openpond 0.2.63 → 0.2.65

package/dist/chunks/chunk-F6UQ6PPK.js

@@ -1,2 +0,0 @@
-import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{c as F}from"./chunk-IR3YJWPW.js";import{Vb as G}from"./chunk-M3FLEZ4E.js";import{l as $,m as z,n as H,o as V,p as R,q as U}from"./chunk-LV3CQ72P.js";import{b as q,d as P}from"./chunk-J572ANHX.js";import{qb as Q,x as h,z as T}from"./chunk-PBZCXPPC.js";import{n as E}from"./chunk-VVNV4NNW.js";import{A as i}from"./chunk-PF56N7ZY.js";import{a}from"./chunk-FTQLQJQV.js";async function K(e){let t=D(await e),[s,m,{QuickJSWASMModule:d}]=await Promise.all([t.importModuleLoader().then(D),t.importFFI(),import("./module-ES6BEMUI-ZVL3F6NW.js").then(D)]),r=await s();r.type="sync";let w=new m(r);return new d(r,w)}a(K,"newQuickJSWASMModuleFromVariant");function D(e){return e&&"default"in e&&e.default?e.default&&"default"in e.default&&e.default.default?e.default.default:e.default:e}a(D,"smartUnwrap");var k=class extends Error{static{a(this,"JavaScriptIsolateExecutionError")}constructor(t){super(t),this.name="JavaScriptIsolateExecutionError"}};async function O(e){te(e),e.signal?.throwIfAborted();let t=a(r=>new Error(`${e.errorPrefix}_${r}`),"error"),s=Date.now()+e.timeoutMs,{default:m}=await import("./dist-SKYG5LZO.js");e.signal?.throwIfAborted();let d=await K(m);e.signal?.throwIfAborted();try{return F.withScope(r=>{let w=r.manage(d.newRuntime());w.setMemoryLimit(33554432),w.setMaxStackSize(262144),w.setInterruptHandler(()=>Date.now()>=s||e.signal?.aborted===!0);let n=r.manage(w.newContext());e.deterministic&&r.manage(n.unwrapResult(n.evalCode("globalThis.Date = undefined; Math.random = () => { throw new Error('environment_random_requires_explicit_seed'); };")));let _=r.manage(n.unwrapResult(n.evalCode("JSON.stringify"))),o=r.manage(n.unwrapResult(n.evalCode(`JSON.parse(${JSON.stringify(JSON.stringify(e.value))})`))),p=r.manage(n.unwrapResult(n.evalCode(e.source,`${e.errorPrefix}.mjs`,{type:"module"}))),v=r.manage(X(n,p,s,e.signal,t)),x=r.manage(n.getProp(v,e.exportName));if(n.typeof(x)!=="function")throw t("function_export_missing");let M=r.manage(n.unwrapResult(n.callFunction(x,n.undefined,o))),b=r.manage(X(n,M,s,e.signal,t)),c=r.manage(n.unwrapResult(n.callFunction(_,n.undefined,b)));if(n.typeof(c)!=="string")throw t("result_not_json");let f=n.getString(c);if(new TextEncoder().encode(f).byteLength>e.maxResultBytes)throw t("result_too_large");if(e.signal?.throwIfAborted(),Date.now()>=s)throw t("timeout");let g=JSON.parse(f);return h(g,e.maxResultBytes),g})}catch(r){throw e.signal?.throwIfAborted(),new k(r instanceof Error?r.message:"JavaScript execution failed.")}}a(O,"executeJavaScriptIsolate");function te(e){if(!Number.isInteger(e.timeoutMs)||e.timeoutMs<1||e.timeoutMs>3e5)throw new Error(`${e.errorPrefix}_timeout_invalid`);if(!Number.isInteger(e.maxResultBytes)||e.maxResultBytes<1||e.maxResultBytes>4194304)throw new Error(`${e.errorPrefix}_result_limit_invalid`);if(new TextEncoder().encode(e.source).byteLength>524288)throw new Error(`${e.errorPrefix}_source_too_large`);h(e.value,4194304)}a(te,"validateJavaScriptIsolateInput");function X(e,t,s,m,d){for(;;){if(m?.throwIfAborted(),Date.now()>=s)throw d("timeout");let r=e.getPromiseState(t);if(r.type==="fulfilled")return r.notAPromise?r.value.dup():r.value;if(r.type==="rejected")return e.unwrapResult(r);if(e.runtime.executePendingJobs(1).unwrap()===0)throw d("unresolved_promise")}}a(X,"settle");function I(e,t){let s=t.scope==="arguments"?e.action?.arguments:e[t.scope];for(let m of t.path){if(!s||typeof s!="object"||!Object.hasOwn(s,m))throw new Error("environment_service_reference_missing");s=s[m]}return s}a(I,"resolveValue");async function Y(e,t){if(!e.executionServices?.length)return t.controller(e);if(!Number.isSafeInteger(e.timeoutMs)||e.timeoutMs<1||e.timeoutMs>3e4)throw new Error("environment_timeout_invalid");h(e.value,4194304);let s=e.executionServices.map(n=>z.parse(n));if(s.length>16)throw new Error("environment_service_limit");if(new Set(s.map(n=>n.id)).size!==s.length)throw new Error("environment_service_identity_collision");let m=Date.now(),d=new AbortController,r=e.signal?AbortSignal.any([e.signal,d.signal]):d.signal,w=setTimeout(()=>d.abort(new Error("environment_timeout")),e.timeoutMs);try{let n=Object.create(null);for(let o of s){if(r.throwIfAborted(),o.operation!==e.operation||o.operation==="step"&&o.toolName!==e.value.action?.name)throw new Error("environment_service_operation_mismatch");let p=Math.min(o.timeoutMs,e.timeoutMs-(Date.now()-m));if(p<1)throw new Error("environment_timeout");let v=new AbortController,x=AbortSignal.any([r,v.signal]),M=setTimeout(()=>v.abort(new Error("environment_service_timeout")),p),b=Date.now();try{if(o.kind==="sqlite.v1"){if(!t.sql)throw new Error("environment_sql_executor_required");let c=I(e.value,o.sql);if(typeof c!="string"||!c.length||new TextEncoder().encode(c).byteLength>65536||c.includes("\0")){n[o.id]={status:"rejected",code:"invalid_submission"};continue}let f=$({schemaVersion:"openpond.sqlExecution.v1",snapshot:I(e.value,o.snapshot),sql:c,maxRows:o.maxRows,maxResultBytes:o.maxResultBytes});n[o.id]=await t.sql({request:f,timeoutMs:e.timeoutMs,signal:x})}else{let c=I(e.value,o.source),f=I(e.value,o.cases);if(typeof c!="string"||new TextEncoder().encode(c).byteLength>524288){n[o.id]={status:"rejected",code:"invalid_submission"};continue}if(!Array.isArray(f)||!f.length||f.length>o.maxCases||f.some(u=>!u||typeof u!="object"||!Object.hasOwn(u,"input")))throw new Error("environment_service_cases_invalid");let g=[];for(let u of f){let l=p-(Date.now()-b);if(l<1&&(v.abort(new Error("environment_service_timeout")),x.throwIfAborted()),g.push(await t.candidate({source:c,exportName:o.exportName,value:structuredClone(u.input),timeoutMs:Math.max(1,l),maxResultBytes:o.maxResultBytes,deterministic:!0,errorPrefix:"candidate",signal:x})),new TextEncoder().encode(JSON.stringify({status:"completed",cases:g})).byteLength>o.maxResultBytes)throw new k("candidate_result_too_large")}n[o.id]={status:"completed",cases:g}}}catch(c){if(r.throwIfAborted(),v.signal.aborted)n[o.id]={status:"rejected",code:"timeout"};else if(c instanceof k)n[o.id]={status:"rejected",code:"execution_failed"};else throw c}finally{clearTimeout(M)}}r.throwIfAborted();let _=e.timeoutMs-(Date.now()-m);if(_<1)throw new Error("environment_timeout");return await t.controller({...e,executionServices:void 0,value:{...e.value,services:n},timeoutMs:_,signal:r})}finally{clearTimeout(w)}}a(Y,"executeEnvironmentWithServices");async function re(e){let t=V.parse(e.operation);return Y(e,{candidate:O,controller:a(async s=>R.parse(await O({...s,exportName:t,maxResultBytes:1572864,deterministic:!0,errorPrefix:"environment"})),"controller")})}a(re,"executeJavaScriptEnvironment");var J=class extends Error{static{a(this,"JavaScriptEnvironmentActionError")}code;constructor(t){super(`environment_${t}`),this.code=t,this.name="JavaScriptEnvironmentActionError"}};async function Qe(e){let t=U(e.definition),s=P(e.asset,t.module);if(!Number.isSafeInteger(e.seed))throw new Error("environment_seed_invalid");let m=e.seed;h(e.input,1048576),h(e.initialState,t.maxStateBytes);let d=structuredClone(e.input),r=structuredClone(e.initialState),w=e.execute??re,n=new AbortController,_=e.signal?AbortSignal.any([e.signal,n.signal]):n.signal,o={},p=0,v=!1,x=!1,M=null,b=[],c=null;async function f(u,l){if(v||x)throw new Error("environment_closed");if(M)throw new Error("environment_operation_in_progress");_.throwIfAborted(),l&&(h(l,262144),l=structuredClone(l));let L=E(o),y=l?t.tools.find(S=>S.name===l.name):null;if(u==="step"){if(p>=t.maxSteps)throw new J("step_budget_exhausted");p++}let j=a((S,C)=>b.push({sequence:b.length,operation:u,toolName:l?.name??null,arguments:l?structuredClone(l.arguments):null,observation:S,beforeStateHash:L,afterStateHash:E(o),error:C}),"record");try{if(l&&!y)throw new J("unknown_tool");if(l&&y&&!T(y.inputSchema,l.arguments).valid)throw new J("invalid_arguments");let S=t.executionServices?.filter(N=>N.operation===u&&(u!=="step"||N.toolName===l?.name)),C=w({source:s,operation:u,value:{input:d,initialState:r,seed:m,state:structuredClone(o),sequence:p,action:l?structuredClone(l):null},timeoutMs:Math.min(t.operationTimeoutMs,y?.timeoutMs??t.operationTimeoutMs),signal:_,...S?.length?{executionServices:S}:{}});M=C;let A=R.parse(await C);if(_.throwIfAborted(),h(A.state,t.maxStateBytes),h(A.observation,t.maxObservationBytes),(u==="collect"||y?.sideEffect==="read")&&E(A.state)!==L)throw new Error("environment_read_changed_state");o=structuredClone(A.state);let B=structuredClone(A.observation);return j(B,null),{state:structuredClone(o),observation:B}}catch(S){throw j(null,S instanceof Error?S.message:"environment_execution_failed"),S}finally{M=null}}a(f,"invoke");async function g(){return c||(x=!0,n.abort(new Error("environment_destroyed")),c=(async()=>{try{try{await M}catch{}await w({source:s,operation:"destroy",value:{input:d,initialState:r,seed:m,state:structuredClone(o),sequence:p,action:null},timeoutMs:t.operationTimeoutMs})}finally{o={},v=!0}})(),c)}a(g,"destroy");try{await f("create");let u=await f("reset"),l=E(o),L=a(()=>{if(v)throw new Error("environment_closed");return{definition:{id:t.id,revision:t.revision,contentHash:t.contentHash},inputHash:E(d),seed:m,initialStateHash:l,finalStateHash:E(o),state:structuredClone(o),events:structuredClone(b)}},"snapshot");return{definition:structuredClone(t),observation:u.observation,async step(y){return(await f("step",y)).observation},async collect(){let y=await f("collect");return{...L(),observation:y.observation}},snapshot:L,destroy:g}}catch(u){try{await g()}catch{}throw u}}a(Qe,"createJavaScriptEnvironmentSession");var ne=i.string().trim().min(1).max(500),oe=i.string().regex(/^[a-f0-9]{64}$/),W=i.object({kind:i.literal("model"),id:i.string().trim().min(1).max(191),taskId:ne,admissionHash:oe,model:G,instructions:i.string().max(262144),harness:Q.optional(),input:i.record(i.string(),i.unknown()),policyVisibleContext:i.record(i.string(),i.unknown()),timeoutMs:i.number().int().positive().max(36e5),environment:i.discriminatedUnion("kind",[i.object({kind:i.literal("text")}).strict(),i.object({kind:i.literal("javascript"),definition:H,asset:q,initialState:i.record(i.string(),i.unknown()),seed:i.number().int().safe()}).strict()])}).strict(),Ye=i.discriminatedUnion("kind",[W,i.object({kind:i.literal("profile"),request:i.unknown()}).strict()]);var Z=i.string().trim().min(1).max(200),mt=i.object({receiptId:Z,taskId:Z,seed:i.string(),request:W}).strict(),ee=class extends Error{constructor(s,m,d=409){super(m);this.code=s;this.status=d}code;status;static{a(this,"LocalExperimentError")}};export{k as a,te as b,Y as c,J as d,Qe as e,W as f,Ye as g,ee as h};

package/dist/chunks/chunk-GS6ZUSIN.js

@@ -1,3 +0,0 @@
-import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{a as le,b as Ve,c as De}from"./chunk-2ITZW2CN.js";import{k as re,q as oe,t as se}from"./chunk-OG33QAG6.js";import{d as ie,f as Pe,g as Ee,h as He,i as ze,l as B,n as Oe}from"./chunk-V7KR74NO.js";import{a as Me}from"./chunk-7O6Z3VOZ.js";import{i as D}from"./chunk-5EP2DZLJ.js";import{B as Ae,k as te,m as ne,oa as Ce,ta as we,wa as ae}from"./chunk-YNQWJWIC.js";import{A as Te}from"./chunk-7ZLXFKRI.js";import{h as je}from"./chunk-ABF3W2MO.js";import{f as V}from"./chunk-OE22TAXP.js";import{c as n,d as o,e as c,g as r,h as g}from"./chunk-VVNV4NNW.js";import{A as e}from"./chunk-PF56N7ZY.js";import{a as v}from"./chunk-FTQLQJQV.js";var L=e.string().trim().min(1).max(240),N=e.string().trim().min(1),me=e.string().trim().min(8).max(256),Be=e.record(e.string(),e.unknown()).default({}),At=e.enum(["created","inspecting","awaiting_source_review","awaiting_mapping","awaiting_materialization_approval","materializing","validating","ready","cancelling","cancelled","failed"]),Le=e.object({schemaVersion:e.literal("openpond.huggingFaceDatasetLocator.v1"),repositoryId:e.string().trim().regex(/^[^/\s]+\/[^/\s]+$/).max(500),repositoryUrl:e.string().url().max(2e3),requestedRevision:e.string().trim().min(1).max(256).nullable()}),Ct=e.object({path:e.string().trim().min(1).max(1e3),logicalType:e.string().trim().min(1).max(500),nullable:e.boolean(),sampleValues:e.array(e.unknown()).max(5)}),wt=e.object({configuration:e.string().trim().min(1).max(500),split:e.string().trim().min(1).max(500),rowCount:e.number().int().nonnegative().nullable(),sizeBytes:e.number().int().nonnegative().nullable()}),Mt=e.object({schemaVersion:e.literal("openpond.huggingFaceDatasetInspection.v1"),id:L,locator:Le,resolvedRevision:e.string().trim().min(7).max(256),title:e.string().trim().min(1).max(500),description:e.string().max(2e4).nullable(),declaredLicense:e.string().trim().min(1).max(500).nullable(),gated:e.boolean(),private:e.boolean(),configurations:e.array(e.string().trim().min(1).max(500)).min(1).max(1e3),splits:e.array(wt).min(1).max(1e4),columns:e.array(Ct).min(1).max(1e4),previewRows:e.array(e.record(e.string(),e.unknown())).max(25),sourceFiles:e.array(e.object({path:e.string().trim().min(1).max(2e3),configuration:e.string().trim().min(1).max(500),split:e.string().trim().min(1).max(500),revision:e.string().trim().min(7).max(256),sizeBytes:e.number().int().nonnegative().nullable(),contentHash:me.nullable()})).max(1e5),inspectedAt:N,metadata:Be}),Pt=e.enum(["row_id","cluster_id","prompt","messages","demonstration","chosen","rejected","expected_output","privileged_context","reward","feedback","tag","metadata"]),Et=e.object({sourcePath:e.string().trim().min(1).max(1e3),target:Pt,transform:e.enum(["identity","string","json","messages","numeric","math_final_answer"]),policy:e.enum(["visible","privileged","metadata"]),required:e.boolean()}),qe=e.object({schemaVersion:e.literal("openpond.datasetImportMapping.v1"),sourceSchemaHash:me,configuration:e.string().trim().min(1).max(500),upstreamSplits:e.array(e.string().trim().min(1).max(500)).min(1).max(1e3),preset:e.enum(["prompt_only","prompt_completion","messages","preference","prompt_expected_answer","prompt_privileged_solution","prompt_feedback"]),bindings:e.array(Et).min(1).max(1e4),nullPolicy:e.enum(["reject","drop","preserve"]),invalidRowPolicy:e.enum(["reject_import","drop_with_receipt"]),splitPolicy:e.object({seed:e.number().int(),assignments:e.record(e.string(),te),validationPercent:e.number().min(0).max(100),frozenEvalPercent:e.number().min(0).max(100)}),importerVersion:e.string().trim().min(1).max(200),mappingHash:me}),Ht=e.object({phase:e.enum(["idle","metadata","download","mapping","parquet_write","verification","publish"]),completedBytes:e.number().int().nonnegative(),totalBytes:e.number().int().nonnegative().nullable(),completedRows:e.number().int().nonnegative(),totalRows:e.number().int().nonnegative().nullable(),message:e.string().trim().min(1).max(2e3).nullable()}),Ue=e.object({schemaVersion:e.literal("openpond.datasetImportJob.v1"),id:L,profileId:L,sourceKind:e.enum(["huggingface","uploaded_file"]),status:At,locator:Le.nullable(),inspection:Mt.nullable(),mapping:qe.nullable(),progress:Ht,targetStorageRoot:e.string().trim().min(1).max(4e3).nullable(),tasksetId:L.nullable(),tasksetRevision:e.number().int().positive().nullable(),artifactId:L.nullable(),error:e.string().trim().min(1).max(2e4).nullable(),cancellationRequested:e.boolean(),createdAt:N,updatedAt:N,completedAt:N.nullable(),metadata:Be}),ya=e.object({profileId:L,url:e.string().trim().min(1).max(2e3)}),_a=e.object({name:e.string().trim().min(1).max(500),objective:e.string().trim().min(1).max(2e4),mapping:qe,targetStorageRoot:e.string().trim().min(1).max(4e3).nullable(),licenseApproved:e.boolean().default(!1)});var y=e.string().trim().min(1).max(240),w=e.string().trim().min(1),Ne=e.record(e.string(),e.unknown()).default({}),$e=e.enum(["nominated","needs_review","approved_for_creation","creating","needs_task_review","baselining","ready_for_export","rejected","dismissed","retired","blocked"]),zt=e.object({id:y,kind:e.enum(["repeated_success","accepted_correction","agent_action_recurrence","outcome_linked_change","frontier_cost","expert_label","runtime_feedback"]),sourceRefIds:e.array(y).min(1).max(1e4),occurredAt:w,signature:e.string().trim().min(1).max(2e3),summary:e.string().trim().min(1).max(5e3),confidence:e.number().min(0).max(1),consented:e.boolean(),metadata:Ne}),Ot=e.object({frequency:e.number().min(0).max(1),businessValue:e.number().min(0).max(1),frontierCost:e.number().min(0).max(1),signalQuality:e.number().min(0).max(1),verifiability:e.number().min(0).max(1),repeatability:e.number().min(0).max(1),privacyRisk:e.number().min(0).max(1),overall:e.number().min(0).max(1)}),Vt=e.object({tactic:Me,eligible:e.boolean(),reasons:e.array(e.string().trim().min(1).max(5e3)).min(1).max(100),blockers:e.array(e.string().trim().min(1).max(5e3)).max(100).default([]),requiredSignals:e.array(e.string().trim().min(1).max(500)).max(100).default([]),generatedBy:e.enum(["decision_table","baseline_reassessment"])}),ce=e.object({schemaVersion:e.literal("openpond.taskCandidate.v1"),id:y,profileId:y,status:$e,fingerprint:e.string().trim().min(1).max(512),title:e.string().trim().min(1).max(500),summary:e.string().trim().min(1).max(1e4),workflowSignature:e.string().trim().min(1).max(2e3),evidence:e.array(zt).min(1).max(1e5),scorecard:Ot,recommendation:Vt,mergedIntoId:y.nullable(),createdAt:w,updatedAt:w,metadata:Ne}),q=e.object({schemaVersion:e.literal("openpond.taskMinerConfig.v1"),enabled:e.boolean().default(!1),localOnly:e.boolean().default(!0),observationWindowDays:e.number().int().min(1).max(365).default(30),minimumRecurrence:e.number().int().min(2).max(100).default(3),clustering:e.literal("hybrid_deterministic_first").default("hybrid_deterministic_first"),consentRequired:e.boolean().default(!0)}),ja=e.object({items:e.array(ce),generatedAt:w,config:q}),Ta=e.object({status:$e.optional(),mergeIntoId:y.nullable().optional()}),Aa=e.object({profileId:y,sourceIds:e.array(y).max(1e5).default([]),sessionIds:e.array(y).max(1e5).default([]),config:q.optional()}),Fe=e.object({schemaVersion:e.literal("openpond.taskMinerRun.v1"),id:y,profileId:y,status:e.enum(["queued","running","cancelling","cancelled","succeeded","failed"]),config:q,sourceIds:e.array(y).max(1e5),sessionIds:e.array(y).max(1e5).default([]),progress:e.object({stage:e.enum(["queued","ingesting","preparing","clustering","persisting","complete"]),processedSources:e.number().int().nonnegative(),totalSources:e.number().int().nonnegative(),candidatesFound:e.number().int().nonnegative(),skippedSources:e.number().int().nonnegative().default(0)}),candidateIds:e.array(y).max(1e5),cancelRequested:e.boolean(),error:e.string().trim().min(1).max(2e4).nullable(),createdAt:w,startedAt:w.nullable(),completedAt:w.nullable(),updatedAt:w});var Dt=e.enum(["draft","active","completed","archived"]),Bt=e.enum(["draft","ready","queued","running","candidate","accepted","rejected","no_signal","failed","cancelled"]),Ge=e.enum(["seed","daily_residual","rank_candidate","weekly_rollup","full_refresh"]),Lt=e.enum(["base_model","previous_release","seed_release","accepted_daily_head","accepted_seed"]),qt=e.enum(["seed_taskset","nightly_selection","daily_cohort_union","eligible_task_pool"]),Ut=e.enum(["live_client","evaluation_attempt","replay_evidence","manual"]),Nt=e.object({id:n,ordinal:e.number().int().nonnegative(),label:e.string().trim().min(1).max(100),role:Ge,parentRule:Lt,taskSource:qt,trainableRank:e.number().int().positive().max(1024),minimumTasks:e.number().int().nonnegative().max(1e5),maximumTasks:e.number().int().positive().max(1e5)}).strict().superRefine((t,i)=>{t.minimumTasks>t.maximumTasks&&i.addIssue({code:"custom",path:["minimumTasks"],message:"The minimum task count cannot exceed the maximum task count."