequestHash)&&t.addIssue({code:`custom`,path:[`configurationHash`],message:`A flat Experiment requires its sealed configuration hash.`})}),c({operationId:k,profileRepositoryId:k,definitionId:k,modelId:k,localRuntime:It.shape.localRuntime,maxOutputTokens:It.shape.maxOutputTokens,temperature:It.shape.temperature,topP:It.shape.topP,reasoningEffort:d([`off`,`low`,`medium`,`high`,`xhigh`,`max`]).optional(),maximumCostUsd:u().finite().positive().max(1e4),externalDatasetBinding:pe.optional(),sourceCandidate:It.shape.sourceCandidate}).strict(),c({request:zt,manifest:gt,maximumCostUsd:u().finite().positive().max(1e4),graders:n(Xt).min(1).max(100)}).strict();var $t=c({id:w,purpose:f().trim().min(1).max(500),audience:d([`orchestrator`,`environment`,`privileged_scorer`,`trainer`,`infrastructure`]),required:l(),ttlSeconds:u().int().positive().max(86400),scopes:n(f().trim().min(1).max(500)).max(100).default([])}).strict(),en=c({declarationId:w,leaseRef:f().trim().min(8).max(1e3),audience:$t.shape.audience,expiresAt:me}).strict(),tn=c({adapterId:w,placement:d([`local`,`remote`,`colocated`,`provider_native`]),capabilityReceipt:C,runtimeVersion:f().trim().min(1).max(200),dataPlane:c({provider:w,dataPlaneId:w,cellId:w,runnerPoolId:w,runtimeImageDigest:f().regex(/^sha256:[a-f0-9]{64}$/),capabilityReceipt:C}).strict().nullable()}).strict(),nn=c({adapterId:w,kind:d([`local`,`ssh`,`managed`,`custom`]),deviceOrPool:f().trim().min(1).max(1e3),capabilityReceipt:C,provider:w.nullable()}).strict(),rn=c({adapterId:w,workerVersion:f().trim().min(1).max(200),workerImageDigest:f().regex(/^sha256:[a-f0-9]{64}$/).nullable(),upstreamRevision:f().trim().min(1).max(500),capabilityReceipt:C}).strict(),an=c({schemaVersion:a(`openpond.harnessRunManifest.v1`),executionMode:a(`qualification`).optional(),qualificationSamples:u().int().min(1).max(32).optional(),id:w,harnessRelease:S,datasetRelease:S,evidenceSets:n(S).max(1e4),model:c({source:f().trim().min(1).max(200),revision:f().trim().min(1).max(500),artifactHash:C.nullable(),tokenizerRevision:f().trim().min(1).max(500),chatTemplateHash:C}).strict(),recipe:c({method:f().trim().min(1).max(100),version:f().trim().min(1).max(200),configHash:C}).strict(),runtimeTarget:tn,computeTarget:nn,engine:rn,resolvedBundleHash:C,secretLeaseRefs:n(en).max(1e3),approval:c({approvalHash:C,approvedAt:me,maximumSpendUsd:u().nonnegative().nullable()}).strict(),createdAt:me}).strict().extend({contentHash:C}).strict();c({schemaVersion:a(`openpond.resolvedTrainingBundle.v1`),executionMode:a(`qualification`).optional(),qualificationSamples:u().int().min(1).max(32).optional(),projection:a(`trainer`),harnessRelease:S,datasetRelease:S,evidenceSetRelease:S.nullable(),files:n(c({path:f().trim().min(1).max(2e3),sha256:C,sizeBytes:u().int().nonnegative()}).strict()).max(1e5)}).strict().extend({contentHash:C}).strict();var on=c({name:a(`adamw`).default(`adamw`),weightDecay:u().nonnegative().max(1).default(0),beta1:u().positive().lt(1).default(.9),beta2:u().positive().lt(1).default(.999),epsilon:u().positive().max(.01).default(1e-8)}).strict().default({name:`adamw`,weightDecay:0,beta1:.9,beta2:.999,epsilon:1e-8}),A=f().trim().min(1).max(240),j=f().trim().min(1),sn=f().trim().min(8).max(256),cn=o(f(),p()).default({}),M=A.nullable(),ln=d([`slash_train`,`session_menu`,`bulk_selection`,`training_page`,`task_candidate`]),un=d([`defaults`,`customize`]),dn=d([`automated`,`manual`]);c({schemaVersion:a(`openpond.trainingSourceEstimate.v1`),sessionId:A,messageCount:u().int().nonnegative(),estimatedTokens:u().int().nonnegative(),textBytes:u().int().nonnegative()}),c({query:f().max(500).default(``),offset:u().int().nonnegative().default(0),limit:u().int().min(1).max(100).default(20),candidates:n(c({sessionId:A,title:f().trim().min(1).max(500),updatedAt:j})).max(500).default([])});var fn=c({sessionId:A,title:f().trim().min(1).max(500),updatedAt:j,snippet:f().max(2e3).nullable()});c({schemaVersion:a(`openpond.trainingChatSearchResult.v1`),query:f(),offset:u().int().nonnegative(),limit:u().int().positive(),total:u().int().nonnegative(),hasMore:l(),indexedChats:u().int().nonnegative(),totalChats:u().int().nonnegative(),indexing:l(),entries:n(fn)});var pn=d([`policy_failure`,`grader_failure`,`environment_failure`,`infrastructure_failure`,`timeout`,`cancelled`]),mn=c({graderId:A.nullable(),path:f().trim().min(1).max(1e3),language:d([`javascript`,`typescript`,`python`,`json`,`text`]),content:f().max(2e6),sha256:sn,declaredSha256:sn.nullable(),integrity:d([`verified`,`unverified`,`mismatch`])}),hn=c({protocolVersion:f().trim().min(1).max(200),module:f().trim().min(1).max(1e3),moduleSha256:sn,command:n(f().max(2e3)).max(100),cwd:f().trim().min(1).max(4e3),maxTurns:u().int().positive().max(100)});c({schemaVersion:a(`openpond.tasksetGraderDetails.v1`),taskset:_,sourceTasksetId:A,graders:n(Ie).max(1e3),runtime:hn.nullable(),sources:n(mn).max(1e3),unavailableReason:f().max(2e3).nullable()});var gn=c({schemaVersion:a(`openpond.taskAttempt.v1`),id:A,tasksetId:A,taskId:A,split:je,attempt:u().int().nonnegative(),seed:u().int(),modelRef:b.nullable(),startedAt:j,completedAt:j,output:o(f(),p()),runtimeEventRefs:n(A).max(1e4).default([]),artifactRefs:n(A).max(1e4).default([]),privilegedOutcomeRef:M,infrastructureError:f().trim().min(1).max(1e4).nullable(),costUsd:u().nonnegative().nullable(),latencyMs:u().int().nonnegative(),userInterventions:u().int().nonnegative().default(0),metadata:cn}),_n=c({schemaVersion:a(`openpond.taskAttemptArtifact.v1`),id:A,tasksetId:A,taskId:A,attemptId:A,kind:d([`raw_model_response`,`runtime_trace`,`environment_state`,`grader_evidence`,`output_artifact`]),path:f().trim().min(1).max(4e3),mediaType:f().trim().min(1).max(200).nullable().optional(),sha256:sn,sizeBytes:u().int().nonnegative(),createdAt:j,metadata:cn}),vn=c({graderId:A,graderVersion:f().trim().min(1).max(100),score:u().min(0).max(1).nullable(),passed:l(),hardGate:l(),rewardEligible:l(),feedback:f().trim().max(2e4).nullable(),evidenceRefs:n(A).max(1e4).default([]),judge:b.nullable().optional().default(null),calibrationStatus:d([`not_applicable`,`pending`,`passed`,`failed`])}),yn=c({schemaVersion:a(`openpond.gradeResult.v1`),id:A,attemptId:A,graderSetHash:sn,score:u().min(0).max(1).nullable(),passed:l(),components:n(vn).min(1).max(1e3),failureClass:pn.nullable(),feedback:n(f().trim().min(1).max(2e4)).max(1e3).default([]),rewardEligible:l(),rewardComposition:Ce.optional(),createdAt:j});d([`stable_hash_top_n`,`rft_easy_curriculum_v1`]);var bn=c({schemaVersion:a(`openpond.graderAuditReport.v1`),id:A,tasksetId:A,tasksetHash:sn,fixtureRefs:n(A).min(1).max(1e5),gradeRefs:n(A).min(1).max(1e5),passed:l(),hackingChecksPassed:l(),leakageChecksPassed:l(),infrastructureSafetyPassed:l(),failures:n(c({fixtureId:A,label:Je,gradeId:A,reason:f().trim().min(1).max(5e3)})).max(1e5),createdAt:j});function xn(e,t=`New model`){let n=e?.trim().split(/\s+/).filter(Boolean)??[];return n.length===0?t:n.slice(0,5).join(` `)}var Sn=c({schemaVersion:a(`openpond.baseModelPreference.v1`),modelId:A,revision:f().trim().min(1).max(256).nullable(),tokenizerRevision:f().trim().min(1).max(256).nullable(),chatTemplateHash:f().trim().min(8).max(256).nullable(),modelAssetId:M,source:d([`managed`,`local`,`builtin`])}),Cn=c({schemaVersion:a(`openpond.taskCreationRequest.v1`),id:A,profileId:A,surface:ln,mode:un,entryMode:dn.default(`manual`),resourceIntent:d([`workproduct`,`dataset`]).default(`workproduct`),buildIntent:He.default(`demonstrations`),buildSpecification:Ue.nullable().default(null),objective:f().trim().min(1).max(2e4).nullable(),methodHint:d([`sft`,`dpo`,`grpo`,`ppo`]).nullable().default(null),preferredBaseModelId:A.nullable().default(null),preferredBaseModel:Sn.nullable().default(null),sourceIds:n(A).max(1e5),candidateId:M,analysisModel:b.nullable(),analysisReasoningEffort:se.nullable().default(null),createImproveRunId:M.default(null),targetIntent:c({kind:d([`agent`,`skill`,`extension`,`model`,`configuration`]).nullable(),id:M,displayName:f().trim().min(1).max(500).nullable(),operation:d([`create`,`improve`])}).default({kind:`model`,id:null,displayName:null,operation:`create`}),disclosure:c({status:d([`not_required`,`pending`,`approved`,`declined`]),content:a(`raw_excerpts`),sourceIds:n(A).max(1e5),providerModel:b.nullable(),approvalId:M,approvedAt:j.nullable()}).default({status:`not_required`,content:`raw_excerpts`,sourceIds:[],providerModel:null,approvalId:null,approvedAt:null}),createdAt:j}),wn=c({schemaVersion:a(`openpond.taskCreationSnapshot.v1`),id:A,request:Cn,state:d([`planning`,`awaiting_disclosure_approval`,`awaiting_questions`,`recommendation_ready`,`awaiting_materialization_approval`,`materializing`,`validating`,`ready`,`blocked`,`failed`,`cancelled`]),proposal:xt.nullable(),materializedTasksetId:M,disclosureApprovalId:M,materializationApprovalId:M,blockingQuestions:n(c({id:A,kind:d([`objective`,`consent`,`success_signal`,`privacy_licensing`,`interpretation`]),prompt:f().trim().min(1).max(5e3),answer:f().trim().min(1).max(2e4).nullable()})).default([]),transcript:n(c({id:A,role:d([`user`,`assistant`,`system`,`tool`]),text:f().max(1e5),createdAt:j})).max(1e4).default([]),repairHistory:n(Ge).max(1e3).default([]),blockedReason:f().trim().min(1).max(1e4).nullable(),createdAt:j,updatedAt:j});c({schemaVersion:a(`openpond.taskCreationTranscript.v1`),creationId:A,profileId:A,messages:wn.shape.transcript,updatedAt:j});var Tn=f().trim().min(1).max(240),En=f().trim().min(1),Dn=f().trim().min(8).max(256),On=o(f(),p()).default({}),kn=d([`created`,`inspecting`,`awaiting_source_review`,`awaiting_mapping`,`awaiting_materialization_approval`,`materializing`,`validating`,`ready`,`cancelling`,`cancelled`,`failed`]),An=c({schemaVersion:a(`openpond.huggingFaceDatasetLocator.v1`),repositoryId:f().trim().regex(/^[^/\s]+\/[^/\s]+$/).max(500),repositoryUrl:f().url().max(2e3),requestedRevision:f().trim().min(1).max(256).nullable()}),jn=c({path:f().trim().min(1).max(1e3),logicalType:f().trim().min(1).max(500),nullable:l(),sampleValues:n(p()).max(5)}),Mn=c({configuration:f().trim().min(1).max(500),split:f().trim().min(1).max(500),rowCount:u().int().nonnegative().nullable(),sizeBytes:u().int().nonnegative().nullable()}),Nn=c({schemaVersion:a(`openpond.huggingFaceDatasetInspection.v1`),id:Tn,locator:An,resolvedRevision:f().trim().min(7).max(256),title:f().trim().min(1).max(500),description:f().max(2e4).nullable(),declaredLicense:f().trim().min(1).max(500).nullable(),gated:l(),private:l(),configurations:n(f().trim().min(1).max(500)).min(1).max(1e3),splits:n(Mn).min(1).max(1e4),columns:n(jn).min(1).max(1e4),previewRows:n(o(f(),p())).max(25),sourceFiles:n(c({path:f().trim().min(1).max(2e3),configuration:f().trim().min(1).max(500),split:f().trim().min(1).max(500),revision:f().trim().min(7).max(256),sizeBytes:u().int().nonnegative().nullable(),contentHash:Dn.nullable()})).max(1e5),inspectedAt:En,metadata:On}),Pn=d([`row_id`,`cluster_id`,`prompt`,`messages`,`demonstration`,`chosen`,`rejected`,`expected_output`,`privileged_context`,`reward`,`feedback`,`tag`,`metadata`]),Fn=c({sourcePath:f().trim().min(1).max(1e3),target:Pn,transform:d([`identity`,`string`,`json`,`messages`,`numeric`,`math_final_answer`]),policy:d([`visible`,`privileged`,`metadata`]),required:l()}),In=c({schemaVersion:a(`openpond.datasetImportMapping.v1`),sourceSchemaHash:Dn,configuration:f().trim().min(1).max(500),upstreamSplits:n(f().trim().min(1).max(500)).min(1).max(1e3),preset:d([`prompt_only`,`prompt_completion`,`messages`,`preference`,`prompt_expected_answer`,`prompt_privileged_solution`,`prompt_feedback`]),bindings:n(Fn).min(1).max(1e4),nullPolicy:d([`reject`,`drop`,`preserve`]),invalidRowPolicy:d([`reject_import`,`drop_with_receipt`]),splitPolicy:c({seed:u().int(),assignments:o(f(),qe),validationPercent:u().min(0).max(100),frozenEvalPercent:u().min(0).max(100)}),importerVersion:f().trim().min(1).max(200),mappingHash:Dn}),Ln=c({phase:d([`idle`,`metadata`,`download`,`mapping`,`parquet_write`,`verification`,`publish`]),completedBytes:u().int().nonnegative(),totalBytes:u().int().nonnegative().nullable(),completedRows:u().int().nonnegative(),totalRows:u().int().nonnegative().nullable(),message:f().trim().min(1).max(2e3).nullable()}),Rn=c({schemaVersion:a(`openpond.datasetImportJob.v1`),id:Tn,profileId:Tn,sourceKind:d([`huggingface`,`uploaded_file`]),status:kn,locator:An.nullable(),inspection:Nn.nullable(),mapping:In.nullable(),progress:Ln,targetStorageRoot:f().trim().min(1).max(4e3).nullable(),tasksetId:Tn.nullable(),tasksetRevision:u().int().positive().nullable(),artifactId:Tn.nullable(),error:f().trim().min(1).max(2e4).nullable(),cancellationRequested:l(),createdAt:En,updatedAt:En,completedAt:En.nullable(),metadata:On});c({profileId:Tn,url:f().trim().min(1).max(2e3)}),c({name:f().trim().min(1).max(500),objective:f().trim().min(1).max(2e4),mapping:In,targetStorageRoot:f().trim().min(1).max(4e3).nullable(),licenseApproved:l().default(!1)});var N=f().trim().min(1).max(240),P=f().trim().min(1),zn=o(f(),p()).default({}),Bn=d([`nominated`,`needs_review`,`approved_for_creation`,`creating`,`needs_task_review`,`baselining`,`ready_for_export`,`rejected`,`dismissed`,`retired`,`blocked`]),Vn=c({id:N,kind:d([`repeated_success`,`accepted_correction`,`agent_action_recurrence`,`outcome_linked_change`,`frontier_cost`,`expert_label`,`runtime_feedback`]),sourceRefIds:n(N).min(1).max(1e4),occurredAt:P,signature:f().trim().min(1).max(2e3),summary:f().trim().min(1).max(5e3),confidence:u().min(0).max(1),consented:l(),metadata:zn}),Hn=c({frequency:u().min(0).max(1),businessValue:u().min(0).max(1),frontierCost:u().min(0).max(1),signalQuality:u().min(0).max(1),verifiability:u().min(0).max(1),repeatability:u().min(0).max(1),privacyRisk:u().min(0).max(1),overall:u().min(0).max(1)}),Un=c({tactic:_t,eligible:l(),reasons:n(f().trim().min(1).max(5e3)).min(1).max(100),blockers:n(f().trim().min(1).max(5e3)).max(100).default([]),requiredSignals:n(f().trim().min(1).max(500)).max(100).default([]),generatedBy:d([`decision_table`,`baseline_reassessment`])}),Wn=c({schemaVersion:a(`openpond.taskCandidate.v1`),id:N,profileId:N,status:Bn,fingerprint:f().trim().min(1).max(512),title:f().trim().min(1).max(500),summary:f().trim().min(1).max(1e4),workflowSignature:f().trim().min(1).max(2e3),evidence:n(Vn).min(1).max(1e5),scorecard:Hn,recommendation:Un,mergedIntoId:N.nullable(),createdAt:P,updatedAt:P,metadata:zn}),Gn=c({schemaVersion:a(`openpond.taskMinerConfig.v1`),enabled:l().default(!1),localOnly:l().default(!0),observationWindowDays:u().int().min(1).max(365).default(30),minimumRecurrence:u().int().min(2).max(100).default(3),clustering:a(`hybrid_deterministic_first`).default(`hybrid_deterministic_first`),consentRequired:l().default(!0)});c({items:n(Wn),generatedAt:P,config:Gn}),c({status:Bn.optional(),mergeIntoId:N.nullable().optional()}),c({profileId:N,sourceIds:n(N).max(1e5).default([]),sessionIds:n(N).max(1e5).default([]),config:Gn.optional()});var Kn=c({schemaVersion:a(`openpond.taskMinerRun.v1`),id:N,profileId:N,status:d([`queued`,`running`,`cancelling`,`cancelled`,`succeeded`,`failed`]),config:Gn,sourceIds:n(N).max(1e5),sessionIds:n(N).max(1e5).default([]),progress:c({stage:d([`queued`,`ingesting`,`preparing`,`clustering`,`persisting`,`complete`]),processedSources:u().int().nonnegative(),totalSources:u().int().nonnegative(),candidatesFound:u().int().nonnegative(),skippedSources:u().int().nonnegative().default(0)}),candidateIds:n(N).max(1e5),cancelRequested:l(),error:f().trim().min(1).max(2e4).nullable(),createdAt:P,startedAt:P.nullable(),completedAt:P.nullable(),updatedAt:P});function F(e){let t;return new te({type:`object`,get shape(){return t??=e()}})}var qn=f().trim().min(1).max(200),Jn=c({operationId:qn,execution:S,executionKind:a(`recorded_evidence`).optional(),graders:n(Ee).min(1).max(100),maximumCostUsd:u().finite().min(1e-6).max(1e4),mappings:n(c({graderId:qn,fields:Kt}).strict()).max(100).optional()}).strict().superRefine((e,t)=>{new Set(e.graders.map(e=>e.id)).size!==e.graders.length&&t.addIssue({code:`custom`,path:[`graders`],message:`Select each grader once.`}),e.mappings&&(new Set(e.mappings.map(e=>e.graderId)).size!==e.mappings.length||e.mappings.some(t=>!e.graders.some(e=>e.id===t.graderId)))&&t.addIssue({code:`custom`,path:[`mappings`],message:`Map each selected grader at most once.`})});c({schemaVersion:a(`openpond.experimentScoringPass.v1`),id:qn,teamId:qn,revision:u().int().positive(),request:Jn,graders:n(Xt).min(1).max(100),status:d([`queued`,`running`,`cancelling`,`completed`,`failed`,`cancelled`]),totalCount:u().int().min(1).max(1e4),counts:c({pending:u().int().nonnegative(),running:u().int().nonnegative(),scored:u().int().nonnegative(),unavailable:u().int().nonnegative(),failed:u().int().nonnegative(),cancelled:u().int().nonnegative()}).strict(),resultAvailable:l(),createdAt:t(),startedAt:t().nullable(),completedAt:t().nullable(),error:c({code:qn,message:f().max(2e3)}).strict().nullable()}).strict().extend({contentHash:f().regex(/^[a-f0-9]{64}$/)});var Yn=c({sourcePass:S,graders:Jn.shape.graders,mappings:Jn.shape.mappings,maximumCostUsd:u().finite().positive().max(1e4)}).strict(),Xn=c({externalDatasetBinding:pe,scoring:Yn.optional(),modelId:f().min(1).max(300),maximumCostUsd:u().finite().positive().max(1e4)}).strict(),Zn=c({source:d([`local`,`github`,`openpond_git`]),repositoryId:f().trim().min(1),profileId:f().trim().min(1)}).strict(),I=f().min(1).max(240),L=f().regex(/^[a-f0-9]{64}$/),Qn=c({schemaVersion:a(`openpond.experimentImprovement.v1`),id:I,operationId:I,requestHash:L,actorId:I,teamId:I,projectId:I.nullable(),ownerWorkspaceId:I,ownerRevision:u().int().nonnegative(),ownerSourceRevision:f().min(1).max(500),baseRelease:S,profileRef:Zn,baseProfileSourceRevision:f().min(1).max(500),evidence:c({manifest:xe,result:be}).strict(),component:c({kind:d([`skill`,`workflow`,`agent`,`instruction`]),path:f().min(1).max(2e3),workflowId:I.optional()}).strict(),mode:d([`manual`,`llm_assisted`]),status:d([`draft`,`authoring_queued`,`authoring`,`frozen`,`testing`,`ready_for_review`,`adopting`,`accepted`,`rolled_back`,`declined`,`failed`,`cancelled`]),revision:u().int().positive(),limits:c({maximumCostUsd:u().finite().positive(),maximumAuthoringSteps:u().int().min(1).max(500),maximumDurationMs:u().int().min(1e3).max(864e5)}).strict(),authoringSteps:u().int().nonnegative().max(500),work:c({sessionId:I,turnId:I,revision:u().int().positive(),startedAt:f().datetime()}).strict().nullable(),frozen:c({release:S,sourceRevision:f().min(1).max(500),profileSourceRevision:f().min(1).max(500),authoringHash:L,partitionHash:L,executableSourceHash:L,protectedClosureHash:L,candidateRevision:u().int().positive(),diff:n(c({path:f(),before:L.nullable(),after:L.nullable()}).strict()).max(1e4)}).strict().nullable(),test:c({recipe:Xn,recipeHash:f().regex(/^[a-f0-9]{64}$/),baselineOperationId:f(),candidateOperationId:f(),cancelRequestedAt:f().datetime().nullable(),dispatching:c({kind:d([`baseline`,`candidate`]),token:I,pid:u().int().positive(),processStart:f().nullable(),startedAt:f().datetime()}).strict().nullable()}).strict().nullable(),comparison:c({baseline:c({manifest:xe,result:be}).strict(),candidate:c({manifest:xe,result:be}).strict(),frozenHash:L,humanProjection:ot.optional()}).strict().nullable(),adoptionIntent:c({operationId:I,expectedOwnerRevision:u().int().nonnegative(),frozenHash:L,humanProjection:ot.optional()}).strict().nullable(),acceptance:S.nullable(),rollback:S.nullable(),error:f().max(2e3).nullable(),createdAt:f().datetime(),updatedAt:f().datetime(),contentHash:L}).strict();function $n(e){let t=Qn.parse(e),{contentHash:n,...r}=t;if(x(r)!==n)throw Error(`Improvement state integrity failed.`);if(ye(t.evidence),t.evidence.manifest.teamId!==t.teamId||t.evidence.result.status!==`completed`||!t.evidence.result.cases.some(e=>e.status===`completed`))throw Error(`Improvement requires usable completed owner evidence.`);if(t.test&&(x(t.test.recipe)!==t.test.recipeHash||ve(t.test.recipe.externalDatasetBinding).profileId!==t.profileRef.profileId))throw Error(`The test recipe changed its exact selected Profile.`);if(t.comparison?.humanProjection){let e=ct(t.comparison.humanProjection);if(e.teamId!==t.teamId||e.projectId!==t.projectId||x(e.baseline)!==x(t.comparison.baseline)||x(e.candidate)!==x(t.comparison.candidate))throw Error(`Human comparison projection differs from its exact owner evidence.`)}if(t.comparison&&(ye(t.comparison.baseline),ye(t.comparison.candidate),!t.frozen||t.comparison.frozenHash!==x(t.frozen)||!Te(t.comparison.baseline,t.comparison.candidate).comparable))throw Error(`Improvement comparison is not bound to a comparable frozen candidate.`);if(t.status===`ready_for_review`&&(!t.frozen||!t.comparison))throw Error(`Improvement is missing qualification.`);if(t.status===`adopting`&&(!t.adoptionIntent||!t.frozen||t.adoptionIntent.frozenHash!==x(t.frozen)))throw Error(`Improvement adoption intent is not bound to its tested revision.`);if(t.status===`accepted`&&(!t.acceptance||!t.comparison||!t.frozen))throw Error(`Improvement acceptance is missing its exact qualification receipt.`);if(t.status===`rolled_back`&&(!t.acceptance||!t.rollback))throw Error(`Profile rollback is missing its durable owner receipt.`);return t}var er=c({states:n(Qn).max(50),nextCursor:I.nullable()}).strict(),tr=c({operationId:I,ownerWorkspaceId:I,expectedOwnerRevision:u().int().nonnegative(),baseRelease:S,profileRef:Zn,profileSourceRevision:f().min(1).max(500),projectId:I.nullable(),evidence:S,component:Qn.shape.component,mode:Qn.shape.mode,limits:Qn.shape.limits}).strict(),nr=i(`operation`,[c({operation:a(`list`),limit:u().int().min(1).max(50).default(30),cursor:I.optional()}).strict(),c({operation:a(`options`),evidence:S}).strict(),c({operation:a(`start`),request:tr}).strict(),c({operation:d([`read`,`humanComparisonOptions`]),id:I}).strict(),c({operation:a(`instructions`),id:I}).strict(),c({operation:a(`edit`),id:I,revision:u().int().positive(),expectedFileHash:L,text:f().max(25e4)}).strict(),c({operation:a(`mode`),id:I,revision:u().int().positive(),mode:d([`manual`,`llm_assisted`])}).strict(),c({operation:a(`author`),id:I,revision:u().int().positive(),modelId:I,prompt:f().min(1).max(2e4)}).strict(),c({operation:d([`freeze`,`finishAuthoring`,`useCandidate`,`discard`,`cancel`,`rollback`]),id:I,revision:u().int().positive()}).strict(),c({operation:a(`test`),id:I,revision:u().int().positive(),recipe:Xn}).strict(),c({operation:a(`compare`),id:I,revision:u().int().positive(),baseline:S,candidate:S,humanSelections:rt.optional()}).strict()]),rr=c({teamId:I,evidence:S,scoring:Yn.optional(),profiles:n(c({id:L,name:f(),profileRef:Zn,ownerWorkspaceId:I,ownerRevision:u().int().nonnegative(),baseRelease:S,profileSourceRevision:f(),components:n(c({kind:d([`skill`,`instruction`,`agent`,`workflow`]),path:f(),agentId:f().optional(),workflowId:f().optional(),label:f(),binding:pe}).strict()).max(500)}).strict()).max(100)}).strict(),ir=c({candidateId:I,actorId:I,teamId:I,projectId:I.nullable(),candidateRevision:u().int().positive(),recipeHash:L,baseline:c({manifest:xe,result:be}).strict(),candidate:c({manifest:xe,result:be}).strict(),contentHash:L}).strict();function ar(e){let t=ir.parse(e),{contentHash:n,...r}=t;if(x(r)!==n)throw Error(`Accepted Human discovery integrity failed.`);for(let e of[t.baseline,t.candidate])if(ye(e),e.manifest.teamId!==t.teamId||e.manifest.lineage?.humanProjection)throw Error(`Discovery must retain original owner evidence.`);return t}var R=f().trim().min(1).max(240),z=f().regex(/^[a-f0-9]{64}$/),or=c({schemaVersion:a(`openpond.advancedRefinerEvaluation.v1`),operationId:R,actorId:R,teamId:R,projectId:R.nullable(),mode:d([`review_quality`,`downstream_adaptation`]),profileRef:c({source:d([`local`,`github`,`openpond_git`]),repositoryId:R,profileId:R}).strict(),sourceRevision:f().min(1).max(500),baselineRelease:S,refinerRelease:S,evidence:S,externalDatasetBinding:pe,adaptationDatasetBinding:pe,privateClosureHash:z,adaptationTaskIds:n(R).min(1).max(1e4),holdoutTaskIds:n(R).min(1).max(1e4),adaptationSplit:d([`train`,`validation`,`test`,`frozen_eval`]),holdoutSplit:d([`train`,`validation`,`test`,`frozen_eval`]),environmentPolicy:i(`kind`,[c({kind:a(`resettable`),environmentHash:z}).strict(),c({kind:a(`frozen_observations`),environmentHash:z}).strict()]),maximumCostUsd:u().finite().positive().max(1e4),maximumDurationMs:u().int().min(1e3).max(864e5),maximumModelSteps:u().int().min(1).max(1e3),contentHash:z}).strict();function sr(e){let t=or.parse(e),{contentHash:n,...r}=t;if(x(r)!==n)throw Error(`Advanced Refiner evaluation admission changed.`);let i=ve(t.externalDatasetBinding);if(i.profileId!==t.profileRef.profileId||i.protectedProfileClosureHash!==t.privateClosureHash)throw Error(`Advanced evaluation changed its Profile or protected evaluator closure.`);if(t.adaptationSplit===t.holdoutSplit||new Set([...t.adaptationTaskIds,...t.holdoutTaskIds]).size!==t.adaptationTaskIds.length+t.holdoutTaskIds.length)throw Error(`Adaptation and private holdout must be disjoint ordered cases.`);let a=ve(t.adaptationDatasetBinding);if(a.profileId!==i.profileId||a.packageHash!==i.packageHash||x(a.dataset)!==x(i.dataset)||x(a.target)!==x(i.target)||a.protectedProfileClosureHash!==i.protectedProfileClosureHash)throw Error(`This native advanced adapter needs separate adaptation and holdout recipes from the same exact released package and target.`);if(a.split!==t.adaptationSplit||i.split!==t.holdoutSplit)throw Error(`Advanced split policies differ from their exact admitted recipes.`);let o=new Set(i.population.map(e=>e.taskId)),s=new Set(a.population.map(e=>e.taskId));if(t.holdoutTaskIds.some(e=>!o.has(e))||t.adaptationTaskIds.some(e=>!s.has(e)))throw Error(`Advanced evaluation selected cases outside its exact Dataset recipe.`);return t}var cr=c({pin:or,modelId:R,profileId:R,model:Me,seed:u().int(),reasoningEffort:d([`none`,`off`,`low`,`medium`,`high`,`xhigh`,`max`]).nullable()}).strict(),lr=i(`operation`,[c({operation:a(`start`),request:cr}).strict(),c({operation:d([`read`,`cancel`,`resume`]),id:R}).strict(),c({operation:a(`list`)}).strict()]),ur=c({taskId:R,sourceCaseHash:z,reviewSessionId:R,reviewTurnId:R,inputHarness:S,trigger:S,decision:d([`no_action`,`route_deterministically`,`queue_refiner`]),outcome:S.nullable(),proposal:S.nullable(),validations:n(S).max(100),grade:c({id:R,contentHash:z,score:u().min(0).max(1).nullable(),passed:l(),failureClass:f().nullable()}).strict(),resetComplete:l()}).strict(),dr=c({schemaVersion:a(`openpond.advancedRefinerReviewReceipt.v1`),id:R,runId:R,pinHash:z,ownerActorId:R,teamId:R,privateClosureHash:z,refinerRelease:S,cases:n(ur).min(1).max(1e4),reviewScore:u().min(0).max(1).nullable(),completedCases:u().int().nonnegative(),noActionCases:u().int().nonnegative(),routedCases:u().int().nonnegative(),proposalCases:u().int().nonnegative(),costUsd:u().finite().nonnegative().nullable(),cleanupComplete:l(),createdAt:f().datetime(),contentHash:z}).strict(),fr=rr.shape.profiles.element,pr=rr.extend({profiles:n(fr.extend({components:n(fr.shape.components.element.extend({adaptationSplits:n(c({split:d([`train`,`validation`,`test`,`frozen_eval`]),taskIds:n(R)})),environmentHash:z,resettable:l(),reviewQuality:l()}))})),refinerReleases:n(S),activeRefiner:S.nullable(),models:n(c({id:R,profileId:R,name:f(),location:d([`local`,`hosted`])}))}),mr=c({operation:a(`prepare`),evidence:S,profileOptionId:R,componentLabel:R,mode:d([`review_quality`,`downstream_adaptation`]),adaptationSplit:d([`train`,`validation`,`test`,`frozen_eval`]),operationId:R,refinerRelease:S,maximumCostUsd:u().positive(),maximumDurationMs:u().int().positive(),maximumModelSteps:u().int().positive()}).strict();r([lr,mr,c({operation:a(`options`),evidence:S}).strict()]);var hr=c({modelRequests:u().int().nonnegative(),knownSpendUsd:u().nonnegative(),totalSpendUsd:u().nonnegative().nullable(),heldUsd:u().nonnegative(),uncertainRequests:u().int().nonnegative()}).strict(),B=y.nullable(),V=u().int().nonnegative(),gr=f().regex(/^[a-f0-9]{40}$/),_r=c({name:f().trim().min(1).max(160),startedAt:m,completedAt:m,durationMs:u().nonnegative(),clock:d([`monotonic`,`provider`,`wall`]),outcome:d([`succeeded`,`failed`,`cancelled`])}).strict(),vr=c({schemaVersion:a(`openpond.correlatedTelemetryReceipt.v1`),stage:d([`training`,`evaluation`,`deployment`,`inference`]),correlation:c({modelRunId:B,modelVersionId:B,policyVersion:V.nullable(),taskId:B,rolloutGroupId:B,providerResourceId:B,deploymentId:B,inferenceRequestId:B}).strict(),spans:n(_r).max(1e4),usage:c({promptTokens:V.nullable(),generatedTokens:V.nullable(),gpuSeconds:u().nonnegative().nullable(),workerActiveSeconds:u().nonnegative().nullable(),optimizerSteps:V.nullable(),rolloutGroups:V.nullable(),successfulTrajectories:V.nullable(),failedTrajectories:V.nullable(),peakGpuMemoryBytes:V.nullable(),peakGpuUtilizationPercent:u().min(0).max(100).nullable()}).strict(),resource:c({provider:y,resourceIds:n(y).max(1e3),gpuType:f().trim().min(1).max(300).nullable(),gpuCount:V.nullable(),baseProfileId:B,baseRepository:f().trim().min(1).max(512).nullable(),baseRevision:gr.nullable(),adapterContentHash:v.nullable()}).strict(),cost:c({currency:a(`USD`),providerReportedUsd:u().nonnegative().nullable(),quotedHourlyUsd:u().nonnegative().nullable(),estimatedUsd:u().nonnegative().nullable(),methodologyVersion:f().trim().min(1).max(160),pricingInputs:o(f().trim().min(1).max(160),u().finite()),unitEstimates:o(f().trim().min(1).max(160),u().nonnegative())}).strict(),recordedAt:m,contentHash:v}).strict(),H=f().trim().min(1).max(240),U=f().regex(/^[a-f0-9]{64}$/),W=f().datetime({offset:!0}),G=c({id:H,contentHash:U}).strict(),yr=G.extend({revision:u().int().positive()}).strict(),br=c({providerId:H,modelId:f().trim().min(1).max(300)}).strict(),xr=d([`correction`,`sibling_verification`,`cumulative_known`,`development`,`retained`,`frozen_final`,`training_eligible`]),Sr=c({taskId:H,taskContentHash:U,panelRole:xr,passLabel:H.nullable(),optimizerEligible:l(),criticalInvariantIds:n(H).max(100).default([])}).strict().superRefine((e,t)=>{e.optimizerEligible!==(e.panelRole===`correction`)&&t.addIssue({code:`custom`,path:[`optimizerEligible`],message:`Only correction cases may be optimizer eligible.`})}),Cr=c({schemaVersion:a(`openpond.continualBenchIssuePacket.v1`),id:H,revision:u().int().positive(),contentHash:U,familyId:H,familyLabel:f().trim().min(1).max(500),severity:d([`low`,`medium`,`high`,`critical`]),cases:n(Sr).min(2).max(1e4),duplicateEvidence:G,leakageEvidence:G,priorExposureEvidence:G,observedAt:W,reviewedAt:W,queuedAt:W.nullable(),runAt:W.nullable(),createdBy:H,createdAt:W}).strict().superRefine((e,t)=>{e.cases.some(e=>e.panelRole===`correction`)||t.addIssue({code:`custom`,path:[`cases`],message:`An issue packet requires a correction case.`}),e.cases.some(e=>e.panelRole===`sibling_verification`)||t.addIssue({code:`custom`,path:[`cases`],message:`An issue packet requires a sibling-verification case.`}),new Set(e.cases.map(e=>e.taskId)).size!==e.cases.length&&t.addIssue({code:`custom`,path:[`cases`],message:`Issue-packet task ids must be unique.`})}),wr=c({id:H,role:xr,passLabel:H.nullable(),taskset:yr,familyIds:n(H).min(1).max(1e5),taskCount:u().int().positive().max(1e6),sealedAt:W}).strict().superRefine((e,t)=>{let n=e.role===`correction`||e.role===`sibling_verification`||e.role===`cumulative_known`;n!==(e.passLabel!==null)&&t.addIssue({code:`custom`,path:[`passLabel`],message:n?`This panel role requires a pass label.`:`This panel role is series-wide and cannot carry a pass label.`})}),Tr=i(`kind`,[c({kind:a(`base_model`),id:H,revision:f().trim().min(1).max(500),contentHash:U}).strict(),c({kind:a(`model_version`),id:H,contentHash:U}).strict(),c({kind:a(`external_reference`),id:H,model:br,providerRevision:f().trim().min(1).max(500),contentHash:U}).strict()]),Er=c({scheduleEntryId:H,ordinal:u().int().nonnegative(),label:H,role:d([`seed`,`daily_residual`,`rank_candidate`,`weekly_rollup`,`full_refresh`]),parentRule:d([`base_model`,`previous_release`,`seed_release`,`accepted_daily_head`,`accepted_seed`]),trainableRank:u().int().positive().max(1024),correctionPanelIds:n(H).min(1).max(1e3),optimizerGroupsPerTask:u().int().positive().max(1e3),trajectoriesPerGroup:u().int().min(4).max(1e3)}).strict(),Dr=c({revision:u().int().positive(),requireAllAttemptsTerminal:a(!0),criticalCorrectionPassRate:u().min(0).max(1),siblingPassRate:u().min(0).max(1),behavioralRetentionRate:u().min(0).max(1),maximumRetainedRegressionPoints:u().min(0).max(100),blockCriticalPriorRegression:l(),authorizedClaims:n(d([`systems_complete`,`correction_absorbed`,`issue_generalized`,`continually_current`,`frontier_pareto_result`])).min(1).max(5)}).strict(),Or=c({schemaVersion:a(`openpond.continualBenchmarkProtocol.v1`),id:H,revision:u().int().positive(),contentHash:U,ownerId:H,createdAt:W,sealedAt:W,creationReceipt:G,predecessorSeries:c({id:H,revision:u().int().positive(),contentHash:U}).strict().nullable(),scenarioPack:G,issueFamilyLedger:G,issuePacketRelease:G,panels:n(wr).min(6).max(1e4),grader:G,judge:c({release:G,rubricRelease:G,calibrationRelease:G,model:br}).strict().nullable(),policies:c({base:Tr,master:Tr.nullable(),externalReferences:n(Tr).min(1).max(20)}).strict(),invariants:c({systemPromptHash:U,toolSchema:G,application:G,harness:G,runtime:G,workerImage:G,autoRefiner:c({enabled:l(),release:G.nullable()}).strict()}).strict(),schedule:n(Er).min(1).max(1e3),evaluation:c({seeds:n(u().int()).min(3).max(100),repetitions:u().int().min(3).max(20),powerRule:G,pairedBootstrapSamples:u().int().min(1e3).max(1e6),confidenceLevel:a(.95)}).strict(),resources:c({maximumTrainingGpuSeconds:u().int().positive(),maximumEvaluationGpuSeconds:u().int().positive(),maximumProviderSpendUsd:u().positive(),maximumTotalSpendUsd:u().positive(),maximumConcurrentGroups:u().int().positive().max(1e3)}).strict(),currencyThresholds:Dr}).strict().superRefine((e,t)=>{let n=e.panels.map(e=>e.id);new Set(n).size!==n.length&&t.addIssue({code:`custom`,path:[`panels`],message:`Protocol panel ids must be unique.`});for(let n of[`development`,`retained`,`frozen_final`,`training_eligible`])e.panels.filter(e=>e.role===n).length!==1&&t.addIssue({code:`custom`,path:[`panels`],message:`A protocol requires exactly one ${n} panel.`});[...e.schedule].sort((e,t)=>e.ordinal-t.ordinal).some((e,t)=>e.ordinal!==t)&&t.addIssue({code:`custom`,path:[`schedule`],message:`Protocol schedule ordinals must be contiguous from zero.`});let r=new Set(n);e.schedule.some(e=>e.correctionPanelIds.some(e=>!r.has(e)))&&t.addIssue({code:`custom`,path:[`schedule`],message:`Every scheduled correction panel must belong to the protocol.`})}),kr=d([`draft`,`active`,`completed`,`archived`]),Ar=d([`draft`,`ready`,`queued`,`running`,`candidate`,`accepted`,`rejected`,`no_signal`,`failed`,`cancelled`]),jr=d([`seed`,`daily_residual`,`rank_candidate`,`weekly_rollup`,`full_refresh`]),Mr=d([`base_model`,`previous_release`,`seed_release`,`accepted_daily_head`,`accepted_seed`]),Nr=d([`seed_taskset`,`nightly_selection`,`daily_cohort_union`,`eligible_task_pool`]),Pr=d([`live_client`,`evaluation_attempt`,`replay_evidence`,`manual`]),Fr=c({id:y,ordinal:u().int().nonnegative(),label:f().trim().min(1).max(100),role:jr,parentRule:Mr,taskSource:Nr,trainableRank:u().int().positive().max(1024),minimumTasks:u().int().nonnegative().max(1e5),maximumTasks:u().int().positive().max(1e5)}).strict().superRefine((e,t)=>{e.minimumTasks>e.maximumTasks&&t.addIssue({code:`custom`,path:[`minimumTasks`],message:`The minimum task count cannot exceed the maximum task count.`});let{parentRules:n,taskSource:r}={seed:{parentRules:[`base_model`],taskSource:`seed_taskset`},daily_residual:{parentRules:[`previous_release`,`accepted_daily_head`],taskSource:`nightly_selection`},rank_candidate:{parentRules:[`seed_release`],taskSource:`eligible_task_pool`},weekly_rollup:{parentRules:[`seed_release`,`accepted_seed`],taskSource:`daily_cohort_union`},full_refresh:{parentRules:[`base_model`],taskSource:`eligible_task_pool`}}[e.role];n.includes(e.parentRule)||t.addIssue({code:`custom`,path:[`parentRule`],message:`${e.role} requires parent rule ${n.join(` or `)}.`}),e.taskSource!==r&&t.addIssue({code:`custom`,path:[`taskSource`],message:`${e.role} requires task source ${r}.`}),e.ordinal===0&&e.role!==`seed`&&t.addIssue({code:`custom`,path:[`role`],message:`The first series entry must be the seed.`})}),Ir=c({id:y,version:u().int().positive(),requireCheckpoint:l(),requireAppliedOptimizerUpdate:l(),minimumCurrentCohortMeanImprovement:u().finite(),maximumRetainedMeanRegression:u().finite().nonnegative(),blockCriticalInvariantRegression:l(),automaticDailyAdvancement:l()}).strict(),Lr=c({schemaVersion:a(`openpond.modelComparisonSeries.v1`),id:y,profileId:y,modelProjectId:y,name:f().trim().min(1).max(500),objective:f().trim().min(1).max(2e4),status:kr,revision:u().int().positive(),productionBinding:c({role:a(`chat_manual`),roleTargetId:y}).strict(),baseModel:c({id:y,revision:f().trim().min(1).max(256)}).strict(),seedTaskset:_,eligibleTaskPool:_,evaluationTasksets:c({development:_,retained:_,frozenFinal:_}).strict(),grader:g,benchmarkProtocol:Or.nullable().default(null),automaticEvaluation:c({enabled:l()}).strict().default({enabled:!1}),residualProfile:c({profileId:y,serializedEnvelopeRank:u().int().positive().max(1024),maximumEnabledRank:u().int().positive().max(1024),topology:a(`uniform_block_masked`)}).strict(),schedule:n(Fr).min(1).max(1e3),scheduleSealedAt:m.nullable(),advancementPolicy:Ir,executionPolicy:c({startWhenReady:a(!1)}).strict(),acceptedSeedEntryId:y.nullable(),acceptedDailyHeadEntryId:y.nullable(),promotedBindingId:y.nullable(),createdBy:y,createdAt:m,updatedAt:m}).strict().superRefine((e,t)=>{[...e.schedule].sort((e,t)=>e.ordinal-t.ordinal).some((e,t)=>e.ordinal!==t)&&t.addIssue({code:`custom`,path:[`schedule`],message:`Schedule ordinals must be unique and contiguous from zero.`}),(new Set(e.schedule.map(e=>e.id)).size!==e.schedule.length||new Set(e.schedule.map(e=>e.label)).size!==e.schedule.length)&&t.addIssue({code:`custom`,path:[`schedule`],message:`Schedule ids and labels must be unique.`}),e.residualProfile.maximumEnabledRank>e.residualProfile.serializedEnvelopeRank&&t.addIssue({code:`custom`,path:[`residualProfile`],message:`Maximum enabled rank cannot exceed the serialized envelope.`});let n=Object.values(e.evaluationTasksets);new Set(n.map(e=>`${e.id}:${e.revision}:${e.contentHash}`)).size!==n.length&&t.addIssue({code:`custom`,path:[`evaluationTasksets`],message:`Development, retained, and frozen-final cohorts require distinct immutable Taskset releases.`});let r=e.schedule.find(e=>e.role===`seed`);if(!r||e.schedule.filter(e=>e.role===`seed`).length!==1){t.addIssue({code:`custom`,path:[`schedule`],message:`A series requires exactly one seed entry.`});return}let i=e.schedule.filter(e=>e.role===`rank_candidate`);i.length===1&&t.addIssue({code:`custom`,path:[`schedule`],message:`A causal rank comparison requires at least two rank candidates.`}),new Set(i.map(e=>e.trainableRank)).size!==i.length&&t.addIssue({code:`custom`,path:[`schedule`],message:`Causal rank candidates must use distinct trainable ranks.`}),[r.trainableRank+e.schedule.filter(e=>e.role===`daily_residual`).reduce((e,t)=>e+t.trainableRank,0),...e.schedule.filter(e=>e.role===`weekly_rollup`).map(e=>r.trainableRank+e.trainableRank),...e.schedule.filter(e=>e.role===`rank_candidate`).map(e=>r.trainableRank+e.trainableRank),...e.schedule.filter(e=>e.role===`full_refresh`).map(e=>e.trainableRank)].some(t=>t>e.residualProfile.maximumEnabledRank)&&t.addIssue({code:`custom`,path:[`schedule`],message:`A declared branch exceeds the maximum enabled rank.`})}),Rr=c({source:Pr,taskIds:n(y).min(1).max(1e5),observedFrom:m,observedTo:m,reviewedAt:m,reviewedBy:y,sourceTaskset:_}).strict(),zr=Rr.extend({derivedFamilyKeys:n(y).min(1).max(1e5),familyDerivation:c({algorithm:y,sourceTasksetHash:v}).strict(),selectionHash:v}).strict(),Br=i(`kind`,[c({kind:a(`base_model`),id:y,revision:f().trim().min(1).max(256)}).strict(),c({kind:a(`model_version`),id:y,contentHash:v}).strict()]),Vr=c({id:y,branchOrdinal:u().int().nonnegative(),rank:u().int().positive().max(1024),offsetStart:u().int().nonnegative(),offsetEnd:u().int().positive(),optimizationRole:d([`frozen`,`trainable`]),artifactLineageId:y.nullable()}).strict().superRefine((e,t)=>{e.offsetEnd-e.offsetStart!==e.rank&&t.addIssue({code:`custom`,path:[`offsetEnd`],message:`Block offsets must span exactly the declared rank.`})}),Hr=c({disposition:d([`advance`,`hold`,`no_signal`]),policy:c({id:y,version:u().int().positive()}).strict(),reasonCodes:n(y).min(1).max(100),summary:f().trim().min(1).max(5e3),decidedBy:y,decidedAt:m}).strict(),Ur=c({evaluationRunId:y,modelVersionId:y,taskset:_,grader:g,cohortRole:d([`current`,`correction`,`sibling_verification`,`cumulative_known`,`development`,`retained`,`prior_disclosed`,`frozen_final`]),panelId:y.nullable().default(null),protocol:c({id:y,revision:u().int().positive(),contentHash:v}).strict().nullable().default(null)}).strict(),Wr=c({attemptOrdinal:u().int().positive(),trainingPlanId:y.nullable(),modelRunId:y.nullable(),terminalStatus:d([`failed`,`cancelled`]),queuedAt:m.nullable(),startedAt:m.nullable(),completedAt:m}).strict(),Gr=c({seriesId:y,entryId:y,scheduleEntryId:y,ordinal:u().int().nonnegative(),releaseHash:v}).strict(),Kr=c({schemaVersion:a(`openpond.modelComparisonSeriesEntry.v1`),id:y,seriesId:y,profileId:y,modelProjectId:y,scheduleEntryId:y,releaseHash:v,ordinal:u().int().nonnegative(),label:f().trim().min(1).max(100),role:jr,branch:d([`daily`,`rank_candidate`,`weekly_rollup`,`full_refresh`]),status:Ar,parent:Br,taskset:_,sourceTasksets:n(_).min(1).max(1e3),taskSelection:zr.nullable(),trainableRank:u().int().positive().max(1024),serializedEnvelopeRank:u().int().positive().max(1024),enabledCumulativeRank:u().int().positive().max(1024),trainableBlockId:y,residualBlocks:n(Vr).min(1).max(1024),attemptOrdinal:u().int().positive().default(1),priorRunAttempts:n(Wr).max(100).default([]),trainingPlanId:y.nullable(),modelRunId:y.nullable(),modelVersionId:y.nullable(),evaluations:n(Ur).max(1e5),decision:Hr.nullable(),promotionBindingId:y.nullable(),queuedAt:m.nullable(),startedAt:m.nullable(),completedAt:m.nullable(),createdAt:m,updatedAt:m}).strict().superRefine((e,t)=>{let n=[...e.residualBlocks].sort((e,t)=>e.offsetStart-t.offsetStart);n.some((e,t)=>e.offsetStart!==(t===0?0:n[t-1].offsetEnd))&&t.addIssue({code:`custom`,path:[`residualBlocks`],message:`Residual block offsets must be contiguous from zero.`});let r=n.at(-1)?.offsetEnd??0;(r!==e.enabledCumulativeRank||r>e.serializedEnvelopeRank)&&t.addIssue({code:`custom`,path:[`enabledCumulativeRank`],message:`Enabled rank must match the block inventory and fit the envelope.`});let i=e.residualBlocks.filter(e=>e.optimizationRole===`trainable`);(i.length!==1||i[0]?.id!==e.trainableBlockId||i[0]?.rank!==e.trainableRank)&&t.addIssue({code:`custom`,path:[`trainableBlockId`],message:`Exactly one block must be trainable during the Run and match trainableRank.`});let a=e.priorRunAttempts.map(e=>e.attemptOrdinal);(e.priorRunAttempts.length!==e.attemptOrdinal-1||a.some((e,t)=>e!==t+1))&&t.addIssue({code:`custom`,path:[`priorRunAttempts`],message:`Prior Run attempts must be a contiguous history before the active attempt.`})});c({seriesId:y,scheduleEntryId:y,taskSelection:Rr.nullable(),expectedSeriesRevision:u().int().positive()}).strict();var qr=d([`base_reference`,`lora_adapter`]),Jr=c({baseModel:c({source:a(`huggingface`),repoId:y,revision:f().regex(/^[a-f0-9]{40}$/),configHash:v,tokenizerHash:v,licenseId:y,gated:l()}).strict(),processor:c({repository:y,revision:f().regex(/^[a-f0-9]{40}$/),configHash:v}).strict()}).strict().superRefine((e,t)=>{(e.baseModel.repoId!==e.processor.repository||e.baseModel.revision!==e.processor.revision)&&t.addIssue({code:`custom`,path:[`processor`],message:`Reward Model processor must match the pinned scorer base model.`})}),Yr=c({schemaVersion:a(`openpond.modelVersion.v1`),id:y,modelId:y,profileId:y,version:u().int().nonnegative(),kind:qr,status:d([`available`,`failed`]),baseModel:Sn,taskset:_,comparisonSeriesEntry:Gr.nullable().optional(),evaluationTasksetRef:_.nullable().optional(),releaseGraph:c({resolvedBundleHash:v,profileRelease:_,harnessRelease:g,agentRelease:g.nullable(),grader:c({id:y,contentHash:v}).strict()}).strict(),artifactLineageId:y.nullable(),adapterStatus:d([`not_trained`,`trained`]),createdAt:m,contentHash:v}).strict().superRefine((e,t)=>{e.kind===`base_reference`&&(e.version!==0||e.adapterStatus!==`not_trained`||e.artifactLineageId!==null)&&t.addIssue({code:`custom`,message:`A base-reference Model Version must be version 0 without adapter lineage.`}),e.kind===`lora_adapter`&&(e.version===0||e.adapterStatus!==`trained`||!e.artifactLineageId)&&t.addIssue({code:`custom`,message:`A LoRA Model Version must be version 1+ with adapter lineage.`})}),Xr=c({schemaVersion:a(`openpond.rewardModelVersion.v1`),id:y,modelId:y,profileId:y,version:u().int().positive(),role:a(`reward`),status:d([`available`,`failed`]),scope:d([`synthetic_smoke`,`human_preference`]),baseModel:Sn,runtime:Jr.nullable().default(null),taskset:_,preferenceDatasetRelease:g,releaseGraph:c({resolvedBundleHash:v,profileRelease:_,harnessRelease:g,grader:c({id:y,contentHash:v}).strict()}).strict(),artifacts:c({checkpoint:g.extend({objectRef:f().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),files:n(c({path:f().trim().min(1).max(512),sizeBytes:u().int().positive(),sha256:v}).strict()).min(4).max(128)}).strict(),adapter:g,scalarHead:g,bucketHead:g.nullable(),processorRelease:g}).strict(),qualificationReport:g.nullable(),createdAt:m,contentHash:v}).strict(),Zr=c({schemaVersion:a(`openpond.rewardModelRunReceipt.v1`),provider:y,providerRunId:y,resolvedBundleHash:v,finalCheckpoint:g.extend({objectRef:f().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/)}).strict(),adapter:g,scalarHead:g,bucketHead:g.nullable(),processorRelease:g,optimizerEvidence:g,managedExecutionReceipt:g.nullable().default(null),parameterDeltaHash:v,cleanup:c({computeReleased:l(),providerTerminalObserved:l()}).strict(),contentHash:v}).strict(),Qr=s(e=>{if(!e||typeof e!=`object`||Array.isArray(e))return e;let t=e;if(t.tasksetRelease!==void 0)return e;let n=t.taskset;if(!n||typeof n!=`object`||Array.isArray(n))return e;let r=n;return typeof r.id!=`string`||typeof r.contentHash!=`string`?e:{...t,tasksetRelease:{id:r.id,contentHash:r.contentHash}}},c({schemaVersion:a(`openpond.rewardModelRun.v1`),id:y,rewardModelId:y,rewardModelVersionId:y.nullable(),profileId:y,role:a(`reward`),scope:d([`synthetic_smoke`,`human_preference`]),status:d([`prepared`,`running`,`succeeded`,`failed`,`cancelled`]),taskset:_,tasksetRelease:g,preferenceDatasetRelease:g,recipeRelease:g,destinationId:y,quote:c({maximumSpendUsd:u().positive(),hourlyCostUsd:u().nonnegative().nullable()}).strict(),managedRunId:y.nullable(),progress:c({completedSteps:u().int().nonnegative(),totalSteps:u().int().positive(),latestLoss:u().finite().nullable()}).strict(),receipt:Zr.nullable(),qualificationReport:g.nullable(),accruedSpendUsd:u().nonnegative().nullable(),failureOwner:d([`authoring`,`admission`,`provider`,`runner`,`artifact`,`qualification`,`cleanup`]).nullable(),failure:f().trim().min(1).max(5e3).nullable(),startedAt:m,completedAt:m.nullable(),updatedAt:m}).strict()).superRefine((e,t)=>{e.status===`succeeded`&&!e.receipt&&t.addIssue({code:`custom`,message:`A successful Reward Model Run requires a canonical receipt.`}),e.status!==`succeeded`&&e.rewardModelVersionId!==null&&t.addIssue({code:`custom`,message:`Only a successful Reward Model Run can publish a Reward Model Version.`})}),$r=c({schemaVersion:a(`openpond.modelRunReceipt.v1`),provider:y,providerRunId:y,assignmentHash:v,resultHash:v,transcriptHash:v,traceHash:v.nullable(),resolvedBundleHash:v,artifactPath:f().trim().min(1).max(2e3),cleanup:c({computeReleased:l(),tunnelClosed:l()}).strict(),telemetry:vr.nullable().default(null),contentHash:v}).strict(),K=c({inputTokens:u().int().nonnegative(),outputTokens:u().int().nonnegative(),totalTokens:u().int().nonnegative(),costUsd:u().nonnegative().nullable()}).strict(),ei=f().regex(/^(?:[a-f0-9]{40}|[a-f0-9]{64})$/),ti=c({phase:d([`baseline`,`adaptation`,`candidate_adaptation`,`candidate`]),taskId:y,attemptId:y,sessionId:y.nullable(),turnId:y.nullable(),passed:l(),score:u().min(0).max(1).nullable(),failureClass:pn.nullable(),inputTokens:u().int().nonnegative(),outputTokens:u().int().nonnegative(),totalTokens:u().int().nonnegative(),latencyMs:u().int().nonnegative(),costUsd:u().nonnegative().nullable(),startedAt:m}).strict(),ni=c({targetTaskCount:u().int().nonnegative(),comparedTaskCount:u().int().nonnegative(),passedTaskCount:u().int().nonnegative(),failedTaskCount:u().int().nonnegative(),lowerTaskCount:u().int().nonnegative(),higherTaskCount:u().int().nonnegative(),unchangedTaskCount:u().int().nonnegative()}).strict(),ri=c({target:a(`all_tasks_lower`),targetTaskCount:u().int().nonnegative(),comparedTaskCount:u().int().nonnegative(),passedTaskCount:u().int().nonnegative(),failedTaskCount:u().int().nonnegative(),lowerTaskCount:u().int().nonnegative(),higherTaskCount:u().int().nonnegative(),unchangedTaskCount:u().int().nonnegative(),baselineTokens:u().int().nonnegative(),refinedTokens:u().int().nonnegative(),tokenDelta:u().int(),tokenDeltaPercent:u().finite().nullable(),complete:l(),passed:l(),cohorts:c({adaptation:ni,heldOut:ni}).strict()}).strict(),ii=c({baseline:K,adaptation:K,candidateAdaptation:K,candidate:K,refiner:K,grader:K}).strict(),ai=c({usage:ii,observedSpendUsd:u().nonnegative(),attempts:n(ti).max(1e4)}).strict(),oi=c({id:y,contentHash:v,artifactPath:f().trim().min(1).max(2e3)}).strict(),si=c({schemaVersion:a(`openpond.modelEvaluationReceipt.v1`),benchmarkId:y,resultManifest:c({id:y,contentHash:v,artifactPath:f().trim().min(1).max(2e3)}).strict(),stages:c({baseline:g,adaptation:g,refiner:g,candidateAdaptation:g,candidate:g,comparison:g}).strict(),usage:ii,quality:c({baselinePassRate:u().min(0).max(1),candidatePassRate:u().min(0).max(1),adaptationBaselinePassRate:u().min(0).max(1),adaptationCandidatePassRate:u().min(0).max(1),adaptationCandidatePassed:l(),heldOutCandidatePassed:l(),passed:l()}).strict(),foregroundTokenDelta:u().int(),foregroundTokenDeltaPercent:u().finite().nullable(),taskEfficiency:ri.optional(),efficiency:c({grossForegroundTokenSavings:u().int(),overheadTokens:u().int().nonnegative(),firstPassNetTokenSavings:u().int(),breakEvenReuseCount:u().int().nonnegative().nullable(),amortizedTokenSavings:u().int(),amortizedReuseCount:u().int().positive()}).strict(),budget:c({maximumSpendUsd:u().nonnegative(),observedSpendUsd:u().nonnegative(),enforced:l()}).strict(),evidenceSnapshot:g,lineage:c({adaptationEvidenceHash:v,refinerInputHash:v,refinerOutcomeHash:v,validationHash:v,applyReceiptHash:v,candidateRelease:g,valid:l()}).strict(),invalidReasons:n(f().trim().min(1).max(2e3)).max(100),attempts:n(ti).max(1e4).default([]),terminalClassification:d([`improved`,`no_improvement`,`regressed`,`inconclusive`,`cancelled`,`infrastructure_failure`]),profileGit:c({ref:f().trim().min(1).max(2e3),commit:ei,baseCommit:ei}).strict().nullable(),contentHash:v}).strict(),ci=c({schemaVersion:a(`openpond.modelEvaluationStopReceipt.v1`),benchmarkId:y,terminalClassification:a(`inconclusive`),stopReason:d([`candidate_harness_unchanged`,`candidate_lineage_invalid`]),reason:f().trim().min(1).max(5e3),stoppedAfter:a(`refiner`),baselineHarness:g,candidateHarness:g,refiner:c({id:y,contentHash:v,outcomeCount:u().int().nonnegative()}).strict(),usage:ii,budget:c({maximumSpendUsd:u().nonnegative(),observedSpendUsd:u().nonnegative(),enforced:l()}).strict(),evidenceSnapshot:g,attempts:n(ti).max(1e4),contentHash:v}).strict(),li=c({level:a(.95),lower:u().finite(),upper:u().finite()}).strict().superRefine((e,t)=>{e.lower>e.upper&&t.addIssue({code:`custom`,message:`A confidence interval's lower bound cannot exceed its upper bound.`})}),ui=c({schemaVersion:a(`openpond.modelComparisonBenchmarkReceipt.v1`),benchmarkId:a(`model-comparison`),target:c({kind:d([`base_model`,`model_version`,`external_reference`]),label:f().trim().min(1).max(300),modelVersionId:y.nullable(),model:b.nullable()}).strict(),taskset:_,grader:g,sampling:c({seeds:n(u().int()).min(1).max(100),repetitions:u().int().positive().max(20)}).strict(),deterministic:c({attemptedTaskCount:u().int().nonnegative(),completedTaskCount:u().int().nonnegative(),passedTaskCount:u().int().nonnegative(),failedTaskCount:u().int().nonnegative(),meanScore:u().finite().nullable(),passRate:u().min(0).max(1).nullable(),passRateCi95:li.nullable()}).strict(),judge:c({judgeModel:b,judgeRelease:g,rubricRelease:g,calibrationRelease:g,score:u().min(0).max(100),scoreCi95:li,comparisonCount:u().int().positive(),wins:u().int().nonnegative(),ties:u().int().nonnegative(),losses:u().int().nonnegative(),referenceLabel:f().trim().min(1).max(300)}).strict().superRefine((e,t)=>{e.wins+e.ties+e.losses!==e.comparisonCount&&t.addIssue({code:`custom`,message:`Judge win, tie, and loss counts must equal comparisonCount.`})}).nullable(),attempts:n(c({attemptId:y.nullable().default(null),taskId:y,seed:u().int(),repetition:u().int().nonnegative(),status:d([`succeeded`,`failed`]),deterministicScore:u().finite().nullable(),passed:l().nullable(),judgeScore:u().min(0).max(100).nullable(),judgePreference:d([`win`,`tie`,`loss`]).nullable(),transcriptHash:v.nullable(),traceHash:v.nullable(),transcriptArtifact:c({artifactPath:f().trim().min(1).max(2e3),jsonPointer:f().trim().min(1).max(1e3)}).strict().nullable().default(null),traceArtifact:c({artifactPath:f().trim().min(1).max(2e3),jsonPointer:f().trim().min(1).max(1e3)}).strict().nullable().default(null),latencyMs:u().int().nonnegative().nullable().default(null),failureClass:pn.nullable()}).strict()).max(1e5),usage:c({policy:K.nullable(),judge:K.nullable(),observedSpendUsd:u().nonnegative().nullable(),evaluationGpuSeconds:u().nonnegative().nullable().default(null)}).strict(),managedServing:c({jobId:y,terminalState:d([`completed`,`cancelled`,`failed`,`budget_exhausted`]),sourcePolicyVersion:u().int().nonnegative(),sourceAdapterSha256:v.nullable(),servedPolicyVersion:u().int().nonnegative().nullable(),servedAdapterSha256:v.nullable(),accruedSpendUsd:u().nonnegative(),cleanupAttestationHash:v,resourceCount:u().int().nonnegative(),activeResourceCount:a(0)}).strict().nullable().default(null),evidenceSnapshot:c({id:y,contentHash:v,artifactPath:f().trim().min(1).max(2e3)}).strict(),completedAt:m,contentHash:v}).strict().superRefine((e,t)=>{let n=e.deterministic;n.completedTaskCount>n.attemptedTaskCount&&t.addIssue({code:`custom`,message:`Completed task count cannot exceed attempted task count.`}),n.passedTaskCount+n.failedTaskCount!==n.completedTaskCount&&t.addIssue({code:`custom`,message:`Passed and failed task counts must equal completed task count.`}),n.completedTaskCount===0!=(n.meanScore===null)&&t.addIssue({code:`custom`,message:`Deterministic mean score is available exactly when tasks completed.`}),n.completedTaskCount===0!=(n.passRate===null)&&t.addIssue({code:`custom`,message:`Deterministic pass rate is available exactly when tasks completed.`})}),di=c({advancedEvaluation:or.optional(),benchmarkId:a(`harness-refiner`),model:b,upstreamModel:c({providerId:y,modelId:f().trim().min(1).max(300),revision:f().trim().min(1).max(300),pricing:c({version:f().trim().min(1).max(300),source:f().trim().min(1).max(300),effectiveAt:m,inputUsdPerMillionTokens:u().nonnegative(),cachedInputUsdPerMillionTokens:u().nonnegative(),outputUsdPerMillionTokens:u().nonnegative()}).strict().optional()}).strict(),reasoningEffort:f().trim().min(1).max(100).nullable(),seeds:n(u().int()).min(1).max(100),repetitions:u().int().positive().max(20),maximumSpendUsd:u().nonnegative(),attemptPlan:n(c({stage:d([`baseline`,`adaptation`,`candidate_adaptation`,`candidate`]),split:f().trim().min(1).max(100),taskIds:n(y).min(1).max(1e5),attemptCount:u().int().positive()}).strict()).length(4)}).strict(),fi=c({benchmarkId:a(`model-comparison`),target:c({kind:d([`base_model`,`model_version`,`external_reference`]),label:f().trim().min(1).max(300),modelVersionId:y.nullable(),model:b.nullable()}).strict(),grader:g,judge:c({model:b,release:g,rubricRelease:g,calibrationRelease:g,referenceLabel:f().trim().min(1).max(300)}).strict().nullable(),seeds:n(u().int()).min(1).max(100),repetitions:u().int().positive().max(20),maximumSpendUsd:u().nonnegative(),series:c({id:y,protocol:c({id:y,revision:u().int().positive(),contentHash:v}).strict()}).strict().nullable().default(null),panel:c({id:y,role:d([`correction`,`sibling_verification`,`cumulative_known`,`development`,`retained`,`frozen_final`]),passLabel:y.nullable()}).strict().nullable().default(null),comparisonPair:c({entryId:y,parent:Br,candidateModelVersionId:y}).strict().nullable().default(null),attemptPlan:n(c({stage:a(`comparison`),split:f().trim().min(1).max(100),taskIds:n(y).min(1).max(1e5),attemptCount:u().int().positive()}).strict()).length(1)}).strict(),pi=r([di,fi]),mi=c({schemaVersion:a(`openpond.modelRun.v1`),id:y,modelId:y,modelVersionId:y.nullable(),profileId:y,kind:d([`rollout_smoke`,`training`,`evaluation`]),status:d([`prepared`,`running`,`succeeded`,`failed`,`cancelled`]),method:d([`sft`,`dpo`,`grpo`,`ppo`,`sdft`,`opd`,`opsd`,`sdpo`]).nullable(),destinationId:y.nullable(),taskset:_,comparisonSeriesEntry:Gr.nullable().optional(),evaluationTasksetRef:_.nullable().optional(),harnessRelease:g.nullable().optional(),quote:c({maximumSpendUsd:u().nonnegative(),hourlyCostUsd:u().nonnegative().nullable()}).strict().nullable(),evaluation:pi.nullable().default(null),evaluationProgress:c({stage:d([`baseline`,`adaptation`,`refiner`,`candidate_adaptation`,`candidate`,`comparison`]),completedAttempts:u().int().nonnegative(),totalAttempts:u().int().positive(),accounting:ai.nullable().default(null),advancedAccounting:hr.optional(),evidenceSnapshot:oi.nullable().default(null)}).strict().nullable().default(null),reward:c({raw:u(),components:o(f(),u())}).strict().nullable(),receipt:r([$r,si,ci,ui,dr]).nullable(),adapterArtifactLineageId:y.nullable(),failure:f().trim().min(1).max(5e3).nullable(),startedAt:m,completedAt:m.nullable(),updatedAt:m}).strict().superRefine((e,t)=>{e.status===`succeeded`&&!e.receipt&&t.addIssue({code:`custom`,message:`A successful Model Run requires a canonical receipt.`}),e.kind===`evaluation`&&(e.method!==null||e.destinationId!==null||!e.evaluation||e.adapterArtifactLineageId!==null)&&t.addIssue({code:`custom`,message:`An evaluation Model Run requires evaluation configuration and cannot claim training method, destination, or adapter lineage.