});let s={seed:{parentRules:["base_model"],taskSource:"seed_taskset"},daily_residual:{parentRules:["previous_release","accepted_daily_head"],taskSource:"nightly_selection"},rank_candidate:{parentRules:["seed_release"],taskSource:"eligible_task_pool"},weekly_rollup:{parentRules:["seed_release","accepted_seed"],taskSource:"daily_cohort_union"},full_refresh:{parentRules:["base_model"],taskSource:"eligible_task_pool"}},{parentRules:l,taskSource:f}=s[t.role];l.includes(t.parentRule)||i.addIssue({code:"custom",path:["parentRule"],message:`${t.role} requires parent rule ${l.join(" or ")}.`}),t.taskSource!==f&&i.addIssue({code:"custom",path:["taskSource"],message:`${t.role} requires task source ${f}.`}),t.ordinal===0&&t.role!=="seed"&&i.addIssue({code:"custom",path:["role"],message:"The first series entry must be the seed."})}),$t=e.object({id:n,version:e.number().int().positive(),requireCheckpoint:e.boolean(),requireAppliedOptimizerUpdate:e.boolean(),minimumCurrentCohortMeanImprovement:e.number().finite(),maximumRetainedMeanRegression:e.number().finite().nonnegative(),blockCriticalInvariantRegression:e.boolean(),automaticDailyAdvancement:e.boolean()}).strict(),Je=e.object({schemaVersion:e.literal("openpond.modelComparisonSeries.v1"),id:n,profileId:n,modelProjectId:n,name:e.string().trim().min(1).max(500),objective:e.string().trim().min(1).max(2e4),status:Dt,revision:e.number().int().positive(),productionBinding:e.object({role:e.literal("chat_manual"),roleTargetId:n}).strict(),baseModel:e.object({id:n,revision:e.string().trim().min(1).max(256)}).strict(),seedTaskset:g,eligibleTaskPool:g,evaluationTasksets:e.object({development:g,retained:g,frozenFinal:g}).strict(),grader:r,benchmarkProtocol:De.nullable().default(null),automaticEvaluation:e.object({enabled:e.boolean()}).strict().default({enabled:!1}),residualProfile:e.object({profileId:n,serializedEnvelopeRank:e.number().int().positive().max(1024),maximumEnabledRank:e.number().int().positive().max(1024),topology:e.literal("uniform_block_masked")}).strict(),schedule:e.array(Nt).min(1).max(1e3),scheduleSealedAt:c.nullable(),advancementPolicy:$t,executionPolicy:e.object({startWhenReady:e.literal(!1)}).strict(),acceptedSeedEntryId:n.nullable(),acceptedDailyHeadEntryId:n.nullable(),promotedBindingId:n.nullable(),createdBy:n,createdAt:c,updatedAt:c}).strict().superRefine((t,i)=>{[...t.schedule].sort((u,j)=>u.ordinal-j.ordinal).some((u,j)=>u.ordinal!==j)&&i.addIssue({code:"custom",path:["schedule"],message:"Schedule ordinals must be unique and contiguous from zero."}),(new Set(t.schedule.map(u=>u.id)).size!==t.schedule.length||new Set(t.schedule.map(u=>u.label)).size!==t.schedule.length)&&i.addIssue({code:"custom",path:["schedule"],message:"Schedule ids and labels must be unique."}),t.residualProfile.maximumEnabledRank>t.residualProfile.serializedEnvelopeRank&&i.addIssue({code:"custom",path:["residualProfile"],message:"Maximum enabled rank cannot exceed the serialized envelope."});let l=Object.values(t.evaluationTasksets);new Set(l.map(u=>`${u.id}:${u.revision}:${u.contentHash}`)).size!==l.length&&i.addIssue({code:"custom",path:["evaluationTasksets"],message:"Development, retained, and frozen-final cohorts require distinct immutable Taskset releases."});let f=t.schedule.find(u=>u.role==="seed");if(!f||t.schedule.filter(u=>u.role==="seed").length!==1){i.addIssue({code:"custom",path:["schedule"],message:"A series requires exactly one seed entry."});return}let h=t.schedule.filter(u=>u.role==="rank_candidate");h.length===1&&i.addIssue({code:"custom",path:["schedule"],message:"A causal rank comparison requires at least two rank candidates."}),new Set(h.map(u=>u.trainableRank)).size!==h.length&&i.addIssue({code:"custom",path:["schedule"],message:"Causal rank candidates must use distinct trainable ranks."}),[f.trainableRank+t.schedule.filter(u=>u.role==="daily_residual").reduce((u,j)=>u+j.trainableRank,0),...t.schedule.filter(u=>u.role==="weekly_rollup").map(u=>f.trainableRank+u.trainableRank),...t.schedule.filter(u=>u.role==="rank_candidate").map(u=>f.trainableRank+u.trainableRank),...t.schedule.filter(u=>u.role==="full_refresh").map(u=>u.trainableRank)].some(u=>u>t.residualProfile.maximumEnabledRank)&&i.addIssue({code:"custom",path:["schedule"],message:"A declared branch exceeds the maximum enabled rank."})}),We=e.object({source:Ut,taskIds:e.array(n).min(1).max(1e5),observedFrom:c,observedTo:c,reviewedAt:c,reviewedBy:n,sourceTaskset:g}).strict(),Ft=We.extend({derivedFamilyKeys:e.array(n).min(1).max(1e5),familyDerivation:e.object({algorithm:n,sourceTasksetHash:o}).strict(),selectionHash:o}).strict(),ue=e.discriminatedUnion("kind",[e.object({kind:e.literal("base_model"),id:n,revision:e.string().trim().min(1).max(256)}).strict(),e.object({kind:e.literal("model_version"),id:n,contentHash:o}).strict()]),Gt=e.object({id:n,branchOrdinal:e.number().int().nonnegative(),rank:e.number().int().positive().max(1024),offsetStart:e.number().int().nonnegative(),offsetEnd:e.number().int().positive(),optimizationRole:e.enum(["frozen","trainable"]),artifactLineageId:n.nullable()}).strict().superRefine((t,i)=>{t.offsetEnd-t.offsetStart!==t.rank&&i.addIssue({code:"custom",path:["offsetEnd"],message:"Block offsets must span exactly the declared rank."})}),Jt=e.object({disposition:e.enum(["advance","hold","no_signal"]),policy:e.object({id:n,version:e.number().int().positive()}).strict(),reasonCodes:e.array(n).min(1).max(100),summary:e.string().trim().min(1).max(5e3),decidedBy:n,decidedAt:c}).strict(),Wt=e.object({evaluationRunId:n,modelVersionId:n,taskset:g,grader:r,cohortRole:e.enum(["current","correction","sibling_verification","cumulative_known","development","retained","prior_disclosed","frozen_final"]),panelId:n.nullable().default(null),protocol:e.object({id:n,revision:e.number().int().positive(),contentHash:o}).strict().nullable().default(null)}).strict(),Yt=e.object({attemptOrdinal:e.number().int().positive(),trainingPlanId:n.nullable(),modelRunId:n.nullable(),terminalStatus:e.enum(["failed","cancelled"]),queuedAt:c.nullable(),startedAt:c.nullable(),completedAt:c}).strict(),A=e.object({seriesId:n,entryId:n,scheduleEntryId:n,ordinal:e.number().int().nonnegative(),releaseHash:o}).strict(),Ye=e.object({schemaVersion:e.literal("openpond.modelComparisonSeriesEntry.v1"),id:n,seriesId:n,profileId:n,modelProjectId:n,scheduleEntryId:n,releaseHash:o,ordinal:e.number().int().nonnegative(),label:e.string().trim().min(1).max(100),role:Ge,branch:e.enum(["daily","rank_candidate","weekly_rollup","full_refresh"]),status:Bt,parent:ue,taskset:g,sourceTasksets:e.array(g).min(1).max(1e3),taskSelection:Ft.nullable(),trainableRank:e.number().int().positive().max(1024),serializedEnvelopeRank:e.number().int().positive().max(1024),enabledCumulativeRank:e.number().int().positive().max(1024),trainableBlockId:n,residualBlocks:e.array(Gt).min(1).max(1024),attemptOrdinal:e.number().int().positive().default(1),priorRunAttempts:e.array(Yt).max(100).default([]),trainingPlanId:n.nullable(),modelRunId:n.nullable(),modelVersionId:n.nullable(),evaluations:e.array(Wt).max(1e5),decision:Jt.nullable(),promotionBindingId:n.nullable(),queuedAt:c.nullable(),startedAt:c.nullable(),completedAt:c.nullable(),createdAt:c,updatedAt:c}).strict().superRefine((t,i)=>{let s=[...t.residualBlocks].sort((x,k)=>x.offsetStart-k.offsetStart);s.some((x,k)=>x.offsetStart!==(k===0?0:s[k-1].offsetEnd))&&i.addIssue({code:"custom",path:["residualBlocks"],message:"Residual block offsets must be contiguous from zero."});let l=s.at(-1)?.offsetEnd??0;(l!==t.enabledCumulativeRank||l>t.serializedEnvelopeRank)&&i.addIssue({code:"custom",path:["enabledCumulativeRank"],message:"Enabled rank must match the block inventory and fit the envelope."});let f=t.residualBlocks.filter(x=>x.optimizationRole==="trainable");(f.length!==1||f[0]?.id!==t.trainableBlockId||f[0]?.rank!==t.trainableRank)&&i.addIssue({code:"custom",path:["trainableBlockId"],message:"Exactly one block must be trainable during the Run and match trainableRank."});let h=t.priorRunAttempts.map(x=>x.attemptOrdinal);(t.priorRunAttempts.length!==t.attemptOrdinal-1||h.some((x,k)=>x!==k+1))&&i.addIssue({code:"custom",path:["priorRunAttempts"],message:"Prior Run attempts must be a contiguous history before the active attempt."})}),Ea=e.object({seriesId:n,scheduleEntryId:n,taskSelection:We.nullable(),expectedSeriesRevision:e.number().int().positive()}).strict();var M=n.nullable(),C=e.number().int().nonnegative(),Kt=e.string().regex(/^[a-f0-9]{40}$/),Qt=e.object({name:e.string().trim().min(1).max(160),startedAt:c,completedAt:c,durationMs:e.number().nonnegative(),clock:e.enum(["monotonic","provider","wall"]),outcome:e.enum(["succeeded","failed","cancelled"])}).strict(),Ke=e.object({schemaVersion:e.literal("openpond.correlatedTelemetryReceipt.v1"),stage:e.enum(["training","evaluation","deployment","inference"]),correlation:e.object({modelRunId:M,modelVersionId:M,policyVersion:C.nullable(),taskId:M,rolloutGroupId:M,providerResourceId:M,deploymentId:M,inferenceRequestId:M}).strict(),spans:e.array(Qt).max(1e4),usage:e.object({promptTokens:C.nullable(),generatedTokens:C.nullable(),gpuSeconds:e.number().nonnegative().nullable(),workerActiveSeconds:e.number().nonnegative().nullable(),optimizerSteps:C.nullable(),rolloutGroups:C.nullable(),successfulTrajectories:C.nullable(),failedTrajectories:C.nullable(),peakGpuMemoryBytes:C.nullable(),peakGpuUtilizationPercent:e.number().min(0).max(100).nullable()}).strict(),resource:e.object({provider:n,resourceIds:e.array(n).max(1e3),gpuType:e.string().trim().min(1).max(300).nullable(),gpuCount:C.nullable(),baseProfileId:M,baseRepository:e.string().trim().min(1).max(512).nullable(),baseRevision:Kt.nullable(),adapterContentHash:o.nullable()}).strict(),cost:e.object({currency:e.literal("USD"),providerReportedUsd:e.number().nonnegative().nullable(),quotedHourlyUsd:e.number().nonnegative().nullable(),estimatedUsd:e.number().nonnegative().nullable(),methodologyVersion:e.string().trim().min(1).max(160),pricingInputs:e.record(e.string().trim().min(1).max(160),e.number().finite()),unitEstimates:e.record(e.string().trim().min(1).max(160),e.number().nonnegative())}).strict(),recordedAt:c,contentHash:o}).strict();var Zt=e.enum(["base_reference","lora_adapter"]),de=e.object({baseModel:e.object({source:e.literal("huggingface"),repoId:n,revision:e.string().regex(/^[a-f0-9]{40}$/),configHash:o,tokenizerHash:o,licenseId:n,gated:e.boolean()}).strict(),processor:e.object({repository:n,revision:e.string().regex(/^[a-f0-9]{40}$/),configHash:o}).strict()}).strict().superRefine((t,i)=>{(t.baseModel.repoId!==t.processor.repository||t.baseModel.revision!==t.processor.revision)&&i.addIssue({code:"custom",path:["processor"],message:"Reward Model processor must match the pinned scorer base model."})}),et=e.object({schemaVersion:e.literal("openpond.modelVersion.v1"),id:n,modelId:n,profileId:n,version:e.number().int().nonnegative(),kind:Zt,status:e.enum(["available","failed"]),baseModel:B,taskset:g,comparisonSeriesEntry:A.nullable().optional(),evaluationTasksetRef:g.nullable().optional(),releaseGraph:e.object({resolvedBundleHash:o,profileRelease:g,harnessRelease:r,agentRelease:r.nullable(),grader:e.object({id:n,contentHash:o}).strict()}).strict(),artifactLineageId:n.nullable(),adapterStatus:e.enum(["not_trained","trained"]),createdAt:c,contentHash:o}).strict().superRefine((t,i)=>{t.kind==="base_reference"&&(t.version!==0||t.adapterStatus!=="not_trained"||t.artifactLineageId!==null)&&i.addIssue({code:"custom",message:"A base-reference Model Version must be version 0 without adapter lineage."}),t.kind==="lora_adapter"&&(t.version===0||t.adapterStatus!=="trained"||!t.artifactLineageId)&&i.addIssue({code:"custom",message:"A LoRA Model Version must be version 1+ with adapter lineage."})}),tt=e.object({schemaVersion:e.literal("openpond.rewardModelVersion.v1"),id:n,modelId:n,profileId:n,version:e.number().int().positive(),role:e.literal("reward"),status:e.enum(["available","failed"]),scope:e.enum(["synthetic_smoke","human_preference"]),baseModel:B,runtime:de.nullable().default(null),taskset:g,preferenceDatasetRelease:r,releaseGraph:e.object({resolvedBundleHash:o,profileRelease:g,harnessRelease:r,grader:e.object({id:n,contentHash:o}).strict()}).strict(),artifacts:e.object({checkpoint:r.extend({objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),files:e.array(e.object({path:e.string().trim().min(1).max(512),sizeBytes:e.number().int().positive(),sha256:o}).strict()).min(4).max(128)}).strict(),adapter:r,scalarHead:r,bucketHead:r.nullable(),processorRelease:r}).strict(),qualificationReport:r.nullable(),createdAt:c,contentHash:o}).strict(),Xt=e.object({schemaVersion:e.literal("openpond.rewardModelRunReceipt.v1"),provider:n,providerRunId:n,resolvedBundleHash:o,finalCheckpoint:r.extend({objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/)}).strict(),adapter:r,scalarHead:r,bucketHead:r.nullable(),processorRelease:r,optimizerEvidence:r,managedExecutionReceipt:r.nullable().default(null),parameterDeltaHash:o,cleanup:e.object({computeReleased:e.boolean(),providerTerminalObserved:e.boolean()}).strict(),contentHash:o}).strict(),nt=e.preprocess(t=>{if(!t||typeof t!="object"||Array.isArray(t))return t;let i=t;if(i.tasksetRelease!==void 0)return t;let s=i.taskset;if(!s||typeof s!="object"||Array.isArray(s))return t;let l=s;return typeof l.id!="string"||typeof l.contentHash!="string"?t:{...i,tasksetRelease:{id:l.id,contentHash:l.contentHash}}},e.object({schemaVersion:e.literal("openpond.rewardModelRun.v1"),id:n,rewardModelId:n,rewardModelVersionId:n.nullable(),profileId:n,role:e.literal("reward"),scope:e.enum(["synthetic_smoke","human_preference"]),status:e.enum(["prepared","running","succeeded","failed","cancelled"]),taskset:g,tasksetRelease:r,preferenceDatasetRelease:r,recipeRelease:r,destinationId:n,quote:e.object({maximumSpendUsd:e.number().positive(),hourlyCostUsd:e.number().nonnegative().nullable()}).strict(),managedRunId:n.nullable(),progress:e.object({completedSteps:e.number().int().nonnegative(),totalSteps:e.number().int().positive(),latestLoss:e.number().finite().nullable()}).strict(),receipt:Xt.nullable(),qualificationReport:r.nullable(),accruedSpendUsd:e.number().nonnegative().nullable(),failureOwner:e.enum(["authoring","admission","provider","runner","artifact","qualification","cleanup"]).nullable(),failure:e.string().trim().min(1).max(5e3).nullable(),startedAt:c,completedAt:c.nullable(),updatedAt:c}).strict()).superRefine((t,i)=>{t.status==="succeeded"&&!t.receipt&&i.addIssue({code:"custom",message:"A successful Reward Model Run requires a canonical receipt."}),t.status!=="succeeded"&&t.rewardModelVersionId!==null&&i.addIssue({code:"custom",message:"Only a successful Reward Model Run can publish a Reward Model Version."