`}),e.kind===`evaluation`&&e.receipt&&e.receipt.schemaVersion!==`openpond.modelEvaluationReceipt.v1`&&e.receipt.schemaVersion!==`openpond.modelEvaluationStopReceipt.v1`&&e.receipt.schemaVersion!==`openpond.modelComparisonBenchmarkReceipt.v1`&&e.receipt.schemaVersion!==`openpond.advancedRefinerReviewReceipt.v1`&&t.addIssue({code:`custom`,message:`An evaluation Model Run requires an evaluation receipt.`}),e.kind!==`evaluation`&&(e.method===null||e.destinationId===null||e.quote===null)&&t.addIssue({code:`custom`,message:`A training or rollout Model Run requires method, destination, and quote.`}),e.kind===`rollout_smoke`&&e.adapterArtifactLineageId!==null&&t.addIssue({code:`custom`,message:`A rollout smoke cannot claim adapter artifacts.`})});function hi(e){let t=new Map;for(let n of e.attempts){let e=`${n.phase}\u0000${n.taskId}`,r=t.get(e);(!r||n.startedAt>r.startedAt||n.startedAt===r.startedAt&&n.attemptId>r.attemptId)&&t.set(e,n)}let n=[...t.values()],r=[[`adaptation`,`adaptation`,`candidate_adaptation`],[`held_out`,`baseline`,`candidate`]].flatMap(([e,t,r])=>{let i=new Map(n.filter(e=>e.phase===r).map(e=>[e.taskId,e]));return n.filter(e=>e.phase===t).flatMap(t=>{let n=i.get(t.taskId);return n?[{cohort:e,taskId:t.taskId,baseline:t,refined:n,tokenDelta:n.totalTokens-t.totalTokens}]:[]})}),i=new Set(n.filter(e=>e.phase===`baseline`||e.phase===`adaptation`).map(e=>e.taskId)).size,a=e.targetTaskCount??i;if(!Number.isInteger(a)||a<0)throw Error(`Benchmark task-efficiency target must be a non-negative integer.`);let o={adaptation:new Set(n.filter(e=>e.phase===`adaptation`).map(e=>e.taskId)).size,held_out:new Set(n.filter(e=>e.phase===`baseline`).map(e=>e.taskId)).size},s=(e,t)=>({targetTaskCount:t,comparedTaskCount:e.length,passedTaskCount:e.filter(e=>e.tokenDelta<0).length,failedTaskCount:e.filter(e=>e.tokenDelta>=0).length,lowerTaskCount:e.filter(e=>e.tokenDelta<0).length,higherTaskCount:e.filter(e=>e.tokenDelta>0).length,unchangedTaskCount:e.filter(e=>e.tokenDelta===0).length}),ee=s(r,a),c=r.reduce((e,t)=>e+t.baseline.totalTokens,0),te=r.reduce((e,t)=>e+t.refined.totalTokens,0),l=te-c,u=r.length===a;return{summary:ri.parse({target:`all_tasks_lower`,...ee,baselineTokens:c,refinedTokens:te,tokenDelta:l,tokenDeltaPercent:c>0?l/c*100:null,complete:u,passed:u&&ee.passedTaskCount===a,cohorts:{adaptation:s(r.filter(e=>e.cohort===`adaptation`),o.adaptation),heldOut:s(r.filter(e=>e.cohort===`held_out`),o.held_out)}}),pairs:r}}var gi=d([`measuring`,`up_to_date`,`needs_attention`]),_i=d([`fixed`,`retained`,`regressed`,`unresolved`,`unavailable`]),vi=c({evaluationRunId:y,attemptKey:y,artifactPath:f().trim().min(1).max(2e3),jsonPointer:f().trim().min(1).max(1e3),transcriptHash:v.nullable(),traceHash:v.nullable()}).strict(),yi=c({level:a(.95),lower:u().finite(),upper:u().finite()}).strict(),bi=c({panelId:y,panelRole:xr,passLabel:y.nullable(),taskset:_,attempted:u().int().nonnegative(),available:u().int().nonnegative(),passed:u().int().nonnegative(),strictSuccess:u().min(0).max(1).nullable(),strictSuccessCi95:yi.nullable(),judgeScore:u().min(0).max(100).nullable(),judgeScoreCi95:yi.nullable()}).strict(),xi=c({schemaVersion:a(`openpond.modelCurrencySnapshot.v1`),id:y,contentHash:v,seriesId:y,protocol:c({id:y,revision:u().int().positive(),contentHash:v}).strict(),entryId:y,passLabel:y,parent:i(`kind`,[c({kind:a(`base_model`),id:y,revision:f().trim().min(1).max(500)}).strict(),c({kind:a(`model_version`),id:y,contentHash:v}).strict()]),candidate:c({kind:a(`model_version`),id:y,contentHash:v}).strict(),sources:c({evaluationRunIds:n(y).max(1e5),attemptIds:n(y).max(1e6),tasksets:n(_).max(1e4),grader:g,judge:g.nullable(),calibration:g.nullable()}).strict(),matches:n(c({key:y,panelId:y,panelRole:xr,taskId:y,issueFamilyId:y,seed:u().int(),repetition:u().int().nonnegative(),classification:_i,parentPassed:l().nullable(),candidatePassed:l().nullable(),parentAttempt:vi.nullable(),candidateAttempt:vi.nullable()}).strict()).max(1e6),taskIds:c({fixed:n(y).max(1e6),retained:n(y).max(1e6),regressed:n(y).max(1e6),unresolved:n(y).max(1e6),unavailable:n(y).max(1e6)}).strict(),panels:n(bi).max(1e4),metrics:c({knownIssueCoverage:u().min(0).max(1).nullable(),issueFamilyGeneralization:u().min(0).max(1).nullable(),behavioralRetention:u().min(0).max(1).nullable(),currentAcquisitionDelta:u().finite().nullable(),retainedDeltaPoints:u().finite().nullable(),frontierStrictDelta:u().finite().nullable(),pairedPValue:u().min(0).max(1).nullable(),wins:u().int().nonnegative(),ties:u().int().nonnegative(),losses:u().int().nonnegative()}).strict(),statistics:c({matchedAttemptCount:u().int().nonnegative(),strictDelta:u().finite().nullable(),strictDeltaCi95:yi.nullable(),exactPairedBinaryPValue:u().min(0).max(1).nullable(),pairedBootstrapSamples:u().int().positive()}).strict(),criteria:c({allRequiredAttemptsTerminal:l(),criticalCorrectionPassRate:u().min(0).max(1).nullable(),siblingPassRate:u().min(0).max(1).nullable(),behavioralRetentionRate:u().min(0).max(1).nullable(),retainedRegressionPoints:u().finite().nullable(),criticalPriorRegressionCount:u().int().nonnegative(),thresholdRelease:g}).strict(),efficiency:c({inputTokens:u().int().nonnegative(),outputTokens:u().int().nonnegative(),totalTokens:u().int().nonnegative(),costUsd:u().nonnegative().nullable(),latencyMs:u().int().nonnegative(),tokensPerSuccess:u().nonnegative().nullable(),throughputPerHour:u().nonnegative().nullable(),gpuSeconds:u().nonnegative().nullable()}).strict(),invariants:c({systemPromptHash:v,toolSchema:g,application:g,harness:g,runtime:g,grader:g,autoRefinerEnabled:l()}).strict(),evidenceState:gi,evidenceReasons:n(y).max(100),projectedAt:m}).strict(),Si=i(`kind`,[c({kind:a(`base_model`),id:y,revision:f().trim().min(1).max(500),label:f().trim().min(1).max(300)}).strict(),c({kind:a(`model_version`),id:y,label:f().trim().min(1).max(300)}).strict(),c({kind:a(`captured_model`),id:y,label:f().trim().min(1).max(300)}).strict()]),Ci=r([c({source:a(`evaluation_run`).default(`evaluation_run`),evaluationRunId:y,attemptId:y,modelLabel:f().trim().min(1).max(300).nullable().default(null),modelVersionId:y.nullable().default(null),reward:u().finite().nullable(),components:o(f(),u().finite()).default({})}).strict(),c({source:a(`imported_response`),modelLabel:f().trim().min(1).max(300),modelVersionId:y.nullable().default(null),response:p().refine(e=>e!==void 0,`An imported model response is required.`),reward:u().finite().nullable().default(null),components:o(f(),u().finite()).default({})}).strict()]),wi=r([c({taskId:y,target:Si,attempt:Ci}).strict(),c({taskId:y,target:d([`base`,`current`]).default(`current`),attempt:Ci}).strict(),c({taskId:y,target:d([`base`,`current`]).default(`current`),evaluationRunId:y,attemptId:y,reward:u().finite().nullable(),components:o(f(),u().finite()).default({})}).strict().transform(({taskId:e,target:t,...n})=>({taskId:e,target:t,attempt:{source:`evaluation_run`,modelLabel:null,modelVersionId:null,...n}}))]);c({schemaVersion:a(`openpond.continualLearningDailyBatchManifest.v1`),id:y,seriesId:y,scheduleEntryId:y,dayOrdinal:u().int().positive().max(1e4),sourceTaskset:_,taskIds:n(y).min(1).max(1e3),observedAttempts:n(wi).max(1e3).default([]),sourceFileName:f().trim().min(1).max(500).nullable().default(null),availableAt:m}).strict().superRefine((e,t)=>{new Set(e.taskIds).size!==e.taskIds.length&&t.addIssue({code:`custom`,path:[`taskIds`],message:`A daily batch cannot repeat a task.`}),new Set(e.observedAttempts.map(e=>`${e.taskId}:${Oi(e.target)}`)).size!==e.observedAttempts.length&&t.addIssue({code:`custom`,path:[`observedAttempts`],message:`An intake batch cannot repeat a task and Policy target pair.`});let n=new Set(e.taskIds);e.observedAttempts.some(e=>!n.has(e.taskId))&&t.addIssue({code:`custom`,path:[`observedAttempts`],message:`Observed attempts must belong to tasks in the daily batch.`})});var Ti=c({schemaVersion:a(`openpond.continualLearningDailyBatch.v1`),id:y,seriesId:y,scheduleEntryId:y,dayOrdinal:u().int().positive().max(1e4),label:f().trim().min(1).max(100),runMode:d([`automatic`,`manual`]).optional(),source:d([`json_upload`,`sealed_fixture`]),sourceFileName:f().trim().min(1).max(500).nullable(),sourceTaskset:_,tasks:n(c({taskId:y,taskContentHash:v,familyKey:y,disposition:d([`include`,`defer`,`exclude`]).nullable(),oracleReview:d([`pending`,`confirmed`,`needs_correction`]),note:f().trim().max(2e3),observedAttempt:Ci.nullable(),baselineAttempt:Ci.nullable().default(null),responses:n(c({target:Si,attempt:Ci}).strict()).max(100).default([]),stagedAt:m.nullable().default(null),queuedEntry:Gr.nullable().default(null)}).strict()).min(1).max(1e3),intakeEvaluation:c({status:d([`awaiting`,`running`,`ready`,`failed`]),requestedTargets:n(d([`base`,`current`])).max(2).default([`base`,`current`]),requestedPolicies:n(Si).max(100).default([]),runs:n(c({target:Si,runId:y}).strict()).max(100).default([]),baselineRunId:y.nullable(),currentRunId:y.nullable(),currentPolicy:i(`kind`,[c({kind:a(`base_model`),id:y}).strict(),c({kind:a(`model_version`),id:y}).strict()]).nullable(),failure:f().trim().min(1).max(2e3).nullable()}).strict().default({status:`awaiting`,requestedTargets:[`base`,`current`],requestedPolicies:[],runs:[],baselineRunId:null,currentRunId:null,currentPolicy:null,failure:null}),status:d([`pending`,`reviewed`,`queued`]),queuedEntry:Gr.nullable(),reviewedBy:y.nullable(),reviewedAt:m.nullable(),availableAt:m,revision:u().int().positive(),createdAt:m,updatedAt:m}).strict().superRefine((e,t)=>{new Set(e.tasks.map(e=>e.taskId)).size!==e.tasks.length&&t.addIssue({code:`custom`,path:[`tasks`],message:`A daily batch cannot repeat a task.`});for(let[n,r]of e.tasks.entries()){let e=r.responses.map(e=>Di(e.target));new Set(e).size!==e.length&&t.addIssue({code:`custom`,path:[`tasks`,n,`responses`],message:`A task cannot repeat a model response target.`})}let n=e.tasks.every(e=>e.disposition!==null&&e.oracleReview!==`pending`&&(e.oracleReview!==`needs_correction`||e.note.length>0));e.status!==`pending`&&!n&&t.addIssue({code:`custom`,path:[`tasks`],message:`Every task requires a decision and oracle review before the day completes.`}),e.status===`queued`!=(e.queuedEntry!==null)&&t.addIssue({code:`custom`,path:[`queuedEntry`],message:`Only a queued day carries an exact series-entry reference.`}),e.status===`pending`&&(e.reviewedAt!==null||e.reviewedBy!==null)&&t.addIssue({code:`custom`,path:[`reviewedAt`],message:`A pending day cannot carry reviewer completion identity.`}),e.status!==`pending`&&(e.reviewedAt===null||e.reviewedBy===null)&&t.addIssue({code:`custom`,path:[`reviewedAt`],message:`Reviewed identity is required after a day review completes.`})}),Ei=c({schemaVersion:a(`openpond.continualBenchIssueReview.v1`),id:y,seriesId:y,scheduleEntryId:y,passLabel:y,packet:Cr,splitManifest:c({id:y,contentHash:v}).strict(),decisions:n(c({taskId:y,disposition:d([`include`,`defer`,`exclude`]).nullable(),note:f().trim().max(2e3)}).strict()).min(2).max(1e4),status:d([`pending`,`reviewed`,`queued`]),queuedEntry:Gr.nullable(),reviewedBy:y.nullable(),reviewedAt:m.nullable(),revision:u().int().positive(),createdAt:m,updatedAt:m}).strict().superRefine((e,t)=>{let n=[...e.packet.cases.map(e=>e.taskId)].sort(),r=[...e.decisions.map(e=>e.taskId)].sort();JSON.stringify(n)!==JSON.stringify(r)&&t.addIssue({code:`custom`,path:[`decisions`],message:`Review decisions must cover the exact issue-packet cases.`});let i=e.decisions.every(e=>e.disposition!==null&&e.note.length>0);e.status!==`pending`&&!i&&t.addIssue({code:`custom`,path:[`decisions`],message:`Every case requires a disposition and note before review completes.`}),e.status===`queued`!=(e.queuedEntry!==null)&&t.addIssue({code:`custom`,path:[`queuedEntry`],message:`Only a queued review carries an exact series-entry reference.`})}).readonly();function Di(e){return`${e.kind}:${e.id}`}function Oi(e){return typeof e==`string`?e:Di(e)}var q=f().trim().min(1).max(240),J=f().trim().min(8).max(256),ki=d([`grpo`,`dapo`,`gspo-token`]),Ai=c({id:q,revision:f().trim().min(1).max(256),tokenizerRevision:f().trim().min(1).max(256),chatTemplateHash:J}),ji=c({maxRollouts:u().int().positive().max(1e6),maxEnvironmentExecutions:u().int().positive().max(1e6),maxInputTokens:u().int().positive(),maxOutputTokens:u().int().positive(),maxOptimizerSteps:u().int().positive().max(1e6),wallTimeMs:u().int().positive(),maximumCostUsd:u().nonnegative().nullable()}),Mi=c({method:a(`grpo`),groupSize:u().int().min(2).max(128),normalization:a(`group_standardized`),advantageEpsilon:u().positive().max(.01).default(1e-8),loss:ki,clipRange:u().positive().max(1).default(.2),iterations:u().int().min(2).max(16).default(2),microbatchSize:u().int().positive().max(128).default(1),gradientAccumulationSteps:u().int().positive().max(128).default(1),adamw:on}),Ni=c({method:a(`ppo`),valueModel:Ai,gamma:u().min(0).max(1),gaeLambda:u().min(0).max(1),policyClip:u().positive().max(1),valueClip:u().positive().max(10),valueLossCoefficient:u().nonnegative().max(10),ppoEpochs:u().int().positive().max(100),minibatchSize:u().int().positive().max(1e5)}),Pi=i(`method`,[Mi,Ni]),Fi=c({rewardModelVersion:c({id:q,contentHash:J}).strict(),qualificationReport:c({id:q,contentHash:J}).strict().nullable().default(null),evaluationReferences:n(c({id:q,contentHash:J}).strict()).max(1e3).default([]),checkpoint:c({id:q,contentHash:J,objectRef:f().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),files:n(c({path:f().trim().min(1).max(512),sizeBytes:u().int().positive(),sha256:J}).strict()).min(4).max(128)}).strict(),runtime:Jr,processorRelease:c({id:q,contentHash:J}).strict(),rewardComposerRelease:c({id:q,contentHash:J}).strict(),executionReceipt:c({id:q,contentHash:J}).strict()}).strict(),Ii=c({schemaVersion:a(`openpond.policyOptimization.v1`),policyModel:Ai,referenceModel:Ai,dataset:c({tasksetId:q,tasksetHash:J,split:a(`train`),selectionStrategy:d([`stable_hash_top_n`,`rft_easy_curriculum_v1`]),selectionSeed:u().int(),maxExamples:u().int().positive().max(1e5)}),sampler:c({temperature:u().min(0).max(2),topP:u().min(0).max(1),maxOutputTokens:u().int().positive().max(32768),maxTurns:u().int().positive().max(1e3),concurrency:u().int().positive().max(1e3)}),environment:c({id:q,version:f().trim().min(1).max(256),toolContractHash:J}),reward:c({graderId:q,graderHash:J,learnedPreference:Fi.nullable().default(null)}),kl:c({coefficient:u().nonnegative().nullable(),referenceConstraint:a(`fixed_reference`)}),budgets:ji,checkpointEverySteps:u().int().positive(),seed:u().int(),evaluationSplit:a(`frozen_eval`),optimizer:Pi}),Y=f().trim().min(1).max(240),X=f().trim().min(1),Z=f().trim().min(8).max(256),Li=o(f(),p()).default({}),Ri=c({schemaVersion:a(`openpond.localModelChatConfiguration.v1`).default(`openpond.localModelChatConfiguration.v1`),profile:d([`efficient`,`full_harness`,`custom`]).default(`efficient`),systemPromptMode:d([`lean`,`full_harness`,`custom`]).default(`lean`),customSystemPrompt:f().max(2e4).nullable().default(null),contextWindowTokens:u().int().min(128).max(32768).default(1024),maxOutputTokens:u().int().min(1).max(512).default(64),temperature:u().min(0).max(2).default(0),repetitionPenalty:u().min(.5).max(2).default(1.1),noRepeatNgramSize:u().int().min(0).max(10).default(3),compaction:d([`off`,`when_needed`]).default(`when_needed`),keepWarmSeconds:u().int().min(0).max(3600).default(300),updatedAt:X.nullable().default(null)}),zi=Ri.parse({}),Bi=c({schemaVersion:a(`openpond.managedAdapterServingProjection.v1`),teamId:Y.nullable().default(null),source:a(`sandbox_managed_rl`),sourceRef:Y,canonicalArtifactId:Y.nullable(),canonicalModelName:f().trim().min(1).max(256).nullable().default(null),canonicalArtifactState:d([`imported_unvalidated`,`evaluating`,`promotable`,`rejected`,`deleted`]).nullable(),canonicalDeploymentId:Y.nullable(),canonicalDeploymentState:d([`requested`,`deploying`,`ready`,`degraded`,`deleting`,`deleted`,`failed`]).nullable(),state:d([`pending`,`imported`,`ready`,`failed`]),customerBindingAllowed:l().default(!1),artifactContentHash:Z.nullable().default(null),baseProfileId:Y.nullable().default(null),publishedAt:X.nullable(),lastSyncedAt:X,lastError:f().trim().min(1).max(5e3).nullable()});function Vi(e){let t=e.managedServing;return!!(t?.source===`sandbox_managed_rl`&&t.canonicalArtifactState===`promotable`&&t.customerBindingAllowed)}function Hi(e){return e.state===`ready`&&e.canonicalArtifactState===`promotable`&&e.canonicalDeploymentState===`ready`&&e.customerBindingAllowed&&!!e.teamId&&!!e.canonicalArtifactId&&!!e.canonicalDeploymentId}function Ui(e){if(e.promotable&&e.frozenEvaluationArtifactId)return{kind:`source_frozen_evaluation`,evaluationArtifactId:e.frozenEvaluationArtifactId,canonicalArtifactId:null,canonicalDeploymentId:null};let t=e.managedServing;return t?.source===`sandbox_managed_rl`&&Hi(t)&&t.canonicalArtifactId&&t.canonicalDeploymentId?{kind:`sandbox_customer_binding`,evaluationArtifactId:null,canonicalArtifactId:t.canonicalArtifactId,canonicalDeploymentId:t.canonicalDeploymentId}:null}function Wi(e){let t=e.managedServing;return t?Hi(t)&&t.canonicalArtifactId&&t.canonicalDeploymentId?{kind:`sandbox_serving`,canonicalArtifactId:t.canonicalArtifactId,canonicalDeploymentId:t.canonicalDeploymentId}:null:{kind:`source_artifact`,canonicalArtifactId:null,canonicalDeploymentId:null}}var Gi=`openpond.crossSystemOperations.v1`,Ki=`1.0.0`,qi=`openpond.crossSystemTools.v1`,Ji=[`search_crm`,`query_billing`,`search_support`,`run_python`],Yi=d([`train`,`validation`,`frozen_eval`]),Xi=d([`easy`,`medium`,`hard`]),Zi=d([`renewal_risk_v2`]);d([`renewal_exposure`,`collections_prioritization`,`invoice_reconciliation`,`sla_escalation`,`contract_billing_mismatch`]),c({seed:u().int().nonnegative(),split:Yi,difficulty:Xi,scenarioProfile:Zi.optional()});var Qi={anyOf:[{type:`string`,minLength:1,maxLength:256},{type:`null`}],default:null},$i={type:`integer`,minimum:1,maximum:50,default:25},ea=[{name:`search_crm`,description:`Search synthetic CRM accounts and contacts. Results are scoped, projected, paginated, and budgeted.`,parameters:{type:`object`,additionalProperties:!1,required:[`query`,`fields`,`cursor`,`limit`],properties:{query:{type:`string`,minLength:1,maxLength:500},fields:{type:`array`,minItems:1,maxItems:12,uniqueItems:!0,items:{type:`string`,enum:[`account_id`,`name`,`aliases`,`contact_ids`,`contract_value_usd_cents`,`renewal_date`,`tier`,`active_contract_id`,`contract_term_months`]}},cursor:Qi,limit:$i}}},{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:Qi,limit:$i}}},{name:`search_support`,description:`Search synthetic support cases for explicit account IDs, severity, and state filters.`,parameters:{type:`object`,additionalProperties:!1,required:[`account_ids`,`severity`,`state`,`cursor`,`limit`],properties:{account_ids:{type:`array`,minItems:1,maxItems:50,uniqueItems:!0,items:{type:`string`,minLength:1,maxLength:100}},severity:{type:`array`,minItems:1,maxItems:4,uniqueItems:!0,items:{type:`string`,enum:[`P1`,`P2`,`P3`,`P4`]}},state:{type:`array`,minItems:1,maxItems:5,uniqueItems:!0,items:{type:`string`,enum:[`new`,`investigating`,`waiting_customer`,`resolved`,`closed`]}},cursor:Qi,limit:$i}}},{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}}}}],ta=_a({schemaVersion:qi,tools:ea}),na=[`Use only the four registered synthetic Cross-System Operations tools. Contract ${ta}.`,`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(` `);function ra(e=`auto`){return[na,`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(e)}. Maximum tool turns: 15.`,`Registered tool signatures: ${ea.map(ia).join(`; `)}`].join(`
-`)}ra();function ia(e){let t=oa(e.parameters),n=oa(t.properties),r=ca(t.required),i=Object.fromEntries(r.map(e=>[e,aa(n[e])]));return`${e.name}(${JSON.stringify(i)})`}function aa(e){let t=oa(e),n=sa(t.anyOf);if(n.length)return n.map(aa).join(`|`);let r=sa(t.enum);if(r.length)return r.join(`|`);if(t.type===`array`)return[aa(t.items)];if(t.type===`object`){let e=oa(t.properties),n=ca(t.required);return Object.fromEntries(n.map(t=>[t,aa(e[t])]))}return typeof t.type==`string`?t.type:`unknown`}function oa(e){return e&&typeof e==`object`&&!Array.isArray(e)?e:{}}function sa(e){return Array.isArray(e)?e:[]}function ca(e){return sa(e).filter(e=>typeof e==`string`)}var la=d(Ji),ua=i(`kind`,[c({kind:a(`model`),turn:u().int().nonnegative(),content:f().max(1e5)}),c({kind:a(`tool_call`),turn:u().int().nonnegative(),callId:f().trim().min(1).max(240),name:la,arguments:o(f(),p())}),c({kind:a(`tool_result`),turn:u().int().nonnegative(),callId:f().trim().min(1).max(240),name:la,ok:l(),result:p(),rows:u().int().nonnegative(),bytes:u().int().nonnegative(),durationMs:u().int().nonnegative(),error:f().trim().min(1).max(2e4).nullable()}),c({kind:a(`final`),turn:u().int().nonnegative(),content:f().max(1e5)})]),da=c({schemaVersion:a(Gi),id:f().trim().min(1).max(240),worldId:f().trim().min(1).max(240),taskId:f().trim().min(1).max(240),toolContractHash:a(ta),modelRef:c({providerId:f().trim().min(1),modelId:f().trim().min(1)}).nullable(),status:d([`completed`,`budget_exhausted`,`cancelled`,`infrastructure_failure`]),steps:n(ua).max(1e3),startedAt:f().trim().min(1),completedAt:f().trim().min(1),infrastructureError:f().trim().min(1).max(2e4).nullable(),metadata:o(f(),p()).default({})}),fa=c({schemaVersion:a(Gi),trajectoryId:f().trim().min(1).max(240),outcome:d([`correct`,`incorrect`,`parse_failure`,`budget_exhausted`,`tool_schema_violation`,`infrastructure_failure`,`cancelled`]),reward:u().min(0).max(1.15).nullable(),rewardEligible:l(),exactAnswer:l(),components:c({exactAnswer:u().min(0).max(1),efficiency:u().min(0).max(.1),conciseOutput:u().min(0).max(.05)}),metrics:c({toolCalls:u().int().nonnegative(),rowsRead:u().int().nonnegative(),bytesRead:u().int().nonnegative(),consecutiveRetrievalCalls:u().int().nonnegative(),wallTimeMs:u().int().nonnegative(),parseFailures:u().int().nonnegative(),budgetExhausted:l()}),parsedAnswer:p().nullable(),feedback:n(f().trim().min(1).max(5e3)).max(100)}),pa=c({id:f().trim().min(1).max(240),type:a(`function`),function:c({name:la,arguments:f().max(2e4)})}),ma=c({role:d([`system`,`user`,`assistant`,`tool`]),content:f().max(2e5).nullable(),tool_call_id:f().trim().min(1).max(240).optional(),tool_calls:n(pa).max(4).optional()});c({schemaVersion:a(Gi),id:f().trim().min(1).max(240),taskId:f().trim().min(1).max(240),worldId:f().trim().min(1).max(240),trajectoryId:f().trim().min(1).max(240),toolContractHash:a(ta),approval:c({status:a(`approved`),approvedBy:f().trim().min(1).max(240),approvedAt:f().trim().min(1)}),messages:n(ma).min(3).max(1e3)});var ha=c({tasksetTaskId:f().trim().min(1).max(240),environmentTaskId:f().trim().min(1).max(240),family:d([`renewal_exposure`,`collections_prioritization`,`invoice_reconciliation`,`sla_escalation`,`contract_billing_mismatch`]),prompt:f().trim().min(1).max(1e5),finalAnswer:f().trim().min(1).max(2e5),trajectoryId:f().trim().min(1).max(240),trajectoryHash:f().trim().min(8).max(256),toolNames:n(la).min(1).max(100),toolCallCount:u().int().positive(),messageCount:u().int().min(3),reward:u().min(0).max(1.15),messages:n(ma).min(3).max(1e3)}),ga=c({status:a(`approved`),approvedBy:f().trim().min(1).max(240),approvedAt:f().trim().min(1),previewHash:f().trim().min(8).max(256),trajectoryCount:u().int().positive()});c({schemaVersion:a(`openpond.crossSystemExpertBootstrapPreview.v1`),tasksetId:f().trim().min(1).max(240),tasksetHash:f().trim().min(8).max(256),tasksetRevision:u().int().positive(),previewHash:f().trim().min(8).max(256),toolContractHash:a(ta),status:d([`ready_for_review`,`approved`]),approval:ga.nullable(),tasks:n(ha).min(1).max(100)});function _a(e){let t=va(e),n=2166136261,r=2654435769;for(let e=0;e<t.length;e+=1){let i=t.charCodeAt(e);n=Math.imul(n^i,16777619),r=Math.imul(r^i,2246822507)}return`${qi}:${ya(n)}${ya(r)}`}function va(e){return Array.isArray(e)?`[${e.map(va).join(`,`)}]`:e&&typeof e==`object`?`{${Object.entries(e).sort(([e],[t])=>e.localeCompare(t)).map(([e,t])=>`${JSON.stringify(e)}:${va(t)}`).join(`,`)}}`:JSON.stringify(e)}function ya(e){return(e>>>0).toString(16).padStart(8,`0`)}var ba=c({schemaVersion:a(`openpond.singleTurnPolicyTrajectory.v1`),id:Y,taskId:Y,status:d([`completed`,`infrastructure_failure`]),promptHash:Z,responseText:f().max(2e6),infrastructureError:f().trim().min(1).max(1e4).nullable(),startedAt:X,completedAt:X,metadata:Li}),xa=c({schemaVersion:a(`openpond.ppoTrajectory.v1`),id:Y,taskId:Y,policyModelId:Y,referenceModelId:Y,valueModelId:Y,steps:n(c({index:u().int().nonnegative(),observationHash:Z,actionTokenIds:n(u().int().nonnegative()).min(1),terminated:l(),truncated:l(),reward:u(),policyLogProbability:u(),referenceLogProbability:u(),valuePrediction:u(),return:u(),advantage:u(),mask:u().min(0).max(1)})).min(1).max(1e5),createdAt:X}),Sa=c({schemaVersion:a(`openpond.exactAnswerVerifierResult.v1`),outcome:d([`correct`,`incorrect`,`parse_failure`,`infrastructure_failure`]),graderSetHash:Z,score:u().min(0).max(1).nullable(),passed:l(),rewardEligible:l(),expectedAnswerHash:Z,extractedAnswer:f().max(2e4).nullable(),feedback:n(f().trim().min(1).max(2e4)).max(1e3)}),Ca=c({schemaVersion:a(`openpond.rolloutTrajectoryReceipt.v1`),id:Y,jobId:Y,planId:Y,tasksetId:Y,tasksetHash:Z,taskId:Y,split:a(`train`),correlationId:Y,provider:Y,providerTrace:c({invocationId:Y,experimentId:Y,rolloutId:Y,runId:Y,rowId:Y}),optimizerMethod:d([`grpo`,`ppo`]).default(`grpo`),evidenceLevels:c({requested:d([`trajectory`,`aggregate`,`provider_reported`]),observed:d([`trajectory`,`aggregate`,`provider_reported`]),providerReported:d([`trajectory`,`aggregate`,`provider_reported`])}).default({requested:`trajectory`,observed:`trajectory`,providerReported:`provider_reported`}),budgetUsage:c({rollouts:u().int().nonnegative(),environmentExecutions:u().int().nonnegative(),inputTokens:u().int().nonnegative(),outputTokens:u().int().nonnegative(),optimizerSteps:u().int().nonnegative(),costUsd:u().nonnegative()}).default({rollouts:0,environmentExecutions:0,inputTokens:0,outputTokens:0,optimizerSteps:0,costUsd:0}),environment:c({id:Y,version:f().trim().min(1).max(256),worldId:Y,worldHash:Z,toolContractHash:Z}),policy:c({modelId:Y,checkpointId:Y.nullable(),completionParametersHash:Z}),status:d([`received`,`running`,`succeeded`,`failed`]),failureClass:d([`policy_failure`,`parse_failure`,`tool_schema_violation`,`budget_exhausted`,`cancelled`,`environment_failure`,`infrastructure_failure`]).nullable(),reward:c({eligible:l(),raw:u().min(0).max(1.15).nullable(),normalized:u().min(0).max(1).nullable(),components:o(f(),u())}),trajectory:r([da,ba,xa]).nullable(),verifier:r([fa,Sa]).nullable(),providerStatus:o(f(),p()).default({}),receivedAt:X,startedAt:X.nullable(),completedAt:X.nullable(),updatedAt:X}),wa=`dataset-exact-answer`,Ta=`dataset-exact-answer-v1`,Ea=`no-tools-v1`,Q=d([`sft`,`dpo`,`grpo`,`ppo`,`sdft`,`opd`,`opsd`,`sdpo`]),Da=d([`lora`,`full`]),$=d([`openpond_managed`]),Oa=F(()=>({schemaVersion:a(`openpond.sftRecipe.v1`),method:a(`sft`),parameterization:a(`lora`),baseModel:c({id:Y,revision:f().trim().min(1).max(256),tokenizerRevision:f().trim().min(1).max(256),chatTemplateHash:Z}),dataset:c({trainSplit:a(`train`),validationSplit:d([`validation`,`frozen_eval`]),completionOnly:l(),maxSequenceLength:u().int().positive().max(32768),maxExamples:u().int().positive().max(1e5).default(1e3),selectionStrategy:a(`stable_hash_top_n`).default(`stable_hash_top_n`),selectionSeed:u().int().default(17)}),lora:c({rank:u().int().positive().max(256),alpha:u().positive().max(1024),dropout:u().min(0).max(1),targetModules:n(Y).min(1).max(100)}),optimizer:c({learningRate:u().positive(),epochs:u().positive().max(100),maxSteps:u().int().positive().max(1e6),batchSize:u().int().positive().max(1e4),gradientAccumulationSteps:u().int().positive().max(1e4),seed:u().int()}),resourceLimits:c({cpuThreads:u().int().positive().max(256),memoryBytes:u().int().positive(),wallTimeMs:u().int().positive()})}));F(()=>({id:Y,input:o(f(),p()),expectedOutput:o(f(),p()),tags:n(Y).max(100).default([])})),F(()=>({id:Y,prompt:f().min(1).max(5e5),chosen:f().min(1).max(5e5),rejected:f().min(1).max(5e5),sourceRefs:n(Y).min(1).max(1e4)})),F(()=>({id:Y,input:o(f(),p()),tags:n(Y).max(100).default([])})),F(()=>({schemaVersion:a(`openpond.rewardModelRecipe.v1`),method:a(`reward_model`),parameterization:a(`lora_with_scalar_head`),runScope:d([`synthetic_smoke`,`human_preference`]),baseModel:c({id:Y,revision:f().trim().min(1).max(256),tokenizerRevision:f().trim().min(1).max(256),processorRevision:f().trim().min(1).max(256),chatTemplateHash:Z}).strict(),tasksetRelease:g,preferenceDatasetRelease:g,processorRelease:g,input:c({kind:a(`structured_text`),serialization:d([`scenario_input_and_candidate_json_v1`,`visible_agent_trajectory_v1`]),maxCharacters:u().int().positive().max(5e5)}).strict().default({kind:`structured_text`,serialization:`scenario_input_and_candidate_json_v1`,maxCharacters:32e3}),lora:c({rank:u().int().positive().max(256),alpha:u().positive().max(1024),dropout:u().min(0).max(1),targetModules:n(Y).min(1).max(100)}).strict(),heads:c({scalar:a(`pooled_hidden_state_linear`),bucket:d([`none`,`three_class`]).default(`three_class`)}).strict(),loss:c({ranking:a(`bradley_terry`),rankingWeight:u().positive(),bucketWeight:u().nonnegative(),tieWeight:u().nonnegative()}).strict(),optimizer:c({learningRate:u().positive(),maxSteps:u().int().positive().max(1e5),batchSize:u().int().positive().max(1e4),gradientAccumulationSteps:u().int().positive().max(1e4),seed:u().int(),checkpointEverySteps:u().int().positive().max(1e5)}).strict(),resourceLimits:c({wallTimeMs:u().int().positive().max(864e5),maxExamples:u().int().positive().max(1e5),maxInputCharacters:u().int().positive().max(5e5).default(32e3),maxImagePixels:u().int().positive().max(1e8).optional(),maximumSpendUsd:u().nonnegative().max(1e5)}).strict()})).superRefine((e,t)=>{e.runScope===`synthetic_smoke`&&(e.optimizer.maxSteps>100||e.resourceLimits.maximumSpendUsd>10)&&t.addIssue({code:`custom`,path:[`resourceLimits`],message:`Synthetic-smoke Reward Model runs are limited to 100 steps and USD 10.