})}),en=e.object({schemaVersion:e.literal("openpond.modelRunReceipt.v1"),provider:n,providerRunId:n,assignmentHash:o,resultHash:o,transcriptHash:o,traceHash:o.nullable(),resolvedBundleHash:o,artifactPath:e.string().trim().min(1).max(2e3),cleanup:e.object({computeReleased:e.boolean(),tunnelClosed:e.boolean()}).strict(),telemetry:Ke.nullable().default(null),contentHash:o}).strict(),P=e.object({inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),totalTokens:e.number().int().nonnegative(),costUsd:e.number().nonnegative().nullable()}).strict(),Qe=e.string().regex(/^(?:[a-f0-9]{40}|[a-f0-9]{64})$/),pe=e.object({phase:e.enum(["baseline","adaptation","candidate_adaptation","candidate"]),taskId:n,attemptId:n,sessionId:n.nullable(),turnId:n.nullable(),passed:e.boolean(),score:e.number().min(0).max(1).nullable(),failureClass:ie.nullable(),inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),totalTokens:e.number().int().nonnegative(),latencyMs:e.number().int().nonnegative(),costUsd:e.number().nonnegative().nullable(),startedAt:c}).strict(),Ze=e.object({targetTaskCount:e.number().int().nonnegative(),comparedTaskCount:e.number().int().nonnegative(),passedTaskCount:e.number().int().nonnegative(),failedTaskCount:e.number().int().nonnegative(),lowerTaskCount:e.number().int().nonnegative(),higherTaskCount:e.number().int().nonnegative(),unchangedTaskCount:e.number().int().nonnegative()}).strict(),at=e.object({target:e.literal("all_tasks_lower"),targetTaskCount:e.number().int().nonnegative(),comparedTaskCount:e.number().int().nonnegative(),passedTaskCount:e.number().int().nonnegative(),failedTaskCount:e.number().int().nonnegative(),lowerTaskCount:e.number().int().nonnegative(),higherTaskCount:e.number().int().nonnegative(),unchangedTaskCount:e.number().int().nonnegative(),baselineTokens:e.number().int().nonnegative(),refinedTokens:e.number().int().nonnegative(),tokenDelta:e.number().int(),tokenDeltaPercent:e.number().finite().nullable(),complete:e.boolean(),passed:e.boolean(),cohorts:e.object({adaptation:Ze,heldOut:Ze}).strict()}).strict(),be=e.object({baseline:P,adaptation:P,candidateAdaptation:P,candidate:P,refiner:P,grader:P}).strict(),tn=e.object({usage:be,observedSpendUsd:e.number().nonnegative(),attempts:e.array(pe).max(1e4)}).strict(),nn=e.object({id:n,contentHash:o,artifactPath:e.string().trim().min(1).max(2e3)}).strict(),an=e.object({schemaVersion:e.literal("openpond.modelEvaluationReceipt.v1"),benchmarkId:n,resultManifest:e.object({id:n,contentHash:o,artifactPath:e.string().trim().min(1).max(2e3)}).strict(),stages:e.object({baseline:r,adaptation:r,refiner:r,candidateAdaptation:r,candidate:r,comparison:r}).strict(),usage:be,quality:e.object({baselinePassRate:e.number().min(0).max(1),candidatePassRate:e.number().min(0).max(1),adaptationBaselinePassRate:e.number().min(0).max(1),adaptationCandidatePassRate:e.number().min(0).max(1),adaptationCandidatePassed:e.boolean(),heldOutCandidatePassed:e.boolean(),passed:e.boolean()}).strict(),foregroundTokenDelta:e.number().int(),foregroundTokenDeltaPercent:e.number().finite().nullable(),taskEfficiency:at.optional(),efficiency:e.object({grossForegroundTokenSavings:e.number().int(),overheadTokens:e.number().int().nonnegative(),firstPassNetTokenSavings:e.number().int(),breakEvenReuseCount:e.number().int().nonnegative().nullable(),amortizedTokenSavings:e.number().int(),amortizedReuseCount:e.number().int().positive()}).strict(),budget:e.object({maximumSpendUsd:e.number().nonnegative(),observedSpendUsd:e.number().nonnegative(),enforced:e.boolean()}).strict(),evidenceSnapshot:r,lineage:e.object({adaptationEvidenceHash:o,refinerInputHash:o,refinerOutcomeHash:o,validationHash:o,applyReceiptHash:o,candidateRelease:r,valid:e.boolean()}).strict(),invalidReasons:e.array(e.string().trim().min(1).max(2e3)).max(100),attempts:e.array(pe).max(1e4).default([]),terminalClassification:e.enum(["improved","no_improvement","regressed","inconclusive","cancelled","infrastructure_failure"]),profileGit:e.object({ref:e.string().trim().min(1).max(2e3),commit:Qe,baseCommit:Qe}).strict().nullable(),contentHash:o}).strict(),rn=e.object({schemaVersion:e.literal("openpond.modelEvaluationStopReceipt.v1"),benchmarkId:n,terminalClassification:e.literal("inconclusive"),stopReason:e.enum(["candidate_harness_unchanged","candidate_lineage_invalid"]),reason:e.string().trim().min(1).max(5e3),stoppedAfter:e.literal("refiner"),baselineHarness:r,candidateHarness:r,refiner:e.object({id:n,contentHash:o,outcomeCount:e.number().int().nonnegative()}).strict(),usage:be,budget:e.object({maximumSpendUsd:e.number().nonnegative(),observedSpendUsd:e.number().nonnegative(),enforced:e.boolean()}).strict(),evidenceSnapshot:r,attempts:e.array(pe).max(1e4),contentHash:o}).strict(),Xe=e.object({level:e.literal(.95),lower:e.number().finite(),upper:e.number().finite()}).strict().superRefine((t,i)=>{t.lower>t.upper&&i.addIssue({code:"custom",message:"A confidence interval's lower bound cannot exceed its upper bound."})}),on=e.object({schemaVersion:e.literal("openpond.modelComparisonBenchmarkReceipt.v1"),benchmarkId:e.literal("model-comparison"),target:e.object({kind:e.enum(["base_model","model_version","external_reference"]),label:e.string().trim().min(1).max(300),modelVersionId:n.nullable(),model:V.nullable()}).strict(),taskset:g,grader:r,sampling:e.object({seeds:e.array(e.number().int()).min(1).max(100),repetitions:e.number().int().positive().max(20)}).strict(),deterministic:e.object({attemptedTaskCount:e.number().int().nonnegative(),completedTaskCount:e.number().int().nonnegative(),passedTaskCount:e.number().int().nonnegative(),failedTaskCount:e.number().int().nonnegative(),meanScore:e.number().finite().nullable(),passRate:e.number().min(0).max(1).nullable(),passRateCi95:Xe.nullable()}).strict(),judge:e.object({judgeModel:V,judgeRelease:r,rubricRelease:r,calibrationRelease:r,score:e.number().min(0).max(100),scoreCi95:Xe,comparisonCount:e.number().int().positive(),wins:e.number().int().nonnegative(),ties:e.number().int().nonnegative(),losses:e.number().int().nonnegative(),referenceLabel:e.string().trim().min(1).max(300)}).strict().superRefine((t,i)=>{t.wins+t.ties+t.losses!==t.comparisonCount&&i.addIssue({code:"custom",message:"Judge win, tie, and loss counts must equal comparisonCount."})}).nullable(),attempts:e.array(e.object({attemptId:n.nullable().default(null),taskId:n,seed:e.number().int(),repetition:e.number().int().nonnegative(),status:e.enum(["succeeded","failed"]),deterministicScore:e.number().finite().nullable(),passed:e.boolean().nullable(),judgeScore:e.number().min(0).max(100).nullable(),judgePreference:e.enum(["win","tie","loss"]).nullable(),transcriptHash:o.nullable(),traceHash:o.nullable(),transcriptArtifact:e.object({artifactPath:e.string().trim().min(1).max(2e3),jsonPointer:e.string().trim().min(1).max(1e3)}).strict().nullable().default(null),traceArtifact:e.object({artifactPath:e.string().trim().min(1).max(2e3),jsonPointer:e.string().trim().min(1).max(1e3)}).strict().nullable().default(null),latencyMs:e.number().int().nonnegative().nullable().default(null),failureClass:ie.nullable()}).strict()).max(1e5),usage:e.object({policy:P.nullable(),judge:P.nullable(),observedSpendUsd:e.number().nonnegative().nullable(),evaluationGpuSeconds:e.number().nonnegative().nullable().default(null)}).strict(),managedServing:e.object({jobId:n,terminalState:e.enum(["completed","cancelled","failed","budget_exhausted"]),sourcePolicyVersion:e.number().int().nonnegative(),sourceAdapterSha256:o.nullable(),servedPolicyVersion:e.number().int().nonnegative().nullable(),servedAdapterSha256:o.nullable(),accruedSpendUsd:e.number().nonnegative(),cleanupAttestationHash:o,resourceCount:e.number().int().nonnegative(),activeResourceCount:e.literal(0)}).strict().nullable().default(null),evidenceSnapshot:e.object({id:n,contentHash:o,artifactPath:e.string().trim().min(1).max(2e3)}).strict(),completedAt:c,contentHash:o}).strict().superRefine((t,i)=>{let s=t.deterministic;s.completedTaskCount>s.attemptedTaskCount&&i.addIssue({code:"custom",message:"Completed task count cannot exceed attempted task count."}),s.passedTaskCount+s.failedTaskCount!==s.completedTaskCount&&i.addIssue({code:"custom",message:"Passed and failed task counts must equal completed task count."}),s.completedTaskCount===0!=(s.meanScore===null)&&i.addIssue({code:"custom",message:"Deterministic mean score is available exactly when tasks completed."}),s.completedTaskCount===0!=(s.passRate===null)&&i.addIssue({code:"custom",message:"Deterministic pass rate is available exactly when tasks completed."})}),sn=e.object({advancedEvaluation:re.optional(),benchmarkId:e.literal("harness-refiner"),model:V,upstreamModel:e.object({providerId:n,modelId:e.string().trim().min(1).max(300),revision:e.string().trim().min(1).max(300),pricing:e.object({version:e.string().trim().min(1).max(300),source:e.string().trim().min(1).max(300),effectiveAt:c,inputUsdPerMillionTokens:e.number().nonnegative(),cachedInputUsdPerMillionTokens:e.number().nonnegative(),outputUsdPerMillionTokens:e.number().nonnegative()}).strict().optional()}).strict(),reasoningEffort:e.string().trim().min(1).max(100).nullable(),seeds:e.array(e.number().int()).min(1).max(100),repetitions:e.number().int().positive().max(20),maximumSpendUsd:e.number().nonnegative(),attemptPlan:e.array(e.object({stage:e.enum(["baseline","adaptation","candidate_adaptation","candidate"]),split:e.string().trim().min(1).max(100),taskIds:e.array(n).min(1).max(1e5),attemptCount:e.number().int().positive()}).strict()).length(4)}).strict(),ln=e.object({benchmarkId:e.literal("model-comparison"),target:e.object({kind:e.enum(["base_model","model_version","external_reference"]),label:e.string().trim().min(1).max(300),modelVersionId:n.nullable(),model:V.nullable()}).strict(),grader:r,judge:e.object({model:V,release:r,rubricRelease:r,calibrationRelease:r,referenceLabel:e.string().trim().min(1).max(300)}).strict().nullable(),seeds:e.array(e.number().int()).min(1).max(100),repetitions:e.number().int().positive().max(20),maximumSpendUsd:e.number().nonnegative(),series:e.object({id:n,protocol:e.object({id:n,revision:e.number().int().positive(),contentHash:o}).strict()}).strict().nullable().default(null),panel:e.object({id:n,role:e.enum(["correction","sibling_verification","cumulative_known","development","retained","frozen_final"]),passLabel:n.nullable()}).strict().nullable().default(null),comparisonPair:e.object({entryId:n,parent:ue,candidateModelVersionId:n}).strict().nullable().default(null),attemptPlan:e.array(e.object({stage:e.literal("comparison"),split:e.string().trim().min(1).max(100),taskIds:e.array(n).min(1).max(1e5),attemptCount:e.number().int().positive()}).strict()).length(1)}).strict(),mn=e.union([sn,ln]),it=e.object({schemaVersion:e.literal("openpond.modelRun.v1"),id:n,modelId:n,modelVersionId:n.nullable(),profileId:n,kind:e.enum(["rollout_smoke","training","evaluation"]),status:e.enum(["prepared","running","succeeded","failed","cancelled"]),method:e.enum(["sft","dpo","grpo","ppo","sdft","opd","opsd","sdpo"]).nullable(),destinationId:n.nullable(),taskset:g,comparisonSeriesEntry:A.nullable().optional(),evaluationTasksetRef:g.nullable().optional(),harnessRelease:r.nullable().optional(),quote:e.object({maximumSpendUsd:e.number().nonnegative(),hourlyCostUsd:e.number().nonnegative().nullable()}).strict().nullable(),evaluation:mn.nullable().default(null),evaluationProgress:e.object({stage:e.enum(["baseline","adaptation","refiner","candidate_adaptation","candidate","comparison"]),completedAttempts:e.number().int().nonnegative(),totalAttempts:e.number().int().positive(),accounting:tn.nullable().default(null),advancedAccounting:se.optional(),evidenceSnapshot:nn.nullable().default(null)}).strict().nullable().default(null),reward:e.object({raw:e.number(),components:e.record(e.string(),e.number())}).strict().nullable(),receipt:e.union([en,an,rn,on,oe]).nullable(),adapterArtifactLineageId:n.nullable(),failure:e.string().trim().min(1).max(5e3).nullable(),startedAt:c,completedAt:c.nullable(),updatedAt:c}).strict().superRefine((t,i)=>{t.status==="succeeded"&&!t.receipt&&i.addIssue({code:"custom",message:"A successful Model Run requires a canonical receipt."}),t.kind==="evaluation"&&(t.method!==null||t.destinationId!==null||!t.evaluation||t.adapterArtifactLineageId!==null)&&i.addIssue({code:"custom",message:"An evaluation Model Run requires evaluation configuration and cannot claim training method, destination, or adapter lineage."}),t.kind==="evaluation"&&t.receipt&&t.receipt.schemaVersion!=="openpond.modelEvaluationReceipt.v1"&&t.receipt.schemaVersion!=="openpond.modelEvaluationStopReceipt.v1"&&t.receipt.schemaVersion!=="openpond.modelComparisonBenchmarkReceipt.v1"&&t.receipt.schemaVersion!=="openpond.advancedRefinerReviewReceipt.v1"&&i.addIssue({code:"custom",message:"An evaluation Model Run requires an evaluation receipt."}),t.kind!=="evaluation"&&(t.method===null||t.destinationId===null||t.quote===null)&&i.addIssue({code:"custom",message:"A training or rollout Model Run requires method, destination, and quote."}),t.kind==="rollout_smoke"&&t.adapterArtifactLineageId!==null&&i.addIssue({code:"custom",message:"A rollout smoke cannot claim adapter artifacts."})});function ai(t){let i=new Map;for(let p of t.attempts){let T=`${p.phase}\0${p.taskId}`,S=i.get(T);(!S||p.startedAt>S.startedAt||p.startedAt===S.startedAt&&p.attemptId>S.attemptId)&&i.set(T,p)}let s=[...i.values()],l=[["adaptation","adaptation","candidate_adaptation"],["held_out","baseline","candidate"]].flatMap(([p,T,S])=>{let Tt=new Map(s.filter(R=>R.phase===S).map(R=>[R.taskId,R]));return s.filter(R=>R.phase===T).flatMap(R=>{let ee=Tt.get(R.taskId);return ee?[{cohort:p,taskId:R.taskId,baseline:R,refined:ee,tokenDelta:ee.totalTokens-R.totalTokens}]:[]})}),f=new Set(s.filter(p=>p.phase==="baseline"||p.phase==="adaptation").map(p=>p.taskId)).size,h=t.targetTaskCount??f;if(!Number.isInteger(h)||h<0)throw new Error("Benchmark task-efficiency target must be a non-negative integer.");let x={adaptation:new Set(s.filter(p=>p.phase==="adaptation").map(p=>p.taskId)).size,held_out:new Set(s.filter(p=>p.phase==="baseline").map(p=>p.taskId)).size},k=v((p,T)=>({targetTaskCount:T,comparedTaskCount:p.length,passedTaskCount:p.filter(S=>S.tokenDelta<0).length,failedTaskCount:p.filter(S=>S.tokenDelta>=0).length,lowerTaskCount:p.filter(S=>S.tokenDelta<0).length,higherTaskCount:p.filter(S=>S.tokenDelta>0).length,unchangedTaskCount:p.filter(S=>S.tokenDelta===0).length}),"summarizePairs"),u=k(l,h),j=l.reduce((p,T)=>p+T.baseline.totalTokens,0),ke=l.reduce((p,T)=>p+T.refined.totalTokens,0),Re=ke-j,Ie=l.length===h;return{summary:at.parse({target:"all_tasks_lower",...u,baselineTokens:j,refinedTokens:ke,tokenDelta:Re,tokenDeltaPercent:j>0?Re/j*100:null,complete:Ie,passed:Ie&&u.passedTaskCount===h,cohorts:{adaptation:k(l.filter(p=>p.cohort==="adaptation"),x.adaptation),heldOut:k(l.filter(p=>p.cohort==="held_out"),x.held_out)}}),pairs:l}}v(ai,"summarizeModelEvaluationTaskEfficiency");var cn=e.enum(["measuring","up_to_date","needs_attention"]),un=e.enum(["fixed","retained","regressed","unresolved","unavailable"]),rt=e.object({evaluationRunId:n,attemptKey:n,artifactPath:e.string().trim().min(1).max(2e3),jsonPointer:e.string().trim().min(1).max(1e3),transcriptHash:o.nullable(),traceHash:o.nullable()}).strict(),ge=e.object({level:e.literal(.95),lower:e.number().finite(),upper:e.number().finite()}).strict(),dn=e.object({panelId:n,panelRole:le,passLabel:n.nullable(),taskset:g,attempted:e.number().int().nonnegative(),available:e.number().int().nonnegative(),passed:e.number().int().nonnegative(),strictSuccess:e.number().min(0).max(1).nullable(),strictSuccessCi95:ge.nullable(),judgeScore:e.number().min(0).max(100).nullable(),judgeScoreCi95:ge.nullable()}).strict(),ot=e.object({schemaVersion:e.literal("openpond.modelCurrencySnapshot.v1"),id:n,contentHash:o,seriesId:n,protocol:e.object({id:n,revision:e.number().int().positive(),contentHash:o}).strict(),entryId:n,passLabel:n,parent:e.discriminatedUnion("kind",[e.object({kind:e.literal("base_model"),id:n,revision:e.string().trim().min(1).max(500)}).strict(),e.object({kind:e.literal("model_version"),id:n,contentHash:o}).strict()]),candidate:e.object({kind:e.literal("model_version"),id:n,contentHash:o}).strict(),sources:e.object({evaluationRunIds:e.array(n).max(1e5),attemptIds:e.array(n).max(1e6),tasksets:e.array(g).max(1e4),grader:r,judge:r.nullable(),calibration:r.nullable()}).strict(),matches:e.array(e.object({key:n,panelId:n,panelRole:le,taskId:n,issueFamilyId:n,seed:e.number().int(),repetition:e.number().int().nonnegative(),classification:un,parentPassed:e.boolean().nullable(),candidatePassed:e.boolean().nullable(),parentAttempt:rt.nullable(),candidateAttempt:rt.nullable()}).strict()).max(1e6),taskIds:e.object({fixed:e.array(n).max(1e6),retained:e.array(n).max(1e6),regressed:e.array(n).max(1e6),unresolved:e.array(n).max(1e6),unavailable:e.array(n).max(1e6)}).strict(),panels:e.array(dn).max(1e4),metrics:e.object({knownIssueCoverage:e.number().min(0).max(1).nullable(),issueFamilyGeneralization:e.number().min(0).max(1).nullable(),behavioralRetention:e.number().min(0).max(1).nullable(),currentAcquisitionDelta:e.number().finite().nullable(),retainedDeltaPoints:e.number().finite().nullable(),frontierStrictDelta:e.number().finite().nullable(),pairedPValue:e.number().min(0).max(1).nullable(),wins:e.number().int().nonnegative(),ties:e.number().int().nonnegative(),losses:e.number().int().nonnegative()}).strict(),statistics:e.object({matchedAttemptCount:e.number().int().nonnegative(),strictDelta:e.number().finite().nullable(),strictDeltaCi95:ge.nullable(),exactPairedBinaryPValue:e.number().min(0).max(1).nullable(),pairedBootstrapSamples:e.number().int().positive()}).strict(),criteria:e.object({allRequiredAttemptsTerminal:e.boolean(),criticalCorrectionPassRate:e.number().min(0).max(1).nullable(),siblingPassRate:e.number().min(0).max(1).nullable(),behavioralRetentionRate:e.number().min(0).max(1).nullable(),retainedRegressionPoints:e.number().finite().nullable(),criticalPriorRegressionCount:e.number().int().nonnegative(),thresholdRelease:r}).strict(),efficiency:e.object({inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),totalTokens:e.number().int().nonnegative(),costUsd:e.number().nonnegative().nullable(),latencyMs:e.number().int().nonnegative(),tokensPerSuccess:e.number().nonnegative().nullable(),throughputPerHour:e.number().nonnegative().nullable(),gpuSeconds:e.number().nonnegative().nullable()}).strict(),invariants:e.object({systemPromptHash:o,toolSchema:r,application:r,harness:r,runtime:r,grader:r,autoRefinerEnabled:e.boolean()}).strict(),evidenceState:cn,evidenceReasons:e.array(n).max(100),projectedAt:c}).strict();var $=e.discriminatedUnion("kind",[e.object({kind:e.literal("base_model"),id:n,revision:e.string().trim().min(1).max(500),label:e.string().trim().min(1).max(300)}).strict(),e.object({kind:e.literal("model_version"),id:n,label:e.string().trim().min(1).max(300)}).strict(),e.object({kind:e.literal("captured_model"),id:n,label:e.string().trim().min(1).max(300)}).strict()]),U=e.union([e.object({source:e.literal("evaluation_run").default("evaluation_run"),evaluationRunId:n,attemptId:n,modelLabel:e.string().trim().min(1).max(300).nullable().default(null),modelVersionId:n.nullable().default(null),reward:e.number().finite().nullable(),components:e.record(e.string(),e.number().finite()).default({})}).strict(),e.object({source:e.literal("imported_response"),modelLabel:e.string().trim().min(1).max(300),modelVersionId:n.nullable().default(null),response:e.unknown().refine(t=>t!==void 0,"An imported model response is required."),reward:e.number().finite().nullable().default(null),components:e.record(e.string(),e.number().finite()).default({})}).strict()]),pn=e.union([e.object({taskId:n,target:$,attempt:U}).strict(),e.object({taskId:n,target:e.enum(["base","current"]).default("current"),attempt:U}).strict(),e.object({taskId:n,target:e.enum(["base","current"]).default("current"),evaluationRunId:n,attemptId:n,reward:e.number().finite().nullable(),components:e.record(e.string(),e.number().finite()).default({})}).strict().transform(({taskId:t,target:i,...s})=>({taskId:t,target:i,attempt:{source:"evaluation_run",modelLabel:null,modelVersionId:null,...s}}))]),bi=e.object({schemaVersion:e.literal("openpond.continualLearningDailyBatchManifest.v1"),id:n,seriesId:n,scheduleEntryId:n,dayOrdinal:e.number().int().positive().max(1e4),sourceTaskset:g,taskIds:e.array(n).min(1).max(1e3),observedAttempts:e.array(pn).max(1e3).default([]),sourceFileName:e.string().trim().min(1).max(500).nullable().default(null),availableAt:c}).strict().superRefine((t,i)=>{new Set(t.taskIds).size!==t.taskIds.length&&i.addIssue({code:"custom",path:["taskIds"],message:"A daily batch cannot repeat a task."}),new Set(t.observedAttempts.map(l=>`${l.taskId}:${bn(l.target)}`)).size!==t.observedAttempts.length&&i.addIssue({code:"custom",path:["observedAttempts"],message:"An intake batch cannot repeat a task and Policy target pair."});let s=new Set(t.taskIds);t.observedAttempts.some(l=>!s.has(l.taskId))&&i.addIssue({code:"custom",path:["observedAttempts"],message:"Observed attempts must belong to tasks in the daily batch."})}),st=e.object({schemaVersion:e.literal("openpond.continualLearningDailyBatch.v1"),id:n,seriesId:n,scheduleEntryId:n,dayOrdinal:e.number().int().positive().max(1e4),label:e.string().trim().min(1).max(100),runMode:e.enum(["automatic","manual"]).optional(),source:e.enum(["json_upload","sealed_fixture"]),sourceFileName:e.string().trim().min(1).max(500).nullable(),sourceTaskset:g,tasks:e.array(e.object({taskId:n,taskContentHash:o,familyKey:n,disposition:e.enum(["include","defer","exclude"]).nullable(),oracleReview:e.enum(["pending","confirmed","needs_correction"]),note:e.string().trim().max(2e3),observedAttempt:U.nullable(),baselineAttempt:U.nullable().default(null),responses:e.array(e.object({target:$,attempt:U}).strict()).max(100).default([]),stagedAt:c.nullable().default(null),queuedEntry:A.nullable().default(null)}).strict()).min(1).max(1e3),intakeEvaluation:e.object({status:e.enum(["awaiting","running","ready","failed"]),requestedTargets:e.array(e.enum(["base","current"])).max(2).default(["base","current"]),requestedPolicies:e.array($).max(100).default([]),runs:e.array(e.object({target:$,runId:n}).strict()).max(100).default([]),baselineRunId:n.nullable(),currentRunId:n.nullable(),currentPolicy:e.discriminatedUnion("kind",[e.object({kind:e.literal("base_model"),id:n}).strict(),e.object({kind:e.literal("model_version"),id:n}).strict()]).nullable(),failure:e.string().trim().min(1).max(2e3).nullable()}).strict().default({status:"awaiting",requestedTargets:["base","current"],requestedPolicies:[],runs:[],baselineRunId:null,currentRunId:null,currentPolicy:null,failure:null}),status:e.enum(["pending","reviewed","queued"]),queuedEntry:A.nullable(),reviewedBy:n.nullable(),reviewedAt:c.nullable(),availableAt:c,revision:e.number().int().positive(),createdAt:c,updatedAt:c}).strict().superRefine((t,i)=>{new Set(t.tasks.map(l=>l.taskId)).size!==t.tasks.length&&i.addIssue({code:"custom",path:["tasks"],message:"A daily batch cannot repeat a task."});for(let[l,f]of t.tasks.entries()){let h=f.responses.map(x=>mt(x.target));new Set(h).size!==h.length&&i.addIssue({code:"custom",path:["tasks",l,"responses"],message:"A task cannot repeat a model response target."})}let s=t.tasks.every(l=>l.disposition!==null&&l.oracleReview!=="pending"&&(l.oracleReview!=="needs_correction"||l.note.length>0));t.status!=="pending"&&!s&&i.addIssue({code:"custom",path:["tasks"],message:"Every task requires a decision and oracle review before the day completes."}),t.status==="queued"!=(t.queuedEntry!==null)&&i.addIssue({code:"custom",path:["queuedEntry"],message:"Only a queued day carries an exact series-entry reference."}),t.status==="pending"&&(t.reviewedAt!==null||t.reviewedBy!==null)&&i.addIssue({code:"custom",path:["reviewedAt"],message:"A pending day cannot carry reviewer completion identity."}),t.status!=="pending"&&(t.reviewedAt===null||t.reviewedBy===null)&&i.addIssue({code:"custom",path:["reviewedAt"],message:"Reviewed identity is required after a day review completes."})}),lt=e.object({schemaVersion:e.literal("openpond.continualBenchIssueReview.v1"),id:n,seriesId:n,scheduleEntryId:n,passLabel:n,packet:Ve,splitManifest:e.object({id:n,contentHash:o}).strict(),decisions:e.array(e.object({taskId:n,disposition:e.enum(["include","defer","exclude"]).nullable(),note:e.string().trim().max(2e3)}).strict()).min(2).max(1e4),status:e.enum(["pending","reviewed","queued"]),queuedEntry:A.nullable(),reviewedBy:n.nullable(),reviewedAt:c.nullable(),revision:e.number().int().positive(),createdAt:c,updatedAt:c}).strict().superRefine((t,i)=>{let s=[...t.packet.cases.map(h=>h.taskId)].sort(),l=[...t.decisions.map(h=>h.taskId)].sort();JSON.stringify(s)!==JSON.stringify(l)&&i.addIssue({code:"custom",path:["decisions"],message:"Review decisions must cover the exact issue-packet cases."});let f=t.decisions.every(h=>h.disposition!==null&&h.note.length>0);t.status!=="pending"&&!f&&i.addIssue({code:"custom",path:["decisions"],message:"Every case requires a disposition and note before review completes."}),t.status==="queued"!=(t.queuedEntry!==null)&&i.addIssue({code:"custom",path:["queuedEntry"],message:"Only a queued review carries an exact series-entry reference."})}).readonly();function mt(t){return`${t.kind}:${t.id}`}v(mt,"responseTargetKey");function bn(t){return typeof t=="string"?t:mt(t)}v(bn,"observedTargetKey");var I=e.string().trim().min(1).max(240),_=e.string().trim().min(8).max(256),F=e.enum(["grpo","dapo","gspo-token"]),H=e.object({id:I,revision:e.string().trim().min(1).max(256),tokenizerRevision:e.string().trim().min(1).max(256),chatTemplateHash:_}),ct=e.object({maxRollouts:e.number().int().positive().max(1e6),maxEnvironmentExecutions:e.number().int().positive().max(1e6),maxInputTokens:e.number().int().positive(),maxOutputTokens:e.number().int().positive(),maxOptimizerSteps:e.number().int().positive().max(1e6),wallTimeMs:e.number().int().positive(),maximumCostUsd:e.number().nonnegative().nullable()}),ut=e.object({method:e.literal("grpo"),groupSize:e.number().int().min(2).max(128),normalization:e.literal("group_standardized"),advantageEpsilon:e.number().positive().max(.01).default(1e-8),loss:F,clipRange:e.number().positive().max(1).default(.2),iterations:e.number().int().min(2).max(16).default(2),microbatchSize:e.number().int().positive().max(128).default(1),gradientAccumulationSteps:e.number().int().positive().max(128).default(1),adamw:D}),G=e.object({method:e.literal("ppo"),valueModel:H,gamma:e.number().min(0).max(1),gaeLambda:e.number().min(0).max(1),policyClip:e.number().positive().max(1),valueClip:e.number().positive().max(10),valueLossCoefficient:e.number().nonnegative().max(10),ppoEpochs:e.number().int().positive().max(100),minibatchSize:e.number().int().positive().max(1e5)}),dt=e.discriminatedUnion("method",[ut,G]),J=e.object({rewardModelVersion:e.object({id:I,contentHash:_}).strict(),qualificationReport:e.object({id:I,contentHash:_}).strict().nullable().default(null),evaluationReferences:e.array(e.object({id:I,contentHash:_}).strict()).max(1e3).default([]),checkpoint:e.object({id:I,contentHash:_,objectRef:e.string().regex(/^r2:\/\/[A-Za-z0-9._/-]+$/),files:e.array(e.object({path:e.string().trim().min(1).max(512),sizeBytes:e.number().int().positive(),sha256:_}).strict()).min(4).max(128)}).strict(),runtime:de,processorRelease:e.object({id:I,contentHash:_}).strict(),rewardComposerRelease:e.object({id:I,contentHash:_}).strict(),executionReceipt:e.object({id:I,contentHash:_}).strict()}).strict(),W=e.object({schemaVersion:e.literal("openpond.policyOptimization.v1"),policyModel:H,referenceModel:H,dataset:e.object({tasksetId:I,tasksetHash:_,split:e.literal("train"),selectionStrategy:e.enum(["stable_hash_top_n","rft_easy_curriculum_v1"]),selectionSeed:e.number().int(),maxExamples:e.number().int().positive().max(1e5)}),sampler:e.object({temperature:e.number().min(0).max(2),topP:e.number().min(0).max(1),maxOutputTokens:e.number().int().positive().max(32768),maxTurns:e.number().int().positive().max(1e3),concurrency:e.number().int().positive().max(1e3)}),environment:e.object({id:I,version:e.string().trim().min(1).max(256),toolContractHash:_}),reward:e.object({graderId:I,graderHash:_,learnedPreference:J.nullable().default(null)}),kl:e.object({coefficient:e.number().nonnegative().nullable(),referenceConstraint:e.literal("fixed_reference")}),budgets:ct,checkpointEverySteps:e.number().int().positive(),seed:e.number().int(),evaluationSplit:e.literal("frozen_eval"),optimizer:dt});var a=e.string().trim().min(1).max(240),b=e.string().trim().min(1),m=e.string().trim().min(8).max(256),z=e.record(e.string(),e.unknown()).default({});var he=e.object({schemaVersion:e.literal("openpond.localModelChatConfiguration.v1").default("openpond.localModelChatConfiguration.v1"),profile:e.enum(["efficient","full_harness","custom"]).default("efficient"),systemPromptMode:e.enum(["lean","full_harness","custom"]).default("lean"),customSystemPrompt:e.string().max(2e4).nullable().default(null),contextWindowTokens:e.number().int().min(128).max(32768).default(1024),maxOutputTokens:e.number().int().min(1).max(512).default(64),temperature:e.number().min(0).max(2).default(0),repetitionPenalty:e.number().min(.5).max(2).default(1.1),noRepeatNgramSize:e.number().int().min(0).max(10).default(3),compaction:e.enum(["off","when_needed"]).default("when_needed"),keepWarmSeconds:e.number().int().min(0).max(3600).default(300),updatedAt:b.nullable().default(null)}),pt=he.parse({}),bt=e.object({schemaVersion:e.literal("openpond.managedAdapterServingProjection.v1"),teamId:a.nullable().default(null),source:e.literal("sandbox_managed_rl"),sourceRef:a,canonicalArtifactId:a.nullable(),canonicalModelName:e.string().trim().min(1).max(256).nullable().default(null),canonicalArtifactState:e.enum(["imported_unvalidated","evaluating","promotable","rejected","deleted"]).nullable(),canonicalDeploymentId:a.nullable(),canonicalDeploymentState:e.enum(["requested","deploying","ready","degraded","deleting","deleted","failed"]).nullable(),state:e.enum(["pending","imported","ready","failed"]),customerBindingAllowed:e.boolean().default(!1),artifactContentHash:m.nullable().default(null),baseProfileId:a.nullable().default(null),publishedAt:b.nullable(),lastSyncedAt:b,lastError:e.string().trim().min(1).max(5e3).nullable()});function gn(t){return t.state==="ready"&&t.canonicalArtifactState==="promotable"&&t.canonicalDeploymentState==="ready"&&t.customerBindingAllowed&&!!t.teamId&&!!t.canonicalArtifactId&&!!t.canonicalDeploymentId}v(gn,"managedAdapterProjectionReady");function ji(t){let i=t.managedServing;return i?gn(i)&&i.canonicalArtifactId&&i.canonicalDeploymentId?{kind:"sandbox_serving",canonicalArtifactId:i.canonicalArtifactId,canonicalDeploymentId:i.canonicalDeploymentId}:null:{kind:"source_artifact",canonicalArtifactId:null,canonicalDeploymentId:null}}v(ji,"resolveModelBindingEligibility");var ye="openpond.crossSystemOperations.v1",wi="1.0.0",ht="openpond.crossSystemTools.v1",hn=["search_crm","query_billing","search_support","run_python"],fn=e.enum(["train","validation","frozen_eval"]),vn=e.enum(["easy","medium","hard"]),xn=e.enum(["renewal_risk_v2"]),Mi=e.enum(["renewal_exposure","collections_prioritization","invoice_reconciliation","sla_escalation","contract_billing_mismatch"]),Pi=e.object({seed:e.number().int().nonnegative(),split:fn,difficulty:vn,scenarioProfile:xn.optional()});var fe={anyOf:[{type:"string",minLength:1,maxLength:256},{type:"null"}],default:null},ve={type:"integer",minimum:1,maximum:50,default:25},ft=[{name:"search_crm",description:"Search synthetic CRM accounts and contacts. Results are scoped, projected, paginated, and budgeted.",parameters:{type:"object",additionalProperties:!1,required:["query","fields","cursor","limit"],properties:{query:{type:"string",minLength:1,maxLength:500},fields:{type:"array",minItems:1,maxItems:12,uniqueItems:!0,items:{type:"string",enum:["account_id","name","aliases","contact_ids","contract_value_usd_cents","renewal_date","tier","active_contract_id","contract_term_months"]}},cursor:fe,limit:ve}}},{name:"query_billing",description:"Query synthetic invoices and payments for explicit account IDs and a bounded date range.",parameters:{type:"object",additionalProperties:!1,required:["account_ids","date_range","status","cursor","limit"],properties:{account_ids:{type:"array",minItems:1,maxItems:50,uniqueItems:!0,items:{type:"string",minLength:1,maxLength:100}},date_range:{type:"object",additionalProperties:!1,required:["from","to"],properties:{from:{type:"string",format:"date"},to:{type:"string",format:"date"}}},status:{type:"array",minItems:1,maxItems:6,uniqueItems:!0,items:{type:"string",enum:["open","overdue","paid","disputed","scheduled","void"]}},cursor:fe,limit:ve}}},{name:"search_support",description:"Search synthetic support cases for explicit account IDs, severity, and state filters.",parameters:{type:"object",additionalProperties:!1,required:["account_ids","severity","state","cursor","limit"],properties:{account_ids:{type:"array",minItems:1,maxItems:50,uniqueItems:!0,items:{type:"string",minLength:1,maxLength:100}},severity:{type:"array",minItems:1,maxItems:4,uniqueItems:!0,items:{type:"string",enum:["P1","P2","P3","P4"]}},state:{type:"array",minItems:1,maxItems:5,uniqueItems:!0,items:{type:"string",enum:["new","investigating","waiting_customer","resolved","closed"]}},cursor:fe,limit:ve}}},{name:"run_python",description:"Run standard-library-only Python in the persistent synthetic attempt sandbox with strict limits and no network.",parameters:{type:"object",additionalProperties:!1,required:["code"],properties:{code:{type:"string",minLength:1,maxLength:1e4}}}}],Z=An({schemaVersion:ht,tools:ft}),Sn=[`Use only the four registered synthetic Cross-System Operations tools. Contract ${Z}.`,"Treat every filter as exact. An overdue-only query_billing call returns invoice rows; payment rows are included only when paid is requested.","For cross-system joins, retrieve the bounded projected rows and use run_python for date filtering, set intersection, and arithmetic before answering.","Finish with ANSWER: JSON."].join(" "),yn=15;function _n(t="auto"){return[Sn,"LOCAL TOOL PROTOCOL (strict):","Use only the registered functions below. Never invent, rename, or call any other tool.","For one tool call, output exactly one JSON object and no prose:",'{"type":"tool_call","name":"registered_name","arguments":{}}',"Tool results arrive as exactly one JSON object:",'{"type":"tool_result","name":"registered_name","ok":true,"result":{},"error":null}','After one tool result, either call one registered tool again or answer normally. A JSON final may use {"type":"final","content":"..."}.',`Tool choice: ${JSON.stringify(t)}. Maximum tool turns: ${yn}.`,`Registered tool signatures: ${ft.map(kn).join("; ")}`].join(`
-`)}v(_n,"crossSystemLocalToolSystemPrompt");var Ei=_n();function kn(t){let i=K(t.parameters),s=K(i.properties),l=vt(i.required),f=Object.fromEntries(l.map(h=>[h,Y(s[h])]));return`${t.name}(${JSON.stringify(f)})`}v(kn,"compactToolSignature");function Y(t){let i=K(t),s=xe(i.anyOf);if(s.length)return s.map(Y).join("|");let l=xe(i.enum);if(l.length)return l.join("|");if(i.type==="array")return[Y(i.items)];if(i.type==="object"){let f=K(i.properties),h=vt(i.required);return Object.fromEntries(h.map(x=>[x,Y(f[x])]))}return typeof i.type=="string"?i.type:"unknown"}v(Y,"compactSchema");function K(t){return t&&typeof t=="object"&&!Array.isArray(t)?t:{}}v(K,"recordValue");function xe(t){return Array.isArray(t)?t:[]}v(xe,"arrayValue");function vt(t){return xe(t).filter(i=>typeof i=="string")}v(vt,"stringValues");var Q=e.enum(hn),Rn=e.discriminatedUnion("kind",[e.object({kind:e.literal("model"),turn:e.number().int().nonnegative(),content:e.string().max(1e5)}),e.object({kind:e.literal("tool_call"),turn:e.number().int().nonnegative(),callId:e.string().trim().min(1).max(240),name:Q,arguments:e.record(e.string(),e.unknown())}),e.object({kind:e.literal("tool_result"),turn:e.number().int().nonnegative(),callId:e.string().trim().min(1).max(240),name:Q,ok:e.boolean(),result:e.unknown(),rows:e.number().int().nonnegative(),bytes:e.number().int().nonnegative(),durationMs:e.number().int().nonnegative(),error:e.string().trim().min(1).max(2e4).nullable()}),e.object({kind:e.literal("final"),turn:e.number().int().nonnegative(),content:e.string().max(1e5)})]),xt=e.object({schemaVersion:e.literal(ye),id:e.string().trim().min(1).max(240),worldId:e.string().trim().min(1).max(240),taskId:e.string().trim().min(1).max(240),toolContractHash:e.literal(Z),modelRef:e.object({providerId:e.string().trim().min(1),modelId:e.string().trim().min(1)}).nullable(),status:e.enum(["completed","budget_exhausted","cancelled","infrastructure_failure"]),steps:e.array(Rn).max(1e3),startedAt:e.string().trim().min(1),completedAt:e.string().trim().min(1),infrastructureError:e.string().trim().min(1).max(2e4).nullable(),metadata:e.record(e.string(),e.unknown()).default({})}),St=e.object({schemaVersion:e.literal(ye),trajectoryId:e.string().trim().min(1).max(240),outcome:e.enum(["correct","incorrect","parse_failure","budget_exhausted","tool_schema_violation","infrastructure_failure","cancelled"]),reward:e.number().min(0).max(1.15).nullable(),rewardEligible:e.boolean(),exactAnswer:e.boolean(),components:e.object({exactAnswer:e.number().min(0).max(1),efficiency:e.number().min(0).max(.1),conciseOutput:e.number().min(0).max(.05)}),metrics:e.object({toolCalls:e.number().int().nonnegative(),rowsRead:e.number().int().nonnegative(),bytesRead:e.number().int().nonnegative(),consecutiveRetrievalCalls:e.number().int().nonnegative(),wallTimeMs:e.number().int().nonnegative(),parseFailures:e.number().int().nonnegative(),budgetExhausted:e.boolean()}),parsedAnswer:e.unknown().nullable(),feedback:e.array(e.string().trim().min(1).max(5e3)).max(100)}),In=e.object({id:e.string().trim().min(1).max(240),type:e.literal("function"),function:e.object({name:Q,arguments:e.string().max(2e4)})}),yt=e.object({role:e.enum(["system","user","assistant","tool"]),content:e.string().max(2e5).nullable(),tool_call_id:e.string().trim().min(1).max(240).optional(),tool_calls:e.array(In).max(4).optional()}),Hi=e.object({schemaVersion:e.literal(ye),id:e.string().trim().min(1).max(240),taskId:e.string().trim().min(1).max(240),worldId:e.string().trim().min(1).max(240),trajectoryId:e.string().trim().min(1).max(240),toolContractHash:e.literal(Z),approval:e.object({status:e.literal("approved"),approvedBy:e.string().trim().min(1).max(240),approvedAt:e.string().trim().min(1)}),messages:e.array(yt).min(3).max(1e3)}),jn=e.object({tasksetTaskId:e.string().trim().min(1).max(240),environmentTaskId:e.string().trim().min(1).max(240),family:e.enum(["renewal_exposure","collections_prioritization","invoice_reconciliation","sla_escalation","contract_billing_mismatch"]),prompt:e.string().trim().min(1).max(1e5),finalAnswer:e.string().trim().min(1).max(2e5),trajectoryId:e.string().trim().min(1).max(240),trajectoryHash:e.string().trim().min(8).max(256),toolNames:e.array(Q).min(1).max(100),toolCallCount:e.number().int().positive(),messageCount:e.number().int().min(3),reward:e.number().min(0).max(1.15),messages:e.array(yt).min(3).max(1e3)}),Tn=e.object({status:e.literal("approved"),approvedBy:e.string().trim().min(1).max(240),approvedAt:e.string().trim().min(1),previewHash:e.string().trim().min(8).max(256),trajectoryCount:e.number().int().positive()}),zi=e.object({schemaVersion:e.literal("openpond.crossSystemExpertBootstrapPreview.v1"),tasksetId:e.string().trim().min(1).max(240),tasksetHash:e.string().trim().min(8).max(256),tasksetRevision:e.number().int().positive(),previewHash:e.string().trim().min(8).max(256),toolContractHash:e.literal(Z),status:e.enum(["ready_for_review","approved"]),approval:Tn.nullable(),tasks:e.array(jn).min(1).max(100)});function An(t){let i=Se(t),s=2166136261,l=2654435769;for(let f=0;f<i.length;f+=1){let h=i.charCodeAt(f);s=Math.imul(s^h,16777619),l=Math.imul(l^h,2246822507)}return`${ht}:${gt(s)}${gt(l)}`}v(An,"contractHash");function Se(t){return Array.isArray(t)?