`})});var ka=F(()=>({schemaVersion:a(`openpond.rftRecipe.v1`),method:a(`grpo`),parameterization:a(`lora`),baseModel:c({id:Y,revision:f().trim().min(1).max(256),tokenizerRevision:f().trim().min(1).max(256),chatTemplateHash:Z}),dataset:c({trainSplit:a(`train`),validationSplit:d([`validation`,`frozen_eval`]),maxPromptTokens:u().int().positive().max(32768),maxExamples:u().int().positive().max(1e5).default(1e3),selectionStrategy:d([`stable_hash_top_n`,`rft_easy_curriculum_v1`]).default(`stable_hash_top_n`)}),lora:c({rank:u().int().positive().max(256)}),rollout:c({groupSize:u().int().min(2).max(64),concurrency:u().int().positive().max(32),maxTurns:u().int().positive().max(100),maxOutputTokens:u().int().positive().max(8192),temperature:u().min(0).max(2),topP:u().positive().max(1),seed:u().int()}),optimizer:c({learningRate:u().positive(),maxSteps:u().int().positive().max(1e5),clipRange:u().positive().max(1).default(.2),iterations:u().int().min(2).max(16).default(2),microbatchSize:u().int().positive().max(64).default(1),gradientAccumulationSteps:u().int().positive().max(64).default(1),advantageEpsilon:u().positive().max(.01).default(1e-8),adamw:on}),loss:c({method:ki.default(`grpo`),klBeta:u().min(0).nullable().default(null)}).default({method:`grpo`,klBeta:null}),reward:c({graderId:Y,graderHash:Z,environmentId:Y,environmentVersion:f().trim().min(1).max(256),toolContractHash:Z,learnedPreference:Fi.nullable().optional()}),resourceLimits:c({wallTimeMs:u().int().positive().max(864e5),maxGpuSeconds:u().int().positive().max(21600).optional(),warmupDeadlineSeconds:u().int().positive().max(3600).optional(),maxRollouts:u().int().positive().max(1e5),maxPayloadBytes:u().int().positive().max(10485760)}),policyOptimization:Ii.nullable().default(null),continuation:c({schemaVersion:a(`openpond.crossJobContinuationRequest.v1`),parentArtifact:g,sourceArtifact:c({jobId:Y,artifactId:Y,checkpointId:Y,contentHash:Z}).strict(),optimizerMode:d([`continue`,`reset`]),adapterMode:d([`residual_append`,`shared`,`sequential_merge`]).optional()}).strict().optional()})),Aa=F(()=>({schemaVersion:a(`openpond.dpoRecipe.v1`),method:a(`dpo`),parameterization:a(`lora`),policyModel:Ai,referenceModel:Ai,dataset:c({trainSplit:a(`train`),validationSplit:d([`validation`,`frozen_eval`]),maxPairs:u().int().positive().max(1e5),maxPromptTokens:u().int().positive().max(32768),maxCompletionTokens:u().int().positive().max(32768),selectionStrategy:a(`stable_hash_top_n`),selectionSeed:u().int()}),lora:c({rank:u().int().positive().max(256),alpha:u().positive().max(1024),dropout:u().min(0).max(1),targetModules:n(Y).min(1).max(100)}),loss:c({variant:a(`sigmoid`),beta:u().positive().max(10),labelSmoothing:u().min(0).max(.5).default(0)}),optimizer:c({learningRate:u().positive(),epochs:u().positive().max(100),maxSteps:u().int().positive().max(1e6),batchSize:u().int().positive().max(1e4),gradientAccumulationSteps:u().int().positive().max(1e4),seed:u().int()}),referenceLogprobs:c({cacheSchemaVersion:a(`openpond.dpoReferenceLogprobs.v1`),cacheKey:Z,invalidationHash:Z}),resourceLimits:c({cpuThreads:u().int().positive().max(256),memoryBytes:u().int().positive(),wallTimeMs:u().int().positive()})})),ja=F(()=>({schemaVersion:a(`openpond.ppoRecipe.v1`),method:a(`ppo`),parameterization:a(`lora`),policyOptimization:Ii.extend({optimizer:Ni}),lora:c({rank:u().int().positive().max(256),alpha:u().positive().max(1024),dropout:u().min(0).max(1),targetModules:n(Y).min(1).max(100)}),valueHead:c({initialization:a(`policy_hidden_state_linear`),optimizerLearningRate:u().positive(),artifactName:a(`value_head.safetensors`)}),policyLearningRate:u().positive(),resume:c({checkpointId:Y.nullable(),policyHash:Z,referenceHash:Z,valueModelHash:Z,optimizerStateHash:Z.nullable()}),resourceLimits:c({cpuThreads:u().int().positive().max(256),memoryBytes:u().int().positive(),wallTimeMs:u().int().positive()})})),Ma=F(()=>({schemaVersion:a(`openpond.unsupportedRecipe.v1`),method:Q.exclude([`sft`,`dpo`,`grpo`,`ppo`]),parameterization:Da,unsupportedReason:f().trim().min(1).max(5e3)})),Na=r([Oa,Aa,ka,ja,Ma]),Pa=F(()=>({schemaVersion:a(`openpond.trainingDestinationCapabilities.v1`),destinationId:$,available:l(),methods:n(Q),parameterizations:n(Da),modelAllowlist:n(Y).default([]),maxDatasetBytes:u().int().nonnegative().nullable(),environmentPlacements:n(d([`none`,`local`,`remote`,`colocated`,`provider_native`])),nonProduction:l(),unavailableReason:f().trim().min(1).max(5e3).nullable(),checkedAt:X})),Fa=F(()=>({destinationId:$,available:l(),methods:n(Q),parameterizations:n(Da),nonProduction:l(),unavailableReason:f().trim().min(1).max(5e3).nullable()})),Ia=F(()=>({schemaVersion:a(`openpond.baseModelCandidate.v1`),selectionKey:Y,label:f().trim().min(1).max(500),sourceLabel:f().trim().min(1).max(500),preference:Sn,available:l(),nonProduction:l(),unavailableReason:f().trim().min(1).max(5e3).nullable(),methods:n(Q),executionOptions:n(Fa).min(1)})),La=F(()=>({code:Y,severity:d([`warning`,`error`]),path:f().trim().max(2e3).nullable(),message:f().trim().min(1).max(5e3)})),Ra=F(()=>({schemaVersion:a(`openpond.trainingCompatibility.v1`),compatible:l(),destinationId:$,tasksetId:Y,recipeMethod:Q,issues:n(La),checkedAt:X})),za=r([Le,d([`destination_unavailable`,`destination_method_unsupported`,`destination_model_unsupported`,`destination_parameterization_unsupported`,`experimental_destination`])]);F(()=>({method:Q,state:d([`recommended`,`compatible`,`needs_dataset_work`,`destination_unavailable`,`experimental_destination`]),reasonCodes:n(za).default([]),reasons:n(f().trim().min(1).max(5e3)).default([]),destinationId:$.nullable().default(null)})),d([`small`,`standard`,`custom`,`small_experiment`]);var Ba=Ae,Va=F(()=>({schemaVersion:a(`openpond.trainingPlan.v1`),id:Y,modelId:Y,tasksetId:Y,tasksetHash:Z,comparisonSeriesEntry:Gr.nullable().optional(),evaluationTasksetRef:_.nullable().optional(),harnessRelease:g.nullable().optional(),modelImprovementQualification:g.nullable().optional(),destinationId:$,recipe:Na,environmentPlacement:d([`none`,`local`,`remote`,`colocated`,`provider_native`]),compatibility:Ra,dataPolicy:c({exportApproved:l(),approvedSourceIds:n(Y),retentionDays:u().int().nonnegative().nullable(),region:f().trim().min(1).max(200).nullable()}),estimatedCostUsd:u().nonnegative().nullable(),createdAt:X,contentHash:Z})),Ha=F(()=>({path:f().trim().min(1).max(2e3),sha256:Z,sizeBytes:u().int().nonnegative(),role:d([`manifest`,`task_data`,`grader`,`environment`,`recipe`,`policy`,`provenance`])})),Ua=F(()=>({schemaVersion:a(`openpond.trainingBundle.v1`),id:Y,planId:Y,tasksetId:Y,tasksetHash:Z,recipeHash:Z,files:n(Ha).min(1).max(1e6),totalSizeBytes:u().int().nonnegative(),sourceIds:n(Y),excludedSourceIds:n(Y),containsRawChats:a(!1),containsSecrets:a(!1),containsHiddenGraderAssets:a(!1),createdAt:X,contentHash:Z}));F(()=>({schemaVersion:a(`openpond.trainingBundleExport.v1`),manifest:Ua,files:n(c({path:f().trim().min(1).max(2e3),sha256:Z,sizeBytes:u().int().nonnegative(),encoding:a(`base64`),content:f()})).min(1),contentHash:Z})),F(()=>({schemaVersion:a(`openpond.trainingPreparedStart.v1`),plan:Va,bundle:Ua,approvalActor:Y.nullable(),preparedAt:X})),F(()=>({schemaVersion:a(`openpond.trainingApproval.v1`),evaluationTasksetRef:_.nullable().optional(),id:Y,planId:Y,bundleHash:Z,harnessRelease:g.nullable().optional(),modelImprovementQualification:g.nullable().optional(),destinationId:$,modelId:Y,method:Q,parameterization:Da,maximumCostUsd:u().nonnegative().nullable(),approvedBy:Y,approvedAt:X})),F(()=>({schemaVersion:a(`openpond.trainingJobSourceSnapshot.v1`),evaluationTasksetRef:_.nullable().optional(),modelProjectId:Y,sourceProjectRevision:u().int().positive(),profileId:Y,taskset:c({id:Y,revision:u().int().positive(),contentHash:Z}).strict(),tasksetRelease:g,harnessRelease:g,baseModel:Sn,method:Q})).strict();var Wa=d([`queued`,`starting`,`running`,`cancelling`,`cancelled`,`succeeded`,`failed`,`reconciling`]),Ga=F(()=>({schemaVersion:a(`openpond.trainingJob.v1`),id:Y,planId:Y,bundleHash:Z,approvalId:Y,destinationId:$,status:Wa,nonProduction:l(),workerPid:u().int().positive().nullable(),startedAt:X.nullable(),completedAt:X.nullable(),error:f().trim().min(1).max(2e4).nullable(),createdAt:X,updatedAt:X,metadata:Li})),Ka=F(()=>({schemaVersion:a(`openpond.trainingJobEvent.v1`),id:Y,jobId:Y,sequence:u().int().nonnegative(),type:d([`queued`,`start`,`progress`,`metric`,`checkpoint`,`cancel`,`complete`,`failure`,`reconcile`]),timestamp:X,payload:Li})),qa=F(()=>({schemaVersion:a(`openpond.sftStepMetric.v1`),step:u().int().nonnegative(),maxSteps:u().int().positive(),timestamp:X,epoch:u().nonnegative().nullable(),loss:u().nullable(),learningRate:u().nonnegative().nullable(),gradientNorm:u().nonnegative().nullable(),entropy:u().nonnegative().nullable(),meanTokenAccuracy:u().min(0).max(1).nullable(),preferenceAccuracy:u().min(0).max(1).nullable().default(null),preferenceMargin:u().nullable().default(null),chosenReward:u().nullable().default(null),rejectedReward:u().nullable().default(null),chosenLogProbability:u().nullable().default(null),rejectedLogProbability:u().nullable().default(null),reward:u().nullable().default(null),policyLoss:u().nullable().default(null),advantageLoss:u().nullable().default(null),inputTokensSeen:u().int().nonnegative().nullable(),memoryBytes:u().int().nonnegative().nullable(),elapsedSeconds:u().nonnegative().nullable()})),Ja=F(()=>({schemaVersion:a(`openpond.policyOptimizationMetric.v1`),method:d([`grpo`,`ppo`]),step:u().int().nonnegative(),timestamp:X,learningRate:u().nonnegative().nullable().default(null),policyLoss:u().nullable(),valueLoss:u().nullable(),gradientNorm:u().nonnegative().nullable().default(null),meanReward:u().nullable(),meanReturn:u().nullable(),kl:u().nullable(),behaviorPolicyKlPreUpdate:u().nonnegative().nullable().default(null),entropy:u().nullable(),policyClipFraction:u().min(0).max(1).nullable(),behaviorPolicyClipFractionPreUpdate:u().min(0).max(1).nullable().default(null),valueClipFraction:u().min(0).max(1).nullable(),explainedVariance:u().nullable(),rolloutLearnerLag:u().int().nonnegative().nullable(),inputTokens:u().int().nonnegative(),outputTokens:u().int().nonnegative(),environmentExecutions:u().int().nonnegative(),trajectoryCount:u().int().nonnegative().nullable().default(null),costUsd:u().nonnegative().nullable()})),Ya=F(()=>({schemaVersion:a(`openpond.managedTrainingRunEvidence.v2`),provider:Y,providerRunId:Y,state:Y,progress:c({targetOptimizerSteps:u().int().nonnegative(),committedOptimizerSteps:u().int().nonnegative(),skippedOptimizerSteps:u().int().nonnegative()}),reward:c({finalMean:u().nullable(),variance:u().nonnegative().nullable(),minimum:u().nullable(),maximum:u().nullable(),distinctValueCount:u().int().nonnegative(),noSignalGroupCount:u().int().nonnegative(),trajectoryCount:u().int().nonnegative(),eligibleTrajectoryCount:u().int().nonnegative()}),usage:c({inputTokens:u().int().nonnegative(),outputTokens:u().int().nonnegative(),environmentExecutions:u().int().nonnegative()}),resource:c({provider:Y,gpuType:f().trim().min(1).max(300).nullable(),gpuCount:u().int().nonnegative().nullable(),hourlyCostUsd:u().nonnegative().nullable(),durationSeconds:u().nonnegative().nullable(),gpuSeconds:u().nonnegative().nullable()}),cost:c({totalUsd:u().nonnegative().nullable()}),checkpoint:c({id:Y,policyVersion:u().int().nonnegative(),sha256:Z.nullable(),sizeBytes:u().int().nonnegative().nullable()}).nullable(),movement:c({adapterDeltaNorm:u().nonnegative().nullable()}),evaluations:n(c({reference:c({id:Y,contentHash:Z}).nullable().optional(),kind:d([`baseline`,`candidate`]),policyVersion:u().int().nonnegative(),taskCount:u().int().nonnegative().nullable().optional(),targetTaskCount:u().int().nonnegative().nullable().optional(),score:u().nullable(),threshold:u().nullable(),passed:l().nullable()})),canonicalPublication:c({state:Y.nullable(),artifactId:Y.nullable()}),syncedAt:X})),Xa=F(()=>({count:u().int().nonnegative(),scoredCount:u().int().nonnegative(),meanScore:u().min(0).max(1).nullable(),passedCount:u().int().nonnegative(),passRate:u().min(0).max(1).nullable()})),Za=F(()=>({status:d([`scored`,`unavailable`]),score:u().min(0).max(1).nullable(),passed:l(),rewardEligible:l(),failureClass:d([`policy_failure`,`grader_failure`,`environment_failure`,`infrastructure_failure`,`timeout`,`cancelled`]).nullable(),feedback:n(f().trim().min(1).max(2e4)).max(1e3),components:n(c({graderId:Y,score:u().min(0).max(1).nullable(),passed:l(),feedback:f().trim().max(2e4).nullable()})).max(1e3)})),Qa=F(()=>({taskId:Y,input:o(f(),p()),baseOutput:o(f(),p()).nullable(),trainedOutput:o(f(),p()).nullable(),baseGrade:Za.nullable(),trainedGrade:Za.nullable()})),$a=F(()=>({schemaVersion:a(`openpond.trainingEvaluationSummary.v1`),jobId:Y,tasksetId:Y,base:Xa,trained:Xa,meanScoreDelta:u().min(-1).max(1).nullable(),examples:n(Qa).max(1e6)})),eo=F(()=>({schemaVersion:a(`openpond.trainingRunDetail.v1`),job:Ga,events:n(Ka),stepMetrics:n(qa),policyMetrics:n(Ja).default([]),managedEvidence:Ya.nullable().default(null),evaluation:$a.nullable(),generatedAt:X})),to=F(()=>({schemaVersion:a(`openpond.trainingArtifact.v1`),id:Y,jobId:Y,kind:d([`adapter`,`checkpoint`,`metrics`,`log`,`manifest`,`evaluation`]),path:f().trim().min(1).max(4e3),sha256:Z,sizeBytes:u().int().nonnegative(),baseModelId:Y.nullable(),baseModelRevision:f().trim().min(1).max(256).nullable(),tokenizerRevision:f().trim().min(1).max(256).nullable(),chatTemplateHash:Z.nullable(),nonProduction:l(),createdAt:X,metadata:Li})),no=F(()=>({schemaVersion:a(`openpond.modelArtifactLineage.v1`),id:Y,modelId:Y,artifactId:Y,jobId:Y,tasksetId:Y,tasksetHash:Z,graderHash:Z,planHash:Z,bundleHash:Z,recipeHash:Z,workerVersion:f().trim().min(1).max(256),trainerVersion:f().trim().min(1).max(256),importedAt:X,frozenEvaluationArtifactId:Y.nullable(),promotable:l(),pinned:l().default(!1),status:d([`imported`,`rejected`]).default(`imported`),rejectedAt:X.nullable().default(null),rejectionReason:f().trim().min(1).max(5e3).nullable().default(null),chatConfiguration:Ri.default(zi),managedServing:Bi.nullable().default(null)})),ro=d([`chat_manual`,`agent`,`extension`,`authoring_optimizer`]),io=F(()=>({schemaVersion:a(`openpond.modelBinding.v1`),id:Y,profileId:Y,role:ro,roleTargetId:Y,modelArtifactLineageId:Y,tasksetId:Y,evaluationArtifactId:Y.nullable(),status:d([`active`,`rolled_back`]),priorBindingId:Y.nullable(),rollbackTargetBindingId:Y.nullable(),promotedBy:Y,promotedAt:X,rolledBackAt:X.nullable(),metadata:Li}));F(()=>({schemaVersion:a(`openpond.trainingActivity.v1`),profileId:Y,active:l(),activeCounts:c({jobs:u().int().nonnegative(),creations:u().int().nonnegative(),minerRuns:u().int().nonnegative(),datasetImports:u().int().nonnegative()}),revision:Z,generatedAt:X})),F(()=>({schemaVersion:a(`openpond.tasksetOperationalState.v1`),tasksetId:Y,attempts:n(gn),artifacts:n(_n),grades:n(yn),evaluationResults:n(ft).default([]),generatedAt:X}));var ao=F(()=>({schemaVersion:a(`openpond.trainingState.v1`),profileId:Y,sources:n(Be),creations:n(wn),tasksetDrafts:n(We).default([]),tasksets:n(Ke),benchmarkRuns:n(ne()).default([]),benchmarkComparisons:n(ne()).default([]),datasetImports:n(Rn).default([]),datasetArtifacts:n(Fe).default([]),graderAuditReports:n(bn),evaluationResults:n(ne()).default([]),candidates:n(Wn),minerConfig:Gn,minerRuns:n(Kn).default([]),modelProjects:n(Ba).default([]),modelVersions:n(Yr).default([]),modelRuns:n(mi).default([]),comparisonSeries:n(Lr).default([]),comparisonSeriesEntries:n(Kr).default([]),modelCurrencySnapshots:n(xi).default([]),continualBenchIssueReviews:n(Ei).default([]),continualLearningDailyBatches:n(Ti).default([]),rewardModelVersions:n(Xr).default([]),rewardModelRuns:n(Qr).default([]),modelTasksets:n(Ke).default([]),plans:n(Va),bundles:n(Ua),jobs:n(Ga),artifacts:n(to),models:n(no),rolloutReceipts:n(Ca).default([]),modelBindings:n(io).default([]),destinations:n(Pa),baseModelCandidates:n(Ia).default([]),activityRevision:Z.optional(),generatedAt:X}));export{tn as A,Ot as B,Qn as C,xn as D,$n as E,Zt as F,gt as H,Qt as I,Vt as L,Yt as M,Kt as N,nn as O,Xt as P,zt as R,rr as S,ar as T,Ze as U,xt as V,Xe as W,lr as _,io as a,nr as b,eo as c,ta as d,Vi as f,hi as g,mi as h,Ya as i,rn as j,an as k,ao as l,Ui as m,Ta as n,ka as o,Wi as p,Ea as r,Na as s,wa as t,Ki as u,pr as v,Xn as w,er as x,sr as y,wt as z};
+import{C as e,T as t,a as n,b as r,c as i,d as a,g as o,h as s,i as ee,m as c,n as te,o as l,p as u,r as d,s as ne,v as f,x as p}from"./schemas-lO3MoWQr.js";import{a as re,c as m,d as h,i as g,l as _,n as ie,o as v,s as y,t as ae}from"./common-BMjyKlX-.js";import{D as b,O as oe,r as se}from"./settings-D7C-Auyi.js";import{A as ce,D as le,R as ue,U as de,W as fe,_ as pe,c as me,d as x,h as he,i as S,o as C,p as ge,s as w,t as _e,v as ve}from"./chunk-H3JZ3P5T-B0SCoPw4.js";import{A as ye,D as be,E as xe,H as Se,I as Ce,M as we,O as Te,R as Ee,U as De}from"./contracts-BLiqJxBv.js";import{$ as Oe,A as ke,C as Ae,K as je,N as Me,O as Ne,P as Pe,S as Fe,T as Ie,Y as Le,Z as Re,_ as ze,c as Be,f as Ve,g as He,h as Ue,l as We,m as Ge,o as Ke,s as qe,t as Je,w as Ye}from"./chunk-GJRXJL6G-pJp1DyJd.js";var Xe=f().trim().min(1).max(200),Ze=c({sessionId:Xe,turnId:Xe,source:ge,admissionHash:v,modelConfigurationHash:v,traceHash:v,runtimeEventRefs:n(Xe).max(1e4),startedAt:m,completedAt:m,outputHash:v,partialOutputAvailable:l()}).strict().superRefine((e,t)=>{new Set(e.runtimeEventRefs).size!==e.runtimeEventRefs.length&&t.addIssue({code:`custom`,path:[`runtimeEventRefs`],message:`Native event references must be unique.`}),Date.parse(e.completedAt)<Date.parse(e.startedAt)&&t.addIssue({code:`custom`,path:[`completedAt`],message:`Native completion precedes its start.`})});c({schemaVersion:a(`openpond.nativeHarnessExperimentAttempt.v1`),taskId:f().min(1).max(500),status:d([`completed`,`cancelled`,`timed_out`,`policy_failure`]),output:f().max(8388608).nullable(),error:f().max(2e3).nullable(),messages:n(p()).max(4002),collected:l(),environmentCleanupComplete:a(!0),snapshot:ee(),native:Ze}).strict().superRefine((e,t)=>{let n=e.status===`completed`;(n!==e.collected||n!==(e.output!==null)||n!==(e.error===null))&&t.addIssue({code:`custom`,message:`Native status, output, collection and error must agree.`}),n&&(e.native.partialOutputAvailable||e.native.outputHash!==h(e.output))&&t.addIssue({code:`custom`,path:[`native`,`outputHash`],message:`Native output differs from its retained evidence.`})}).extend({contentHash:v}).strict();var Qe=c({schemaVersion:a(`openpond.humanResultView.v1`),executionId:f().min(1),executionHash:f().regex(/^[a-f0-9]{64}$/),selection:n(ue).max(1e4),coverage:c({pending:u().int().nonnegative(),submitted:u().int().nonnegative(),accepted:u().int().nonnegative(),rejected:u().int().nonnegative(),cancelled:u().int().nonnegative()}).strict(),results:n(c({review:ue,decision:ue.nullable(),attemptIds:n(f().min(1)).min(1).max(2),mode:d([`individual`,`pairwise`]),origin:d([`hosted_retained`,`recorded_evidence`,`owner_attested_local`,`native_local`]),status:d([`pending`,`submitted`,`accepted`,`rejected`,`cancelled`]),grader:ue,answers:n(ce).max(100)}).strict()).max(1e4)}).strict().extend({contentHash:f().regex(/^[a-f0-9]{64}$/)}).strict(),$e=f().regex(/^[a-f0-9]{64}$/),et=f().min(1).max(240),tt={feedbackKey:et,criterionId:et},nt=i(`kind`,[c({...tt,kind:a(`normalized_score`),minimum:u().finite(),maximum:u().finite(),passThreshold:u().min(0).max(1).optional()}).strict(),c({...tt,kind:a(`boolean`)}).strict(),c({...tt,kind:a(`category`)}).strict(),c({...tt,kind:a(`preference_score`),tieScore:u().min(0).max(1).nullable()}).strict()]),rt=c({execution:g,resultHash:$e,reviews:n(c({review:ue,decision:ue}).strict()).min(1).max(1e4)}).strict(),it=c({rules:n(nt).min(1).max(100),baseline:rt,candidate:rt}).strict().superRefine((e,t)=>{new Set(e.rules.map(e=>e.feedbackKey)).size!==e.rules.length&&t.addIssue({code:`custom`,message:`Select each Human feedback key once.`})}),at=c({manifest:xe,result:be}).strict(),ot=c({id:et,manifestHash:$e,resultHash:$e}).strict(),st=c({schemaVersion:a(`openpond.humanComparisonProjection.v1`),teamId:et,projectId:et,selections:it,original:c({baseline:ot,candidate:ot}).strict(),humanResults:c({baseline:Qe,candidate:Qe}).strict(),baseline:at,candidate:at}).strict().extend({contentHash:$e}).strict();function ct(e){throw new De(e,409)}function lt(e,t){let n=st.parse(e),{contentHash:r,...i}=n;r!==h(i)&&ct(`human_comparison_projection_invalid`);for(let e of[`baseline`,`candidate`]){let r=ye(n[e]),i=r.manifest.lineage,a=i?.humanProjection,o=n.original[e],s=n.selections[e];(!i||!a||a.originalManifest.id!==o.id||a.originalManifest.contentHash!==o.manifestHash||a.selectionHash!==h(s)||a.rulesHash!==h(n.selections.rules)||h(i.execution)!==h(s.execution)||r.manifest.teamId!==n.teamId||n.humanResults[e].contentHash!==s.resultHash)&&ct(`human_comparison_projection_lineage_changed`);let{contentHash:ee,...c}=n.humanResults[e];if((h(c)!==ee||c.executionId!==s.execution.id||c.executionHash!==s.execution.contentHash||h(c.selection)!==h(s.reviews.map(e=>e.review)))&&ct(`human_comparison_human_result_changed`),t){let n=ye(t[e]);(n.manifest.lineage?.humanProjection||n.manifest.id!==o.id||n.manifest.contentHash!==o.manifestHash||n.result.contentHash!==o.resultHash||h(n.manifest.lineage?.execution)!==h(s.execution))&&ct(`human_comparison_original_changed`)}}return n}var ut=c({adapterId:y,placement:d([`local`,`remote`,`colocated`,`provider_native`]),runtimeVersion:f().trim().min(1).max(200),capabilityReceipt:v}).strict(),dt=c({maxTurns:u().int().positive().max(1e5),timeoutMs:u().int().positive().max(864e5),maxOutputBytes:u().int().positive().max(25e7),maximumSpendUsd:u().nonnegative().nullable()}).strict();c({schemaVersion:a(`openpond.runManifest.v1`),id:y,harnessRelease:g,tasksetRelease:g,model:oe,runtimeTarget:ut,limits:dt,approval:c({id:y,contentHash:v}).strict().nullable(),createdAt:m,metadata:re}).strict().extend({contentHash:v}).strict();var ft=c({schemaVersion:a(`openpond.attemptReceipt.v1`),id:y,runManifest:g,taskId:y,seed:f().trim().min(1).max(500),terminal:l(),failureClass:ae.nullable(),outputHash:v.nullable(),traceHash:v,artifactRefs:n(ie).max(1e5),graderEvidenceRefs:n(ie).max(1e4),startedAt:m,completedAt:m,latencyMs:u().int().nonnegative(),costUsd:u().nonnegative().nullable(),legacyAttemptRef:y.nullable().default(null),metadata:re}).strict().extend({contentHash:v}).strict(),pt=c({schemaVersion:a(`openpond.evaluationResult.v1`),id:y,runManifest:g,harnessRelease:g,tasksetRelease:g,model:oe,receiptRefs:n(g).min(1).max(1e6),attemptCount:u().int().positive(),rewardEligibleCount:u().int().nonnegative(),terminalCount:u().int().nonnegative(),meanScore:u().min(0).max(1).nullable(),failureCounts:o(ae,u().int().nonnegative()),authoredMetric:fe.optional(),metadata:re}).strict().extend({contentHash:v}).strict();c({schemaVersion:a(`openpond.harnessCompatibility.v1`),id:y,baseHarnessRelease:g,candidateHarnessRelease:g,tasksetRelease:g,environmentHash:v,toolContractHash:v,policyHash:v,graderInterfaceHash:v,metadata:re}).strict().extend({contentHash:v}).strict();var mt=i(`kind`,[c({kind:a(`model`),model:oe,configurationHash:v}).strict(),c({kind:a(`fixture`)}).strict()]),ht=i(`kind`,[c({kind:a(`taskset`),environmentRelease:g,verifierSetRelease:g,policyHash:v}).strict(),c({kind:a(`harness`),harnessRelease:g}).strict()]),gt=c({receiptId:y,taskId:y,seed:f().trim().min(1).max(500),fixtureId:y.nullable()}).strict(),_t=c({schemaVersion:a(`openpond.tasksetRunManifest.v1`),id:y,tasksetRelease:g,packageHash:v,execution:ht,profileEvaluation:he.optional(),policy:mt,gradingRole:a(`evaluation`),metricPolicy:de,population:n(gt).min(1).max(1e5),runtimeTarget:ut,limits:dt,createdAt:m,metadata:re}).strict().superRefine((e,t)=>{new Set(e.population.map(e=>e.receiptId)).size!==e.population.length&&t.addIssue({code:`custom`,path:[`population`],message:`Run receipt identities must be unique.`}),e.population.some(t=>t.fixtureId!==null!=(e.policy.kind===`fixture`))&&t.addIssue({code:`custom`,path:[`population`],message:`Only fixture runs require a fixture identity for every member.`}),e.profileEvaluation&&(e.execution.kind!==`harness`||e.execution.harnessRelease.id!==e.profileEvaluation.harnessRelease.id||e.execution.harnessRelease.contentHash!==e.profileEvaluation.harnessRelease.contentHash)&&t.addIssue({code:`custom`,path:[`profileEvaluation`],message:`Profile evaluation must execute its exact bound Harness release.`})}).safeExtend({contentHash:v}),T=f().trim().min(1).max(240),E=f().trim().min(1).max(5e3),vt=d([`no_training`,`prompting`,`retrieval`,`sft`,`preference`,`grpo_rft`,`sdft_opsd`,`sdpo`,`agentic_rl`]),yt=c({id:T,taskIndex:u().int().nonnegative(),label:Ye,output:o(f(),p()),infrastructureError:f().trim().min(1).max(1e4).nullable(),expectedPassed:l(),expectedRewardEligible:l(),metadata:o(f(),p()).default({})}),bt=c({schemaVersion:a(`openpond.capabilityDiagnosis.v1`),summary:f().trim().min(1).max(1e4),stableBehavior:n(E).max(100).default([]),changingKnowledge:n(E).max(100).default([]),requiredContext:n(E).max(100).default([]),requiredTools:n(T).max(100).default([]),intervention:vt,trainingEligible:l(),rationale:n(E).min(1).max(100),confidence:u().min(0).max(1)}),xt=c({id:T,sourceId:T,sourceTurnId:T.nullable(),split:Me,origin:d([`extracted`,`corrected`,`synthetic`,`expert_authored`]),inputPrompt:f().trim().min(1).max(1e5),expectedOutputText:f().trim().min(1).max(2e5).nullable(),rationale:E}),St=c({schemaVersion:a(`openpond.taskDesignProposal.v1`),id:T,name:f().trim().min(1).max(500),objective:f().trim().min(1).max(2e4),diagnosis:bt.default({schemaVersion:`openpond.capabilityDiagnosis.v1`,summary:`Reproduce the selected approved behavior.`,stableBehavior:[],changingKnowledge:[],requiredContext:[],requiredTools:[],intervention:`sft`,trainingEligible:!0,rationale:[`The selected examples were supplied as demonstrations.`],confidence:.5}),taskKind:ke.shape.taskKind,sourceIds:n(T).min(1).max(1e5),assumptions:n(E).max(1e3),successCriteria:n(E).min(1).max(1e3),proposedGraders:n(Ie).max(1e3).default([]),graderFixtures:n(yt).max(1e5).default([]),generatedFiles:n(We).max(1e3).default([]),proposedExamples:n(xt).max(1e5).default([]),proposedMethod:Ge.shape.recommendedMethod,trainingPath:ze.nullable().default(null),policy:Ne,warnings:n(f().trim().min(1)).default([]),createdAt:f().trim().min(1)}),Ct=f().regex(/^[a-f0-9]{64}$/),wt=c({providerId:a(`claude-code`),configurationHash:Ct,executableHash:Ct,version:f().regex(/^2\.\d+\.\d+$/),capabilityHash:Ct,maximumRequestCostUsd:u().finite().positive().max(1e4),requestTimeoutMs:u().int().min(1e3).max(3e5),maximumTurns:u().int().min(1).max(100),holdOpen:l()}).strict();c({id:f().min(1).max(200),receiptId:f().min(1).max(200),sessionId:e(),operationId:f().min(1).max(200),action:d([`message`,`seal`]),text:f().max(262144).optional()}).strict().superRefine((e,t)=>{e.action===`message`&&!e.text?.trim()&&t.addIssue({code:`custom`,message:`A human intervention requires a message.`}),e.action===`seal`&&e.text!==void 0&&t.addIssue({code:`custom`,message:`Sealing does not add a prompt.`})});var Tt=c({provider:a(`claude-code`),sessionId:e(),requestId:f().min(1).max(200),modelId:f().min(1).max(500),runtime:wt,sourceHash:Ct,outputHash:Ct,traceHash:Ct,runtimeEventRefs:n(f().min(1).max(200)).min(1).max(2e4),eventCount:u().int().positive().max(2e4),startedAt:t(),completedAt:t(),cleanupComplete:a(!0)}).strict().superRefine((e,t)=>{(e.runtimeEventRefs.length!==e.eventCount||new Set(e.runtimeEventRefs).size!==e.eventCount)&&t.addIssue({code:`custom`,message:`Process event references must match the exact retained trace population.`})}),Et=c({providerId:a(`codex`),configurationHash:f().regex(/^[a-f0-9]{64}$/),accountHash:f().regex(/^[a-f0-9]{64}$/),reasoningEffort:d([`low`,`medium`,`high`,`xhigh`]),maximumRequestCostUsd:u().finite().positive().max(1e4),requestTimeoutMs:u().int().min(1e3).max(3e5)}).strict(),Dt=c({id:f().trim().min(1).max(240),revision:u().int().positive(),contentHash:f().regex(/^[a-f0-9]{64}$/),targetId:f().trim().min(1).max(240).nullable()}).strict(),Ot=n(c({role:d([`system`,`user`,`assistant`]),content:f().max(1e5)}).strict()).max(100).superRefine((e,t)=>{e.reduce((e,t)=>e+t.content.length,0)>1e5&&t.addIssue({code:`custom`,message:`Evaluation prompt exceeds 100,000 characters.`})}),kt=c({schemaVersion:a(`openpond.experimentAttemptGrade.v1`),selectionHash:f().regex(/^[a-f0-9]{64}$/),grades:n(Se).min(1).max(100),score:u().min(0).max(1).nullable(),passed:l(),rewardEligible:l(),gradingStatus:d([`scored`,`pending`,`unscorable`,`not_configured`]),failureClass:_e.nullable(),contentHash:f().regex(/^[a-f0-9]{64}$/)}).strict().superRefine((e,t)=>{let{contentHash:n,...r}=e;x(r)!