`[${t.map(Se).join(",")}]`:t&&typeof t=="object"?`{${Object.entries(t).sort(([i],[s])=>i.localeCompare(s)).map(([i,s])=>`${JSON.stringify(i)}:${Se(s)}`).join(",")}}`:JSON.stringify(t)}v(Se,"stableJson");function gt(t){return(t>>>0).toString(16).padStart(8,"0")}v(gt,"hex");var Cn=e.object({schemaVersion:e.literal("openpond.singleTurnPolicyTrajectory.v1"),id:a,taskId:a,status:e.enum(["completed","infrastructure_failure"]),promptHash:m,responseText:e.string().max(2e6),infrastructureError:e.string().trim().min(1).max(1e4).nullable(),startedAt:b,completedAt:b,metadata:z}),wn=e.object({schemaVersion:e.literal("openpond.ppoTrajectory.v1"),id:a,taskId:a,policyModelId:a,referenceModelId:a,valueModelId:a,steps:e.array(e.object({index:e.number().int().nonnegative(),observationHash:m,actionTokenIds:e.array(e.number().int().nonnegative()).min(1),terminated:e.boolean(),truncated:e.boolean(),reward:e.number(),policyLogProbability:e.number(),referenceLogProbability:e.number(),valuePrediction:e.number(),return:e.number(),advantage:e.number(),mask:e.number().min(0).max(1)})).min(1).max(1e5),createdAt:b}),Mn=e.object({schemaVersion:e.literal("openpond.exactAnswerVerifierResult.v1"),outcome:e.enum(["correct","incorrect","parse_failure","infrastructure_failure"]),graderSetHash:m,score:e.number().min(0).max(1).nullable(),passed:e.boolean(),rewardEligible:e.boolean(),expectedAnswerHash:m,extractedAnswer:e.string().max(2e4).nullable(),feedback:e.array(e.string().trim().min(1).max(2e4)).max(1e3)}),_t=e.object({schemaVersion:e.literal("openpond.rolloutTrajectoryReceipt.v1"),id:a,jobId:a,planId:a,tasksetId:a,tasksetHash:m,taskId:a,split:e.literal("train"),correlationId:a,provider:a,providerTrace:e.object({invocationId:a,experimentId:a,rolloutId:a,runId:a,rowId:a}),optimizerMethod:e.enum(["grpo","ppo"]).default("grpo"),evidenceLevels:e.object({requested:e.enum(["trajectory","aggregate","provider_reported"]),observed:e.enum(["trajectory","aggregate","provider_reported"]),providerReported:e.enum(["trajectory","aggregate","provider_reported"])}).default({requested:"trajectory",observed:"trajectory",providerReported:"provider_reported"}),budgetUsage:e.object({rollouts:e.number().int().nonnegative(),environmentExecutions:e.number().int().nonnegative(),inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),optimizerSteps:e.number().int().nonnegative(),costUsd:e.number().nonnegative()}).default({rollouts:0,environmentExecutions:0,inputTokens:0,outputTokens:0,optimizerSteps:0,costUsd:0}),environment:e.object({id:a,version:e.string().trim().min(1).max(256),worldId:a,worldHash:m,toolContractHash:m}),policy:e.object({modelId:a,checkpointId:a.nullable(),completionParametersHash:m}),status:e.enum(["received","running","succeeded","failed"]),failureClass:e.enum(["policy_failure","parse_failure","tool_schema_violation","budget_exhausted","cancelled","environment_failure","infrastructure_failure"]).nullable(),reward:e.object({eligible:e.boolean(),raw:e.number().min(0).max(1.15).nullable(),normalized:e.number().min(0).max(1).nullable(),components:e.record(e.string(),e.number())}),trajectory:e.union([xt,Cn,wn]).nullable(),verifier:e.union([St,Mn]).nullable(),providerStatus:e.record(e.string(),e.unknown()).default({}),receivedAt:b,startedAt:b.nullable(),completedAt:b.nullable(),updatedAt:b});function d(t){let i;return new e.ZodObject({type:"object",get shape(){return i??=t()}})}v(d,"deferredObject");var E=e.enum(["sft","dpo","grpo","ppo","sdft","opd","opsd","sdpo"]),X=e.enum(["lora","full"]),O=e.enum(["openpond_managed"]),Pn=d(()=>({schemaVersion:e.literal("openpond.sftRecipe.v1"),method:e.literal("sft"),parameterization:e.literal("lora"),baseModel:e.object({id:a,revision:e.string().trim().min(1).max(256),tokenizerRevision:e.string().trim().min(1).max(256),chatTemplateHash:m}),dataset:e.object({trainSplit:e.literal("train"),validationSplit:e.enum(["validation","frozen_eval"]),completionOnly:e.boolean(),maxSequenceLength:e.number().int().positive().max(32768),maxExamples:e.number().int().positive().max(1e5).default(1e3),selectionStrategy:e.literal("stable_hash_top_n").default("stable_hash_top_n"),selectionSeed:e.number().int().default(17)}),lora:e.object({rank:e.number().int().positive().max(256),alpha:e.number().positive().max(1024),dropout:e.number().min(0).max(1),targetModules:e.array(a).min(1).max(100)}),optimizer:e.object({learningRate:e.number().positive(),epochs:e.number().positive().max(100),maxSteps:e.number().int().positive().max(1e6),batchSize:e.number().int().positive().max(1e4),gradientAccumulationSteps:e.number().int().positive().max(1e4),seed:e.number().int()}),resourceLimits:e.object({cpuThreads:e.number().int().positive().max(256),memoryBytes:e.number().int().positive(),wallTimeMs:e.number().int().positive()})})),cr=d(()=>({id:a,input:e.record(e.string(),e.unknown()),expectedOutput:e.record(e.string(),e.unknown()),tags:e.array(a).max(100).default([])})),ur=d(()=>({id:a,prompt:e.string().min(1).max(5e5),chosen:e.string().min(1).max(5e5),rejected:e.string().min(1).max(5e5),sourceRefs:e.array(a).min(1).max(1e4)})),dr=d(()=>({id:a,input:e.record(e.string(),e.unknown()),tags:e.array(a).max(100).default([])})),pr=d(()=>({schemaVersion:e.literal("openpond.rewardModelRecipe.v1"),method:e.literal("reward_model"),parameterization:e.literal("lora_with_scalar_head"),runScope:e.enum(["synthetic_smoke","human_preference"]),baseModel:e.object({id:a,revision:e.string().trim().min(1).max(256),tokenizerRevision:e.string().trim().min(1).max(256),processorRevision:e.string().trim().min(1).max(256),chatTemplateHash:m}).strict(),tasksetRelease:r,preferenceDatasetRelease:r,processorRelease:r,input:e.object({kind:e.literal("structured_text"),serialization:e.enum(["scenario_input_and_candidate_json_v1","visible_agent_trajectory_v1"]),maxCharacters:e.number().int().positive().max(5e5)}).strict().default({kind:"structured_text",serialization:"scenario_input_and_candidate_json_v1",maxCharacters:32e3}),lora:e.object({rank:e.number().int().positive().max(256),alpha:e.number().positive().max(1024),dropout:e.number().min(0).max(1),targetModules:e.array(a).min(1).max(100)}).strict(),heads:e.object({scalar:e.literal("pooled_hidden_state_linear"),bucket:e.enum(["none","three_class"]).default("three_class")}).strict(),loss:e.object({ranking:e.literal("bradley_terry"),rankingWeight:e.number().positive(),bucketWeight:e.number().nonnegative(),tieWeight:e.number().nonnegative()}).strict(),optimizer:e.object({learningRate:e.number().positive(),maxSteps:e.number().int().positive().max(1e5),batchSize:e.number().int().positive().max(1e4),gradientAccumulationSteps:e.number().int().positive().max(1e4),seed:e.number().int(),checkpointEverySteps:e.number().int().positive().max(1e5)}).strict(),resourceLimits:e.object({wallTimeMs:e.number().int().positive().max(1440*60*1e3),maxExamples:e.number().int().positive().max(1e5),maxInputCharacters:e.number().int().positive().max(5e5).default(32e3),maxImagePixels:e.number().int().positive().max(1e8).optional(),maximumSpendUsd:e.number().nonnegative().max(1e5)}).strict()})).superRefine((t,i)=>{t.runScope==="synthetic_smoke"&&(t.optimizer.maxSteps>100||t.resourceLimits.maximumSpendUsd>10)&&i.addIssue({code:"custom",path:["resourceLimits"],message:"Synthetic-smoke Reward Model runs are limited to 100 steps and USD 10."})}),En=d(()=>({schemaVersion:e.literal("openpond.rftRecipe.v1"),method:e.literal("grpo"),parameterization:e.literal("lora"),baseModel:e.object({id:a,revision:e.string().trim().min(1).max(256),tokenizerRevision:e.string().trim().min(1).max(256),chatTemplateHash:m}),dataset:e.object({trainSplit:e.literal("train"),validationSplit:e.enum(["validation","frozen_eval"]),maxPromptTokens:e.number().int().positive().max(32768),maxExamples:e.number().int().positive().max(1e5).default(1e3),selectionStrategy:e.enum(["stable_hash_top_n","rft_easy_curriculum_v1"]).default("stable_hash_top_n")}),lora:e.object({rank:e.number().int().positive().max(256)}),rollout:e.object({groupSize:e.number().int().min(2).max(64),concurrency:e.number().int().positive().max(32),maxTurns:e.number().int().positive().max(100),maxOutputTokens:e.number().int().positive().max(8192),temperature:e.number().min(0).max(2),topP:e.number().positive().max(1),seed:e.number().int()}),optimizer:e.object({learningRate:e.number().positive(),maxSteps:e.number().int().positive().max(1e5),clipRange:e.number().positive().max(1).default(.2),iterations:e.number().int().min(2).max(16).default(2),microbatchSize:e.number().int().positive().max(64).default(1),gradientAccumulationSteps:e.number().int().positive().max(64).default(1),advantageEpsilon:e.number().positive().max(.01).default(1e-8),adamw:D}),loss:e.object({method:F.default("grpo"),klBeta:e.number().min(0).nullable().default(null)}).default({method:"grpo",klBeta:null}),reward:e.object({graderId:a,graderHash:m,environmentId:a,environmentVersion:e.string().trim().min(1).max(256),toolContractHash:m,learnedPreference:J.nullable().optional()}),resourceLimits:e.object({wallTimeMs:e.number().int().positive().max(1440*60*1e3),maxGpuSeconds:e.number().int().positive().max(21600).optional(),warmupDeadlineSeconds:e.number().int().positive().max(3600).optional(),maxRollouts:e.number().int().positive().max(1e5),maxPayloadBytes:e.number().int().positive().max(10*1024*1024)}),policyOptimization:W.nullable().default(null),continuation:e.object({schemaVersion:e.literal("openpond.crossJobContinuationRequest.v1"),parentArtifact:r,sourceArtifact:e.object({jobId:a,artifactId:a,checkpointId:a,contentHash:m}).strict(),optimizerMode:e.enum(["continue","reset"]),adapterMode:e.enum(["residual_append","shared","sequential_merge"]).optional()}).strict().optional()})),Hn=d(()=>({schemaVersion:e.literal("openpond.dpoRecipe.v1"),method:e.literal("dpo"),parameterization:e.literal("lora"),policyModel:H,referenceModel:H,dataset:e.object({trainSplit:e.literal("train"),validationSplit:e.enum(["validation","frozen_eval"]),maxPairs:e.number().int().positive().max(1e5),maxPromptTokens:e.number().int().positive().max(32768),maxCompletionTokens:e.number().int().positive().max(32768),selectionStrategy:e.literal("stable_hash_top_n"),selectionSeed:e.number().int()}),lora:e.object({rank:e.number().int().positive().max(256),alpha:e.number().positive().max(1024),dropout:e.number().min(0).max(1),targetModules:e.array(a).min(1).max(100)}),loss:e.object({variant:e.literal("sigmoid"),beta:e.number().positive().max(10),labelSmoothing:e.number().min(0).max(.5).default(0)}),optimizer:e.object({learningRate:e.number().positive(),epochs:e.number().positive().max(100),maxSteps:e.number().int().positive().max(1e6),batchSize:e.number().int().positive().max(1e4),gradientAccumulationSteps:e.number().int().positive().max(1e4),seed:e.number().int()}),referenceLogprobs:e.object({cacheSchemaVersion:e.literal("openpond.dpoReferenceLogprobs.v1"),cacheKey:m,invalidationHash:m}),resourceLimits:e.object({cpuThreads:e.number().int().positive().max(256),memoryBytes:e.number().int().positive(),wallTimeMs:e.number().int().positive()})})),zn=d(()=>({schemaVersion:e.literal("openpond.ppoRecipe.v1"),method:e.literal("ppo"),parameterization:e.literal("lora"),policyOptimization:W.extend({optimizer:G}),lora:e.object({rank:e.number().int().positive().max(256),alpha:e.number().positive().max(1024),dropout:e.number().min(0).max(1),targetModules:e.array(a).min(1).max(100)}),valueHead:e.object({initialization:e.literal("policy_hidden_state_linear"),optimizerLearningRate:e.number().positive(),artifactName:e.literal("value_head.safetensors")}),policyLearningRate:e.number().positive(),resume:e.object({checkpointId:a.nullable(),policyHash:m,referenceHash:m,valueModelHash:m,optimizerStateHash:m.nullable()}),resourceLimits:e.object({cpuThreads:e.number().int().positive().max(256),memoryBytes:e.number().int().positive(),wallTimeMs:e.number().int().positive()})})),On=d(()=>({schemaVersion:e.literal("openpond.unsupportedRecipe.v1"),method:E.exclude(["sft","dpo","grpo","ppo"]),parameterization:X,unsupportedReason:e.string().trim().min(1).max(5e3)})),Vn=e.union([Pn,Hn,En,zn,On]),Dn=d(()=>({schemaVersion:e.literal("openpond.trainingDestinationCapabilities.v1"),destinationId:O,available:e.boolean(),methods:e.array(E),parameterizations:e.array(X),modelAllowlist:e.array(a).default([]),maxDatasetBytes:e.number().int().nonnegative().nullable(),environmentPlacements:e.array(e.enum(["none","local","remote","colocated","provider_native"])),nonProduction:e.boolean(),unavailableReason:e.string().trim().min(1).max(5e3).nullable(),checkedAt:b})),Bn=d(()=>({destinationId:O,available:e.boolean(),methods:e.array(E),parameterizations:e.array(X),nonProduction:e.boolean(),unavailableReason:e.string().trim().min(1).max(5e3).nullable()})),Ln=d(()=>({schemaVersion:e.literal("openpond.baseModelCandidate.v1"),selectionKey:a,label:e.string().trim().min(1).max(500),sourceLabel:e.string().trim().min(1).max(500),preference:B,available:e.boolean(),nonProduction:e.boolean(),unavailableReason:e.string().trim().min(1).max(5e3).nullable(),methods:e.array(E),executionOptions:e.array(Bn).min(1)})),qn=d(()=>({code:a,severity:e.enum(["warning","error"]),path:e.string().trim().max(2e3).nullable(),message:e.string().trim().min(1).max(5e3)})),Un=d(()=>({schemaVersion:e.literal("openpond.trainingCompatibility.v1"),compatible:e.boolean(),destinationId:O,tasksetId:a,recipeMethod:E,issues:e.array(qn),checkedAt:b})),Nn=e.union([we,e.enum(["destination_unavailable","destination_method_unsupported","destination_model_unsupported","destination_parameterization_unsupported","experimental_destination"])]),br=d(()=>({method:E,state:e.enum(["recommended","compatible","needs_dataset_work","destination_unavailable","experimental_destination"]),reasonCodes:e.array(Nn).default([]),reasons:e.array(e.string().trim().min(1).max(5e3)).default([]),destinationId:O.nullable().default(null)})),gr=e.enum(["small","standard","custom","small_experiment"]);var $n=Te,It=d(()=>({schemaVersion:e.literal("openpond.trainingPlan.v1"),id:a,modelId:a,tasksetId:a,tasksetHash:m,comparisonSeriesEntry:A.nullable().optional(),evaluationTasksetRef:g.nullable().optional(),harnessRelease:r.nullable().optional(),modelImprovementQualification:r.nullable().optional(),destinationId:O,recipe:Vn,environmentPlacement:e.enum(["none","local","remote","colocated","provider_native"]),compatibility:Un,dataPolicy:e.object({exportApproved:e.boolean(),approvedSourceIds:e.array(a),retentionDays:e.number().int().nonnegative().nullable(),region:e.string().trim().min(1).max(200).nullable()}),estimatedCostUsd:e.number().nonnegative().nullable(),createdAt:b,contentHash:m})),Fn=d(()=>({path:e.string().trim().min(1).max(2e3),sha256:m,sizeBytes:e.number().int().nonnegative(),role:e.enum(["manifest","task_data","grader","environment","recipe","policy","provenance"])})),_e=d(()=>({schemaVersion:e.literal("openpond.trainingBundle.v1"),id:a,planId:a,tasksetId:a,tasksetHash:m,recipeHash:m,files:e.array(Fn).min(1).max(1e6),totalSizeBytes:e.number().int().nonnegative(),sourceIds:e.array(a),excludedSourceIds:e.array(a),containsRawChats:e.literal(!1),containsSecrets:e.literal(!1),containsHiddenGraderAssets:e.literal(!1),createdAt:b,contentHash:m})),hr=d(()=>({schemaVersion:e.literal("openpond.trainingBundleExport.v1"),manifest:_e,files:e.array(e.object({path:e.string().trim().min(1).max(2e3),sha256:m,sizeBytes:e.number().int().nonnegative(),encoding:e.literal("base64"),content:e.string()})).min(1),contentHash:m})),fr=d(()=>({schemaVersion:e.literal("openpond.trainingPreparedStart.v1"),plan:It,bundle:_e,approvalActor:a.nullable(),preparedAt:b})),vr=d(()=>({schemaVersion:e.literal("openpond.trainingApproval.v1"),evaluationTasksetRef:g.nullable().optional(),id:a,planId:a,bundleHash:m,harnessRelease:r.nullable().optional(),modelImprovementQualification:r.nullable().optional(),destinationId:O,modelId:a,method:E,parameterization:X,maximumCostUsd:e.number().nonnegative().nullable(),approvedBy:a,approvedAt:b})),xr=d(()=>({schemaVersion:e.literal("openpond.trainingJobSourceSnapshot.v1"),evaluationTasksetRef:g.nullable().optional(),modelProjectId:a,sourceProjectRevision:e.number().int().positive(),profileId:a,taskset:e.object({id:a,revision:e.number().int().positive(),contentHash:m}).strict(),tasksetRelease:r,harnessRelease:r,baseModel:B,method:E})).strict(),Gn=e.enum(["queued","starting","running","cancelling","cancelled","succeeded","failed","reconciling"]),jt=d(()=>({schemaVersion:e.literal("openpond.trainingJob.v1"),id:a,planId:a,bundleHash:m,approvalId:a,destinationId:O,status:Gn,nonProduction:e.boolean(),workerPid:e.number().int().positive().nullable(),startedAt:b.nullable(),completedAt:b.nullable(),error:e.string().trim().min(1).max(2e4).nullable(),createdAt:b,updatedAt:b,metadata:z})),Jn=d(()=>({schemaVersion:e.literal("openpond.trainingJobEvent.v1"),id:a,jobId:a,sequence:e.number().int().nonnegative(),type:e.enum(["queued","start","progress","metric","checkpoint","cancel","complete","failure","reconcile"]),timestamp:b,payload:z})),Wn=d(()=>({schemaVersion:e.literal("openpond.sftStepMetric.v1"),step:e.number().int().nonnegative(),maxSteps:e.number().int().positive(),timestamp:b,epoch:e.number().nonnegative().nullable(),loss:e.number().nullable(),learningRate:e.number().nonnegative().nullable(),gradientNorm:e.number().nonnegative().nullable(),entropy:e.number().nonnegative().nullable(),meanTokenAccuracy:e.number().min(0).max(1).nullable(),preferenceAccuracy:e.number().min(0).max(1).nullable().default(null),preferenceMargin:e.number().nullable().default(null),chosenReward:e.number().nullable().default(null),rejectedReward:e.number().nullable().default(null),chosenLogProbability:e.number().nullable().default(null),rejectedLogProbability:e.number().nullable().default(null),reward:e.number().nullable().default(null),policyLoss:e.number().nullable().default(null),advantageLoss:e.number().nullable().default(null),inputTokensSeen:e.number().int().nonnegative().nullable(),memoryBytes:e.number().int().nonnegative().nullable(),elapsedSeconds:e.number().nonnegative().nullable()})),Yn=d(()=>({schemaVersion:e.literal("openpond.policyOptimizationMetric.v1"),method:e.enum(["grpo","ppo"]),step:e.number().int().nonnegative(),timestamp:b,learningRate:e.number().nonnegative().nullable().default(null),policyLoss:e.number().nullable(),valueLoss:e.number().nullable(),gradientNorm:e.number().nonnegative().nullable().default(null),meanReward:e.number().nullable(),meanReturn:e.number().nullable(),kl:e.number().nullable(),behaviorPolicyKlPreUpdate:e.number().nonnegative().nullable().default(null),entropy:e.number().nullable(),policyClipFraction:e.number().min(0).max(1).nullable(),behaviorPolicyClipFractionPreUpdate:e.number().min(0).max(1).nullable().default(null),valueClipFraction:e.number().min(0).max(1).nullable(),explainedVariance:e.number().nullable(),rolloutLearnerLag:e.number().int().nonnegative().nullable(),inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),environmentExecutions:e.number().int().nonnegative(),trajectoryCount:e.number().int().nonnegative().nullable().default(null),costUsd:e.number().nonnegative().nullable()})),Kn=d(()=>({schemaVersion:e.literal("openpond.managedTrainingRunEvidence.v2"),provider:a,providerRunId:a,state:a,progress:e.object({targetOptimizerSteps:e.number().int().nonnegative(),committedOptimizerSteps:e.number().int().nonnegative(),skippedOptimizerSteps:e.number().int().nonnegative()}),reward:e.object({finalMean:e.number().nullable(),variance:e.number().nonnegative().nullable(),minimum:e.number().nullable(),maximum:e.number().nullable(),distinctValueCount:e.number().int().nonnegative(),noSignalGroupCount:e.number().int().nonnegative(),trajectoryCount:e.number().int().nonnegative(),eligibleTrajectoryCount:e.number().int().nonnegative()}),usage:e.object({inputTokens:e.number().int().nonnegative(),outputTokens:e.number().int().nonnegative(),environmentExecutions:e.number().int().nonnegative()}),resource:e.object({provider:a,gpuType:e.string().trim().min(1).max(300).nullable(),gpuCount:e.number().int().nonnegative().nullable(),hourlyCostUsd:e.number().nonnegative().nullable(),durationSeconds:e.number().nonnegative().nullable(),gpuSeconds:e.number().nonnegative().nullable()}),cost:e.object({totalUsd:e.number().nonnegative().nullable()}),checkpoint:e.object({id:a,policyVersion:e.number().int().nonnegative(),sha256:m.nullable(),sizeBytes:e.number().int().nonnegative().nullable()}).nullable(),movement:e.object({adapterDeltaNorm:e.number().nonnegative().nullable()}),evaluations:e.array(e.object({reference:e.object({id:a,contentHash:m}).nullable().optional(),kind:e.enum(["baseline","candidate"]),policyVersion:e.number().int().nonnegative(),taskCount:e.number().int().nonnegative().nullable().optional(),targetTaskCount:e.number().int().nonnegative().nullable().optional(),score:e.number().nullable(),threshold:e.number().nullable(),passed:e.boolean().nullable()})),canonicalPublication:e.object({state:a.nullable(),artifactId:a.nullable()}),syncedAt:b})),kt=d(()=>({count:e.number().int().nonnegative(),scoredCount:e.number().int().nonnegative(),meanScore:e.number().min(0).max(1).nullable(),passedCount:e.number().int().nonnegative(),passRate:e.number().min(0).max(1).nullable()})),Rt=d(()=>({status:e.enum(["scored","unavailable"]),score:e.number().min(0).max(1).nullable(),passed:e.boolean(),rewardEligible:e.boolean(),failureClass:e.enum(["policy_failure","grader_failure","environment_failure","infrastructure_failure","timeout","cancelled"]).nullable(),feedback:e.array(e.string().trim().min(1).max(2e4)).max(1e3),components:e.array(e.object({graderId:a,score:e.number().min(0).max(1).nullable(),passed:e.boolean(),feedback:e.string().trim().max(2e4).nullable()})).max(1e3)})),Qn=d(()=>({taskId:a,input:e.record(e.string(),e.unknown()),baseOutput:e.record(e.string(),e.unknown()).nullable(),trainedOutput:e.record(e.string(),e.unknown()).nullable(),baseGrade:Rt.nullable(),trainedGrade:Rt.nullable()})),Zn=d(()=>({schemaVersion:e.literal("openpond.trainingEvaluationSummary.v1"),jobId:a,tasksetId:a,base:kt,trained:kt,meanScoreDelta:e.number().min(-1).max(1).nullable(),examples:e.array(Qn).max(1e6)})),Sr=d(()=>({schemaVersion:e.literal("openpond.trainingRunDetail.v1"),job:jt,events:e.array(Jn),stepMetrics:e.array(Wn),policyMetrics:e.array(Yn).default([]),managedEvidence:Kn.nullable().default(null),evaluation:Zn.nullable(),generatedAt:b})),Xn=d(()=>({schemaVersion:e.literal("openpond.trainingArtifact.v1"),id:a,jobId:a,kind:e.enum(["adapter","checkpoint","metrics","log","manifest","evaluation"]),path:e.string().trim().min(1).max(4e3),sha256:m,sizeBytes:e.number().int().nonnegative(),baseModelId:a.nullable(),baseModelRevision:e.string().trim().min(1).max(256).nullable(),tokenizerRevision:e.string().trim().min(1).max(256).nullable(),chatTemplateHash:m.nullable(),nonProduction:e.boolean(),createdAt:b,metadata:z})),ea=d(()=>({schemaVersion:e.literal("openpond.modelArtifactLineage.v1"),id:a,modelId:a,artifactId:a,jobId:a,tasksetId:a,tasksetHash:m,graderHash:m,planHash:m,bundleHash:m,recipeHash:m,workerVersion:e.string().trim().min(1).max(256),trainerVersion:e.string().trim().min(1).max(256),importedAt:b,frozenEvaluationArtifactId:a.nullable(),promotable:e.boolean(),pinned:e.boolean().default(!1),status:e.enum(["imported","rejected"]).default("imported"),rejectedAt:b.nullable().default(null),rejectionReason:e.string().trim().min(1).max(5e3).nullable().default(null),chatConfiguration:he.default(pt),managedServing:bt.nullable().default(null)})),ta=e.enum(["chat_manual","agent","extension","authoring_optimizer"]),na=d(()=>({schemaVersion:e.literal("openpond.modelBinding.v1"),id:a,profileId:a,role:ta,roleTargetId:a,modelArtifactLineageId:a,tasksetId:a,evaluationArtifactId:a.nullable(),status:e.enum(["active","rolled_back"]),priorBindingId:a.nullable(),rollbackTargetBindingId:a.nullable(),promotedBy:a,promotedAt:b,rolledBackAt:b.nullable(),metadata:z})),yr=d(()=>({schemaVersion:e.literal("openpond.trainingActivity.v1"),profileId:a,active:e.boolean(),activeCounts:e.object({jobs:e.number().int().nonnegative(),creations:e.number().int().nonnegative(),minerRuns:e.number().int().nonnegative(),datasetImports:e.number().int().nonnegative()}),revision:m,generatedAt:b})),_r=d(()=>({schemaVersion:e.literal("openpond.tasksetOperationalState.v1"),tasksetId:a,attempts:e.array(Pe),artifacts:e.array(Ee),grades:e.array(He),evaluationResults:e.array(je).default([]),generatedAt:b})),kr=d(()=>({schemaVersion:e.literal("openpond.trainingState.v1"),profileId:a,sources:e.array(Ae),creations:e.array(Oe),tasksetDrafts:e.array(Ce).default([]),tasksets:e.array(ae),benchmarkRuns:e.array(e.custom()).default([]),benchmarkComparisons:e.array(e.custom()).default([]),datasetImports:e.array(Ue).default([]),datasetArtifacts:e.array(ne).default([]),graderAuditReports:e.array(ze),evaluationResults:e.array(e.custom()).default([]),candidates:e.array(ce),minerConfig:q,minerRuns:e.array(Fe).default([]),modelProjects:e.array($n).default([]),modelVersions:e.array(et).default([]),modelRuns:e.array(it).default([]),comparisonSeries:e.array(Je).default([]),comparisonSeriesEntries:e.array(Ye).default([]),modelCurrencySnapshots:e.array(ot).default([]),continualBenchIssueReviews:e.array(lt).default([]),continualLearningDailyBatches:e.array(st).default([]),rewardModelVersions:e.array(tt).default([]),rewardModelRuns:e.array(nt).default([]),modelTasksets:e.array(ae).default([]),plans:e.array(It),bundles:e.array(_e),jobs:e.array(jt),artifacts:e.array(Xn),models:e.array(ea),rolloutReceipts:e.array(_t).default([]),modelBindings:e.array(na).default([]),destinations:e.array(Dn),baseModelCandidates:e.array(Ln).default([]),activityRevision:m.optional(),generatedAt:b}));export{Le as a,Mt as b,qe as c,Ue as d,ya as e,_a as f,Vt as g,ce as h,q as i,Ta as j,Aa as k,Fe as l,Bt as m,Je as n,Ft as o,Jt as p,Wt as q,A as r,Ye as s,Ea as t,et as u,tt as v,Xt as w,nt as x,an as y,on as z,it as A,ai as B,ot as C,$ as D,bi as E,st as F,lt as G,J as H,he as I,bt as J,gn as K,ji as L,ye as M,wi as N,ht as O,hn as P,Pi as Q,ft as R,Z as S,Ei as T,xt as U,St as V,yt as W,Hi as X,Tn as Y,zi as Z,_t as _,O as $,pr as aa,Vn as ba,Dn as ca,Ln as da,Un as ea,$n as fa,It as ga,_e as ha,hr as ia,fr as ja,vr as ka,xr as la,jt as ma,Jn as na,Wn as oa,Yn as pa,Kn as qa,Zn as ra,Sr as sa,Xn as ta,ea as ua,ta as va,na as wa,_r as xa};

package/dist/chunks/chunk-JSQRLLN6.js

@@ -1,2 +0,0 @@
-import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{D as oe,F as V,G as Ae,J as W,K as De,L as ee,M as P,N as Ee,R as ce,S as qe,T as Ce,_ as de,h as Te,i as xe,n as Se,o as Pe,p as j,q as ye,r as K,s as Z,t as Ie,v as He,x as _e,y as z,z as je}from"./chunk-WSOUMXH3.js";import{s as re,v as Be,w as Me}from"./chunk-NG2L6D7G.js";import{C as Q,E as ve,O as v,P as F,Q as M,R as we,b as ke,ca as $,d as D,ia as Re,mb as ie,nb as ne,o as q,ob as be,p as N,s as C,w as Y,z as U}from"./chunk-J572ANHX.js";import{Ra as fe,Va as he,Wa as ge,Xa as J,Ya as X,t as H}from"./chunk-PBZCXPPC.js";import{m as ue,n as u}from"./chunk-VVNV4NNW.js";import{A as c}from"./chunk-PF56N7ZY.js";import{a as b}from"./chunk-FTQLQJQV.js";function Ke(s){let t=P(s.source);if(t.modelResources||t.learningResources)throw new Error("Binding an ordinary Reward requires an ordinary source package.");let a=ee(t);if(!a.trim())throw new Error("Add collection instructions before selecting a saved Reward.");let e=new Map,i=Ae(t);for(let g of i?.assets??[])e.set(g.id,g);let r=new Set([...t.taskset.tasks.flatMap(g=>g.privilegedContextRef?[g.privilegedContextRef]:[]),...[t.environment.actionSchemaRef,t.environment.observationSchemaRef,t.environment.stateSchemaRef].flatMap(g=>g?[g.id]:[])]);for(let g of r){if(e.has(g))continue;let S=t.files.find(se=>se.asset.id===g);if(!S)throw new Error(`Source Taskset context is missing: ${g}.`);let I=new TextDecoder("utf-8",{fatal:!0}).decode(V(S));e.set(g,ke.parse(M({schemaVersion:"openpond.learningTextAsset.v1",id:g,revision:1,asset:S.asset,text:I})))}for(let g of s.assets){let S=t.files.find(I=>I.asset.id===g.id);if(S&&M({asset:S.asset}).contentHash!==M({asset:g.asset}).contentHash)throw new Error(`Selected Reward conflicts with a source asset: ${g.id}.`);e.set(g.id,g)}let{policy:p,environment:o,environmentRelease:k,tools:w,capabilities:n,verifierSetRelease:R}=t.taskset,l=C.parse(M({schemaVersion:"openpond.taskDefinition.v1",id:`${t.taskset.id}-definition`,revision:1,name:t.taskset.id,description:"",instructions:a,category:"custom",familyNamespace:t.taskset.id,inputSchema:{type:"object"},outputSchema:{type:"object"},rewardBinding:v(s.rewardBinding),harness:null,execution:{policy:p,environment:o,environmentRelease:k,tools:w,capabilities:n,verifierSetRelease:R}})),T=je({...s,assets:[...e.values()],source:{taskset:t.taskset,taskDefinition:l,instructionMode:"per_task",executionResources:{environment:t.environment,verifierSet:t.verifierSet},...i?{execution:i.execution}:{}}}),m=new Map(t.files.map(g=>[g.asset.id,g]));for(let g of T.assets){let S="";for(let I of new TextEncoder().encode(g.text))S+=String.fromCharCode(I);m.set(g.id,{asset:g.asset,base64:btoa(S)})}let{taskset:B,executionResources:f,...h}=T;return Ee({schemaVersion:"openpond.tasksetPackage.v1",taskset:B,...f,modelResources:h,files:[...m.values()]})}b(Ke,"bindOrdinaryModelTasksetReward");var _=c.string().trim().min(1).max(500),O=c.string().regex(/^[a-f0-9]{64}$/),We=Be.pick({portableProjectId:!0,name:!0,objective:!0,defaultBaseModel:!0,defaultDestinationId:!0,sourceRevision:!0,sourceUpdatedAt:!0}).extend({trainingSetup:re.omit({tasksetRef:!0,rewardBindingRef:!0,tasksetRelease:!0})}).strict(),Je=c.object({schemaVersion:c.literal("openpond.tasksetPackagePublication.v1"),operationId:_,modelProjectId:_,expectedProjectEtag:O.nullable(),name:c.string().trim().min(1).max(200),description:c.string().max(5e3),buildIntent:c.enum(["demonstrations","preferences","verifiable_reward","rubric","discovery"]),methodHint:c.enum(["sft","dpo","grpo","ppo"]).nullable(),package:W,modelConfiguration:We.optional(),evaluationPackage:W.optional(),selection:c.enum(["select","attach"]).optional()}).strict().refine(s=>s.selection!=="attach"||s.modelConfiguration===void 0,"Attachment-only publication cannot replace Model configuration.").refine(s=>!s.evaluationPackage||s.modelConfiguration!==void 0&&s.modelConfiguration.trainingSetup.evaluationTasksetRef!=null&&u(s.modelConfiguration.trainingSetup.evaluationTasksetRef)===u(v(s.evaluationPackage.taskset)),"Evaluation package must match the selected immutable evaluation in Model configuration."),Xe=c.object({schemaVersion:c.literal("openpond.tasksetPackageReceipt.v1"),teamId:_,modelProjectId:_,operationId:_,taskset:z,packageHash:O,hostedTasksetId:_,projectEtag:O,project:Me.optional(),selection:c.enum(["select","attach"]).optional(),evaluation:c.object({taskset:z,packageHash:O,hostedTasksetId:_}).strict().optional()}).strict(),Ye=c.object({schemaVersion:c.literal("openpond.tasksetPackageReadback.v1"),teamId:_,modelProjectId:_,package:W}).strict(),Qe=c.object({schemaVersion:c.literal("openpond.tasksetReleasePackageReadback.v1"),teamId:_,release:z,package:W}).strict(),y=class extends Error{constructor(a,e,i){super(i);this.status=a;this.code=e;this.name="OpenPondTasksetPackageError"}status;code;static{b(this,"OpenPondTasksetPackageError")}},$e=class{static{b(this,"OpenPondTasksetPackageClient")}#e;constructor(t){let a=new URL(t.baseUrl);if(!["http:","https:"].includes(a.protocol)||a.username||a.password||a.search||a.hash||!t.apiKey.trim()||!t.teamId.trim())throw new Error("Taskset package client requires a clean HTTP(S) endpoint and workspace credentials.");this.#e={...t,baseUrl:a.toString().replace(/\/+$/,"")}}async publish(t,a={}){let e=Je.parse(t);P(e.package),e.evaluationPackage&&P(e.evaluationPackage);let{evaluationPackage:i,...r}=e,p=i?.contentHash===e.package.contentHash?r:e,o=Xe.parse(await this.#t("","POST",p,a.signal)),k=e.package.taskset;if(o.teamId!==this.#e.teamId||o.modelProjectId!==e.modelProjectId||o.operationId!==e.operationId||o.packageHash!==e.package.contentHash||o.taskset.id!==k.id||o.taskset.revision!==k.revision||o.taskset.contentHash!==k.contentHash)throw new y(502,"package_receipt_mismatch","Taskset publication receipt did not match the requested package and owner.");if(e.selection!==void 0&&o.selection!==e.selection)throw new y(502,"package_receipt_mismatch","Taskset publication did not retain its selection mode.");if(e.evaluationPackage&&(!o.evaluation||o.evaluation.packageHash!==e.evaluationPackage.contentHash||u(o.evaluation.taskset)!==u(v(e.evaluationPackage.taskset))))throw new y(502,"package_receipt_mismatch","Taskset publication did not retain the complete selected evaluation package.");if(e.selection==="attach"&&o.projectEtag!==e.expectedProjectEtag)throw new y(502,"package_receipt_mismatch","Attaching a historical Taskset changed the Model configuration.");if(e.modelConfiguration){let w=e.modelConfiguration,n=o.project,R=De(e.package),l=re.parse({...w.trainingSetup,tasksetRef:v(k),rewardBindingRef:R?v(R):null,tasksetRelease:null});if(!n||n.teamId!==this.#e.teamId||n.etag!==o.projectEtag||![n.id,n.portableProjectId].includes(e.modelProjectId)||n.portableProjectId!==w.portableProjectId||n.name!==w.name||n.objective!==w.objective||n.sourceRevision!==w.sourceRevision||Date.parse(n.sourceUpdatedAt)!==Date.parse(w.sourceUpdatedAt)||u(n.defaultBaseModel)!==u(w.defaultBaseModel)||n.defaultDestinationId!==w.defaultDestinationId||u(n.trainingSetup)!==u(l))throw new y(502,"package_receipt_mismatch","Taskset publication did not retain the requested Model configuration.")}return o}async getByRelease(t,a={}){let e=z.parse(t),i=a.expectedPackageHash===void 0?void 0:O.parse(a.expectedPackageHash),r=Qe.parse(await this.#t(`/releases/${encodeURIComponent(e.id)}/${e.revision}/${e.contentHash}/package`,"GET",void 0,a.signal)),p=P(r.package),o={id:p.taskset.id,revision:p.taskset.revision,contentHash:p.taskset.contentHash};if(r.teamId!==this.#e.teamId||u(r.release)!==u(e)||u(o)!==u(e)||i!==void 0&&p.contentHash!==i)throw new y(502,"package_readback_mismatch","Independent Dataset package differs from the requested workspace or exact release.");return p}async get(t,a,e={}){let i=_.parse(t),r=z.parse(a),p=e.expectedPackageHash===void 0?void 0:O.parse(e.expectedPackageHash),o=Ye.parse(await this.#t(`/${encodeURIComponent(i)}/${encodeURIComponent(r.id)}/${r.revision}/${r.contentHash}`,"GET",void 0,e.signal)),k=o.package.taskset;if(o.teamId!==this.#e.teamId||o.modelProjectId!==i||k.id!==r.id||k.revision!==r.revision||k.contentHash!