==n&&t.addIssue({code:`custom`,message:`Experiment grading integrity failed.`});let i=e.grades.flatMap(e=>e.components);(i.length!==e.grades.length||new Set(i.map(e=>e.graderId)).size!==i.length)&&t.addIssue({code:`custom`,message:`Each selected criterion requires exactly one distinct grader result.`});for(let n of e.grades){let{contentHash:e,...r}=n;x(r)!==e&&t.addIssue({code:`custom`,message:`A retained criterion changed.`});for(let e of n.components){let{contentHash:n,...r}=e;x(r)!==n&&t.addIssue({code:`custom`,message:`A retained grader result changed.`})}}}),D=f().trim().min(1).max(200),O=f().regex(/^[a-f0-9]{64}$/),At=i(`kind`,[c({kind:a(`hosted_chat`),modelId:D,localRuntime:r([c({providerId:a(`custom-openai-compatible`),configurationHash:O,maximumRequestCostUsd:u().finite().positive().max(1e4),requestTimeoutMs:u().int().min(1e3).max(3e5)}).strict(),wt,Et]).optional(),candidate:c({artifactId:D,contentHash:O,preparationId:D,dispatchId:D,baseProfileId:D,workerImageDigest:f().regex(/^sha256:[a-f0-9]{64}$/)}).strict().optional(),harness:ge.optional(),messages:Ot.optional(),maxOutputTokens:u().int().min(1).max(1e6).default(1e5),temperature:u().min(0).max(2).default(0),topP:u().gt(0).max(1).default(1)}).strict(),c({kind:a(`fixture`),fixtureId:D}).strict()]),jt=c({schemaVersion:a(`openpond.modelStarterAttemptRequest.v1`),operationId:D,teamId:D,modelProjectId:D.nullable(),project:Dt.optional(),taskset:Le,taskId:D,environmentSeed:u().int().min(0).max(2147483647).default(0),policy:At}).strict();c({modelProjectId:D.nullable(),taskset:Le,afterTaskId:D.optional(),limit:u().int().min(1).max(100).default(25)}).strict(),c({modelProjectId:D.nullable(),taskset:Le,available:l(),unavailableReason:f().max(1e3).nullable(),models:n(c({id:D,name:f().max(500),maximumOutputTokens:u().int().min(1).max(1e6).optional()}).strict()).max(200),tasks:n(c({id:D,split:le,inputPreview:f().max(500),fixtures:n(c({id:D,label:f().max(200)}).strict()).max(1e3)}).strict()).max(100),nextCursor:D.nullable()}).strict();var Mt=c({schemaVersion:a(`openpond.modelStarterAttemptSummary.v1`),id:D,revision:u().int().positive(),request:jt,status:d([`queued`,`running`,`cancelling`,`completed`,`failed`,`cancelled`]),policySnapshot:c({modelId:D,provider:D,upstreamModelId:f().min(1).max(500),configurationHash:O}).strict().nullable(),createdAt:t(),startedAt:t().nullable(),completedAt:t().nullable(),cleanupComplete:l(),resultAvailable:l(),score:u().min(0).max(1).nullable(),gradingStatus:d([`not_started`,`pending`,`scored`,`unscorable`,`not_configured`]).default(`not_started`),passed:l().nullable(),outputPreview:f().max(500).nullable(),error:c({code:D,message:f().max(2e3)}).strict().nullable()}).strict().superRefine((e,t)=>{e.request.policy.kind===`fixture`!=(e.policySnapshot===null)&&t.addIssue({code:`custom`,path:[`policySnapshot`],message:`Fixture execution must not claim a model identity.`}),e.status===`completed`&&(!e.cleanupComplete||!e.resultAvailable||!e.completedAt)&&t.addIssue({code:`custom`,message:`Completed attempts require retained results and confirmed cleanup.`})});c({modelProjectId:D,afterId:D.optional(),limit:u().int().min(1).max(100).default(25)}).strict(),c({items:n(Mt).max(100),nextCursor:D.nullable()}).strict(),c({schemaVersion:a(`openpond.modelStarterAttemptResult.v1`),attempt:Mt,output:f().max(1048576).nullable(),messages:n(o(f(),p())).max(4002),composition:Ce.nullable(),grade:Se.nullable().optional(),experimentGrade:kt.nullable().optional(),environment:c({status:d([`completed`,`budget_exhausted`,`cancelled`,`timed_out`,`policy_failure`,`environment_failure`]),collected:l(),definition:Le,initialStateHash:O.nullable(),finalStateHash:O.nullable(),attemptHash:O}).strict(),native:Ze.optional(),providerRequestIds:n(D).max(1001),contentHash:O}).strict().superRefine((e,t)=>{let n=e.attempt.request.policy.kind===`hosted_chat`?e.attempt.request.policy.harness:void 0;!!n!=!!e.native&&t.addIssue({code:`custom`,path:[`native`],message:`Standalone Harness results require their exact native evidence.`}),e.native&&(!n||x(n)!==x(e.native.source)||e.native.modelConfigurationHash!==e.attempt.policySnapshot?.configurationHash||e.environment.initialStateHash!==null||e.environment.finalStateHash!==null||e.environment.status===`completed`&&(!e.environment.collected||e.output===null||e.native.partialOutputAvailable||e.native.outputHash!==x(e.output)))&&t.addIssue({code:`custom`,path:[`native`],message:`Native evidence differs from the admitted source, model or output.`}),e.experimentGrade&&((e.grade||e.composition)&&t.addIssue({code:`custom`,message:`Selected Experiment grading cannot claim another grade population.`}),(![`completed`,`failed`].includes(e.attempt.status)||!e.attempt.resultAvailable||e.attempt.score!==e.experimentGrade.score||e.attempt.gradingStatus!==e.experimentGrade.gradingStatus||e.attempt.passed!==(e.experimentGrade.score===null?null:e.experimentGrade.passed))&&t.addIssue({code:`custom`,message:`Experiment grading differs from its retained attempt summary.`})),e.grade&&(e.composition&&t.addIssue({code:`custom`,path:[`grade`],message:`An ordinary grade cannot also claim a Reward composition.`}),(![`completed`,`failed`].includes(e.attempt.status)||!e.attempt.resultAvailable||e.attempt.score!==e.grade.score||e.attempt.gradingStatus!==e.grade.gradingStatus||e.attempt.passed!==(e.grade.score===null?null:e.grade.passed))&&t.addIssue({code:`custom`,path:[`grade`],message:`Ordinary grade differs from its terminal attempt summary.`}))});var Nt=f().trim().min(1).max(240),Pt=f().regex(/^[a-f0-9]{64}$/),Ft=c({maxOutputTokens:u().int().min(1).max(1e6).default(4096),temperature:u().min(0).max(2).default(0),topP:u().gt(0).max(1).default(1)}).strict(),It=c({kind:a(`hosted_harness`),modelId:Nt,maxOutputTokens:Ft.shape.maxOutputTokens.removeDefault().optional(),temperature:Ft.shape.temperature.removeDefault().optional(),topP:Ft.shape.topP.removeDefault().optional(),localRuntime:Et.optional(),candidate:c({artifactId:Nt,contentHash:Pt,preparationId:Nt,dispatchId:Nt,baseProfileId:Nt,workerImageDigest:f().regex(/^sha256:[a-f0-9]{64}$/)}).strict().optional(),sourceCandidate:he.shape.sourceCandidate,reasoningEffort:d([`off`,`low`,`medium`,`high`,`xhigh`,`max`]).optional(),profileRepositoryId:Nt,source:he,modelConfigurationHash:Pt,packageHash:Pt,externalDatasetBinding:pe.optional()}).strict().superRefine((e,t)=>{JSON.stringify(e.externalDatasetBinding??null)!==JSON.stringify(e.source.externalDatasetBinding??null)&&t.addIssue({code:`custom`,message:`The policy and source must bind the same external Dataset recipe.`}),JSON.stringify(e.sourceCandidate??null)!==JSON.stringify(e.source.sourceCandidate??null)&&t.addIssue({code:`custom`,message:`The policy and prepared source must bind the same actual source candidate.`})}),k=f().trim().min(1).max(200),Lt=f().regex(/^[a-f0-9]{64}$/),Rt=i(`kind`,[At.options[0],c({kind:a(`fixture`)}).strict(),It]),zt=c({schemaVersion:a(`openpond.modelTasksetRunRequest.v1`),operationId:k,teamId:k,modelProjectId:k.nullable(),name:f().trim().min(1).max(200).optional(),project:Dt.optional(),taskset:Le,policy:Rt,population:n(gt).min(1).max(1e4)}).strict().superRefine((e,t)=>{new Set(e.population.map(e=>e.receiptId)).size!==e.population.length&&t.addIssue({code:`custom`,path:[`population`],message:`Run receipt identities must be unique.`}),e.population.some(t=>t.fixtureId!==null!=(e.policy.kind===`fixture`))&&t.addIssue({code:`custom`,path:[`population`],message:`Only fixture runs require a fixture identity for every member.`}),e.population.some(e=>!/^(0|[1-9][0-9]*)$/.test(e.seed)||Number(e.seed)>2147483647)&&t.addIssue({code:`custom`,path:[`population`],message:`Hosted environment seeds must be integers from 0 to 2147483647.`}),e.policy.kind===`hosted_harness`&&e.modelProjectId!==null&&t.addIssue({code:`custom`,path:[`modelProjectId`],message:`A Harness target retains its released Profile binding separately from Model artifacts.`}),e.policy.kind===`hosted_harness`&&new Set(e.population.map(e=>Re([e.taskId,e.seed]))).size!==e.population.length&&t.addIssue({code:`custom`,path:[`population`],message:`Harness case identities must be unique for each task and seed.`})}),Bt=c({schemaVersion:a(`openpond.modelTasksetRunSummary.v1`),id:k,revision:u().int().positive(),teamId:k,modelProjectId:k.nullable(),name:f().trim().min(1).max(200).optional(),project:Dt.optional(),operationId:k,taskset:Le,policyKind:d([`hosted_chat`,`fixture`,`hosted_harness`]),manifestHash:Lt,status:d([`queued`,`running`,`cancelling`,`completed`,`failed`,`cancelled`]),totalCount:u().int().min(1).max(1e4),counts:c({pending:u().int().nonnegative(),running:u().int().nonnegative(),completed:u().int().nonnegative(),failed:u().int().nonnegative(),cancelled:u().int().nonnegative()}).strict(),createdAt:t(),startedAt:t().nullable(),completedAt:t().nullable(),cleanupComplete:l(),resultAvailable:l(),score:u().min(0).max(1).nullable(),metricName:f().min(1).max(240),error:c({code:k,message:f().max(2e3)}).strict().nullable()}).strict().superRefine((e,t)=>{Object.values(e.counts).reduce((e,t)=>e+t,0)!==e.totalCount&&t.addIssue({code:`custom`,path:[`counts`],message:`Run counts must account for the complete population.`}),e.status!==`completed`&&e.score!==null&&t.addIssue({code:`custom`,path:[`score`],message:`Only completed evaluations publish a metric value.`}),[`completed`,`failed`,`cancelled`].includes(e.status)&&(!e.completedAt||e.counts.pending||e.counts.running)&&t.addIssue({code:`custom`,message:`Terminal runs require complete population accounting.`}),e.status===`cancelled`&&!e.cleanupComplete&&t.addIssue({code:`custom`,message:`Cancelled runs require confirmed cleanup.`}),e.resultAvailable&&![`completed`,`failed`,`cancelled`].includes(e.status)&&t.addIssue({code:`custom`,message:`Run results require terminal population accounting.`}),e.status===`completed`&&(!e.cleanupComplete||!e.resultAvailable)&&t.addIssue({code:`custom`,message:`Completed evaluations require retained results and confirmed cleanup.`})}),Vt=c({summary:Bt,request:zt,manifest:_t,policySnapshot:Mt.shape.policySnapshot}).strict();c({modelProjectId:k.optional(),projectId:Dt.shape.id.optional(),afterId:k.optional(),limit:u().int().min(1).max(100).default(25)}).strict(),c({items:n(Bt).max(100),nextCursor:k.nullable()}).strict(),c({schemaVersion:a(`openpond.modelTasksetRunResult.v1`),run:Vt,receipts:n(ft).min(1).max(1e4),metric:fe.nullable(),contentHash:Lt}).strict();var Ht=f().min(1).max(200).refine(e=>![`__proto__`,`prototype`,`constructor`].includes(e),`Use an ordinary field name.`),Ut=n(Ht).min(1).max(20),Wt=d([`input`,`output`,`expectedOutput`,`evaluatorContext`]),Gt=c({destination:Wt,path:Ut,source:Wt,sourcePath:n(Ht).max(20)}).strict(),Kt=n(Gt).max(100).superRefine((e,t)=>{for(let[n,r]of e.entries())e.slice(0,n).some(e=>e.destination===r.destination&&(qt(e.path,r.path)||qt(r.path,e.path)))&&t.addIssue({code:`custom`,path:[n,`path`],message:`Mapped destination paths must not overlap.`})});function qt(e,t){return e.length<=t.length&&e.every((e,n)=>e===t[n])}var A=f().trim().min(1).max(200),Jt=f().regex(/^[a-f0-9]{64}$/),Yt=c({id:A,revision:u().int().positive(),contentHash:Jt}).strict(),Xt=c({id:A,version:f().min(1).max(200),contentHash:Jt,name:f().trim().min(1).max(500).optional(),feedbackKey:we,release:Yt.nullable(),mappings:Kt.optional()}).strict(),Zt=c({id:A,version:f().min(1).max(200),contentHash:Jt,mappings:Kt}).strict(),Qt=c({operationId:A,id:A.optional(),expectedRevision:u().int().nonnegative(),request:zt,maximumCostUsd:u().finite().positive().max(1e4),graders:n(Zt).min(1).max(100).optional()}).strict().superRefine((e,t)=>{e.request.name||t.addIssue({code:`custom`,path:[`request`,`name`],message:`Name this Experiment before saving it.`}),!!e.id!=e.expectedRevision>0&&t.addIssue({code:`custom`,path:[`expectedRevision`],message:`An update requires its saved definition revision.`}),Math.round(e.maximumCostUsd*1e6)<1&&t.addIssue({code:`custom`,path:[`maximumCostUsd`],message:`The budget must be at least $0.000001.`}),e.graders&&new Set(e.graders.map(e=>e.id)).size!==e.graders.length&&t.addIssue({code:`custom`,path:[`graders`],message:`Select each grader once.`})});c({schemaVersion:a(`openpond.experimentDefinition.v1`),id:A,teamId:A,revision:u().int().positive(),request:zt,maximumCostUsd:u().finite().positive().max(1e4),graders:n(Xt).min(1).max(100),createdAt:t(),updatedAt:t()}).strict().extend({contentHash:Jt}),c({operationId:A,definition:Yt}).strict(),c({projectId:A.optional(),datasetHash:Jt.optional(),search:f().trim().max(200).optional(),status:d([`queued`,`running`,`cancelling`,`completed`,`failed`,`cancelled`]).optional(),afterId:A.optional(),limit:u().int().min(1).max(100).default(30)}).strict(),c({definition:Yt.nullable(),configurationHash:Jt.optional(),admissionRequestHash:Jt.optional(),sourceExperimentId:A.optional(),maximumCostUsd:u().finite().positive().max(1e4),graders:n(Xt).min(1).max(100)}).strict().superRefine((e,t)=>{e.definition===null&&(!e.configurationHash||!e.admissionRequestHash)&&t.addIssue({code:`custom`,path:[`configurationHash`],message:`A flat Experiment requires its sealed configuration hash.`})}),c({operationId:A,profileRepositoryId:A,definitionId:A,modelId:A,localRuntime:It.shape.localRuntime,maxOutputTokens:It.shape.maxOutputTokens,temperature:It.shape.temperature,topP:It.shape.topP,reasoningEffort:d([`off`,`low`,`medium`,`high`,`xhigh`,`max`]).optional(),maximumCostUsd:u().finite().positive().max(1e4),externalDatasetBinding:pe.optional(),sourceCandidate:It.shape.sourceCandidate}).strict(),c({request:zt,manifest:_t,maximumCostUsd:u().finite().positive().max(1e4),graders:n(Xt).min(1).max(100)}).strict();var $t=c({id:w,purpose:f().trim().min(1).max(500),audience:d([`orchestrator`,`environment`,`privileged_scorer`,`trainer`,`infrastructure`]),required:l(),ttlSeconds:u().int().positive().max(86400),scopes:n(f().trim().min(1).max(500)).max(100).default([])}).strict(),en=c({declarationId:w,leaseRef:f().trim().min(8).max(1e3),audience:$t.shape.audience,expiresAt:me}).strict(),tn=c({adapterId:w,placement:d([`local`,`remote`,`colocated`,`provider_native`]),capabilityReceipt:C,runtimeVersion:f().trim().min(1).max(200),dataPlane:c({provider:w,dataPlaneId:w,cellId:w,runnerPoolId:w,runtimeImageDigest:f().regex(/^sha256:[a-f0-9]{64}$/),capabilityReceipt:C}).strict().nullable()}).strict(),nn=c({adapterId:w,kind:d([`local`,`ssh`,`managed`,`custom`]),deviceOrPool:f().trim().min(1).max(1e3),capabilityReceipt:C,provider:w.nullable()}).strict(),rn=c({adapterId:w,workerVersion:f().trim().min(1).max(200),workerImageDigest:f().regex(/^sha256:[a-f0-9]{64}$/).nullable(),upstreamRevision:f().trim().min(1).max(500),capabilityReceipt:C}).strict(),an=c({schemaVersion:a(`openpond.harnessRunManifest.v1`),executionMode:a(`qualification`).optional(),qualificationSamples:u().int().min(1).max(32).optional(),id:w,harnessRelease:S,datasetRelease:S,evidenceSets:n(S).max(1e4),model:c({source:f().trim().min(1).max(200),revision:f().trim().min(1).max(500),artifactHash:C.nullable(),tokenizerRevision:f().trim().min(1).max(500),chatTemplateHash:C}).strict(),recipe:c({method:f().trim().min(1).max(100),version:f().trim().min(1).max(200),configHash:C}).strict(),runtimeTarget:tn,computeTarget:nn,engine:rn,resolvedBundleHash:C,secretLeaseRefs:n(en).max(1e3),approval:c({approvalHash:C,approvedAt:me,maximumSpendUsd:u().nonnegative().nullable()}).strict(),createdAt:me}).strict().extend({contentHash:C}).strict();c({schemaVersion:a(`openpond.resolvedTrainingBundle.v1`),executionMode:a(`qualification`).optional(),qualificationSamples:u().int().min(1).max(32).optional(),projection:a(`trainer`),harnessRelease:S,datasetRelease:S,evidenceSetRelease:S.nullable(),files:n(c({path:f().trim().min(1).max(2e3),sha256:C,sizeBytes:u().int().nonnegative()}).strict()).max(1e5)}).strict().extend({contentHash:C}).strict();var on=c({name:a(`adamw`).default(`adamw`),weightDecay:u().nonnegative().max(1).default(0),beta1:u().positive().lt(1).default(.9),beta2:u().positive().lt(1).default(.999),epsilon:u().positive().max(.01).default(1e-8)}).strict().default({name:`adamw`,weightDecay:0,beta1:.9,beta2:.999,epsilon:1e-8}),j=f().trim().min(1).max(240),M=f().trim().min(1),sn=f().trim().min(8).max(256),cn=o(f(),p()).default({}),N=j.nullable(),ln=d([`slash_train`,`session_menu`,`bulk_selection`,`training_page`,`task_candidate`]),un=d([`defaults`,`customize`]),dn=d([`automated`,`manual`]);c({schemaVersion:a(`openpond.trainingSourceEstimate.v1`),sessionId:j,messageCount:u().int().nonnegative(),estimatedTokens:u().int().nonnegative(),textBytes:u().int().nonnegative()}),c({query:f().max(500).default(``),offset:u().int().nonnegative().default(0),limit:u().int().min(1).max(100).default(20),candidates:n(c({sessionId:j,title:f().trim().min(1).max(500),updatedAt:M})).max(500).default([])});var fn=c({sessionId:j,title:f().trim().min(1).max(500),updatedAt:M,snippet:f().max(2e3).nullable()});c({schemaVersion:a(`openpond.trainingChatSearchResult.v1`),query:f(),offset:u().int().nonnegative(),limit:u().int().positive(),total:u().int().nonnegative(),hasMore:l(),indexedChats:u().int().nonnegative(),totalChats:u().int().nonnegative(),indexing:l(),entries:n(fn)});var pn=d([`policy_failure`,`grader_failure`,`environment_failure`,`infrastructure_failure`,`timeout`,`cancelled`]),mn=c({graderId:j.nullable(),path:f().trim().min(1).max(1e3),language:d([`javascript`,`typescript`,`python`,`json`,`text`]),content:f().max(2e6),sha256:sn,declaredSha256:sn.nullable(),integrity:d([`verified`,`unverified`,`mismatch`])}),hn=c({protocolVersion:f().trim().min(1).max(200),module:f().trim().min(1).max(1e3),moduleSha256:sn,command:n(f().max(2e3)).max(100),cwd:f().trim().min(1).max(4e3),maxTurns:u().int().positive().max(100)});c({schemaVersion:a(`openpond.tasksetGraderDetails.v1`),taskset:_,sourceTasksetId:j,graders:n(Ie).max(1e3),runtime:hn.nullable(),sources:n(mn).max(1e3),unavailableReason:f().max(2e3).nullable()});var gn=c({schemaVersion:a(`openpond.taskAttempt.v1`),id:j,tasksetId:j,taskId:j,split:Me,attempt:u().int().nonnegative(),seed:u().int(),modelRef:b.nullable(),startedAt:M,completedAt:M,output:o(f(),p()),runtimeEventRefs:n(j).max(1e4).default([]),artifactRefs:n(j).max(1e4).default([]),privilegedOutcomeRef:N,infrastructureError:f().trim().min(1).max(1e4).nullable(),costUsd:u().nonnegative().nullable(),latencyMs:u().int().nonnegative(),userInterventions:u().int().nonnegative().default(0),metadata:cn}),_n=c({schemaVersion:a(`openpond.taskAttemptArtifact.v1`),id:j,tasksetId:j,taskId:j,attemptId:j,kind:d([`raw_model_response`,`runtime_trace`,`environment_state`,`grader_evidence`,`output_artifact`]),path:f().trim().min(1).max(4e3),mediaType:f().trim().min(1).max(200).nullable().optional(),sha256:sn,sizeBytes:u().int().nonnegative(),createdAt:M,metadata:cn}),vn=c({graderId:j,graderVersion:f().trim().min(1).max(100),score:u().min(0).max(1).nullable(),passed:l(),hardGate:l(),rewardEligible:l(),feedback:f().trim().max(2e4).nullable(),evidenceRefs:n(j).max(1e4).default([]),judge:b.nullable().optional().default(null),calibrationStatus:d([`not_applicable`,`pending`,`passed`,`failed`])}),yn=c({schemaVersion:a(`openpond.gradeResult.v1`),id:j,attemptId:j,graderSetHash:sn,score:u().min(0).max(1).nullable(),passed:l(),components:n(vn).min(1).max(1e3),failureClass:pn.nullable(),feedback:n(f().trim().min(1).max(2e4)).max(1e3).default([]),rewardEligible:l(),rewardComposition:Ce.optional(),createdAt:M});d([`stable_hash_top_n`,`rft_easy_curriculum_v1`]);var bn=c({schemaVersion:a(`openpond.graderAuditReport.v1`),id:j,tasksetId:j,tasksetHash:sn,fixtureRefs:n(j).min(1).max(1e5),gradeRefs:n(j).min(1).max(1e5),passed:l(),hackingChecksPassed:l(),leakageChecksPassed:l(),infrastructureSafetyPassed:l(),failures:n(c({fixtureId:j,label:Ye,gradeId:j,reason:f().trim().min(1).max(5e3)})).max(1e5),createdAt:M});function xn(e,t=`New model`){let n=e?.trim().split(/\s+/).filter(Boolean)??[];return n.length===0?t:n.slice(0,5).join(` `)}var Sn=c({schemaVersion:a(`openpond.baseModelPreference.v1`),modelId:j,revision:f().trim().min(1).max(256).nullable(),tokenizerRevision:f().trim().min(1).max(256).nullable(),chatTemplateHash:f().trim().min(8).max(256).nullable(),modelAssetId:N,source:d([`managed`,`local`,`builtin`])}),Cn=c({schemaVersion:a(`openpond.taskCreationRequest.v1`),id:j,profileId:j,surface:ln,mode:un,entryMode:dn.default(`manual`),resourceIntent:d([`workproduct`,`dataset`]).default(`workproduct`),buildIntent:qe.default(`demonstrations`),buildSpecification:Be.nullable().default(null),objective:f().trim().min(1).max(2e4).nullable(),methodHint:d([`sft`,`dpo`,`grpo`,`ppo`]).nullable().default(null),preferredBaseModelId:j.nullable().default(null),preferredBaseModel:Sn.nullable().default(null),sourceIds:n(j).max(1e5),candidateId:N,analysisModel:b.nullable(),analysisReasoningEffort:se.nullable().default(null),createImproveRunId:N.default(null),targetIntent:c({kind:d([`agent`,`skill`,`extension`,`model`,`configuration`]).nullable(),id:N,displayName:f().trim().min(1).max(500).nullable(),operation:d([`create`,`improve`])}).default({kind:`model`,id:null,displayName:null,operation:`create`}),disclosure:c({status:d([`not_required`,`pending`,`approved`,`declined`]),content:a(`raw_excerpts`),sourceIds:n(j).max(1e5),providerModel:b.nullable(),approvalId:N,approvedAt:M.nullable()}).default({status:`not_required`,content:`raw_excerpts`,sourceIds:[],providerModel:null,approvalId:null,approvedAt:null}),createdAt:M}),wn=c({schemaVersion:a(`openpond.taskCreationSnapshot.v1`),id:j,request:Cn,state:d([`planning`,`awaiting_disclosure_approval`,`awaiting_questions`,`recommendation_ready`,`awaiting_materialization_approval`,`materializing`,`validating`,`ready`,`blocked`,`failed`,`cancelled`]),proposal:St.nullable(),materializedTasksetId:N,disclosureApprovalId:N,materializationApprovalId:N,blockingQuestions:n(c({id:j,kind:d([`objective`,`consent`,`success_signal`,`privacy_licensing`,`interpretation`]),prompt:f().trim().min(1).max(5e3),answer:f().trim().min(1).max(2e4).nullable()})).default([]),transcript:n(c({id:j,role:d([`user`,`assistant`,`system`,`tool`]),text:f().max(1e5),createdAt:M})).max(1e4).default([]),repairHistory:n(Ke).max(1e3).default([]),blockedReason:f().trim().min(1).max(1e4).nullable(),createdAt:M,updatedAt:M});c({schemaVersion:a(`openpond.taskCreationTranscript.v1`),creationId:j,profileId:j,messages:wn.shape.transcript,updatedAt:M});var Tn=f().trim().min(1).max(240),En=f().trim().min(1),Dn=f().trim().min(8).max(256),On=o(f(),p()).default({}),kn=d([`created`,`inspecting`,`awaiting_source_review`,`awaiting_mapping`,`awaiting_materialization_approval`,`materializing`,`validating`,`ready`,`cancelling`,`cancelled`,`failed`]),An=c({schemaVersion:a(`openpond.huggingFaceDatasetLocator.v1`),repositoryId:f().trim().regex(/^[^/\s]+\/[^/\s]+$/).max(500),repositoryUrl:f().url().max(2e3),requestedRevision:f().trim().min(1).max(256).nullable()}),jn=c({path:f().trim().min(1).max(1e3),logicalType:f().trim().min(1).max(500),nullable:l(),sampleValues:n(p()).max(5)}),Mn=c({configuration:f().trim().min(1).max(500),split:f().trim().min(1).max(500),rowCount:u().int().nonnegative().nullable(),sizeBytes:u().int().nonnegative().nullable()}),Nn=c({schemaVersion:a(`openpond.huggingFaceDatasetInspection.v1`),id:Tn,locator:An,resolvedRevision:f().trim().min(7).max(256),title:f().trim().min(1).max(500),description:f().max(2e4).nullable(),declaredLicense:f().trim().min(1).max(500).nullable(),gated:l(),private:l(),configurations:n(f().trim().min(1).max(500)).min(1).max(1e3),splits:n(Mn).min(1).max(1e4),columns:n(jn).min(1).max(1e4),previewRows:n(o(f(),p())).max(25),sourceFiles:n(c({path:f().trim().min(1).max(2e3),configuration:f().trim().min(1).max(500),split:f().trim().min(1).max(500),revision:f().trim().min(7).max(256),sizeBytes:u().int().nonnegative().nullable(),contentHash:Dn.nullable()})).max(1e5),inspectedAt:En,metadata:On}),Pn=d([`row_id`,`cluster_id`,`prompt`,`messages`,`demonstration`,`chosen`,`rejected`,`expected_output`,`privileged_context`,`reward`,`feedback`,`tag`,`metadata`]),Fn=c({sourcePath:f().trim().min(1).max(1e3),target:Pn,transform:d([`identity`,`string`,`json`,`messages`,`numeric`,`math_final_answer`]),policy:d([`visible`,`privileged`,`metadata`]),required:l()}),In=c({schemaVersion:a(`openpond.datasetImportMapping.v1`),sourceSchemaHash:Dn,configuration:f().trim().min(1).max(500),upstreamSplits:n(f().trim().min(1).max(500)).min(1).max(1e3),preset:d([`prompt_only`,`prompt_completion`,`messages`,`preference`,`prompt_expected_answer`,`prompt_privileged_solution`,`prompt_feedback`]),bindings:n(Fn).min(1).max(1e4),nullPolicy:d([`reject`,`drop`,`preserve`]),invalidRowPolicy:d([`reject_import`,`drop_with_receipt`]),splitPolicy:c({seed:u().int(),assignments:o(f(),Ae),validationPercent:u().min(0).max(100),frozenEvalPercent:u().min(0).max(100)}),importerVersion:f().trim().min(1).max(200),mappingHash:Dn}),Ln=c({phase:d([`idle`,`metadata`,`download`,`mapping`,`parquet_write`,`verification`,`publish`]),completedBytes:u().int().nonnegative(),totalBytes:u().int().nonnegative().nullable(),completedRows:u().int().nonnegative(),totalRows:u().int().nonnegative().nullable(),message:f().trim().min(1).max(2e3).nullable()}),Rn=c({schemaVersion:a(`openpond.datasetImportJob.v1`),id:Tn,profileId:Tn,sourceKind:d([`huggingface`,`uploaded_file`]),status:kn,locator:An.nullable(),inspection:Nn.nullable(),mapping:In.nullable(),progress:Ln,targetStorageRoot:f().trim().min(1).max(4e3).nullable(),tasksetId:Tn.nullable(),tasksetRevision:u().int().positive().nullable(),artifactId:Tn.nullable(),error:f().trim().min(1).max(2e4).nullable(),cancellationRequested:l(),createdAt:En,updatedAt:En,completedAt:En.nullable(),metadata:On});c({profileId:Tn,url:f().trim().min(1).max(2e3)}),c({name:f().trim().min(1).max(500),objective:f().trim().min(1).max(2e4),mapping:In,targetStorageRoot:f().trim().min(1).max(4e3).nullable(),licenseApproved:l().default(!1)});var P=f().trim().min(1).max(240),F=f().trim().min(1),zn=o(f(),p()).default({}),Bn=d([`nominated`,`needs_review`,`approved_for_creation`,`creating`,`needs_task_review`,`baselining`,`ready_for_export`,`rejected`,`dismissed`,`retired`,`blocked`]),Vn=c({id:P,kind:d([`repeated_success`,`accepted_correction`,`agent_action_recurrence`,`outcome_linked_change`,`frontier_cost`,`expert_label`,`runtime_feedback`]),sourceRefIds:n(P).min(1).max(1e4),occurredAt:F,signature:f().trim().min(1).max(2e3),summary:f().trim().min(1).max(5e3),confidence:u().min(0).max(1),consented:l(),metadata:zn}),Hn=c({frequency:u().min(0).max(1),businessValue:u().min(0).max(1),frontierCost:u().min(0).max(1),signalQuality:u().min(0).max(1),verifiability:u().min(0).max(1),repeatability:u().min(0).max(1),privacyRisk:u().min(0).max(1),overall:u().min(0).max(1)}),Un=c({tactic:vt,eligible:l(),reasons:n(f().trim().min(1).max(5e3)).min(1).max(100),blockers:n(f().trim().min(1).max(5e3)).max(100).default([]),requiredSignals:n(f().trim().min(1).max(500)).max(100).default([]),generatedBy:d([`decision_table`,`baseline_reassessment`])}),Wn=c({schemaVersion:a(`openpond.taskCandidate.v1`),id:P,profileId:P,status:Bn,fingerprint:f().trim().min(1).max(512),title:f().trim().min(1).max(500),summary:f().trim().min(1).max(1e4),workflowSignature:f().trim().min(1).max(2e3),evidence:n(Vn).min(1).max(1e5),scorecard:Hn,recommendation:Un,mergedIntoId:P.nullable(),createdAt:F,updatedAt:F,metadata:zn}),Gn=c({schemaVersion:a(`openpond.taskMinerConfig.v1`),enabled:l().default(!1),localOnly:l().default(!0),observationWindowDays:u().int().min(1).max(365).default(30),minimumRecurrence:u().int().min(2).max(100).default(3),clustering:a(`hybrid_deterministic_first`).default(`hybrid_deterministic_first`),consentRequired:l().default(!0)});c({items:n(Wn),generatedAt:F,config:Gn}),c({status:Bn.optional(),mergeIntoId:P.nullable().optional()}),c({profileId:P,sourceIds:n(P).max(1e5).default([]),sessionIds:n(P).max(1e5).default([]),config:Gn.optional()});var Kn=c({schemaVersion:a(`openpond.taskMinerRun.v1`),id:P,profileId:P,status:d([`queued`,`running`,`cancelling`,`cancelled`,`succeeded`,`failed`]),config:Gn,sourceIds:n(P).max(1e5),sessionIds:n(P).max(1e5).default([]),progress:c({stage:d([`queued`,`ingesting`,`preparing`,`clustering`,`persisting`,`complete`]),processedSources:u().int().nonnegative(),totalSources:u().int().nonnegative(),candidatesFound:u().int().nonnegative(),skippedSources:u().int().nonnegative().default(0)}),candidateIds:n(P).max(1e5),cancelRequested:l(),error:f().trim().min(1).max(2e4).nullable(),createdAt:F,startedAt:F.nullable(),completedAt:F.nullable(),updatedAt:F});function I(e){let t;return new te({type:`object`,get shape(){return t??=e()}})}var qn=f().trim().min(1).max(200),Jn=c({operationId:qn,execution:S,executionKind:a(`recorded_evidence`).optional(),graders:n(Ee).min(1).max(100),maximumCostUsd:u().finite().min(1e-6).max(1e4),mappings:n(c({graderId:qn,fields:Kt}).strict()).max(100).optional()}).strict().superRefine((e,t)=>{new Set(e.graders.map(e=>e.id)).size!==e.graders.length&&t.addIssue({code:`custom`,path:[`graders`],message:`Select each grader once.`}),e.mappings&&(new Set(e.mappings.map(e=>e.graderId)).size!==e.mappings.length||e.mappings.some(t=>!e.graders.some(e=>e.id===t.graderId)))&&t.addIssue({code:`custom`,path:[`mappings`],message:`Map each selected grader at most once.`})});c({schemaVersion:a(`openpond.experimentScoringPass.v1`),id:qn,teamId:qn,revision:u().int().positive(),request:Jn,graders:n(Xt).min(1).max(100),status:d([`queued`,`running`,`cancelling`,`completed`,`failed`,`cancelled`]),totalCount:u().int().min(1).max(1e4),counts:c({pending:u().int().nonnegative(),running:u().int().nonnegative(),scored:u().int().nonnegative(),unavailable:u().int().nonnegative(),failed:u().int().nonnegative(),cancelled:u().int().nonnegative()}).strict(),resultAvailable:l(),createdAt:t(),startedAt:t().nullable(),completedAt:t().nullable(),error:c({code:qn,message:f().max(2e3)}).strict().nullable()}).strict().extend({contentHash:f().regex(/^[a-f0-9]{64}$/)});var Yn=c({sourcePass:S,graders:Jn.shape.graders,mappings:Jn.shape.mappings,maximumCostUsd:u().finite().positive().max(1e4)}).strict(),Xn=c({externalDatasetBinding:pe,scoring:Yn.optional(),modelId:f().min(1).max(300),maximumCostUsd:u().finite().positive().max(1e4)}).strict(),Zn=c({source:d([`local`,`github`,`openpond_git`]),repositoryId:f().trim().min(1),profileId:f().trim().min(1)}).strict(),L=f().min(1).max(240),R=f().regex(/^[a-f0-9]{64}$/),Qn=c({schemaVersion:a(`openpond.experimentImprovement.v1`),id:L,operationId:L,requestHash:R,actorId:L,teamId:L,projectId:L.nullable(),ownerWorkspaceId:L,ownerRevision:u().int().nonnegative(),ownerSourceRevision:f().min(1).max(500),baseRelease:S,profileRef:Zn,baseProfileSourceRevision:f().min(1).max(500),evidence:c({manifest:xe,result:be}).strict(),component:c({kind:d([`skill`,`workflow`,`agent`,`instruction`]),path:f().min(1).max(2e3),workflowId:L.optional()}).strict(),mode:d([`manual`,`llm_assisted`]),status:d([`draft`,`authoring_queued`,`authoring`,`frozen`,`testing`,`ready_for_review`,`adopting`,`accepted`,`rolled_back`,`declined`,`failed`,`cancelled`]),revision:u().int().positive(),limits:c({maximumCostUsd:u().finite().positive(),maximumAuthoringSteps:u().int().min(1).max(500),maximumDurationMs:u().int().min(1e3).max(864e5)}).strict(),authoringSteps:u().int().nonnegative().max(500),work:c({sessionId:L,turnId:L,revision:u().int().positive(),startedAt:f().datetime()}).strict().nullable(),frozen:c({release:S,sourceRevision:f().min(1).max(500),profileSourceRevision:f().min(1).max(500),authoringHash:R,partitionHash:R,executableSourceHash:R,protectedClosureHash:R,candidateRevision:u().int().positive(),diff:n(c({path:f(),before:R.nullable(),after:R.nullable()}).strict()).max(1e4)}).strict().nullable(),test:c({recipe:Xn,recipeHash:f().regex(/^[a-f0-9]{64}$/),baselineOperationId:f(),candidateOperationId:f(),cancelRequestedAt:f().datetime().nullable(),dispatching:c({kind:d([`baseline`,`candidate`]),token:L,pid:u().int().positive(),processStart:f().nullable(),startedAt:f().datetime()}).strict().nullable()}).strict().nullable(),comparison:c({baseline:c({manifest:xe,result:be}).strict(),candidate:c({manifest:xe,result:be}).strict(),frozenHash:R,humanProjection:st.optional()}).strict().nullable(),adoptionIntent:c({operationId:L,expectedOwnerRevision:u().int().nonnegative(),frozenHash:R,humanProjection:st.optional()}).strict().nullable(),acceptance:S.nullable(),rollback:S.nullable(),error:f().max(2e3).nullable(),createdAt:f().datetime(),updatedAt:f().datetime(),contentHash:R}).strict();function $n(e){let t=Qn.parse(e),{contentHash:n,...r}=t;if(x(r)!==n)throw Error(`Improvement state integrity failed.`);if(ye(t.evidence),t.evidence.manifest.teamId!==t.teamId||t.evidence.result.status!==`completed`||!t.evidence.result.cases.some(e=>e.status===`completed`))throw Error(`Improvement requires usable completed owner evidence.`);if(t.test&&(x(t.test.recipe)!==t.test.recipeHash||ve(t.test.recipe.externalDatasetBinding).profileId!==t.profileRef.profileId))throw Error(`The test recipe changed its exact selected Profile.`);if(t.comparison?.humanProjection){let e=lt(t.comparison.humanProjection);if(e.teamId!==t.teamId||e.projectId!==t.projectId||x(e.baseline)!==x(t.comparison.baseline)||x(e.candidate)!==x(t.comparison.candidate))throw Error(`Human comparison projection differs from its exact owner evidence.`)}if(t.comparison&&(ye(t.comparison.baseline),ye(t.comparison.candidate),!t.frozen||t.comparison.frozenHash!==x(t.frozen)||!Te(t.comparison.baseline,t.comparison.candidate).comparable))throw Error(`Improvement comparison is not bound to a comparable frozen candidate.`);if(t.status===`ready_for_review`&&(!t.frozen||!t.comparison))throw Error(`Improvement is missing qualification.`);if(t.status===`adopting`&&(!t.adoptionIntent||!t.frozen||t.adoptionIntent.frozenHash!==x(t.frozen)))throw Error(`Improvement adoption intent is not bound to its tested revision.`);if(t.status===`accepted`&&(!t.acceptance||!t.comparison||!t.frozen))throw Error(`Improvement acceptance is missing its exact qualification receipt.`);if(t.status===`rolled_back`&&(!t.acceptance||!t.rollback))throw Error(`Profile rollback is missing its durable owner receipt.`);return t}var er=c({states:n(Qn).max(50),nextCursor:L.nullable()}).strict(),tr=c({operationId:L,ownerWorkspaceId:L,expectedOwnerRevision:u().int().nonnegative(),baseRelease:S,profileRef:Zn,profileSourceRevision:f().min(1).max(500),projectId:L.nullable(),evidence:S,component:Qn.shape.component,mode:Qn.shape.mode,limits:Qn.shape.limits}).strict(),nr=i(`operation`,[c({operation:a(`list`),limit:u().int().min(1).max(50).default(30),cursor:L.optional()}).strict(),c({operation:a(`options`),evidence:S}).strict(),c({operation:a(`start`),request:tr}).strict(),c({operation:d([`read`,`humanComparisonOptions`]),id:L}).strict(),c({operation:a(`instructions`),id:L}).strict(),c({operation:a(`edit`),id:L,revision:u().int().positive(),expectedFileHash:R,text:f().max(25e4)}).strict(),c({operation:a(`mode`),id:L,revision:u().int().positive(),mode:d([`manual`,`llm_assisted`])}).strict(),c({operation:a(`author`),id:L,revision:u().int().positive(),modelId:L,prompt:f().min(1).max(2e4)}).strict(),c({operation:d([`freeze`,`finishAuthoring`,`useCandidate`,`discard`,`cancel`,`rollback`]),id:L,revision:u().int().positive()}).strict(),c({operation:a(`test`),id:L,revision:u().int().positive(),recipe:Xn}).strict(),c({operation:a(`compare`),id:L,revision:u().int().positive(),baseline:S,candidate:S,humanSelections:it.optional()}).strict()]),rr=c({teamId:L,evidence:S,scoring:Yn.optional(),profiles:n(c({id:R,name:f(),profileRef:Zn,ownerWorkspaceId:L,ownerRevision:u().int().nonnegative(),baseRelease:S,profileSourceRevision:f(),components:n(c({kind:d([`skill`,`instruction`,`agent`,`workflow`]),path:f(),agentId:f().optional(),workflowId:f().optional(),label:f(),binding:pe}).strict()).max(500)}).strict()).max(100)}).strict(),ir=c({candidateId:L,actorId:L,teamId:L,projectId:L.nullable(),candidateRevision:u().int().positive(),recipeHash:R,baseline:c({manifest:xe,result:be}).strict(),candidate:c({manifest:xe,result:be}).strict(),contentHash:R}).strict();function ar(e){let t=ir.parse(e),{contentHash:n,...r}=t;if(x(r)!==n)throw Error(`Accepted Human discovery integrity failed.`);for(let e of[t.baseline,t.candidate])if(ye(e),e.manifest.teamId!==t.teamId||e.manifest.lineage?.humanProjection)throw Error(`Discovery must retain original owner evidence.`);return t}var z=f().trim().min(1).max(240),B=f().regex(/^[a-f0-9]{64}$/),or=c({schemaVersion:a(`openpond.advancedRefinerEvaluation.v1`),operationId:z,actorId:z,teamId:z,projectId:z.nullable(),mode:d([`review_quality`,`downstream_adaptation`]),profileRef:c({source:d([`local`,`github`,`openpond_git`]),repositoryId:z,profileId:z}).strict(),sourceRevision:f().min(1).max(500),baselineRelease:S,refinerRelease:S,evidence:S,externalDatasetBinding:pe,adaptationDatasetBinding:pe,privateClosureHash:B,adaptationTaskIds:n(z).min(1).max(1e4),holdoutTaskIds:n(z).min(1).max(1e4),adaptationSplit:d([`train`,`validation`,`test`,`frozen_eval`]),holdoutSplit:d([`train`,`validation`,`test`,`frozen_eval`]),environmentPolicy:i(`kind`,[c({kind:a(`resettable`),environmentHash:B}).strict(),c({kind:a(`frozen_observations`),environmentHash:B}).strict()]),maximumCostUsd:u().finite().positive().max(1e4),maximumDurationMs:u().int().min(1e3).max(864e5),maximumModelSteps:u().int().min(1).max(1e3),contentHash:B}).strict();function sr(e){let t=or.parse(e),{contentHash:n,...r}=t;if(x(r)!==n)throw Error(`Advanced Refiner evaluation admission changed.`);let i=ve(t.externalDatasetBinding);if(i.profileId!==t.profileRef.profileId||i.protectedProfileClosureHash!==t.privateClosureHash)throw Error(`Advanced evaluation changed its Profile or protected evaluator closure.`);if(t.adaptationSplit===t.holdoutSplit||new Set([...t.adaptationTaskIds,...t.holdoutTaskIds]).size!==t.adaptationTaskIds.length+t.holdoutTaskIds.length)throw Error(`Adaptation and private holdout must be disjoint ordered cases.`);let a=ve(t.adaptationDatasetBinding);if(a.profileId!==i.profileId||a.packageHash!==i.packageHash||x(a.dataset)!==x(i.dataset)||x(a.target)!==x(i.target)||a.protectedProfileClosureHash!==i.protectedProfileClosureHash)throw Error(`This native advanced adapter needs separate adaptation and holdout recipes from the same exact released package and target.`);if(a.split!==t.adaptationSplit||i.split!==t.holdoutSplit)throw Error(`Advanced split policies differ from their exact admitted recipes.`);let o=new Set(i.population.map(e=>e.taskId)),s=new Set(a.population.map(e=>e.taskId));if(t.holdoutTaskIds.some(e=>!o.has(e))||t.adaptationTaskIds.some(e=>!s.has(e)))throw Error(`Advanced evaluation selected cases outside its exact Dataset recipe.`);return t}var cr=c({pin:or,modelId:z,profileId:z,model:Oe,seed:u().int(),reasoningEffort:d([`none`,`off`,`low`,`medium`,`high`,`xhigh`,`max`]).nullable()}).strict(),lr=i(`operation`,[c({operation:a(`start`),request:cr}).strict(),c({operation:d([`read`,`cancel`,`resume`]),id:z}).strict(),c({operation:a(`list`)}).strict()]),ur=c({taskId:z,sourceCaseHash:B,reviewSessionId:z,reviewTurnId:z,inputHarness:S,trigger:S,decision:d([`no_action`,`route_deterministically`,`queue_refiner`]),outcome:S.nullable(),proposal:S.nullable(),validations:n(S).max(100),grade:c({id:z,contentHash:B,score:u().min(0).max(1).nullable(),passed:l(),failureClass:f().nullable()}).strict(),resetComplete:l()}).strict(),dr=c({schemaVersion:a(`openpond.advancedRefinerReviewReceipt.v1`),id:z,runId:z,pinHash:B,ownerActorId:z,teamId:z,privateClosureHash:B,refinerRelease:S,cases:n(ur).min(1).max(1e4),reviewScore:u().min(0).max(1).nullable(),completedCases:u().int().nonnegative(),noActionCases:u().int().nonnegative(),routedCases:u().int().nonnegative(),proposalCases:u().int().nonnegative(),costUsd:u().finite().nonnegative().nullable(),cleanupComplete:l(),createdAt:f().datetime(),contentHash:B}).strict(),fr=rr.shape.profiles.element,pr=rr.extend({profiles:n(fr.extend({components:n(fr.shape.components.element.extend({adaptationSplits:n(c({split:d([`train`,`validation`,`test`,`frozen_eval`]),taskIds:n(z)})),environmentHash:B,resettable:l(),reviewQuality:l()}))})),refinerReleases:n(S),activeRefiner:S.nullable(),models:n(c({id:z,profileId:z,name:f(),location:d([`local`,`hosted`])}))}),mr=c({operation:a(`prepare`),evidence:S,profileOptionId:z,componentLabel:z,mode:d([`review_quality`,`downstream_adaptation`]),adaptationSplit:d([`train`,`validation`,`test`,`frozen_eval`]),operationId:z,refinerRelease:S,maximumCostUsd:u().positive(),maximumDurationMs:u().int().positive(),maximumModelSteps:u().int().positive()}).strict();r([lr,mr,c({operation:a(`options`),evidence:S}).strict()]);var hr=c({modelRequests:u().int().nonnegative(),knownSpendUsd:u().nonnegative(),totalSpendUsd:u().nonnegative().nullable(),heldUsd:u().nonnegative(),uncertainRequests:u().int().nonnegative()}).strict(),V=y.nullable(),H=u().int().nonnegative(),gr=f().regex(/^[a-f0-9]{40}$/),_r=c({name:f().trim().min(1).max(160),startedAt:m,completedAt:m,durationMs:u().nonnegative(),clock:d([`monotonic`,`provider`,`wall`]),outcome:d([`succeeded`,`failed`,`cancelled`])}).strict(),vr=c({schemaVersion:a(`openpond.correlatedTelemetryReceipt.v1`),stage:d([`training`,`evaluation`,`deployment`,`inference`]),correlation:c({modelRunId:V,modelVersionId:V,policyVersion:H.nullable(),taskId:V,rolloutGroupId:V,providerResourceId:V,deploymentId:V,inferenceRequestId:V}).strict(),spans:n(_r).max(1e4),usage:c({promptTokens:H.nullable(),generatedTokens:H.nullable(),gpuSeconds:u().nonnegative().nullable(),workerActiveSeconds:u().nonnegative().nullable(),optimizerSteps:H.nullable(),rolloutGroups:H.nullable(),successfulTrajectories:H.nullable(),failedTrajectories:H.nullable(),peakGpuMemoryBytes:H.nullable(),peakGpuUtilizationPercent:u().min(0).max(100).nullable()}).strict(),resource:c({provider:y,resourceIds:n(y).max(1e3),gpuType:f().trim().min(1).max(300).nullable(),gpuCount:H.nullable(),baseProfileId:V,baseRepository:f().trim().min(1).max(512).nullable(),baseRevision:gr.nullable(),adapterContentHash:v.nullable()}).strict(),cost:c({currency:a(`USD`),providerReportedUsd:u().nonnegative().nullable(),quotedHourlyUsd:u().nonnegative().nullable(),estimatedUsd:u().nonnegative().nullable(),methodologyVersion:f().trim().min(1).max(160),pricingInputs:o(f().trim().min(1).max(160),u().finite()),unitEstimates:o(f().trim().min(1).max(160),u().nonnegative())}).strict(),recordedAt:m,contentHash:v}).strict(),U=f().trim().min(1).max(240),W=f().regex(/^[a-f0-9]{64}$/),G=f().datetime({offset:!0}),K=c({id:U,contentHash:W}).strict(),yr=K.extend({revision:u().int().positive()}).strict(),br=c({providerId:U,modelId:f().trim().min(1).max(300)}).strict(),xr=d([`correction`,`sibling_verification`,`cumulative_known`,`development`,`retained`,`frozen_final`,`training_eligible`]),Sr=c({taskId:U,taskContentHash:W,panelRole:xr,passLabel:U.nullable(),optimizerEligible:l(),criticalInvariantIds:n(U).max(100).default([])}).strict().superRefine((e,t)=>{e.optimizerEligible!==(e.panelRole===`correction`)&&t.addIssue({code:`custom`,path:[`optimizerEligible`],message:`Only correction cases may be optimizer eligible.`})}),Cr=c({schemaVersion:a(`openpond.continualBenchIssuePacket.v1`),id:U,revision:u().int().positive(),contentHash:W,familyId:U,familyLabel:f().trim().min(1).max(500),severity:d([`low`,`medium`,`high`,`critical`]),cases:n(Sr).min(2).max(1e4),duplicateEvidence:K,leakageEvidence:K,priorExposureEvidence:K,observedAt:G,reviewedAt:G,queuedAt:G.nullable(),runAt:G.nullable(),createdBy:U,createdAt:G}).strict().superRefine((e,t)=>{e.cases.some(e=>e.panelRole===`correction`)||t.addIssue({code:`custom`,path:[`cases`],message:`An issue packet requires a correction case.`}),e.cases.some(e=>e.panelRole===`sibling_verification`)||t.addIssue({code:`custom`,path:[`cases`],message:`An issue packet requires a sibling-verification case.`}),new Set(e.cases.map(e=>e.taskId)).size!==e.cases.length&&t.addIssue({code:`custom`,path:[`cases`],message:`Issue-packet task ids must be unique.`})}),wr=c({id:U,role:xr,passLabel:U.nullable(),taskset:yr,familyIds:n(U).min(1).max(1e5),taskCount:u().int().positive().max(1e6),sealedAt:G}).strict().superRefine((e,t)=>{let n=e.role===`correction`||e.role===`sibling_verification`||e.role===`cumulative_known`;n!==(e.passLabel!==null)&&t.addIssue({code:`custom`,path:[`passLabel`],message:n?`This panel role requires a pass label.`:`This panel role is series-wide and cannot carry a pass label.`})}),Tr=i(`kind`,[c({kind:a(`base_model`),id:U,revision:f().trim().min(1).max(500),contentHash:W}).strict(),c({kind:a(`model_version`),id:U,contentHash:W}).strict(),c({kind:a(`external_reference`),id:U,model:br,providerRevision:f().trim().min(1).max(500),contentHash:W}).strict()]),Er=c({scheduleEntryId:U,ordinal:u().int().nonnegative(),label:U,role:d([`seed`,`daily_residual`,`rank_candidate`,`weekly_rollup`,`full_refresh`]),parentRule:d([`base_model`,`previous_release`,`seed_release`,`accepted_daily_head`,`accepted_seed`]),trainableRank:u().int().positive().max(1024),correctionPanelIds:n(U).min(1).max(1e3),optimizerGroupsPerTask:u().int().positive().max(1e3),trajectoriesPerGroup:u().int().min(4).max(1e3)}).strict(),Dr=c({revision:u().int().positive(),requireAllAttemptsTerminal:a(!0),criticalCorrectionPassRate:u().min(0).max(1),siblingPassRate:u().min(0).max(1),behavioralRetentionRate:u().min(0).max(1),maximumRetainedRegressionPoints:u().min(0).max(100),blockCriticalPriorRegression:l(),authorizedClaims:n(d([`systems_complete`,`correction_absorbed`,`issue_generalized`,`continually_current`,`frontier_pareto_result`])).min(1).max(5)}).strict(),Or=c({schemaVersion:a(`openpond.continualBenchmarkProtocol.v1`),id:U,revision:u().int().positive(),contentHash:W,ownerId:U,createdAt:G,sealedAt:G,creationReceipt:K,predecessorSeries:c({id:U,revision:u().int().positive(),contentHash:W}).strict().nullable(),scenarioPack:K,issueFamilyLedger:K,issuePacketRelease:K,panels:n(wr).min(6).max(1e4),grader:K,judge:c({release:K,rubricRelease:K,calibrationRelease:K,model:br}).strict().nullable(),policies:c({base:Tr,master:Tr.nullable(),externalReferences:n(Tr).min(1).max(20)}).strict(),invariants:c({systemPromptHash:W,toolSchema:K,application:K,harness:K,runtime:K,workerImage:K,autoRefiner:c({enabled:l(),release:K.nullable()}).strict()}).strict(),schedule:n(Er).min(1).max(1e3),evaluation:c({seeds:n(u().int()).min(3).max(100),repetitions:u().int().min(3).max(20),powerRule:K,pairedBootstrapSamples:u().int().min(1e3).max(1e6),confidenceLevel:a(.95)}).strict(),resources:c({maximumTrainingGpuSeconds:u().int().positive(),maximumEvaluationGpuSeconds:u().int().positive(),maximumProviderSpendUsd:u().positive(),maximumTotalSpendUsd:u().positive(),maximumConcurrentGroups:u().int().positive().max(1e3)}).strict(),currencyThresholds:Dr}).strict().superRefine((e,t)=>{let n=e.panels.map(e=>e.id);new Set(n).size!==n.length&&t.addIssue({code:`custom`,path:[`panels`],message:`Protocol panel ids must be unique.`});for(let n of[`development`,`retained`,`frozen_final`,`training_eligible`])e.panels.filter(e=>e.role===n).length!==1&&t.addIssue({code:`custom`,path:[`panels`],message:`A protocol requires exactly one ${n} panel.`});[...e.schedule].sort((e,t)=>e.ordinal-t.ordinal).some((e,t)=>e.ordinal!==t)&&t.addIssue({code:`custom`,path:[`schedule`],message:`Protocol schedule ordinals must be contiguous from zero.`});let r=new Set(n);e.schedule.some(e=>e.correctionPanelIds.some(e=>!r.has(e)))&&t.addIssue({code:`custom`,path:[`schedule`],message:`Every scheduled correction panel must belong to the protocol.`})}),kr=d([`draft`,`active`,`completed`,`archived`]),Ar=d([`draft`,`ready`,`queued`,`running`,`candidate`,`accepted`,`rejected`,`no_signal`,`failed`,`cancelled`]),jr=d([`seed`,`daily_residual`,`rank_candidate`,`weekly_rollup`,`full_refresh`]),Mr=d([`base_model`,`previous_release`,`seed_release`,`accepted_daily_head`,`accepted_seed`]),Nr=d([`seed_taskset`,`nightly_selection`,`daily_cohort_union`,`eligible_task_pool`]),Pr=d([`live_client`,`evaluation_attempt`,`replay_evidence`,`manual`]),Fr=c({id:y,ordinal:u().int().nonnegative(),label:f().trim().min(1).max(100),role:jr,parentRule:Mr,taskSource:Nr,trainableRank:u().int().positive().max(1024),minimumTasks:u().int().nonnegative().max(1e5),maximumTasks:u().int().positive().max(1e5)}).strict().superRefine((e,t)=>{e.minimumTasks>e.maximumTasks&&t.addIssue({code:`custom`,path:[`minimumTasks`],message:`The minimum task count cannot exceed the maximum task count.`});let{parentRules:n,taskSource:r}={seed:{parentRules:[`base_model`],taskSource:`seed_taskset`},daily_residual:{parentRules:[`previous_release`,`accepted_daily_head`],taskSource:`nightly_selection`},rank_candidate:{parentRules:[`seed_release`],taskSource:`eligible_task_pool`},weekly_rollup:{parentRules:[`seed_release`,`accepted_seed`],taskSource:`daily_cohort_union`},full_refresh:{parentRules:[`base_model`],taskSource:`eligible_task_pool`}}[e.role];n.includes(e.parentRule)||t.addIssue({code:`custom`,path:[`parentRule`],message:`${e.role} requires parent rule ${n.join(` or `)}.`}),e.taskSource!==r&&t.addIssue({code:`custom`,path:[`taskSource`],message:`${e.role} requires task source ${r}.`}),e.ordinal===0&&e.role!==`seed`&&t.addIssue({code:`custom`,path:[`role`],message:`The first series entry must be the seed.`})}),Ir=c({id:y,version:u().int().positive(),requireCheckpoint:l(),requireAppliedOptimizerUpdate:l(),minimumCurrentCohortMeanImprovement:u().finite(),maximumRetainedMeanRegression:u().finite().nonnegative(),blockCriticalInvariantRegression:l(),automaticDailyAdvancement:l()}).strict(),Lr=c({schemaVersion:a(`openpond.modelComparisonSeries.v1`),id:y,profileId:y,modelProjectId:y,name:f().trim().min(1).max(500),objective:f().trim().min(1).max(2e4),status:kr,revision:u().int().positive(),productionBinding:c({role:a(`chat_manual`),roleTargetId:y}).strict(),baseModel:c({id:y,revision:f().trim().min(1).max(256)}).strict(),seedTaskset:_,eligibleTaskPool:_,evaluationTasksets:c({development:_,retained:_,frozenFinal:_}).strict(),grader:g,benchmarkProtocol:Or.nullable().default(null),automaticEvaluation:c({enabled:l()}).strict().default({enabled:!1}),residualProfile:c({profileId:y,serializedEnvelopeRank:u().int().positive().max(1024),maximumEnabledRank:u().int().positive().max(1024),topology:a(`uniform_block_masked`)}).strict(),schedule:n(Fr).min(1).max(1e3),scheduleSealedAt:m.nullable(),advancementPolicy:Ir,executionPolicy:c({startWhenReady:a(!1)}).strict(),acceptedSeedEntryId:y.nullable(),acceptedDailyHeadEntryId:y.nullable(),promotedBindingId:y.nullable(),createdBy:y,createdAt:m,updatedAt:m}).strict().superRefine((e,t)=>{[...e.schedule].sort((e,t)=>e.ordinal-t.ordinal).some((e,t)=>e.ordinal!==t)&&t.addIssue({code:`custom`,path:[`schedule`],message:`Schedule ordinals must be unique and contiguous from zero.`}),(new Set(e.schedule.map(e=>e.id)).size!==e.schedule.length||new Set(e.schedule.map(e=>e.label)).size!==e.schedule.length)&&t.addIssue({code:`custom`,path:[`schedule`],message:`Schedule ids and labels must be unique.`}),e.residualProfile.maximumEnabledRank>e.residualProfile.serializedEnvelopeRank&&t.addIssue({code:`custom`,path:[`residualProfile`],message:`Maximum enabled rank cannot exceed the serialized envelope.`});let n=Object.values(e.evaluationTasksets);new Set(n.map(e=>`${e.id}:${e.revision}:${e.contentHash}`)).size!==n.length&&t.addIssue({code:`custom`,path:[`evaluationTasksets`],message:`Development, retained, and frozen-final cohorts require distinct immutable Taskset releases.`});let r=e.schedule.find(e=>e.role===`seed`);if(!r||e.schedule.filter(e=>e.role===`seed`).length!==1){t.addIssue({code:`custom`,path:[`schedule`],message:`A series requires exactly one seed entry.`});return}let i=e.schedule.filter(e=>e.role===`rank_candidate`);i.length===1&&t.addIssue({code:`custom`,path:[`schedule`],message:`A causal rank comparison requires at least two rank candidates.`}),new Set(i.map(e=>e.trainableRank)).size!==i.length&&t.addIssue({code:`custom`,path:[`schedule`],message:`Causal rank candidates must use distinct trainable ranks.`}),[r.trainableRank+e.schedule.filter(e=>e.role===`daily_residual`).reduce((e,t)=>e+t.trainableRank,0),...e.schedule.filter(e=>e.role===`weekly_rollup`).map(e=>r.trainableRank+e.trainableRank),...e.schedule.filter(e=>e.role===`rank_candidate`).map(e=>r.trainableRank+e.trainableRank),...e.schedule.filter(e=>e.role===`full_refresh`).map(e=>e.trainableRank)].some(t=>t>e.residualProfile.maximumEnabledRank)&&t.addIssue({code:`custom`,path:[`schedule`],message:`A declared branch exceeds the maximum enabled rank.`})}),Rr=c({source:Pr,taskIds:n(y).min(1).max(1e5),observedFrom:m,observedTo:m,reviewedAt:m,reviewedBy:y,sourceTaskset:_}).strict(),zr=Rr.extend({derivedFamilyKeys:n(y).min(1).max(1e5),familyDerivation:c({algorithm:y,sourceTasksetHash:v}).strict(),selectionHash:v}).strict(),Br=i(`kind`,[c({kind:a(`base_model`),id:y,revision:f().trim().min(1).max(256)}).strict(),c({kind:a(`model_version`),id:y,contentHash:v}).strict()]),Vr=c({id:y,branchOrdinal:u().int().nonnegative(),rank:u().int().positive().max(1024),offsetStart:u().int().nonnegative(),offsetEnd:u().int().positive(),optimizationRole:d([`frozen`,`trainable`]),artifactLineageId:y.nullable()}).strict().superRefine((e,t)=>{e.offsetEnd-e.offsetStart!==e.rank&&t.addIssue({code:`custom`,path:[`offsetEnd`],message:`Block offsets must span exactly the declared rank.`})}),Hr=c({disposition:d([`advance`,`hold`,`no_signal`]),policy:c({id:y,version:u().int().positive()}).strict(),reasonCodes:n(y).min(1).max(100),summary:f().trim().min(1).max(5e3),decidedBy:y,decidedAt:m}).strict(),Ur=c({evaluationRunId:y,modelVersionId:y,taskset:_,grader:g,cohortRole:d([`current`,`correction`,`sibling_verification`,`cumulative_known`,`development`,`retained`,`prior_disclosed`,`frozen_final`]),panelId:y.nullable().default(null),protocol:c({id:y,revision:u().int().positive(),contentHash:v}).strict().nullable().default(null)}).strict(),Wr=c({attemptOrdinal:u().int().positive(),trainingPlanId:y.nullable(),modelRunId:y.nullable(),terminalStatus:d([`failed`,`cancelled`]),queuedAt:m.nullable(),startedAt:m.nullable(),completedAt:m}).strict(),Gr=c({seriesId:y,entryId:y,scheduleEntryId:y,ordinal:u().int().nonnegative(),releaseHash:v}).strict(),Kr=c({schemaVersion:a(`openpond.modelComparisonSeriesEntry.v1`),id:y,seriesId:y,profileId:y,modelProjectId:y,scheduleEntryId:y,releaseHash:v,ordinal:u().int().nonnegative(),label:f().trim().min(1).max(100),role:jr,branch:d([`daily`,`rank_candidate`,`weekly_rollup`,`full_refresh`]),status:Ar,parent:Br,taskset:_,sourceTasksets:n(_).min(1).max(1e3),taskSelection:zr.nullable(),trainableRank:u().int().positive().max(1024),serializedEnvelopeRank:u().int().positive().max(1024),enabledCumulativeRank:u().int().positive().max(1024),trainableBlockId:y,residualBlocks:n(Vr).min(1).max(1024),attemptOrdinal:u().int().positive().default(1),priorRunAttempts:n(Wr).max(100).default([]),trainingPlanId:y.nullable(),modelRunId:y.nullable(),modelVersionId:y.nullable(),evaluations:n(Ur).max(1e5),decision:Hr.nullable(),promotionBindingId:y.nullable(),queuedAt:m.nullable(),startedAt:m.nullable(),completedAt:m.nullable(),createdAt:m,updatedAt:m}).strict().superRefine((e,t)=>{let n=[...e.residualBlocks].sort((e,t)=>e.offsetStart-t.offsetStart);n.some((e,t)=>e.offsetStart!==(t===0?0:n[t-1].offsetEnd))&&t.addIssue({code:`custom`,path:[`residualBlocks`],message:`Residual block offsets must be contiguous from zero.`});let r=n.at(-1)?.offsetEnd??0;(r!==e.enabledCumulativeRank||r>e.serializedEnvelopeRank)&&t.addIssue({code:`custom`,path:[`enabledCumulativeRank`],message:`Enabled rank must match the block inventory and fit the envelope.`});let i=e.residualBlocks.filter(e=>e.optimizationRole===`trainable`);(i.length!==1||i[0]?.id!==e.trainableBlockId||i[0]?.rank!==e.trainableRank)&&t.addIssue({code:`custom`,path:[`trainableBlockId`],message:`Exactly one block must be trainable during the Run and match trainableRank.`});let a=e.priorRunAttempts.map(e=>e.attemptOrdinal);(e.priorRunAttempts.length!==e.attemptOrdinal-1||a.some((e,t)=>e!==t+1))&&t.addIssue({code:`custom`,path:[`priorRunAttempts`],message:`Prior Run attempts must be a contiguous history before the active attempt.`})});c({seriesId:y,scheduleEntryId:y,taskSelection:Rr.nullable(),expectedSeriesRevision:u().int().positive()}).strict();var qr=d([`base_reference`,`lora_adapter`]),Jr=c({baseModel:c({source:a(`huggingface`),repoId:y,revision:f().regex(/^[a-f0-9]{40}$/),configHash:v,tokenizerHash:v,licenseId:y,gated:l()}).strict(),processor:c({repository:y,revision:f().regex(/^[a-f0-9]{40}$/),configHash:v}).strict()}).strict().superRefine((e,t)=>{(e.baseModel.repoId!==e.processor.repository||e.baseModel.revision!==e.processor.revision)&&t.addIssue({code:`custom`,path:[`processor`],message:`Reward Model processor must match the pinned scorer base model.`})}),Yr=c({schemaVersion:a(`openpond.modelVersion.v1`),id:y,modelId:y,profileId:y,version:u().int().nonnegative(),kind:qr,status:d([`available`,`failed`]),baseModel:Sn,taskset:_,comparisonSeriesEntry:Gr.nullable().optional(),evaluationTasksetRef:_.nullable().optional(),releaseGraph:c({resolvedBundleHash:v,profileRelease:_,harnessRelease:g,agentRelease:g.nullable(),grader:c({id:y,contentHash:v}).strict()}).strict(),artifactLineageId:y.nullable(),adapterStatus:d([`not_trained`,`trained`]),createdAt:m,contentHash:v}).strict().superRefine((e,t)=>{e.kind===`base_reference`&&(e.version!==0||e.adapterStatus!==`not_trained`||e.artifactLineageId!==null)&&t.addIssue({code:`custom`,message:`A base-reference Model Version must be version 0 without adapter lineage.`}),e.kind===`lora_adapter`&&(e.version===0||e.adapterStatus!==`trained`||!e.artifactLineageId)&&t.addIssue({code:`custom`,message:`A LoRA Model Version must be version 1+ with adapter lineage.`})}),Xr=c({schemaVersion:a(`openpond.rewardModelVersion.v1`),id:y,modelId:y,profileId:y,version:u().int().positive(),role:a(`reward`),status:d([`available`,`failed`]),scope:d([`synthetic_smoke`,`human_preference`]),baseModel:Sn,runtime:Jr.nullable().default(null),taskset:_,preferenceDatasetRelease:g,releaseGraph:c({resolvedBundleHash:v,profileRelease:_,harnessRelease:g,grader:c({id:y,contentHash:v}).strict()}).strict(),artifacts:c({checkpoint:g.extend({objectRef:f().