==r.contentHash)throw new y(502,"package_readback_mismatch","Taskset package did not match the requested release and owner.");if(p!==void 0&&o.package.contentHash!==p)throw new y(502,"package_readback_mismatch","Taskset package did not match the saved publication receipt.");return P(o.package)}async#t(t,a,e,i){let r=e===void 0?void 0:JSON.stringify(e);if(r&&new TextEncoder().encode(r).byteLength>oe)throw new y(413,"package_too_large","Taskset publication exceeds the 64 MiB envelope.");let p=await(this.#e.fetch??globalThis.fetch)(`${this.#e.baseUrl}/v1/taskset-packages${t}`,{method:a,headers:{Authorization:`Bearer ${this.#e.apiKey}`,"X-OpenPond-Team-Id":this.#e.teamId,Accept:"application/json",...r?{"Content-Type":"application/json"}:{}},body:r,signal:i,redirect:"error"}),o=p.body?.getReader();if(!o)throw new y(p.status,"empty_response","Taskset package response was empty.");let k=new TextDecoder("utf-8",{fatal:!0}),w=0,n="";try{for(;;){let l=await o.read();if(l.done)break;if(w+=l.value.byteLength,w>oe)throw new y(413,"package_too_large","Taskset response exceeds the 64 MiB envelope.");n+=k.decode(l.value,{stream:!0})}n+=k.decode()}finally{await o.cancel(),o.releaseLock()}let R=JSON.parse(n);if(!p.ok){let l=c.object({code:c.string(),message:c.string()}).safeParse(R);throw new y(p.status,l.success?l.data.code:"package_request_failed",l.success?l.data.message:`Taskset package request failed (${p.status}).`)}return R}};var te=N.shape.id,Le=c.object({schemaVersion:c.literal("openpond.modelBatchReviewInspection.v1"),packageHash:c.string().regex(/^[a-f0-9]{64}$/),definition:C,binding:fe,evidence:c.array(U).min(1).max(1e4),decisions:c.array(ve).min(1).max(1e4)}).strict(),pe=c.object({schemaVersion:c.literal("openpond.modelBatchReviewRequest.v1"),operationId:te,modelId:te,expectedModelRevision:c.number().int().positive(),tasksetRef:N,rewardBindingRef:N.nullable(),definition:c.object({name:c.string().trim().min(1).max(500).optional(),instructions:c.string().trim().min(1).max(2e4).optional(),inputSchema:q.optional(),outputSchema:q.optional(),requiredOutputs:C.shape.requiredOutputs}).strict(),examples:c.array(c.object({evidence:N,input:q.optional(),expected:q.nullable().optional(),evaluatorContext:q.nullable().optional(),proposedTarget:q.nullable().optional()}).strict()).max(1e4)}).strict(),Fe=c.object({schemaVersion:c.literal("openpond.modelBatchReviewReceipt.v1"),operationId:te,modelId:te,source:Y,evidence:c.array(U).min(1).max(1e4),proposals:c.array(Q).max(1e4)}).strict();function ze(s){let{request:t,package:a}=s,e=a.learningResources;if(!e)throw new H("model_review_batch_required",422);let i=$(a.taskset),r=`model-review-${u([s.scope,t.modelId,t.operationId])}`,p=new Map(t.examples.map(d=>[u(d.evidence),d])),o=new Set(e.evidence.map(d=>u(v(d)))),k=new Map(e.decisions.map(d=>[u(d.evidence),d]));if(p.size!==t.examples.length||t.examples.some(d=>!o.has(u(d.evidence))))throw new H("model_review_evidence_mismatch",409);for(let d of[s.binding,...s.rewards,...s.assets])we(d);J(s.binding,s.rewards);let w=new Map;for(let d of s.binding.sources){let x=s.rewards.find(Ue=>F(v(Ue),d.reward)),{contentHash:E,...L}=x;w.set(x.contentHash,he({...L,id:`${r}-reward-${x.contentHash}`,revision:1}))}let n=[...w.values()],{contentHash:R,recipeRef:l,...T}=s.binding,m=ge({...T,id:`${r}-binding`,revision:1,sources:s.binding.sources.map(d=>({...d,reward:v(w.get(d.reward.contentHash))}))},n),B=n.flatMap(d=>[...d.assets,..."verifierRef"in d.implementation?[d.implementation.verifierRef]:[],..."rubricRef"in d.implementation?[d.implementation.rubricRef]:[],..."inputContract"in d.implementation?[d.implementation.inputContract]:[]]),f=[...new Map(B.map(d=>{let x=s.assets.find(E=>E.id===d.id);if(!x)throw new H("model_review_reward_asset_missing",422);if(D(x,d),d.visibility==="policy")throw new H("reward_asset_visibility_invalid",422);return[x.id,x]})).values()],{environmentRelease:h,verifierSetRelease:g,...S}=i.definition.execution,{contentHash:I,...se}=i.definition,A=C.parse(M({...se,...t.definition,id:`${r}-definition`,revision:1,rewardBinding:v(m),execution:{...S,policy:{...S.policy,hiddenGraderRefs:X(m,n).filter(d=>d.privileged).map(d=>d.id)}}})),G=Y.parse(M({schemaVersion:"openpond.learningSource.v1",id:r,revision:1,name:`${A.name} review`,kind:"direct",taskDefinition:v(A),enabled:!0,allowedSplits:[...new Set(e.evidence.map(d=>d.submission.split))],mapping:null,adapterVersion:null,reviewOrigin:{modelId:t.modelId,packageHash:a.contentHash,taskset:v(a.taskset),batch:v(e.batch),rewardBinding:v(s.binding)}})),le=e.evidence.map(d=>{let x=p.get(u(v(d))),E=`review-${d.contentHash}`,L=u({instructions:A.instructions,inputSchema:A.inputSchema,outputSchema:A.outputSchema,requiredOutputs:A.requiredOutputs??[],input:x?.input??d.submission.input})!==u({instructions:i.definition.instructions,inputSchema:i.definition.inputSchema,outputSchema:i.definition.outputSchema,requiredOutputs:i.definition.requiredOutputs??[],input:d.submission.input});return U.parse(M({schemaVersion:"openpond.taskEvidence.v1",id:be(G.id,d.submission.exampleId,E),revision:1,source:v(G),submission:{...d.submission,sourceId:G.id,taskDefinition:v(A),attemptId:E,observedOutput:L?null:d.submission.observedOutput,idempotencyKey:`${r}-${d.contentHash}`,...x?.input===void 0?{}:{input:x.input},...x?.expected===void 0?{}:{expected:x.expected},...x?.evaluatorContext===void 0?{}:{evaluatorContext:x.evaluatorContext}},supersedes:v(d),correctionFeedbackId:null,receivedAt:s.now}))}),Ne=le.flatMap(d=>{let x=p.get(u(d.supersedes)),E=k.get(u(d.supersedes)),L=x?.proposedTarget===void 0?E.approvedTarget:x.proposedTarget;return L===null?[]:[Q.parse({schemaVersion:"openpond.taskFeedbackRecord.v1",id:`feedback-${u([r,d.id])}`,revision:1,submission:{schemaVersion:"openpond.taskFeedback.v1",sourceId:G.id,idempotencyKey:`${r}-${d.id}`,exampleId:d.submission.exampleId,attemptId:d.submission.attemptId,expectedEvidenceHash:d.contentHash,occurredAt:s.now,kind:"target_correction",value:L,note:"Proposed target for the revised task. Grade and review before admission."},status:"pending_review",evidence:v(d),createdAt:s.now,review:null})]});return{source:G,definition:A,binding:m,rewards:n,assets:f,evidence:le,proposals:Ne}}b(ze,"prepareModelBatchReview");var Ve=b(s=>`model-batch-review-${u([s.modelId,s.operationId])}`,"operation");function At(s){let t=P(s);if(!t.learningResources)throw new H("model_review_batch_required",422);let a=$(t.taskset);return Le.parse({schemaVersion:"openpond.modelBatchReviewInspection.v1",packageHash:t.contentHash,definition:a.definition,binding:a.binding,evidence:t.learningResources.evidence,decisions:t.learningResources.decisions})}b(At,"inspectModelBatchPackage");async function Ze(s,t){let a=pe.parse(t),e=await s.operation(Ve(a));if(!e)return null;if(e.requestHash!==u(a))throw new H("learning_idempotency_conflict",409);return Oe(s,a,e.resources)}b(Ze,"findModelBatchReview");async function Dt(s,t){let a=pe.parse(t.request),e=await Ze(s,a);if(e)return e;let i=P(t.package);if(!i.learningResources)throw new H("model_review_batch_required",422);let r=$(i.taskset),p=a.rewardBindingRef,o=!p||F(p,v(r.binding)),k=o?r.binding:await ne(s,"binding",p),w=o?r.rewards:await Promise.all(k.sources.map(m=>ne(s,"reward",m.reward))),n=new Set(w.flatMap(m=>[...m.assets,..."verifierRef"in m.implementation?[m.implementation.verifierRef]:[],..."rubricRef"in m.implementation?[m.implementation.rubricRef]:[],..."inputContract"in m.implementation?[m.implementation.inputContract]:[]]).map(m=>m.id)),R=o?i.learningResources.assets:await Promise.all([...n].map(m=>ie(s,"asset",m,1))),l=ze({...t,request:a,package:i,binding:k,rewards:w,assets:R}),T=[];for(let m of l.assets){let B=await s.get("asset",m.id,1);if(B&&B.contentHash!==m.contentHash)throw new H("model_review_asset_conflict",409);B||await s.put("asset",m,0)}for(let m of l.rewards)await s.put("reward",m,0);await s.put("binding",l.binding,0),await s.put("definition",l.definition,0),await s.put("source",l.source,0,{parentId:a.modelId}),T.push({kind:"source",id:l.source.id,revision:1});for(let m of l.evidence)await s.put("evidence",m,0,{parentId:l.source.id}),T.push({kind:"evidence",id:m.id,revision:1});for(let m of l.proposals)await s.put("feedback",m,0,{parentId:m.evidence.id,status:m.status}),T.push({kind:"feedback",id:m.id,revision:1});return await s.saveOperation(Ve(a),{requestHash:u(a),resources:T}),Oe(s,a,T)}b(Dt,"beginModelBatchReview");async function Oe(s,t,a){let e=await Promise.all(a.map(i=>ie(s,i.kind,i.id,i.revision)));return Fe.parse({schemaVersion:"openpond.modelBatchReviewReceipt.v1",operationId:t.operationId,modelId:t.modelId,source:e.find(i=>i.schemaVersion==="openpond.learningSource.v1"),evidence:e.filter(i=>i.schemaVersion==="openpond.taskEvidence.v1"),proposals:e.filter(i=>i.schemaVersion==="openpond.taskFeedbackRecord.v1")})}b(Oe,"readReceipt");function Ge(s){let t=s.taskset.metadata.ordinaryAuthoring,a=c.record(c.string(),c.array(c.string().min(1)).max(1e5)).parse(t?.judgeCalibrationFixtures??{});return s.taskset.graders.map(e=>{let i={...e,label:e.id,metadata:{portableGrader:e}};if(e.kind==="custom_verifier")return j.parse({...i,module:e.verifierRef.path,exportName:e.exportName??"verify",metadata:{...i.metadata,portableVerifierRef:e.verifierRef}});if(e.kind==="human"||e.kind==="model_judge"){let r=s.files.find(k=>k.asset.id===e.rubricRef.id);if(!r)throw new Error(`Imported rubric is missing: ${e.id}.`);let p=new TextDecoder("utf-8",{fatal:!0}).decode(V(r)),o={...i.metadata,portableRubricRef:e.rubricRef};if(e.kind==="human")return j.parse({...i,rubric:p,metadata:o});if(!e.model)throw new Error(`Imported judge ${e.id} requires its declared model before local preparation.`);return j.parse({...i,rubric:p,metadata:o,judge:e.model,temperature:e.temperature??0,calibrationStatus:"pending",calibrationFixtureRefs:a[e.id]??[]})}return j.parse({...i,kind:e.kind==="artifact"?"file":e.kind,config:e.kind==="artifact"?{...e.config,...e.config.refIncludes===void 0?{}:{pathIncludes:e.config.refIncludes}}:e.config})})}b(Ge,"importedPackageGraders");function ae(s,t,a=[]){let e=J(s,t);return X(s,t).map(i=>{let r=e.find(({source:o})=>o.graderId===i.id).reward,p={...i,label:r.name,metadata:{rewardBinding:v(s)}};if(i.kind==="model_judge"){if(!i.model||i.calibrationStatus!=="passed"||!r.calibrationCheckRef||!r.fixtureSetRef)throw new Error(`Reward ${i.id} needs a retained passing calibration check before preparation.`);let o=a.find(n=>n.id===i.rubricRef.id),k=a.find(n=>n.id===r.fixtureSetRef.id);if(!o||!k)throw new Error(`Reward ${i.id} is missing its immutable rubric or calibration fixtures.`);let w=Re.array().min(1).max(50).parse(JSON.parse(D(k,r.fixtureSetRef)));return j.parse({...p,rubric:D(o,i.rubricRef),judge:i.model,temperature:i.temperature??0,calibrationFixtureRefs:w.map(n=>n.id),metadata:{...p.metadata,portableGrader:i,portableRubricRef:i.rubricRef,calibrationEvidenceHash:r.calibrationCheckRef.contentHash,calibrationCheckRef:r.calibrationCheckRef}})}if(i.kind==="custom_verifier")return j.parse({...p,module:i.verifierRef.path,exportName:i.exportName??"verify",metadata:{...p.metadata,portableVerifierRef:i.verifierRef}});if(i.kind==="human"){let o=a.find(k=>k.id===i.rubricRef.id);if(!o)throw new Error(`Reward ${i.id} is missing its immutable rubric.`);return j.parse({...p,rubric:D(o,i.rubricRef),metadata:{...p.metadata,portableGrader:i,portableRubricRef:i.rubricRef}})}return j.parse({...p,kind:i.kind==="artifact"?"file":i.kind})})}b(ae,"projectLearningBatchGraders");function et(s){return`graders/reward-${s}.js`}b(et,"learningVerifierModule");function ta(s){let t=P(s.package),a=t.taskset,e=t.modelResources,i=t.learningResources?$(a):void 0,r=Se.parse({schemaVersion:"openpond.uploadedFileDatasetSource.v1",kind:"uploaded_file",id:`package-source-${u({profileId:s.profileId,packageHash:t.contentHash})}`,profileId:s.profileId,title:s.name,sourceHash:t.contentHash,occurredAt:s.createdAt,licensingStatus:"pending",secretScanStatus:"pending",piiScanStatus:"pending",originalFileNames:["taskset-package.json"],mediaTypes:["application/json"],sourceFileHashes:[t.contentHash],totalBytes:new TextEncoder().encode(JSON.stringify(t)).byteLength,parserVersion:"openpond.tasksetPackage.v1",metadata:{}}),p=qe({profileId:s.profileId,id:`${a.id}-import`,name:s.name,now:s.createdAt}),o=[];for(let f of a.graders)if(f.kind==="custom_verifier"){let h=e?.assets.find(I=>I.id===f.verifierRef.id),g=t.files.find(I=>I.asset.id===f.verifierRef.id);if(!g)throw new Error(`Imported verifier source is missing: ${f.id}.`);let S=f.verifierRef.path;o.some(I=>I.path===S)||o.push({path:S,role:"verifier",content:h?D(h,f.verifierRef):new TextDecoder("utf-8",{fatal:!0}).decode(V(g))})}let k=a.metadata.ordinaryAuthoring??a.metadata.starterAuthoring,w=ye.array().max(1e5).parse(k?.graderFixtures??[]),n=a.environment.kind==="text"?"chat":a.environment.kind==="custom_program"?"program":a.environment.kind,R=c.record(c.string(),c.record(c.string(),c.string())).parse(a.metadata.workAssetStagingPaths??{}),l=Pe.array().max(1e4).parse(a.metadata.environmentResources??[]),T=l.filter(f=>f.visibility==="policy_visible").map(f=>{let h=t.files.filter(g=>g.asset.path===f.path);if(h.length!==1||h[0].asset.visibility!=="policy")throw new Error(`Policy environment resource requires one public package file: ${f.path}.`);return h[0].asset}),m=new Set(T.map(f=>f.id)),B=ce.parse({schemaVersion:"openpond.taskset.v1",id:a.id,revision:a.revision,profileId:s.profileId,profileRelease:null,createImproveRunId:null,name:s.name,objective:ee(t),status:"needs_review",sourceRefs:[r],datasetArtifact:null,policy:a.policy,environment:{...p.environment,kind:n,entrypoint:a.environment.entrypoint,resources:l,stateful:a.environment.stateful,deterministicSeeds:a.environment.deterministicSeeds,toolNames:a.tools.map(f=>f.name),defaultTimeoutMs:a.environment.defaultTimeoutMs,networkPolicy:a.environment.networkPolicy,metadata:{runtimeSourceTasksetId:tt(s.profileId,t.contentHash),portableEnvironment:a.environment,portableTools:a.tools,portableExecutionResources:{environment:t.environment,verifierSet:t.verifierSet}}},capabilities:{...p.capabilities,taskKind:n==="chat"?"chat":n==="program"?"custom_program":"single_agent",compatibleMethods:["none"],requiresTools:a.tools.length>0,requiresState:a.environment.stateful,rewardKinds:[...new Set(a.graders.map(f=>f.kind==="human"?"human":f.kind==="model_judge"?"model_judge":"deterministic"))]},tasks:a.tasks.map(f=>({...f,schemaVersion:"openpond.taskData.v1",sourceRefs:[r.id],assets:[...new Map([...f.artifactRefs,...(f.requiredOutputs??[]).flatMap(h=>h.submissionMode==="structured_json"&&h.schemaRef?[h.schemaRef]:[]),...T].map(h=>[h.id,h])).values()].map(h=>({id:h.id,sourceRefId:r.id,artifactRef:h.path,fileName:h.path.split("/").at(-1),...m.has(h.id)?{stagingPath:h.path}:R[f.id]?.[h.id]?{stagingPath:R[f.id][h.id]}:{},mediaType:h.mediaType,sha256:h.contentHash,sizeBytes:h.sizeBytes,split:f.split,metadata:{portableAsset:h,...m.has(h.id)?{sharedEnvironmentResource:{id:l.find(g=>g.path===h.path).id,path:h.path}}:{}}})),...f.requiredOutputs?{requiredOutputs:f.requiredOutputs.map(h=>({...h,schemaRef:h.schemaRef?.id??null,maxBytes:h.maxBytes??void 0}))}:{},metadata:{portableTaskRecord:f,exampleOrigin:"imported"}})),graders:e?ae(e.rewardBinding,e.rewards,e.assets):i?ae(i.binding,i.rewards,t.learningResources.assets):Ge(t),graderFixtures:w,learningSignals:p.learningSignals,...a.metrics?{metrics:a.metrics}:{},authoringProvenance:{schemaVersion:"openpond.taskAuthoringProvenance.v1",model:null,modelConfig:{},skillHash:u("openpond-package-import-v1"),promptTemplateVersion:"package-import-v1",buildIntent:"discovery",buildSpecification:null,evidenceHashes:[t.contentHash],tasksetSdkVersion:"package-import-v1",sourceCommit:null,repairHistory:[],createdAt:s.createdAt},readiness:null,contentHash:"00000000",createdAt:s.createdAt,updatedAt:s.createdAt,metadata:{importedPackageHash:t.contentHash,...a.metadata.tasksetOutputContract===void 0?{}:{tasksetOutputContract:a.metadata.tasksetOutputContract},...i?{learning:i}:{},portableFileInventory:t.files.map(f=>({asset:f.asset,sourcePath:f.asset.path})),...e?{taskDefinition:v(e.taskDefinition),rewardBinding:v(e.rewardBinding),rewardExecution:{binding:e.rewardBinding,rewards:e.rewards}}:{},portableCapabilities:a.capabilities,derivedPortableMetadata:a.metadata,...a.metadata.modelTasksetDerivation?{modelTasksetDerivation:a.metadata.modelTasksetDerivation}:{}}});return{package:t,taskset:ce.parse({...B,contentHash:Ce(B)}),generatedFiles:o}}b(ta,"prepareImportedTasksetPackage");function tt(s,t){return`package-${u({profileId:s,packageHash:t})}`}b(tt,"tasksetPackageSourceId");var me=c.object({draft:K,filePaths:c.array(c.object({assetId:c.string().min(1),paths:c.array(xe).min(1).max(1e4)}).strict()).max(1e4)}).strict();function at(s){let t=P(s.source),a=K.parse(s.sourceDraft),e=Te.parse(s.preparation);if(t.contentHash!==e.sourcePackageHash||!F(v(t.taskset),e.sourceTasksetRef))throw new Error("Taskset draft preparation differs from its source package.");if(t.learningResources)throw new Error("Reviewed Tasksets require their reviewed authoring workflow.");if(!!t.modelResources!=(e.authoringGraph==="bound"))throw new Error("Taskset draft authoring graph differs from its source.");let i=de.parse(a.metadata.portableFileInventory??[]),r=b(n=>Z(n)?`source-artifacts/${t.contentHash}/${u(n)}/${n.split("/").at(-1)}`:n,"sourcePath");if([...a.tasks.flatMap(n=>(n.assets??[]).map(R=>R.artifactRef)),...(a.environment.resources??[]).map(n=>n.path),...t.taskset.tasks.flatMap(n=>(n.requiredOutputs??[]).flatMap(R=>R.schemaRef?[R.schemaRef.path]:[])),...a.metrics.customAggregator?[a.metrics.customAggregator.module]:[],...a.graders.flatMap(n=>n.kind==="custom_verifier"?[n.module]:[])].some(Z))throw new Error("A referenced source asset occupies a reserved Taskset authoring path.");let o=t.files.map(n=>{let R=new Set([n.asset.path]);for(let l of i)l.asset.id===n.asset.id&&R.add(l.sourcePath);for(let l of a.tasks)for(let T of l.assets??[])T.id===n.asset.id&&R.add(T.artifactRef);for(let l of t.taskset.graders)if(l.kind==="custom_verifier"&&l.verifierRef.id===n.asset.id){let T=a.graders.find(m=>m.id===l.id);T?.kind==="custom_verifier"&&R.add(T.module)}return{assetId:n.asset.id,paths:[...R].map(r)}}),k={...a.metadata};if(t.modelResources)for(let n of["taskDefinition","rewardBinding","rewardExecution","derivedPortableMetadata","modelTasksetDerivation","importedPackageHash"])delete k[n];let w=K.parse({...a,id:e.draftId,revision:1,status:"draft",graders:a.graders.map(n=>({...n,metadata:{...n.metadata,portableGrader:t.taskset.graders.find(R=>R.id===n.id)}})),environment:{...a.environment,metadata:{...a.environment.metadata,portableExecutionResources:{environment:t.environment}}},modelScope:{modelId:e.lineage.owner.modelId,expectedModelRevision:s.expectedModelRevision,source:e},publishedTasksetRef:e.tasksetRevision>1?e.sourceTasksetRef:null,metadata:{...k,modelTasksetAuthoring:e.lineage,portableFileInventory:i.map(n=>({...n,sourcePath:r(n.sourcePath),asset:{...n.asset,id:Z(n.sourcePath)?`source-artifact-${u({packageHash:t.contentHash,assetId:n.asset.id})}`:n.asset.id,path:r(n.asset.path)}}))}});return me.parse({draft:w,filePaths:o})}b(at,"prepareTasksetDraftSource");function st(s){let t=me.parse(s.initialized),a=P(s.source);if(t.draft.modelScope?.source?.sourcePackageHash!==a.contentHash)throw new Error("Taskset draft lost its source preparation.");let e=new Map;for(let r of t.filePaths){let p=a.files.find(o=>o.asset.id===r.assetId);if(!p)throw new Error("Taskset draft source asset is missing.");for(let o of r.paths){let k=e.get(o);if(k&&k.contentHash!==p.asset.contentHash)throw new Error("Taskset draft source paths conflict.");e.set(o,{path:o,contentHash:p.asset.contentHash,sizeBytes:p.asset.sizeBytes,base64:p.base64})}}for(let r of Ie(t.draft))if(!e.has(r.relativePath)){let p=new TextEncoder().encode(r.source);e.set(r.relativePath,{path:r.relativePath,contentHash:ue(p),sizeBytes:p.byteLength,base64:Buffer.from(p).toString("base64")})}let i=t.draft;if(i.metrics.customAggregator){let r=e.get(i.metrics.customAggregator.module);if(!r)throw new Error("Declared Taskset metric module is missing.");_e(r),i=K.parse({...i,metrics:{...i.metrics,customAggregator:{...i.metrics.customAggregator,contentHash:r.contentHash}}})}return He({schemaVersion:"openpond.tasksetDraftWorkspace.v1",draft:i,files:[...e.values()]})}b(st,"materializeTasksetDraftWorkspace");export{Ke as a,We as b,Je as c,Xe as d,Ye as e,Qe as f,y as g,$e as h,Le as i,pe as j,Fe as k,At as l,Ze as m,Dt as n,ae as o,et as p,ta as q,tt as r,me as s,at as t,st as u};