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),files:n(c({path:f().trim().min(1).max(512),sizeBytes:u().int().positive(),sha256:v}).strict()).min(4).max(128)}).strict(),adapter:g,scalarHead:g,bucketHead:g.nullable(),processorRelease:g}).strict(),qualificationReport:g.nullable(),createdAt:m,contentHash:v}).strict(),Zr=c({schemaVersion:a(`openpond.rewardModelRunReceipt.v1`),provider:y,providerRunId:y,resolvedBundleHash:v,finalCheckpoint:g.extend({objectRef:f().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/)}).strict(),adapter:g,scalarHead:g,bucketHead:g.nullable(),processorRelease:g,optimizerEvidence:g,managedExecutionReceipt:g.nullable().default(null),parameterDeltaHash:v,cleanup:c({computeReleased:l(),providerTerminalObserved:l()}).strict(),contentHash:v}).strict(),Qr=s(e=>{if(!e||typeof e!=`object`||Array.isArray(e))return e;let t=e;if(t.tasksetRelease!==void 0)return e;let n=t.taskset;if(!n||typeof n!=`object`||Array.isArray(n))return e;let r=n;return typeof r.id!=`string`||typeof r.contentHash!=`string`?e:{...t,tasksetRelease:{id:r.id,contentHash:r.contentHash}}},c({schemaVersion:a(`openpond.rewardModelRun.v1`),id:y,rewardModelId:y,rewardModelVersionId:y.nullable(),profileId:y,role:a(`reward`),scope:d([`synthetic_smoke`,`human_preference`]),status:d([`prepared`,`running`,`succeeded`,`failed`,`cancelled`]),taskset:_,tasksetRelease:g,preferenceDatasetRelease:g,recipeRelease:g,destinationId:y,quote:c({maximumSpendUsd:u().positive(),hourlyCostUsd:u().nonnegative().nullable()}).strict(),managedRunId:y.nullable(),progress:c({completedSteps:u().int().nonnegative(),totalSteps:u().int().positive(),latestLoss:u().finite().nullable()}).strict(),receipt:Zr.nullable(),qualificationReport:g.nullable(),accruedSpendUsd:u().nonnegative().nullable(),failureOwner:d([`authoring`,`admission`,`provider`,`runner`,`artifact`,`qualification`,`cleanup`]).nullable(),failure:f().trim().min(1).max(5e3).nullable(),startedAt:m,completedAt:m.nullable(),updatedAt:m}).strict()).superRefine((e,t)=>{e.status===`succeeded`&&!e.receipt&&t.addIssue({code:`custom`,message:`A successful Reward Model Run requires a canonical receipt.`}),e.status!==`succeeded`&&e.rewardModelVersionId!==null&&t.addIssue({code:`custom`,message:`Only a successful Reward Model Run can publish a Reward Model Version.`})}),$r=c({schemaVersion:a(`openpond.modelRunReceipt.v1`),provider:y,providerRunId:y,assignmentHash:v,resultHash:v,transcriptHash:v,traceHash:v.nullable(),resolvedBundleHash:v,artifactPath:f().trim().min(1).max(2e3),cleanup:c({computeReleased:l(),tunnelClosed:l()}).strict(),telemetry:vr.nullable().default(null),contentHash:v}).strict(),q=c({inputTokens:u().int().nonnegative(),outputTokens:u().int().nonnegative(),totalTokens:u().int().nonnegative(),costUsd:u().nonnegative().nullable()}).strict(),ei=f().regex(/^(?:[a-f0-9]{40}|[a-f0-9]{64})$/),ti=c({phase:d([`baseline`,`adaptation`,`candidate_adaptation`,`candidate`]),taskId:y,attemptId:y,sessionId:y.nullable(),turnId:y.nullable(),passed:l(),score:u().min(0).max(1).nullable(),failureClass:pn.nullable(),inputTokens:u().int().nonnegative(),outputTokens:u().int().nonnegative(),totalTokens:u().int().nonnegative(),latencyMs:u().int().nonnegative(),costUsd:u().nonnegative().nullable(),startedAt:m}).strict(),ni=c({targetTaskCount:u().int().nonnegative(),comparedTaskCount:u().int().nonnegative(),passedTaskCount:u().int().nonnegative(),failedTaskCount:u().int().nonnegative(),lowerTaskCount:u().int().nonnegative(),higherTaskCount:u().int().nonnegative(),unchangedTaskCount:u().int().nonnegative()}).strict(),ri=c({target:a(`all_tasks_lower`),targetTaskCount:u().int().nonnegative(),comparedTaskCount:u().int().nonnegative(),passedTaskCount:u().int().nonnegative(),failedTaskCount:u().int().nonnegative(),lowerTaskCount:u().int().nonnegative(),higherTaskCount:u().int().nonnegative(),unchangedTaskCount:u().int().nonnegative(),baselineTokens:u().int().nonnegative(),refinedTokens:u().int().nonnegative(),tokenDelta:u().int(),tokenDeltaPercent:u().finite().nullable(),complete:l(),passed:l(),cohorts:c({adaptation:ni,heldOut:ni}).strict()}).strict(),ii=c({baseline:q,adaptation:q,candidateAdaptation:q,candidate:q,refiner:q,grader:q}).strict(),ai=c({usage:ii,observedSpendUsd:u().nonnegative(),attempts:n(ti).max(1e4)}).strict(),oi=c({id:y,contentHash:v,artifactPath:f().trim().min(1).max(2e3)}).strict(),si=c({schemaVersion:a(`openpond.modelEvaluationReceipt.v1`),benchmarkId:y,resultManifest:c({id:y,contentHash:v,artifactPath:f().trim().min(1).max(2e3)}).strict(),stages:c({baseline:g,adaptation:g,refiner:g,candidateAdaptation:g,candidate:g,comparison:g}).strict(),usage:ii,quality:c({baselinePassRate:u().min(0).max(1),candidatePassRate:u().min(0).max(1),adaptationBaselinePassRate:u().min(0).max(1),adaptationCandidatePassRate:u().min(0).max(1),adaptationCandidatePassed:l(),heldOutCandidatePassed:l(),passed:l()}).strict(),foregroundTokenDelta:u().int(),foregroundTokenDeltaPercent:u().finite().nullable(),taskEfficiency:ri.optional(),efficiency:c({grossForegroundTokenSavings:u().int(),overheadTokens:u().int().nonnegative(),firstPassNetTokenSavings:u().int(),breakEvenReuseCount:u().int().nonnegative().nullable(),amortizedTokenSavings:u().int(),amortizedReuseCount:u().int().positive()}).strict(),budget:c({maximumSpendUsd:u().nonnegative(),observedSpendUsd:u().nonnegative(),enforced:l()}).strict(),evidenceSnapshot:g,lineage:c({adaptationEvidenceHash:v,refinerInputHash:v,refinerOutcomeHash:v,validationHash:v,applyReceiptHash:v,candidateRelease:g,valid:l()}).strict(),invalidReasons:n(f().trim().min(1).max(2e3)).max(100),attempts:n(ti).max(1e4).default([]),terminalClassification:d([`improved`,`no_improvement`,`regressed`,`inconclusive`,`cancelled`,`infrastructure_failure`]),profileGit:c({ref:f().trim().min(1).max(2e3),commit:ei,baseCommit:ei}).strict().nullable(),contentHash:v}).strict(),ci=c({schemaVersion:a(`openpond.modelEvaluationStopReceipt.v1`),benchmarkId:y,terminalClassification:a(`inconclusive`),stopReason:d([`candidate_harness_unchanged`,`candidate_lineage_invalid`]),reason:f().trim().min(1).max(5e3),stoppedAfter:a(`refiner`),baselineHarness:g,candidateHarness:g,refiner:c({id:y,contentHash:v,outcomeCount:u().int().nonnegative()}).strict(),usage:ii,budget:c({maximumSpendUsd:u().nonnegative(),observedSpendUsd:u().nonnegative(),enforced:l()}).strict(),evidenceSnapshot:g,attempts:n(ti).max(1e4),contentHash:v}).strict(),li=c({level:a(.95),lower:u().finite(),upper:u().finite()}).strict().superRefine((e,t)=>{e.lower>e.upper&&t.addIssue({code:`custom`,message:`A confidence interval's lower bound cannot exceed its upper bound.`})}),ui=c({schemaVersion:a(`openpond.modelComparisonBenchmarkReceipt.v1`),benchmarkId:a(`model-comparison`),target:c({kind:d([`base_model`,`model_version`,`external_reference`]),label:f().trim().min(1).max(300),modelVersionId:y.nullable(),model:b.nullable()}).strict(),taskset:_,grader:g,sampling:c({seeds:n(u().int()).min(1).max(100),repetitions:u().int().positive().max(20)}).strict(),deterministic:c({attemptedTaskCount:u().int().nonnegative(),completedTaskCount:u().int().nonnegative(),passedTaskCount:u().int().nonnegative(),failedTaskCount:u().int().nonnegative(),meanScore:u().finite().nullable(),passRate:u().min(0).max(1).nullable(),passRateCi95:li.nullable()}).strict(),judge:c({judgeModel:b,judgeRelease:g,rubricRelease:g,calibrationRelease:g,score:u().min(0).max(100),scoreCi95:li,comparisonCount:u().int().positive(),wins:u().int().nonnegative(),ties:u().int().nonnegative(),losses:u().int().nonnegative(),referenceLabel:f().trim().min(1).max(300)}).strict().superRefine((e,t)=>{e.wins+e.ties+e.losses!==e.comparisonCount&&t.addIssue({code:`custom`,message:`Judge win, tie, and loss counts must equal comparisonCount.`})}).nullable(),attempts:n(c({attemptId:y.nullable().default(null),taskId:y,seed:u().int(),repetition:u().int().nonnegative(),status:d([`succeeded`,`failed`]),deterministicScore:u().finite().nullable(),passed:l().nullable(),judgeScore:u().min(0).max(100).nullable(),judgePreference:d([`win`,`tie`,`loss`]).nullable(),transcriptHash:v.nullable(),traceHash:v.nullable(),transcriptArtifact:c({artifactPath:f().trim().min(1).max(2e3),jsonPointer:f().trim().min(1).max(1e3)}).strict().nullable().default(null),traceArtifact:c({artifactPath:f().trim().min(1).max(2e3),jsonPointer:f().trim().min(1).max(1e3)}).strict().nullable().default(null),latencyMs:u().int().nonnegative().nullable().default(null),failureClass:pn.nullable()}).strict()).max(1e5),usage:c({policy:q.nullable(),judge:q.nullable(),observedSpendUsd:u().nonnegative().nullable(),evaluationGpuSeconds:u().nonnegative().nullable().default(null)}).strict(),managedServing:c({jobId:y,terminalState:d([`completed`,`cancelled`,`failed`,`budget_exhausted`]),sourcePolicyVersion:u().int().nonnegative(),sourceAdapterSha256:v.nullable(),servedPolicyVersion:u().int().nonnegative().nullable(),servedAdapterSha256:v.nullable(),accruedSpendUsd:u().nonnegative(),cleanupAttestationHash:v,resourceCount:u().int().nonnegative(),activeResourceCount:a(0)}).strict().nullable().default(null),evidenceSnapshot:c({id:y,contentHash:v,artifactPath:f().trim().min(1).max(2e3)}).strict(),completedAt:m,contentHash:v}).strict().superRefine((e,t)=>{let n=e.deterministic;n.completedTaskCount>n.attemptedTaskCount&&t.addIssue({code:`custom`,message:`Completed task count cannot exceed attempted task count.`}),n.passedTaskCount+n.failedTaskCount!==n.completedTaskCount&&t.addIssue({code:`custom`,message:`Passed and failed task counts must equal completed task count.`}),n.completedTaskCount===0!=(n.meanScore===null)&&t.addIssue({code:`custom`,message:`Deterministic mean score is available exactly when tasks completed.`}),n.completedTaskCount===0!=(n.passRate===null)&&t.addIssue({code:`custom`,message:`Deterministic pass rate is available exactly when tasks completed.`})}),di=c({advancedEvaluation:or.optional(),benchmarkId:a(`harness-refiner`),model:b,upstreamModel:c({providerId:y,modelId:f().trim().min(1).max(300),revision:f().trim().min(1).max(300),pricing:c({version:f().trim().min(1).max(300),source:f().trim().min(1).max(300),effectiveAt:m,inputUsdPerMillionTokens:u().nonnegative(),cachedInputUsdPerMillionTokens:u().nonnegative(),outputUsdPerMillionTokens:u().nonnegative()}).strict().optional()}).strict(),reasoningEffort:f().trim().min(1).max(100).nullable(),seeds:n(u().int()).min(1).max(100),repetitions:u().int().positive().max(20),maximumSpendUsd:u().nonnegative(),attemptPlan:n(c({stage:d([`baseline`,`adaptation`,`candidate_adaptation`,`candidate`]),split:f().trim().min(1).max(100),taskIds:n(y).min(1).max(1e5),attemptCount:u().int().positive()}).strict()).length(4)}).strict(),fi=c({benchmarkId:a(`model-comparison`),target:c({kind:d([`base_model`,`model_version`,`external_reference`]),label:f().trim().min(1).max(300),modelVersionId:y.nullable(),model:b.nullable()}).strict(),grader:g,judge:c({model:b,release:g,rubricRelease:g,calibrationRelease:g,referenceLabel:f().trim().min(1).max(300)}).strict().nullable(),seeds:n(u().int()).min(1).max(100),repetitions:u().int().positive().max(20),maximumSpendUsd:u().nonnegative(),series:c({id:y,protocol:c({id:y,revision:u().int().positive(),contentHash:v}).strict()}).strict().nullable().default(null),panel:c({id:y,role:d([`correction`,`sibling_verification`,`cumulative_known`,`development`,`retained`,`frozen_final`]),passLabel:y.nullable()}).strict().nullable().default(null),comparisonPair:c({entryId:y,parent:Br,candidateModelVersionId:y}).strict().nullable().default(null),attemptPlan:n(c({stage:a(`comparison`),split:f().trim().min(1).max(100),taskIds:n(y).min(1).max(1e5),attemptCount:u().int().positive()}).strict()).length(1)}).strict(),pi=r([di,fi]),mi=c({schemaVersion:a(`openpond.modelRun.v1`),id:y,modelId:y,modelVersionId:y.nullable(),profileId:y,kind:d([`rollout_smoke`,`training`,`evaluation`]),status:d([`prepared`,`running`,`succeeded`,`failed`,`cancelled`]),method:d([`sft`,`dpo`,`grpo`,`ppo`,`sdft`,`opd`,`opsd`,`sdpo`]).nullable(),destinationId:y.nullable(),taskset:_,comparisonSeriesEntry:Gr.nullable().optional(),evaluationTasksetRef:_.nullable().optional(),harnessRelease:g.nullable().optional(),quote:c({maximumSpendUsd:u().nonnegative(),hourlyCostUsd:u().nonnegative().nullable()}).strict().nullable(),evaluation:pi.nullable().default(null),evaluationProgress:c({stage:d([`baseline`,`adaptation`,`refiner`,`candidate_adaptation`,`candidate`,`comparison`]),completedAttempts:u().int().nonnegative(),totalAttempts:u().int().positive(),accounting:ai.nullable().default(null),advancedAccounting:hr.optional(),evidenceSnapshot:oi.nullable().default(null)}).strict().nullable().default(null),reward:c({raw:u(),components:o(f(),u())}).strict().nullable(),receipt:r([$r,si,ci,ui,dr]).nullable(),adapterArtifactLineageId:y.nullable(),failure:f().trim().min(1).max(5e3).nullable(),startedAt:m,completedAt:m.nullable(),updatedAt:m}).strict().superRefine((e,t)=>{e.status===`succeeded`&&!e.receipt&&t.addIssue({code:`custom`,message:`A successful Model Run requires a canonical receipt.`}),e.kind===`evaluation`&&(e.method!==null||e.destinationId!==null||!e.evaluation||e.adapterArtifactLineageId!==null)&&t.addIssue({code:`custom`,message:`An evaluation Model Run requires evaluation configuration and cannot claim training method, destination, or adapter lineage.`}),e.kind===`evaluation`&&e.receipt&&e.receipt.schemaVersion!==`openpond.modelEvaluationReceipt.v1`&&e.receipt.schemaVersion!==`openpond.modelEvaluationStopReceipt.v1`&&e.receipt.schemaVersion!==`openpond.modelComparisonBenchmarkReceipt.v1`&&e.receipt.schemaVersion!==`openpond.advancedRefinerReviewReceipt.v1`&&t.addIssue({code:`custom`,message:`An evaluation Model Run requires an evaluation receipt.`}),e.kind!==`evaluation`&&(e.method===null||e.destinationId===null||e.quote===null)&&t.addIssue({code:`custom`,message:`A training or rollout Model Run requires method, destination, and quote.`}),e.kind===`rollout_smoke`&&e.adapterArtifactLineageId!==null&&t.addIssue({code:`custom`,message:`A rollout smoke cannot claim adapter artifacts.`})});function hi(e){let t=new Map;for(let n of e.attempts){let e=`${n.phase}\u0000${n.taskId}`,r=t.get(e);(!r||n.startedAt>r.startedAt||n.startedAt===r.startedAt&&n.attemptId>r.attemptId)&&t.set(e,n)}let n=[...t.values()],r=[[`adaptation`,`adaptation`,`candidate_adaptation`],[`held_out`,`baseline`,`candidate`]].flatMap(([e,t,r])=>{let i=new Map(n.filter(e=>e.phase===r).map(e=>[e.taskId,e]));return n.filter(e=>e.phase===t).flatMap(t=>{let n=i.get(t.taskId);return n?[{cohort:e,taskId:t.taskId,baseline:t,refined:n,tokenDelta:n.totalTokens-t.totalTokens}]:[]})}),i=new Set(n.filter(e=>e.phase===`baseline`||e.phase===`adaptation`).map(e=>e.taskId)).size,a=e.targetTaskCount??i;if(!Number.isInteger(a)||a<0)throw Error(`Benchmark task-efficiency target must be a non-negative integer.`);let o={adaptation:new Set(n.filter(e=>e.phase===`adaptation`).map(e=>e.taskId)).size,held_out:new Set(n.filter(e=>e.phase===`baseline`).map(e=>e.taskId)).size},s=(e,t)=>({targetTaskCount:t,comparedTaskCount:e.length,passedTaskCount:e.filter(e=>e.tokenDelta<0).length,failedTaskCount:e.filter(e=>e.tokenDelta>=0).length,lowerTaskCount:e.filter(e=>e.tokenDelta<0).length,higherTaskCount:e.filter(e=>e.tokenDelta>0).length,unchangedTaskCount:e.filter(e=>e.tokenDelta===0).length}),ee=s(r,a),c=r.reduce((e,t)=>e+t.baseline.totalTokens,0),te=r.reduce((e,t)=>e+t.refined.totalTokens,0),l=te-c,u=r.length===a;return{summary:ri.parse({target:`all_tasks_lower`,...ee,baselineTokens:c,refinedTokens:te,tokenDelta:l,tokenDeltaPercent:c>0?l/c*100:null,complete:u,passed:u&&ee.passedTaskCount===a,cohorts:{adaptation:s(r.filter(e=>e.cohort===`adaptation`),o.adaptation),heldOut:s(r.filter(e=>e.cohort===`held_out`),o.held_out)}}),pairs:r}}var gi=d([`measuring`,`up_to_date`,`needs_attention`]),_i=d([`fixed`,`retained`,`regressed`,`unresolved`,`unavailable`]),vi=c({evaluationRunId:y,attemptKey:y,artifactPath:f().trim().min(1).max(2e3),jsonPointer:f().trim().min(1).max(1e3),transcriptHash:v.nullable(),traceHash:v.nullable()}).strict(),yi=c({level:a(.95),lower:u().finite(),upper:u().finite()}).strict(),bi=c({panelId:y,panelRole:xr,passLabel:y.nullable(),taskset:_,attempted:u().int().nonnegative(),available:u().int().nonnegative(),passed:u().int().nonnegative(),strictSuccess:u().min(0).max(1).nullable(),strictSuccessCi95:yi.nullable(),judgeScore:u().min(0).max(100).nullable(),judgeScoreCi95:yi.nullable()}).strict(),xi=c({schemaVersion:a(`openpond.modelCurrencySnapshot.v1`),id:y,contentHash:v,seriesId:y,protocol:c({id:y,revision:u().int().positive(),contentHash:v}).strict(),entryId:y,passLabel:y,parent:i(`kind`,[c({kind:a(`base_model`),id:y,revision:f().trim().min(1).max(500)}).strict(),c({kind:a(`model_version`),id:y,contentHash:v}).strict()]),candidate:c({kind:a(`model_version`),id:y,contentHash:v}).strict(),sources:c({evaluationRunIds:n(y).max(1e5),attemptIds:n(y).max(1e6),tasksets:n(_).max(1e4),grader:g,judge:g.nullable(),calibration:g.nullable()}).strict(),matches:n(c({key:y,panelId:y,panelRole:xr,taskId:y,issueFamilyId:y,seed:u().int(),repetition:u().int().nonnegative(),classification:_i,parentPassed:l().nullable(),candidatePassed:l().nullable(),parentAttempt:vi.nullable(),candidateAttempt:vi.nullable()}).strict()).max(1e6),taskIds:c({fixed:n(y).max(1e6),retained:n(y).max(1e6),regressed:n(y).max(1e6),unresolved:n(y).max(1e6),unavailable:n(y).max(1e6)}).strict(),panels:n(bi).max(1e4),metrics:c({knownIssueCoverage:u().min(0).max(1).nullable(),issueFamilyGeneralization:u().min(0).max(1).nullable(),behavioralRetention:u().min(0).max(1).nullable(),currentAcquisitionDelta:u().finite().nullable(),retainedDeltaPoints:u().finite().nullable(),frontierStrictDelta:u().finite().nullable(),pairedPValue:u().min(0).max(1).nullable(),wins:u().int().nonnegative(),ties:u().int().nonnegative(),losses:u().int().nonnegative()}).strict(),statistics:c({matchedAttemptCount:u().int().nonnegative(),strictDelta:u().finite().nullable(),strictDeltaCi95:yi.nullable(),exactPairedBinaryPValue:u().min(0).max(1).nullable(),pairedBootstrapSamples:u().int().positive()}).strict(),criteria:c({allRequiredAttemptsTerminal:l(),criticalCorrectionPassRate:u().min(0).max(1).nullable(),siblingPassRate:u().min(0).max(1).nullable(),behavioralRetentionRate:u().min(0).max(1).nullable(),retainedRegressionPoints:u().finite().nullable(),criticalPriorRegressionCount:u().int().nonnegative(),thresholdRelease:g}).strict(),efficiency:c({inputTokens:u().int().nonnegative(),outputTokens:u().int().nonnegative(),totalTokens:u().int().nonnegative(),costUsd:u().nonnegative().nullable(),latencyMs:u().int().nonnegative(),tokensPerSuccess:u().nonnegative().nullable(),throughputPerHour:u().nonnegative().nullable(),gpuSeconds:u().nonnegative().nullable()}).strict(),invariants:c({systemPromptHash:v,toolSchema:g,application:g,harness:g,runtime:g,grader:g,autoRefinerEnabled:l()}).strict(),evidenceState:gi,evidenceReasons:n(y).max(100),projectedAt:m}).strict(),Si=i(`kind`,[c({kind:a(`base_model`),id:y,revision:f().trim().min(1).max(500),label:f().trim().min(1).max(300)}).strict(),c({kind:a(`model_version`),id:y,label:f().trim().min(1).max(300)}).strict(),c({kind:a(`captured_model`),id:y,label:f().trim().min(1).max(300)}).strict()]),Ci=r([c({source:a(`evaluation_run`).default(`evaluation_run`),evaluationRunId:y,attemptId:y,modelLabel:f().trim().min(1).max(300).nullable().default(null),modelVersionId:y.nullable().default(null),reward:u().finite().nullable(),components:o(f(),u().finite()).default({})}).strict(),c({source:a(`imported_response`),modelLabel:f().trim().min(1).max(300),modelVersionId:y.nullable().default(null),response:p().refine(e=>e!==void 0,`An imported model response is required.`),reward:u().finite().nullable().default(null),components:o(f(),u().finite()).default({})}).strict()]),wi=r([c({taskId:y,target:Si,attempt:Ci}).strict(),c({taskId:y,target:d([`base`,`current`]).default(`current`),attempt:Ci}).strict(),c({taskId:y,target:d([`base`,`current`]).default(`current`),evaluationRunId:y,attemptId:y,reward:u().finite().nullable(),components:o(f(),u().finite()).default({})}).strict().transform(({taskId:e,target:t,...n})=>({taskId:e,target:t,attempt:{source:`evaluation_run`,modelLabel:null,modelVersionId:null,...n}}))]);c({schemaVersion:a(`openpond.continualLearningDailyBatchManifest.v1`),id:y,seriesId:y,scheduleEntryId:y,dayOrdinal:u().int().positive().max(1e4),sourceTaskset:_,taskIds:n(y).min(1).max(1e3),observedAttempts:n(wi).max(1e3).default([]),sourceFileName:f().trim().min(1).max(500).nullable().default(null),availableAt:m}).strict().superRefine((e,t)=>{new Set(e.taskIds).size!==e.taskIds.length&&t.addIssue({code:`custom`,path:[`taskIds`],message:`A daily batch cannot repeat a task.`}),new Set(e.observedAttempts.map(e=>`${e.taskId}:${Oi(e.target)}`)).size!==e.observedAttempts.length&&t.addIssue({code:`custom`,path:[`observedAttempts`],message:`An intake batch cannot repeat a task and Policy target pair.`});let n=new Set(e.taskIds);e.observedAttempts.some(e=>!n.has(e.taskId))&&t.addIssue({code:`custom`,path:[`observedAttempts`],message:`Observed attempts must belong to tasks in the daily batch.`})});var Ti=c({schemaVersion:a(`openpond.continualLearningDailyBatch.v1`),id:y,seriesId:y,scheduleEntryId:y,dayOrdinal:u().int().positive().max(1e4),label:f().trim().min(1).max(100),runMode:d([`automatic`,`manual`]).optional(),source:d([`json_upload`,`sealed_fixture`]),sourceFileName:f().trim().min(1).max(500).nullable(),sourceTaskset:_,tasks:n(c({taskId:y,taskContentHash:v,familyKey:y,disposition:d([`include`,`defer`,`exclude`]).nullable(),oracleReview:d([`pending`,`confirmed`,`needs_correction`]),note:f().trim().max(2e3),observedAttempt:Ci.nullable(),baselineAttempt:Ci.nullable().default(null),responses:n(c({target:Si,attempt:Ci}).strict()).max(100).default([]),stagedAt:m.nullable().default(null),queuedEntry:Gr.nullable().default(null)}).strict()).min(1).max(1e3),intakeEvaluation:c({status:d([`awaiting`,`running`,`ready`,`failed`]),requestedTargets:n(d([`base`,`current`])).max(2).default([`base`,`current`]),requestedPolicies:n(Si).max(100).default([]),runs:n(c({target:Si,runId:y}).strict()).max(100).default([]),baselineRunId:y.nullable(),currentRunId:y.nullable(),currentPolicy:i(`kind`,[c({kind:a(`base_model`),id:y}).strict(),c({kind:a(`model_version`),id:y}).strict()]).nullable(),failure:f().trim().min(1).max(2e3).nullable()}).strict().default({status:`awaiting`,requestedTargets:[`base`,`current`],requestedPolicies:[],runs:[],baselineRunId:null,currentRunId:null,currentPolicy:null,failure:null}),status:d([`pending`,`reviewed`,`queued`]),queuedEntry:Gr.nullable(),reviewedBy:y.nullable(),reviewedAt:m.nullable(),availableAt:m,revision:u().int().positive(),createdAt:m,updatedAt:m}).strict().superRefine((e,t)=>{new Set(e.tasks.map(e=>e.taskId)).size!==e.tasks.length&&t.addIssue({code:`custom`,path:[`tasks`],message:`A daily batch cannot repeat a task.`});for(let[n,r]of e.tasks.entries()){let e=r.responses.map(e=>Di(e.target));new Set(e).size!==e.length&&t.addIssue({code:`custom`,path:[`tasks`,n,`responses`],message:`A task cannot repeat a model response target.`})}let n=e.tasks.every(e=>e.disposition!==null&&e.oracleReview!==`pending`&&(e.oracleReview!==`needs_correction`||e.note.length>0));e.status!==`pending`&&!n&&t.addIssue({code:`custom`,path:[`tasks`],message:`Every task requires a decision and oracle review before the day completes.`}),e.status===`queued`!=(e.queuedEntry!==null)&&t.addIssue({code:`custom`,path:[`queuedEntry`],message:`Only a queued day carries an exact series-entry reference.`}),e.status===`pending`&&(e.reviewedAt!==null||e.reviewedBy!==null)&&t.addIssue({code:`custom`,path:[`reviewedAt`],message:`A pending day cannot carry reviewer completion identity.`}),e.status!==`pending`&&(e.reviewedAt===null||e.reviewedBy===null)&&t.addIssue({code:`custom`,path:[`reviewedAt`],message:`Reviewed identity is required after a day review completes.`})}),Ei=c({schemaVersion:a(`openpond.continualBenchIssueReview.v1`),id:y,seriesId:y,scheduleEntryId:y,passLabel:y,packet:Cr,splitManifest:c({id:y,contentHash:v}).strict(),decisions:n(c({taskId:y,disposition:d([`include`,`defer`,`exclude`]).nullable(),note:f().trim().max(2e3)}).strict()).min(2).max(1e4),status:d([`pending`,`reviewed`,`queued`]),queuedEntry:Gr.nullable(),reviewedBy:y.nullable(),reviewedAt:m.nullable(),revision:u().int().positive(),createdAt:m,updatedAt:m}).strict().superRefine((e,t)=>{let n=[...e.packet.cases.map(e=>e.taskId)].sort(),r=[...e.decisions.map(e=>e.taskId)].sort();JSON.stringify(n)!==JSON.stringify(r)&&t.addIssue({code:`custom`,path:[`decisions`],message:`Review decisions must cover the exact issue-packet cases.`});let i=e.decisions.every(e=>e.disposition!==null&&e.note.length>0);e.status!==`pending`&&!i&&t.addIssue({code:`custom`,path:[`decisions`],message:`Every case requires a disposition and note before review completes.`}),e.status===`queued`!=(e.queuedEntry!==null)&&t.addIssue({code:`custom`,path:[`queuedEntry`],message:`Only a queued review carries an exact series-entry reference.`})}).readonly();function Di(e){return`${e.kind}:${e.id}`}function Oi(e){return typeof e==`string`?e:Di(e)}var J=f().trim().min(1).max(240),Y=f().trim().min(8).max(256),ki=d([`grpo`,`dapo`,`gspo-token`]),Ai=c({id:J,revision:f().trim().min(1).max(256),tokenizerRevision:f().trim().min(1).max(256),chatTemplateHash:Y}),ji=c({maxRollouts:u().int().positive().max(1e6),maxEnvironmentExecutions:u().int().positive().max(1e6),maxInputTokens:u().int().positive(),maxOutputTokens:u().int().positive(),maxOptimizerSteps:u().int().positive().max(1e6),wallTimeMs:u().int().positive(),maximumCostUsd:u().nonnegative().nullable()}),Mi=c({method:a(`grpo`),groupSize:u().int().min(2).max(128),normalization:a(`group_standardized`),advantageEpsilon:u().positive().max(.01).default(1e-8),loss:ki,clipRange:u().positive().max(1).default(.2),iterations:u().int().min(2).max(16).default(2),microbatchSize:u().int().positive().max(128).default(1),gradientAccumulationSteps:u().int().positive().max(128).default(1),adamw:on}),Ni=c({method:a(`ppo`),valueModel:Ai,gamma:u().min(0).max(1),gaeLambda:u().min(0).max(1),policyClip:u().positive().max(1),valueClip:u().positive().max(10),valueLossCoefficient:u().nonnegative().max(10),ppoEpochs:u().int().positive().max(100),minibatchSize:u().int().positive().max(1e5)}),Pi=i(`method`,[Mi,Ni]),Fi=c({rewardModelVersion:c({id:J,contentHash:Y}).strict(),qualificationReport:c({id:J,contentHash:Y}).strict().nullable().default(null),evaluationReferences:n(c({id:J,contentHash:Y}).strict()).max(1e3).default([]),checkpoint:c({id:J,contentHash:Y,objectRef:f().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),files:n(c({path:f().trim().min(1).max(512),sizeBytes:u().int().positive(),sha256:Y}).strict()).min(4).max(128)}).strict(),runtime:Jr,processorRelease:c({id:J,contentHash:Y}).strict(),rewardComposerRelease:c({id:J,contentHash:Y}).strict(),executionReceipt:c({id:J,contentHash:Y}).strict()}).strict(),Ii=c({schemaVersion:a(`openpond.policyOptimization.v1`),policyModel:Ai,referenceModel:Ai,dataset:c({tasksetId:J,tasksetHash:Y,split:a(`train`),selectionStrategy:d([`stable_hash_top_n`,`rft_easy_curriculum_v1`]),selectionSeed:u().int(),maxExamples:u().int().positive().max(1e5)}),sampler:c({temperature:u().min(0).max(2),topP:u().min(0).max(1),maxOutputTokens:u().int().positive().max(32768),maxTurns:u().int().positive().max(1e3),concurrency:u().int().positive().max(1e3)}),environment:c({id:J,version:f().trim().min(1).max(256),toolContractHash:Y}),reward:c({graderId:J,graderHash:Y,learnedPreference:Fi.nullable().default(null)}),kl:c({coefficient:u().nonnegative().nullable(),referenceConstraint:a(`fixed_reference`)}),budgets:ji,checkpointEverySteps:u().int().positive(),seed:u().int(),evaluationSplit:a(`frozen_eval`),optimizer:Pi}),X=f().trim().min(1).max(240),Z=f().trim().min(1),Q=f().trim().min(8).max(256),Li=o(f(),p()).default({}),Ri=c({schemaVersion:a(`openpond.localModelChatConfiguration.v1`).default(`openpond.localModelChatConfiguration.v1`),profile:d([`efficient`,`full_harness`,`custom`]).default(`efficient`),systemPromptMode:d([`lean`,`full_harness`,`custom`]).default(`lean`),customSystemPrompt:f().max(2e4).nullable().default(null),contextWindowTokens:u().int().min(128).max(32768).default(1024),maxOutputTokens:u().int().min(1).max(512).default(64),temperature:u().min(0).max(2).default(0),repetitionPenalty:u().min(.5).max(2).default(1.1),noRepeatNgramSize:u().int().min(0).max(10).default(3),compaction:d([`off`,`when_needed`]).default(`when_needed`),keepWarmSeconds:u().int().min(0).max(3600).default(300),updatedAt:Z.nullable().default(null)}),zi=Ri.parse({}),Bi=c({schemaVersion:a(`openpond.managedAdapterServingProjection.v1`),teamId:X.nullable().default(null),source:a(`sandbox_managed_rl`),sourceRef:X,canonicalArtifactId:X.nullable(),canonicalModelName:f().trim().min(1).max(256).nullable().default(null),canonicalArtifactState:d([`imported_unvalidated`,`evaluating`,`promotable`,`rejected`,`deleted`]).nullable(),canonicalDeploymentId:X.nullable(),canonicalDeploymentState:d([`requested`,`deploying`,`ready`,`degraded`,`deleting`,`deleted`,`failed`]).nullable(),state:d([`pending`,`imported`,`ready`,`failed`]),customerBindingAllowed:l().default(!1),artifactContentHash:Q.nullable().default(null),baseProfileId:X.nullable().default(null),publishedAt:Z.nullable(),lastSyncedAt:Z,lastError:f().trim().min(1).max(5e3).nullable()});function Vi(e){let t=e.managedServing;return!!(t?.source===`sandbox_managed_rl`&&t.canonicalArtifactState===`promotable`&&t.customerBindingAllowed)}function Hi(e){return e.state===`ready`&&e.canonicalArtifactState===`promotable`&&e.canonicalDeploymentState===`ready`&&e.customerBindingAllowed&&!!e.teamId&&!!e.canonicalArtifactId&&!!e.canonicalDeploymentId}function Ui(e){if(e.promotable&&e.frozenEvaluationArtifactId)return{kind:`source_frozen_evaluation`,evaluationArtifactId:e.frozenEvaluationArtifactId,canonicalArtifactId:null,canonicalDeploymentId:null};let t=e.managedServing;return t?.source===`sandbox_managed_rl`&&Hi(t)&&t.canonicalArtifactId&&t.canonicalDeploymentId?{kind:`sandbox_customer_binding`,evaluationArtifactId:null,canonicalArtifactId:t.canonicalArtifactId,canonicalDeploymentId:t.canonicalDeploymentId}:null}function Wi(e){let t=e.managedServing;return t?Hi(t)&&t.canonicalArtifactId&&t.canonicalDeploymentId?{kind:`sandbox_serving`,canonicalArtifactId:t.canonicalArtifactId,canonicalDeploymentId:t.canonicalDeploymentId}:null:{kind:`source_artifact`,canonicalArtifactId:null,canonicalDeploymentId:null}}var Gi=`openpond.crossSystemOperations.v1`,Ki=`1.0.0`,qi=`openpond.crossSystemTools.v1`,Ji=[`search_crm`,`query_billing`,`search_support`,`run_python`],Yi=d([`train`,`validation`,`frozen_eval`]),Xi=d([`easy`,`medium`,`hard`]),Zi=d([`renewal_risk_v2`]);d([`renewal_exposure`,`collections_prioritization`,`invoice_reconciliation`,`sla_escalation`,`contract_billing_mismatch`]),c({seed:u().int().nonnegative(),split:Yi,difficulty:Xi,scenarioProfile:Zi.optional()});var Qi={anyOf:[{type:`string`,minLength:1,maxLength:256},{type:`null`}],default:null},$i={type:`integer`,minimum:1,maximum:50,default:25},ea=[{name:`search_crm`,description:`Search synthetic CRM accounts and contacts. Results are scoped, projected, paginated, and budgeted.`,parameters:{type:`object`,additionalProperties:!1,required:[`query`,`fields`,`cursor`,`limit`],properties:{query:{type:`string`,minLength:1,maxLength:500},fields:{type:`array`,minItems:1,maxItems:12,uniqueItems:!0,items:{type:`string`,enum:[`account_id`,`name`,`aliases`,`contact_ids`,`contract_value_usd_cents`,`renewal_date`,`tier`,`active_contract_id`,`contract_term_months`]}},cursor:Qi,limit:$i}}},{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:Qi,limit:$i}}},{name:`search_support`,description:`Search synthetic support cases for explicit account IDs, severity, and state filters.`,parameters:{type:`object`,additionalProperties:!1,required:[`account_ids`,`severity`,`state`,`cursor`,`limit`],properties:{account_ids:{type:`array`,minItems:1,maxItems:50,uniqueItems:!0,items:{type:`string`,minLength:1,maxLength:100}},severity:{type:`array`,minItems:1,maxItems:4,uniqueItems:!0,items:{type:`string`,enum:[`P1`,`P2`,`P3`,`P4`]}},state:{type:`array`,minItems:1,maxItems:5,uniqueItems:!0,items:{type:`string`,enum:[`new`,`investigating`,`waiting_customer`,`resolved`,`closed`]}},cursor:Qi,limit:$i}}},{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}}}}],ta=_a({schemaVersion:qi,tools:ea}),na=[`Use only the four registered synthetic Cross-System Operations tools. Contract ${ta}.`,`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(` `);function ra(e=`auto`){return[na,`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(e)}. Maximum tool turns: 15.`,`Registered tool signatures: ${ea.map(ia).join(`; `)}`].join(`
+`)}ra();function ia(e){let t=oa(e.parameters),n=oa(t.properties),r=ca(t.required),i=Object.fromEntries(r.map(e=>[e,aa(n[e])]));return`${e.name}(${JSON.stringify(i)})`}function aa(e){let t=oa(e),n=sa(t.anyOf);if(n.length)return n.map(aa).join(`|`);let r=sa(t.enum);if(r.length)return r.join(`|`);if(t.type===`array`)return[aa(t.items)];if(t.type===`object`){let e=oa(t.properties),n=ca(t.required);return Object.fromEntries(n.map(t=>[t,aa(e[t])]))}return typeof t.type==`string`?t.type:`unknown`}function oa(e){return e&&typeof e==`object`&&!Array.isArray(e)?e:{}}function sa(e){return Array.isArray(e)?e:[]}function ca(e){return sa(e).filter(e=>typeof e==`string`)}var la=d(Ji),ua=i(`kind`,[c({kind:a(`model`),turn:u().int().nonnegative(),content:f().max(1e5)}),c({kind:a(`tool_call`),turn:u().int().nonnegative(),callId:f().trim().min(1).max(240),name:la,arguments:o(f(),p())}),c({kind:a(`tool_result`),turn:u().int().nonnegative(),callId:f().trim().min(1).max(240),name:la,ok:l(),result:p(),rows:u().int().nonnegative(),bytes:u().int().nonnegative(),durationMs:u().int().nonnegative(),error:f().trim().min(1).max(2e4).nullable()}),c({kind:a(`final`),turn:u().int().nonnegative(),content:f().max(1e5)})]),da=c({schemaVersion:a(Gi),id:f().trim().min(1).max(240),worldId:f().trim().min(1).max(240),taskId:f().trim().min(1).max(240),toolContractHash:a(ta),modelRef:c({providerId:f().trim().min(1),modelId:f().trim().min(1)}).nullable(),status:d([`completed`,`budget_exhausted`,`cancelled`,`infrastructure_failure`]),steps:n(ua).max(1e3),startedAt:f().trim().min(1),completedAt:f().trim().min(1),infrastructureError:f().trim().min(1).max(2e4).nullable(),metadata:o(f(),p()).default({})}),fa=c({schemaVersion:a(Gi),trajectoryId:f().trim().min(1).max(240),outcome:d([`correct`,`incorrect`,`parse_failure`,`budget_exhausted`,`tool_schema_violation`,`infrastructure_failure`,`cancelled`]),reward:u().min(0).max(1.15).nullable(),rewardEligible:l(),exactAnswer:l(),components:c({exactAnswer:u().min(0).max(1),efficiency:u().min(0).max(.1),conciseOutput:u().min(0).max(.05)}),metrics:c({toolCalls:u().int().nonnegative(),rowsRead:u().int().nonnegative(),bytesRead:u().int().nonnegative(),consecutiveRetrievalCalls:u().int().nonnegative(),wallTimeMs:u().int().nonnegative(),parseFailures:u().int().nonnegative(),budgetExhausted:l()}),parsedAnswer:p().nullable(),feedback:n(f().trim().min(1).max(5e3)).max(100)}),pa=c({id:f().trim().min(1).max(240),type:a(`function`),function:c({name:la,arguments:f().max(2e4)})}),ma=c({role:d([`system`,`user`,`assistant`,`tool`]),content:f().max(2e5).nullable(),tool_call_id:f().trim().min(1).max(240).optional(),tool_calls:n(pa).max(4).optional()});c({schemaVersion:a(Gi),id:f().trim().min(1).max(240),taskId:f().trim().min(1).max(240),worldId:f().trim().min(1).max(240),trajectoryId:f().trim().min(1).max(240),toolContractHash:a(ta),approval:c({status:a(`approved`),approvedBy:f().trim().min(1).max(240),approvedAt:f().trim().min(1)}),messages:n(ma).min(3).max(1e3)});var ha=c({tasksetTaskId:f().trim().min(1).max(240),environmentTaskId:f().trim().min(1).max(240),family:d([`renewal_exposure`,`collections_prioritization`,`invoice_reconciliation`,`sla_escalation`,`contract_billing_mismatch`]),prompt:f().trim().min(1).max(1e5),finalAnswer:f().trim().min(1).max(2e5),trajectoryId:f().trim().min(1).max(240),trajectoryHash:f().trim().min(8).max(256),toolNames:n(la).min(1).max(100),toolCallCount:u().int().positive(),messageCount:u().int().min(3),reward:u().min(0).max(1.15),messages:n(ma).min(3).max(1e3)}),ga=c({status:a(`approved`),approvedBy:f().trim().min(1).max(240),approvedAt:f().trim().min(1),previewHash:f().trim().min(8).max(256),trajectoryCount:u().int().positive()});c({schemaVersion:a(`openpond.crossSystemExpertBootstrapPreview.v1`),tasksetId:f().trim().min(1).max(240),tasksetHash:f().trim().min(8).max(256),tasksetRevision:u().int().positive(),previewHash:f().trim().min(8).max(256),toolContractHash:a(ta),status:d([`ready_for_review`,`approved`]),approval:ga.nullable(),tasks:n(ha).min(1).max(100)});function _a(e){let t=va(e),n=2166136261,r=2654435769;for(let e=0;e<t.length;e+=1){let i=t.charCodeAt(e);n=Math.imul(n^i,16777619),r=Math.imul(r^i,2246822507)}return`${qi}:${ya(n)}${ya(r)}`}function va(e){return Array.isArray(e)?`[${e.map(va).join(`,`)}]`:e&&typeof e==`object`?`{${Object.entries(e).sort(([e],[t])=>e.localeCompare(t)).map(([e,t])=>`${JSON.stringify(e)}:${va(t)}`).join(`,`)}}`:JSON.stringify(e)}function ya(e){return(e>>>0).toString(16).padStart(8,`0`)}var ba=c({schemaVersion:a(`openpond.singleTurnPolicyTrajectory.v1`),id:X,taskId:X,status:d([`completed`,`infrastructure_failure`]),promptHash:Q,responseText:f().max(2e6),infrastructureError:f().trim().min(1).max(1e4).nullable(),startedAt:Z,completedAt:Z,metadata:Li}),xa=c({schemaVersion:a(`openpond.ppoTrajectory.v1`),id:X,taskId:X,policyModelId:X,referenceModelId:X,valueModelId:X,steps:n(c({index:u().int().nonnegative(),observationHash:Q,actionTokenIds:n(u().int().nonnegative()).min(1),terminated:l(),truncated:l(),reward:u(),policyLogProbability:u(),referenceLogProbability:u(),valuePrediction:u(),return:u(),advantage:u(),mask:u().min(0).max(1)})).min(1).max(1e5),createdAt:Z}),Sa=c({schemaVersion:a(`openpond.exactAnswerVerifierResult.v1`),outcome:d([`correct`,`incorrect`,`parse_failure`,`infrastructure_failure`]),graderSetHash:Q,score:u().min(0).max(1).nullable(),passed:l(),rewardEligible:l(),expectedAnswerHash:Q,extractedAnswer:f().max(2e4).nullable(),feedback:n(f().trim().min(1).max(2e4)).max(1e3)}),Ca=c({schemaVersion:a(`openpond.rolloutTrajectoryReceipt.v1`),id:X,jobId:X,planId:X,tasksetId:X,tasksetHash:Q,taskId:X,split:a(`train`),correlationId:X,provider:X,providerTrace:c({invocationId:X,experimentId:X,rolloutId:X,runId:X,rowId:X}),optimizerMethod:d([`grpo`,`ppo`]).default(`grpo`),evidenceLevels:c({requested:d([`trajectory`,`aggregate`,`provider_reported`]),observed:d([`trajectory`,`aggregate`,`provider_reported`]),providerReported:d([`trajectory`,`aggregate`,`provider_reported`])}).default({requested:`trajectory`,observed:`trajectory`,providerReported:`provider_reported`}),budgetUsage:c({rollouts:u().int().nonnegative(),environmentExecutions:u().int().nonnegative(),inputTokens:u().int().nonnegative(),outputTokens:u().int().nonnegative(),optimizerSteps:u().int().nonnegative(),costUsd:u().nonnegative()}).default({rollouts:0,environmentExecutions:0,inputTokens:0,outputTokens:0,optimizerSteps:0,costUsd:0}),environment:c({id:X,version:f().trim().min(1).max(256),worldId:X,worldHash:Q,toolContractHash:Q}),policy:c({modelId:X,checkpointId:X.nullable(),completionParametersHash:Q}),status:d([`received`,`running`,`succeeded`,`failed`]),failureClass:d([`policy_failure`,`parse_failure`,`tool_schema_violation`,`budget_exhausted`,`cancelled`,`environment_failure`,`infrastructure_failure`]).nullable(),reward:c({eligible:l(),raw:u().min(0).max(1.15).nullable(),normalized:u().min(0).max(1).nullable(),components:o(f(),u())}),trajectory:r([da,ba,xa]).nullable(),verifier:r([fa,Sa]).nullable(),providerStatus:o(f(),p()).default({}),receivedAt:Z,startedAt:Z.nullable(),completedAt:Z.nullable(),updatedAt:Z}),wa=`dataset-exact-answer`,Ta=`dataset-exact-answer-v1`,Ea=`no-tools-v1`,$=d([`sft`,`dpo`,`grpo`,`ppo`,`sdft`,`opd`,`opsd`,`sdpo`]),Da=d([`lora`,`full`]),Oa=d([`openpond_managed`]);I(()=>({id:X,input:o(f(),p()),expectedOutput:o(f(),p()),tags:n(X).max(100).default([])})),I(()=>({id:X,prompt:f().min(1).max(5e5),chosen:f().min(1).max(5e5),rejected:f().min(1).max(5e5),sourceRefs:n(X).min(1).max(1e4)})),I(()=>({id:X,input:o(f(),p()),tags:n(X).max(100).default([])})),I(()=>({schemaVersion:a(`openpond.rewardModelRecipe.v1`),method:a(`reward_model`),parameterization:a(`lora_with_scalar_head`),runScope:d([`synthetic_smoke`,`human_preference`]),baseModel:c({id:X,revision:f().trim().min(1).max(256),tokenizerRevision:f().trim().min(1).max(256),processorRevision:f().trim().min(1).max(256),chatTemplateHash:Q}).strict(),tasksetRelease:g,preferenceDatasetRelease:g,processorRelease:g,input:c({kind:a(`structured_text`),serialization:d([`scenario_input_and_candidate_json_v1`,`visible_agent_trajectory_v1`]),maxCharacters:u().int().positive().max(5e5)}).strict().default({kind:`structured_text`,serialization:`scenario_input_and_candidate_json_v1`,maxCharacters:32e3}),lora:c({rank:u().int().positive().max(256),alpha:u().positive().max(1024),dropout:u().min(0).max(1),targetModules:n(X).min(1).max(100)}).strict(),heads:c({scalar:a(`pooled_hidden_state_linear`),bucket:d([`none`,`three_class`]).default(`three_class`)}).strict(),loss:c({ranking:a(`bradley_terry`),rankingWeight:u().positive(),bucketWeight:u().nonnegative(),tieWeight:u().nonnegative()}).strict(),optimizer:c({learningRate:u().positive(),maxSteps:u().int().positive().max(1e5),batchSize:u().int().positive().max(1e4),gradientAccumulationSteps:u().int().positive().max(1e4),seed:u().int(),checkpointEverySteps:u().int().positive().max(1e5)}).strict(),resourceLimits:c({wallTimeMs:u().int().positive().max(864e5),maxExamples:u().int().positive().max(1e5),maxInputCharacters:u().int().positive().max(5e5).default(32e3),maxImagePixels:u().int().positive().max(1e8).optional(),maximumSpendUsd:u().nonnegative().max(1e5)}).strict()})).superRefine((e,t)=>{e.runScope===`synthetic_smoke`&&(e.optimizer.maxSteps>100||e.resourceLimits.maximumSpendUsd>10)&&t.addIssue({code:`custom`,path:[`resourceLimits`],message:`Synthetic-smoke Reward Model runs are limited to 100 steps and USD 10.`})});var ka=I(()=>({schemaVersion:a(`openpond.rftRecipe.v1`),method:a(`grpo`),parameterization:a(`lora`),baseModel:c({id:X,revision:f().trim().min(1).max(256),tokenizerRevision:f().trim().min(1).max(256),chatTemplateHash:Q}),dataset:c({trainSplit:a(`train`),validationSplit:d([`validation`,`frozen_eval`]),maxPromptTokens:u().int().positive().max(32768),maxExamples:u().int().positive().max(1e5).default(1e3),selectionStrategy:d([`stable_hash_top_n`,`rft_easy_curriculum_v1`]).default(`stable_hash_top_n`)}),lora:c({rank:u().int().positive().max(256)}),rollout:c({groupSize:u().int().min(2).max(64),concurrency:u().int().positive().max(32),maxTurns:u().int().positive().max(100),maxOutputTokens:u().int().positive().max(8192),temperature:u().min(0).max(2),topP:u().positive().max(1),seed:u().int()}),optimizer:c({learningRate:u().positive(),maxSteps:u().int().positive().max(1e5),clipRange:u().positive().max(1).default(.2),iterations:u().int().min(2).max(16).default(2),microbatchSize:u().int().positive().max(64).default(1),gradientAccumulationSteps:u().int().positive().max(64).default(1),advantageEpsilon:u().positive().max(.01).default(1e-8),adamw:on}),loss:c({method:ki.default(`grpo`),klBeta:u().min(0).nullable().default(null)}).default({method:`grpo`,klBeta:null}),reward:c({graderId:X,graderHash:Q,environmentId:X,environmentVersion:f().trim().min(1).max(256),toolContractHash:Q,learnedPreference:Fi.nullable().optional()}),resourceLimits:c({wallTimeMs:u().int().positive().max(864e5),maxGpuSeconds:u().int().positive().max(21600).optional(),warmupDeadlineSeconds:u().int().positive().max(3600).optional(),maxRollouts:u().int().positive().max(1e5),maxPayloadBytes:u().int().positive().max(10485760)}),policyOptimization:Ii.nullable().default(null),continuation:c({schemaVersion:a(`openpond.crossJobContinuationRequest.v1`),parentArtifact:g,sourceArtifact:c({jobId:X,artifactId:X,checkpointId:X,contentHash:Q}).strict(),optimizerMode:d([`continue`,`reset`]),adapterMode:d([`residual_append`,`shared`,`sequential_merge`]).optional()}).strict().optional()})),Aa=I(()=>({schemaVersion:a(`openpond.dpoRecipe.v1`),method:a(`dpo`),parameterization:a(`lora`),policyModel:Ai,referenceModel:Ai,dataset:c({trainSplit:a(`train`),validationSplit:d([`validation`,`frozen_eval`]),maxPairs:u().int().positive().max(1e5),maxPromptTokens:u().int().positive().max(32768),maxCompletionTokens:u().int().positive().max(32768),selectionStrategy:a(`stable_hash_top_n`),selectionSeed:u().int()}),lora:c({rank:u().int().positive().max(256),alpha:u().positive().max(1024),dropout:u().min(0).max(1),targetModules:n(X).min(1).max(100)}),loss:c({variant:a(`sigmoid`),beta:u().positive().max(10),labelSmoothing:u().min(0).max(.5).default(0)}),optimizer:c({learningRate:u().positive(),epochs:u().positive().max(100),maxSteps:u().int().positive().max(1e6),batchSize:u().int().positive().max(1e4),gradientAccumulationSteps:u().int().positive().max(1e4),seed:u().int()}),referenceLogprobs:c({cacheSchemaVersion:a(`openpond.dpoReferenceLogprobs.v1`),cacheKey:Q,invalidationHash:Q}),resourceLimits:c({cpuThreads:u().int().positive().max(256),memoryBytes:u().int().positive(),wallTimeMs:u().int().positive()})})),ja=I(()=>({schemaVersion:a(`openpond.ppoRecipe.v1`),method:a(`ppo`),parameterization:a(`lora`),policyOptimization:Ii.extend({optimizer:Ni}),lora:c({rank:u().int().positive().max(256),alpha:u().positive().max(1024),dropout:u().min(0).max(1),targetModules:n(X).min(1).max(100)}),valueHead:c({initialization:a(`policy_hidden_state_linear`),optimizerLearningRate:u().positive(),artifactName:a(`value_head.safetensors`)}),policyLearningRate:u().positive(),resume:c({checkpointId:X.nullable(),policyHash:Q,referenceHash:Q,valueModelHash:Q,optimizerStateHash:Q.nullable()}),resourceLimits:c({cpuThreads:u().int().positive().max(256),memoryBytes:u().int().positive(),wallTimeMs:u().int().positive()})})),Ma=I(()=>({schemaVersion:a(`openpond.unsupportedRecipe.v1`),method:$.exclude([`sft`,`dpo`,`grpo`,`ppo`]),parameterization:Da,unsupportedReason:f().trim().min(1).max(5e3)})),Na=r([Je,Aa,ka,ja,Ma]),Pa=I(()=>({schemaVersion:a(`openpond.trainingDestinationCapabilities.v1`),destinationId:Oa,available:l(),methods:n($),parameterizations:n(Da),modelAllowlist:n(X).default([]),maxDatasetBytes:u().int().nonnegative().nullable(),environmentPlacements:n(d([`none`,`local`,`remote`,`colocated`,`provider_native`])),nonProduction:l(),unavailableReason:f().trim().min(1).max(5e3).nullable(),checkedAt:Z})),Fa=I(()=>({destinationId:Oa,available:l(),methods:n($),parameterizations:n(Da),nonProduction:l(),unavailableReason:f().trim().min(1).max(5e3).nullable()})),Ia=I(()=>({schemaVersion:a(`openpond.baseModelCandidate.v1`),selectionKey:X,label:f().trim().min(1).max(500),sourceLabel:f().trim().min(1).max(500),preference:Sn,available:l(),nonProduction:l(),unavailableReason:f().trim().min(1).max(5e3).nullable(),methods:n($),executionOptions:n(Fa).min(1)})),La=I(()=>({code:X,severity:d([`warning`,`error`]),path:f().trim().max(2e3).nullable(),message:f().trim().min(1).max(5e3)})),Ra=I(()=>({schemaVersion:a(`openpond.trainingCompatibility.v1`),compatible:l(),destinationId:Oa,tasksetId:X,recipeMethod:$,issues:n(La),checkedAt:Z})),za=r([He,d([`destination_unavailable`,`destination_method_unsupported`,`destination_model_unsupported`,`destination_parameterization_unsupported`,`experimental_destination`])]);I(()=>({method:$,state:d([`recommended`,`compatible`,`needs_dataset_work`,`destination_unavailable`,`experimental_destination`]),reasonCodes:n(za).default([]),reasons:n(f().trim().min(1).max(5e3)).default([]),destinationId:Oa.nullable().default(null)})),d([`small`,`standard`,`custom`,`small_experiment`]);var Ba=je,Va=I(()=>({schemaVersion:a(`openpond.trainingPlan.v1`),id:X,modelId:X,tasksetId:X,tasksetHash:Q,comparisonSeriesEntry:Gr.nullable().optional(),evaluationTasksetRef:_.nullable().optional(),harnessRelease:g.nullable().optional(),modelImprovementQualification:g.nullable().optional(),destinationId:Oa,recipe:Na,environmentPlacement:d([`none`,`local`,`remote`,`colocated`,`provider_native`]),compatibility:Ra,dataPolicy:c({exportApproved:l(),approvedSourceIds:n(X),retentionDays:u().int().nonnegative().nullable(),region:f().trim().min(1).max(200).nullable()}),estimatedCostUsd:u().nonnegative().nullable(),createdAt:Z,contentHash:Q})),Ha=I(()=>({path:f().trim().min(1).max(2e3),sha256:Q,sizeBytes:u().int().nonnegative(),role:d([`manifest`,`task_data`,`grader`,`environment`,`recipe`,`policy`,`provenance`])})),Ua=I(()=>({schemaVersion:a(`openpond.trainingBundle.v1`),id:X,planId:X,tasksetId:X,tasksetHash:Q,recipeHash:Q,files:n(Ha).min(1).max(1e6),totalSizeBytes:u().int().nonnegative(),sourceIds:n(X),excludedSourceIds:n(X),containsRawChats:a(!1),containsSecrets:a(!1),containsHiddenGraderAssets:a(!1),createdAt:Z,contentHash:Q}));I(()=>({schemaVersion:a(`openpond.trainingBundleExport.v1`),manifest:Ua,files:n(c({path:f().trim().min(1).max(2e3),sha256:Q,sizeBytes:u().int().nonnegative(),encoding:a(`base64`),content:f()})).min(1),contentHash:Q})),I(()=>({schemaVersion:a(`openpond.trainingPreparedStart.v1`),plan:Va,bundle:Ua,approvalActor:X.nullable(),preparedAt:Z})),I(()=>({schemaVersion:a(`openpond.trainingApproval.v1`),evaluationTasksetRef:_.nullable().optional(),id:X,planId:X,bundleHash:Q,harnessRelease:g.nullable().optional(),modelImprovementQualification:g.nullable().optional(),destinationId:Oa,modelId:X,method:$,parameterization:Da,maximumCostUsd:u().nonnegative().nullable(),approvedBy:X,approvedAt:Z})),I(()=>({schemaVersion:a(`openpond.trainingJobSourceSnapshot.v1`),evaluationTasksetRef:_.nullable().optional(),modelProjectId:X,sourceProjectRevision:u().int().positive(),profileId:X,taskset:c({id:X,revision:u().int().positive(),contentHash:Q}).strict(),tasksetRelease:g,harnessRelease:g,baseModel:Sn,method:$})).strict();var Wa=d([`queued`,`starting`,`running`,`cancelling`,`cancelled`,`succeeded`,`failed`,`reconciling`]),Ga=I(()=>({schemaVersion:a(`openpond.trainingJob.v1`),id:X,planId:X,bundleHash:Q,approvalId:X,destinationId:Oa,status:Wa,nonProduction:l(),workerPid:u().int().positive().nullable(),startedAt:Z.nullable(),completedAt:Z.nullable(),error:f().trim().min(1).max(2e4).nullable(),createdAt:Z,updatedAt:Z,metadata:Li})),Ka=I(()=>({schemaVersion:a(`openpond.trainingJobEvent.v1`),id:X,jobId:X,sequence:u().int().nonnegative(),type:d([`queued`,`start`,`progress`,`metric`,`checkpoint`,`cancel`,`complete`,`failure`,`reconcile`]),timestamp:Z,payload:Li})),qa=I(()=>({schemaVersion:a(`openpond.sftStepMetric.v1`),step:u().int().nonnegative(),maxSteps:u().int().positive(),timestamp:Z,epoch:u().nonnegative().nullable(),loss:u().nullable(),learningRate:u().nonnegative().nullable(),gradientNorm:u().nonnegative().nullable(),entropy:u().nonnegative().nullable(),meanTokenAccuracy:u().min(0).max(1).nullable(),preferenceAccuracy:u().min(0).max(1).nullable().default(null),preferenceMargin:u().nullable().default(null),chosenReward:u().nullable().default(null),rejectedReward:u().nullable().default(null),chosenLogProbability:u().nullable().default(null),rejectedLogProbability:u().nullable().default(null),reward:u().nullable().default(null),policyLoss:u().nullable().default(null),advantageLoss:u().nullable().default(null),inputTokensSeen:u().int().nonnegative().nullable(),memoryBytes:u().int().nonnegative().nullable(),elapsedSeconds:u().nonnegative().nullable()})),Ja=I(()=>({schemaVersion:a(`openpond.policyOptimizationMetric.v1`),method:d([`grpo`,`ppo`]),step:u().int().nonnegative(),timestamp:Z,learningRate:u().nonnegative().nullable().default(null),policyLoss:u().nullable(),valueLoss:u().nullable(),gradientNorm:u().nonnegative().nullable().default(null),meanReward:u().nullable(),meanReturn:u().nullable(),kl:u().nullable(),behaviorPolicyKlPreUpdate:u().nonnegative().nullable().default(null),entropy:u().nullable(),policyClipFraction:u().min(0).max(1).nullable(),behaviorPolicyClipFractionPreUpdate:u().min(0).max(1).nullable().default(null),valueClipFraction:u().min(0).max(1).nullable(),explainedVariance:u().nullable(),rolloutLearnerLag:u().int().nonnegative().nullable(),inputTokens:u().int().nonnegative(),outputTokens:u().int().nonnegative(),environmentExecutions:u().int().nonnegative(),trajectoryCount:u().int().nonnegative().nullable().default(null),costUsd:u().nonnegative().nullable()})),Ya=I(()=>({schemaVersion:a(`openpond.managedTrainingRunEvidence.v2`),provider:X,providerRunId:X,state:X,progress:c({targetOptimizerSteps:u().int().nonnegative(),committedOptimizerSteps:u().int().nonnegative(),skippedOptimizerSteps:u().int().nonnegative()}),reward:c({finalMean:u().nullable(),variance:u().nonnegative().nullable(),minimum:u().nullable(),maximum:u().nullable(),distinctValueCount:u().int().nonnegative(),noSignalGroupCount:u().int().nonnegative(),trajectoryCount:u().int().nonnegative(),eligibleTrajectoryCount:u().int().nonnegative()}),usage:c({inputTokens:u().int().nonnegative(),outputTokens:u().int().nonnegative(),environmentExecutions:u().int().nonnegative()}),resource:c({provider:X,gpuType:f().trim().min(1).max(300).nullable(),gpuCount:u().int().nonnegative().nullable(),hourlyCostUsd:u().nonnegative().nullable(),durationSeconds:u().nonnegative().nullable(),gpuSeconds:u().nonnegative().nullable()}),cost:c({totalUsd:u().nonnegative().nullable()}),checkpoint:c({id:X,policyVersion:u().int().nonnegative(),sha256:Q.nullable(),sizeBytes:u().int().nonnegative().nullable()}).nullable(),movement:c({adapterDeltaNorm:u().nonnegative().nullable()}),evaluations:n(c({reference:c({id:X,contentHash:Q}).nullable().optional(),kind:d([`baseline`,`candidate`]),policyVersion:u().int().nonnegative(),taskCount:u().int().nonnegative().nullable().optional(),targetTaskCount:u().int().nonnegative().nullable().optional(),score:u().nullable(),threshold:u().nullable(),passed:l().nullable()})),canonicalPublication:c({state:X.nullable(),artifactId:X.nullable()}),syncedAt:Z})),Xa=I(()=>({count:u().int().nonnegative(),scoredCount:u().int().nonnegative(),meanScore:u().min(0).max(1).nullable(),passedCount:u().int().nonnegative(),passRate:u().min(0).max(1).nullable()})),Za=I(()=>({status:d([`scored`,`unavailable`]),score:u().min(0).max(1).nullable(),passed:l(),rewardEligible:l(),failureClass:d([`policy_failure`,`grader_failure`,`environment_failure`,`infrastructure_failure`,`timeout`,`cancelled`]).nullable(),feedback:n(f().trim().min(1).max(2e4)).max(1e3),components:n(c({graderId:X,score:u().min(0).max(1).nullable(),passed:l(),feedback:f().trim().max(2e4).nullable()})).max(1e3)})),Qa=I(()=>({taskId:X,input:o(f(),p()),baseOutput:o(f(),p()).nullable(),trainedOutput:o(f(),p()).nullable(),baseGrade:Za.nullable(),trainedGrade:Za.nullable()})),$a=I(()=>({schemaVersion:a(`openpond.trainingEvaluationSummary.v1`),jobId:X,tasksetId:X,base:Xa,trained:Xa,meanScoreDelta:u().min(-1).max(1).nullable(),examples:n(Qa).max(1e6)})),eo=I(()=>({schemaVersion:a(`openpond.trainingRunDetail.v1`),job:Ga,events:n(Ka),stepMetrics:n(qa),policyMetrics:n(Ja).default([]),managedEvidence:Ya.nullable().default(null),evaluation:$a.nullable(),generatedAt:Z})),to=I(()=>({schemaVersion:a(`openpond.trainingArtifact.v1`),id:X,jobId:X,kind:d([`adapter`,`checkpoint`,`metrics`,`log`,`manifest`,`evaluation`]),path:f().trim().min(1).max(4e3),sha256:Q,sizeBytes:u().int().nonnegative(),baseModelId:X.nullable(),baseModelRevision:f().trim().min(1).max(256).nullable(),tokenizerRevision:f().trim().min(1).max(256).nullable(),chatTemplateHash:Q.nullable(),nonProduction:l(),createdAt:Z,metadata:Li})),no=I(()=>({schemaVersion:a(`openpond.modelArtifactLineage.v1`),id:X,modelId:X,artifactId:X,jobId:X,tasksetId:X,tasksetHash:Q,graderHash:Q,planHash:Q,bundleHash:Q,recipeHash:Q,workerVersion:f().trim().min(1).max(256),trainerVersion:f().trim().min(1).max(256),importedAt:Z,frozenEvaluationArtifactId:X.nullable(),promotable:l(),pinned:l().default(!1),status:d([`imported`,`rejected`]).default(`imported`),rejectedAt:Z.nullable().default(null),rejectionReason:f().trim().min(1).max(5e3).nullable().default(null),chatConfiguration:Ri.default(zi),managedServing:Bi.nullable().default(null)})),ro=d([`chat_manual`,`agent`,`extension`,`authoring_optimizer`]),io=I(()=>({schemaVersion:a(`openpond.modelBinding.v1`),id:X,profileId:X,role:ro,roleTargetId:X,modelArtifactLineageId:X,tasksetId:X,evaluationArtifactId:X.nullable(),status:d([`active`,`rolled_back`]),priorBindingId:X.nullable(),rollbackTargetBindingId:X.nullable(),promotedBy:X,promotedAt:Z,rolledBackAt:Z.nullable(),metadata:Li}));I(()=>({schemaVersion:a(`openpond.trainingActivity.v1`),profileId:X,active:l(),activeCounts:c({jobs:u().int().nonnegative(),creations:u().int().nonnegative(),minerRuns:u().int().nonnegative(),datasetImports:u().int().nonnegative()}),revision:Q,generatedAt:Z})),I(()=>({schemaVersion:a(`openpond.tasksetOperationalState.v1`),tasksetId:X,attempts:n(gn),artifacts:n(_n),grades:n(yn),evaluationResults:n(pt).default([]),generatedAt:Z}));var ao=I(()=>({schemaVersion:a(`openpond.trainingState.v1`),profileId:X,sources:n(Pe),creations:n(wn),tasksetDrafts:n(Ve).default([]),tasksets:n(Ue),benchmarkRuns:n(ne()).default([]),benchmarkComparisons:n(ne()).default([]),datasetImports:n(Rn).default([]),datasetArtifacts:n(Fe).default([]),graderAuditReports:n(bn),evaluationResults:n(ne()).default([]),candidates:n(Wn),minerConfig:Gn,minerRuns:n(Kn).default([]),modelProjects:n(Ba).default([]),modelVersions:n(Yr).default([]),modelRuns:n(mi).default([]),comparisonSeries:n(Lr).default([]),comparisonSeriesEntries:n(Kr).default([]),modelCurrencySnapshots:n(xi).default([]),continualBenchIssueReviews:n(Ei).default([]),continualLearningDailyBatches:n(Ti).default([]),rewardModelVersions:n(Xr).default([]),rewardModelRuns:n(Qr).default([]),modelTasksets:n(Ue).default([]),plans:n(Va),bundles:n(Ua),jobs:n(Ga),artifacts:n(to),models:n(no),rolloutReceipts:n(Ca).default([]),modelBindings:n(io).default([]),destinations:n(Pa),baseModelCandidates:n(Ia).default([]),activityRevision:Q.optional(),generatedAt:Z}));export{tn as A,kt as B,Qn as C,xn as D,$n as E,Zt as F,_t as H,Qt as I,Vt as L,Yt as M,Kt as N,nn as O,Xt as P,zt as R,rr as S,ar as T,Qe as U,St as V,Ze as W,lr as _,io as a,nr as b,eo as c,ta as d,Vi as f,hi as g,mi as h,Ya as i,rn as j,an as k,ao as l,Ui as m,Ta as n,ka as o,Wi as p,Ea as r,Na as s,wa as t,Ki as u,pr as v,Xn as w,er as x,sr as y,Tt as z};