package/dist/chunks/chunk-LAOLJDVB.js

@@ -1,2 +0,0 @@
-import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{$ as J,Eb as z,Fb as U,Hb as F,Mb as G,Nb as Q,Ob as W,Pb as P,Y as S,aa as K,da as q,t as R}from"./chunk-PBZCXPPC.js";import{g as $,n as r}from"./chunk-VVNV4NNW.js";import{A as e}from"./chunk-PF56N7ZY.js";import{a as I}from"./chunk-FTQLQJQV.js";var X=e.object({schemaVersion:e.literal("openpond.humanResultView.v1"),executionId:e.string().min(1),executionHash:e.string().regex(/^[a-f0-9]{64}$/),selection:e.array(S).max(1e4),coverage:e.object({pending:e.number().int().nonnegative(),submitted:e.number().int().nonnegative(),accepted:e.number().int().nonnegative(),rejected:e.number().int().nonnegative(),cancelled:e.number().int().nonnegative()}).strict(),results:e.array(e.object({review:S,decision:S.nullable(),attemptIds:e.array(e.string().min(1)).min(1).max(2),mode:e.enum(["individual","pairwise"]),origin:e.enum(["hosted_retained","recorded_evidence","owner_attested_local","native_local"]),status:e.enum(["pending","submitted","accepted","rejected","cancelled"]),grader:S,answers:e.array(J).max(100)}).strict()).max(1e4)}).strict(),M=X.extend({contentHash:e.string().regex(/^[a-f0-9]{64}$/)}).strict();function Y(o){let m=o.reviews.map(s=>{let i=K.parse(s),{contentHash:p,...f}=i;if(r(f)!==p||!i.evidence.attempts.some(v=>v.experimentId===o.executionId))throw new R("human_result_selection_mismatch",409);return i});if(new Set(m.map(s=>s.id)).size!==m.length)throw new R("human_result_duplicate_review",422);let n=m.map(s=>{let i=s.decisions.at(-1);return{review:{id:s.id,revision:s.revision,contentHash:s.contentHash},decision:i?{id:i.id,revision:i.revision,contentHash:i.contentHash}:null,attemptIds:s.evidence.attempts.map(p=>p.attemptId),mode:s.form.mode,origin:s.evidence.attempts.some(p=>p.localSource?.kind==="owner_attested_local")?"owner_attested_local":s.evidence.attempts.some(p=>p.localSource?.kind==="native_local")?"native_local":s.evidence.attempts.some(p=>p.recordedSource)?"recorded_evidence":"hosted_retained",status:s.status==="accepted"?"accepted":s.status==="rejected"?"rejected":s.status==="cancelled"?"cancelled":s.status==="submitted"?"submitted":"pending",grader:s.evidence.grader,answers:s.status==="accepted"&&i?.outcome==="accept"?i.answers:[]}}),y={pending:0,submitted:0,accepted:0,rejected:0,cancelled:0};for(let s of n)y[s.status]++;return M.parse(q(X.parse({schemaVersion:"openpond.humanResultView.v1",executionId:o.executionId,executionHash:o.executionHash,selection:n.map(s=>s.review),coverage:y,results:n})))}I(Y,"createHumanResultView");function Z(o,m){let n=K.parse(o),{contentHash:y,...s}=n,i=n.decisions.at(-1);if(y!==r(s)||y!==m.reviewHash||n.status!=="accepted"||!i||i.outcome!=="accept"||i.contentHash!==m.decisionHash)throw new R("human_accepted_revision_unavailable",409);let p=i.submissions.map(f=>{let v=n.submissions.find(w=>w.id===f.id&&w.contentHash===f.contentHash);if(!v||v.evidenceHash!==r(n.evidence)||v.formHash!==r(n.form))throw new R("human_accepted_lineage_invalid",409);return{id:v.id,revision:v.revision,contentHash:v.contentHash}});return{schemaVersion:"openpond.acceptedHumanEvidence.v1",scope:n.scope,projectId:n.projectId,review:{id:n.id,revision:n.revision,contentHash:y},sourceOrigins:n.evidence.attempts.map(f=>f.localSource?f.localSource.kind:f.recordedSource?"recorded_evidence":"hosted_retained"),decision:{id:i.id,revision:i.revision,contentHash:i.contentHash},submissions:p,evidence:n.evidence,answers:i.answers,...n.execution?.result?{execution:{id:n.execution.executionId,result:n.execution.result,configurationHash:n.execution.configurationHash}}:{},...i.taskProposalHash?{taskProposal:n.submissions.find(f=>f.taskProposal?.contentHash===i.taskProposalHash).taskProposal}:{}}}I(Z,"acceptedHumanEvidence");var T=e.string().regex(/^[a-f0-9]{64}$/),V=e.string().min(1).max(240),L={feedbackKey:V,criterionId:V},he=e.discriminatedUnion("kind",[e.object({...L,kind:e.literal("normalized_score"),minimum:e.number().finite(),maximum:e.number().finite(),passThreshold:e.number().min(0).max(1).optional()}).strict(),e.object({...L,kind:e.literal("boolean")}).strict(),e.object({...L,kind:e.literal("category")}).strict(),e.object({...L,kind:e.literal("preference_score"),tieScore:e.number().min(0).max(1).nullable()}).strict()]),ee=e.object({execution:$,resultHash:T,reviews:e.array(e.object({review:S,decision:S}).strict()).min(1).max(1e4)}).strict(),ie=e.object({rules:e.array(he).min(1).max(100),baseline:ee,candidate:ee}).strict().superRefine((o,m)=>{new Set(o.rules.map(n=>n.feedbackKey)).size!==o.rules.length&&m.addIssue({code:"custom",message:"Select each Human feedback key once."})}),ne=e.object({manifest:F,result:G}).strict(),te=e.object({id:V,manifestHash:T,resultHash:T}).strict(),ae=e.object({schemaVersion:e.literal("openpond.humanComparisonProjection.v1"),teamId:V,projectId:V,selections:ie,original:e.object({baseline:te,candidate:te}).strict(),humanResults:e.object({baseline:M,candidate:M}).strict(),baseline:ne,candidate:ne}).strict(),se=ae.extend({contentHash:T}).strict();function d(o){throw new R(o,409)}I(d,"fail");async function Ve(o){let m=ie.parse(o.selections);await o.authorize();let n={baseline:P(o.baseline),candidate:P(o.candidate)},y=new Set([m.baseline.execution.id,m.candidate.execution.id]);y.size!==2&&d("human_comparison_distinct_executions_required");let s=0;async function i(H){let l=n[H],j=m[H],_=l.manifest.lineage?.execution;l.manifest.lineage?.humanProjection&&d("human_comparison_recursive_projection"),(!_||r(_)!==r(j.execution)||l.manifest.teamId!==o.teamId)&&d("human_comparison_execution_changed");let E=[],A=new Map;for(let a of j.reviews){let t=await o.readReview(a.review);(t.scope!==o.teamId||t.projectId!==o.projectId||t.kind!=="grade"||t.revision!==a.review.revision||t.id!==a.review.id)&&d("human_comparison_review_scope_changed");let b=Z(t,{reviewHash:a.review.contentHash,decisionHash:a.decision.contentHash});s+=new TextEncoder().encode(JSON.stringify({answers:b.answers,form:t.form})).byteLength,s>4194304&&d("human_comparison_projection_too_large"),r(b.decision)!==r(a.decision)&&d("human_comparison_decision_changed"),t.evidence.attempts.some(k=>!y.has(k.experimentId))&&d("human_comparison_foreign_attempt");let u=t.evidence.attempts.filter(k=>k.experimentId===_.id);u.length!==1&&d("human_comparison_attempt_ambiguous");let c=u[0],g=await o.caseIdentity(c),h=l.result.cases.find(k=>z(k.identity)===z(g));(!h||g.caseId!==c.taskId||!h.traceRef||h.traceRef.contentHash!==c.trace?.contentHash||h.output===null)&&d("human_comparison_receipt_changed");let x=z(g)+":"+t.evidence.grader.id;A.has(x)&&d("human_comparison_duplicate_assessment"),A.set(x,{review:t,accepted:b,attempt:c}),E.push(t)}let D=Y({executionId:_.id,executionHash:_.contentHash,reviews:E});(D.contentHash!==j.resultHash||D.coverage.accepted!==E.length)&&d("human_comparison_result_changed");let N=new Map(m.rules.map(a=>[a.feedbackKey,a])),oe=l.manifest.evaluators.map(a=>{let t=N.get(a.feedbackKey);if(!t)return a;let b=E.filter(c=>c.evidence.grader.id===a.release.id&&String(c.evidence.grader.revision)===String(a.release.revision)&&c.evidence.grader.contentHash===a.release.contentHash).map(c=>c.form);(!b.length||new Set(b.map(r)).size!==1)&&d("human_comparison_form_changed");let u=b[0].criteria.find(c=>c.id===t.criterionId);return u||d("human_comparison_criterion_unavailable"),(t.kind==="normalized_score"&&(u.kind!=="score"||t.minimum!==u.minimum||t.maximum!==u.maximum||t.maximum<=t.minimum)||t.kind==="boolean"&&u.kind!=="boolean"||t.kind==="category"&&u.kind!=="category"||t.kind==="preference_score"&&(u.kind!=="preference"||b[0].mode!=="pairwise"))&&d("human_comparison_rule_changed"),{...a,output:t.kind==="boolean"?"boolean":t.kind==="category"?"category":"score",categories:u.kind==="category"?u.options:[],configurationHash:r({original:a.configurationHash,rule:t,formHash:r(b[0])})}});m.rules.some(a=>!l.manifest.evaluators.some(t=>t.feedbackKey===a.feedbackKey))&&d("human_comparison_grader_unavailable");let{contentHash:ce,...re}=l.manifest,C={originalManifest:{id:l.manifest.id,contentHash:l.manifest.contentHash},selectionHash:r(j),rulesHash:r(m.rules)},B=Q({...re,id:U(C.originalManifest,C.selectionHash,C.rulesHash),lineage:{...l.manifest.lineage,humanProjection:C},evaluators:oe}),de=l.result.cases.map(a=>{let t=a.feedback.map(u=>{let c=N.get(u.feedbackKey);if(!c)return u;let g=A.get(z(a.identity)+":"+u.evaluator.id);(!g||g.review.evidence.grader.contentHash!==u.evaluator.contentHash)&&d("human_comparison_assessment_missing");let h=g.accepted.answers.find(k=>k.criterionId===c.criterionId),x=null;if(h&&!h.abstain&&(c.kind==="normalized_score"&&typeof h.value=="number"&&(x=(h.value-c.minimum)/(c.maximum-c.minimum)),c.kind==="boolean"&&typeof h.value=="boolean"&&(x=h.value),c.kind==="category"&&typeof h.value=="string"&&(x=h.value),c.kind==="preference_score")){let k=g.review.evidence.attempts.findIndex(O=>O.experimentId===_.id&&O.attemptId===g.attempt.attemptId);x=h.value==="tie"?c.tieScore:h.value==="A"||h.value==="B"?h.value===(k===0?"A":"B")?1:0:null}return{...u,status:x===null?"unavailable":"scored",value:x,passed:x===null?null:c.kind==="boolean"?x:c.kind==="normalized_score"&&c.passThreshold!==void 0?Number(x)>=c.passThreshold:null,reasoning:h?.note??null,evidenceRefs:[{id:g.accepted.review.id,contentHash:g.accepted.review.contentHash,mediaType:"application/vnd.openpond.human-review+json",sizeBytes:null},{id:g.accepted.decision.id,contentHash:g.accepted.decision.contentHash,mediaType:"application/vnd.openpond.human-decision+json",sizeBytes:null}]}}),b=a.output!==null&&t.every(u=>u.status==="scored");return{...a,feedback:t,status:b?"completed":a.status,error:b?null:a.error}}),{contentHash:me,...ue}=l.result,le=W({...ue,manifest:{id:B.id,contentHash:B.contentHash},cases:de},B);for(let a of E){let t=await o.readReview({id:a.id,revision:a.revision,contentHash:a.contentHash});(t.contentHash!==a.contentHash||t.status!=="accepted")&&d("human_comparison_review_changed")}return await o.authorize(),{evidence:{manifest:B,result:le},humanResult:D}}I(i,"lane");let p=await i("baseline"),f=await i("candidate");for(let H of[...m.baseline.reviews,...m.candidate.reviews]){let l=await o.readReview(H.review);(l.contentHash!==H.review.contentHash||l.status!=="accepted"||l.decisions.at(-1)?.contentHash!==H.decision.contentHash)&&d("human_comparison_review_changed")}await o.authorize();let v={baseline:{id:n.baseline.manifest.id,manifestHash:n.baseline.manifest.contentHash,resultHash:n.baseline.result.contentHash},candidate:{id:n.candidate.manifest.id,manifestHash:n.candidate.manifest.contentHash,resultHash:n.candidate.result.contentHash}},w=ae.parse({schemaVersion:"openpond.humanComparisonProjection.v1",teamId:o.teamId,projectId:o.projectId,selections:m,original:v,humanResults:{baseline:p.humanResult,candidate:f.humanResult},baseline:p.evidence,candidate:f.evidence});return new TextEncoder().encode(JSON.stringify(w)).byteLength>8388608&&d("human_comparison_projection_too_large"),se.parse({...w,contentHash:r(w)})}I(Ve,"projectAcceptedHumanComparison");function Ce(o,m){let n=se.parse(o),{contentHash:y,...s}=n;y!==r(s)&&d("human_comparison_projection_invalid");for(let i of["baseline","candidate"]){let p=P(n[i]),f=p.manifest.lineage,v=f?.humanProjection,w=n.original[i],H=n.selections[i];(!f||!v||v.originalManifest.id!==w.id||v.originalManifest.contentHash!==w.manifestHash||v.selectionHash!==r(H)||v.rulesHash!==r(n.selections.rules)||r(f.execution)!==r(H.execution)||p.manifest.teamId!==n.teamId||n.humanResults[i].contentHash!==H.resultHash)&&d("human_comparison_projection_lineage_changed");let{contentHash:l,...j}=n.humanResults[i];if((r(j)!==l||j.executionId!==H.execution.id||j.executionHash!==H.execution.contentHash||r(j.selection)!==r(H.reviews.map(_=>_.review)))&&d("human_comparison_human_result_changed"),m){let _=P(m[i]);(_.manifest.lineage?.humanProjection||_.manifest.id!==w.id||_.manifest.contentHash!==w.manifestHash||_.result.contentHash!==w.resultHash||r(_.manifest.lineage?.execution)!==r(H.execution))&&d("human_comparison_original_changed")}}return n}I(Ce,"verifyHumanComparisonProjection");export{M as a,Y as b,ie as c,se as d,Ve as e,Ce as f};

package/dist/chunks/chunk-LKKMLBOH.js

@@ -1,3 +0,0 @@
-import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
-import{a as n}from"./chunk-FTQLQJQV.js";var s="127.0.0.1";var d="0.2.63",h=["You are OpenPond Chat. Respond in the user's language. If the user's latest message is language-neutral, ambiguous, or only a short test, respond in English. Be concise and directly answer the latest request. Do not use emojis. Use Markdown when it improves scanability.","If you emit reasoning or thinking content, keep it sparse and user-readable: at most one short sentence for major progress, decisions, or blockers. Omit reasoning for routine searches, reads, tool calls, and obvious next steps. Do not restate the user request, narrate every action, or include code blocks, code excerpts, diffs, raw tool payloads, or markdown examples in reasoning. Put necessary code or exact snippets only in the final assistant answer.","When the user asks to show or preview an image that is available as a workspace path or signed image URL, use Markdown image syntax like ![description](path-or-url) instead of a bare path or raw HTML.","Final responses:","- After substantial work, including research, analysis, document creation, and coding, finish with a concise, self-contained wrap-up. The user should understand the outcome without reading progress updates or tool output.","- Lead with the result and summarize the meaningful changes or findings. Link relevant deliverables and supporting sources when available, using links the user can open.","- Report what was actually checked and what the evidence establishes when verification matters. Distinguish completed work from anything unfinished, unverified, or still running; never imply that starting an operation confirms its success.","- If genuinely blocked, state what was completed, the blocker, and the specific input or action needed to continue. Continue authorized work until complete or genuinely blocked before giving the wrap-up.","- For code changes, explain what changed and why, reference important files when useful, and report relevant checks and their results. Use Changed Files or Verification sections only when they make the answer easier to read.","- Match the length and structure to the task and respect the user's requested output format. Simple questions and small actions need a direct answer, without a separate recap or fixed headings. Avoid repeating the work log or adding generic next steps.","Do not mention raw tool JSON, internal repo paths, or origin/remote URLs unless the user explicitly asks for them or they are necessary to explain a git/deploy failure."].join(`
-`);import{randomUUID as c}from"node:crypto";import i from"node:path";import{fileURLToPath as p}from"node:url";function u(){return new Date().toISOString()}n(u,"now");function l(e){if(typeof e=="string")return e;if(e instanceof Error)return e.message||e.name;if(e==null)return"";try{return JSON.stringify(e)}catch{return String(e)}}n(l,"textFromUnknown");function k(e){if(!e||typeof e!="object")return"";let t=e;for(let r of["delta","text","output","chunk","message"])if(typeof t[r]=="string")return t[r];return t.delta&&typeof t.delta=="object"?l(t.delta):""}n(k,"extractDelta");function x(e){return{id:c(),timestamp:u(),...e}}n(x,"event");function E(e){let t=e.replace(/^https?:\/\//,""),[r,o]=t.split(":");return{host:r||s,port:o?Number.parseInt(o,10):17874}}n(E,"parseListen");function b(e){return(process.argv[1]?i.resolve(process.argv[1]):"")===i.resolve(p(e))}n(b,"isCliEntrypoint");export{s as a,d as b,h as c,u as d,l as e,k as f,x as g,E as h,b as i};