@tangle-network/agent-eval 0.201.0 → 0.202.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +5 -0
- package/dist/adapters/http.d.ts +1 -1
- package/dist/benchmarks/index.d.ts +1 -1
- package/dist/benchmarks/index.js +1 -1
- package/dist/campaign/index.d.ts +2 -2
- package/dist/campaign/index.js +2 -2
- package/dist/{campaign-PmTWZi72.js → campaign-XXmOU0R0.js} +236 -3
- package/dist/campaign-XXmOU0R0.js.map +1 -0
- package/dist/contract/index.d.ts +3 -3
- package/dist/{define-agent-eval-1YuvJpOs.d.ts → define-agent-eval-DQXLPqWc.d.ts} +2 -2
- package/dist/{define-agent-eval-1YuvJpOs.d.ts.map → define-agent-eval-DQXLPqWc.d.ts.map} +1 -1
- package/dist/{index-EkRNsLr4.d.ts → index-D1ZX28rV.d.ts} +3 -3
- package/dist/{index-EkRNsLr4.d.ts.map → index-D1ZX28rV.d.ts.map} +1 -1
- package/dist/{index-DKmrHA02.d.ts → index-giYkbitB.d.ts} +47 -2
- package/dist/index-giYkbitB.d.ts.map +1 -0
- package/dist/index.d.ts +2 -2
- package/dist/meta-eval/index.d.ts +48 -5
- package/dist/meta-eval/index.d.ts.map +1 -1
- package/dist/meta-eval/index.js +85 -21
- package/dist/meta-eval/index.js.map +1 -1
- package/dist/openapi.json +1 -1
- package/docs/evaluator-construction.md +71 -0
- package/package.json +1 -1
- package/dist/campaign-PmTWZi72.js.map +0 -1
- package/dist/index-DKmrHA02.d.ts.map +0 -1
package/CHANGELOG.md
CHANGED
|
@@ -4,6 +4,11 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
|
|
|
4
4
|
|
|
5
5
|
---
|
|
6
6
|
|
|
7
|
+
## [0.202.0] — 2026-09-28
|
|
8
|
+
|
|
9
|
+
- feat(meta-eval): audit planning and canonical schemas for evaluator construction (#888)
|
|
10
|
+
- fix(test): stop hosted receivers by process group and give loop tests room (#891)
|
|
11
|
+
|
|
7
12
|
## [0.201.0] — 2026-09-28
|
|
8
13
|
|
|
9
14
|
- chore(deps): update all dependencies to latest (#889)
|
package/dist/adapters/http.d.ts
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import { m as CostReceipt } from "../cost-ledger-DbQdN3nO.js";
|
|
2
2
|
import { R as Scenario, d as DispatchContext, f as DispatchFn } from "../types-DfvQOe91.js";
|
|
3
|
-
import "../index-
|
|
3
|
+
import "../index-D1ZX28rV.js";
|
|
4
4
|
//#region src/adapters/http.d.ts
|
|
5
5
|
export interface HttpDispatchOptions<TScenario extends Scenario, _TArtifact> {
|
|
6
6
|
/** Static endpoint URL. Mutually exclusive with `resolveUrl`. */
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import { c as RunSplitTag } from "../run-record-DF-fuu31.js";
|
|
2
2
|
import { a as CampaignResult } from "../types-DfvQOe91.js";
|
|
3
|
-
import "../index-
|
|
3
|
+
import "../index-giYkbitB.js";
|
|
4
4
|
import { Yt as CampaignStorage } from "../search-kernel-C9CXvFR7.js";
|
|
5
5
|
import { a as BenchmarkFamily, c as BenchmarkSource, i as BenchmarkEvaluation, l as BenchmarkTaskKind, n as BenchmarkAdapter, o as BenchmarkResponder, r as BenchmarkDatasetItem, s as BenchmarkScenario, t as BENCHMARK_SPLIT_SEED, u as deterministicSplit } from "../types-C4hH69a9.js";
|
|
6
6
|
//#region src/benchmarks/calibration.d.ts
|
package/dist/benchmarks/index.js
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { t as __exportAll } from "../rolldown-runtime-8H4AJuhK.js";
|
|
2
2
|
import { rt as runCampaign } from "../llm-judge-C-0P1WiT.js";
|
|
3
3
|
import { S as fsCampaignStorage } from "../external-optimizer-subprocess-CkyUUV7h.js";
|
|
4
|
-
import "../campaign-
|
|
4
|
+
import "../campaign-XXmOU0R0.js";
|
|
5
5
|
import { join, resolve } from "node:path";
|
|
6
6
|
import { readFile, readdir, stat } from "node:fs/promises";
|
|
7
7
|
import { parse } from "smol-toml";
|
package/dist/campaign/index.d.ts
CHANGED
|
@@ -4,7 +4,7 @@ import { t as SeriesDistribution } from "../descriptive-B2iPaT9J.js";
|
|
|
4
4
|
import { $ as paretoSignificanceGate, B as PowerPreflight, G as BuildEvidenceVectorOptions, H as powerPreflight, J as ParetoSignificanceGateOptions, K as EvidenceVector, Q as paretoPolicy, U as AxisEvidence, V as PowerPreflightOptions, W as AxisVerdict, X as PromotionPolicy, Y as PromotionObjective, Z as buildEvidenceVector, _ as sequentialPairedGate, a as detectScale, c as pairHoldout, d as SequentialDecision, f as SequentialObservation, g as sequentialDecide, h as SequentialStreamState, i as PairedHoldout, l as SequentialDecideFn, m as SequentialPairedGateOptions, n as HeldoutSignificance, o as dimensionRegressions, p as SequentialPairedGate, q as ObjectiveSource, r as HeldoutSignificanceOptions, s as heldoutSignificance, t as DimensionRegression, u as SequentialDecideOptions } from "../statistical-heldout-HswMvzbP.js";
|
|
5
5
|
import { $ as SearchTokenAccounting, A as SearchLedgerHash, B as SearchOperationStartedEvent, C as SearchEdgeOperator, D as SearchFailureReason, E as SearchExecutionIdentity, F as SearchNodeRegisteredEvent, G as SearchSplit, H as SearchProposerKind, I as SearchNodeStatus, J as SearchSurfaceEffect, K as SearchSplitTasks, L as SearchOpenedEvent, M as SearchNodeDecidedEvent, N as SearchNodeDecision, O as SearchLedgerEntry, P as SearchNodeRef, Q as SearchTaskOutcome, R as SearchOperationKind, S as SearchEdgeAttribution, T as SearchEstimateMethod, U as SearchReservation, V as SearchProposer, W as SearchSourceRef, X as SearchSurfaceKind, Y as SearchSurfaceEvidence, Z as SearchTask, _ as SearchClaimPower, a as SearchAttemptAccounting, b as SearchClosedEvent, c as SearchCancelReason, d as SearchCellCancelledEvent, et as SearchTraceRef, f as SearchCellSettledEvent, g as SearchClaimFinalist, h as SearchClaimEstimator, i as SearchArtifactRef, j as SearchModelIdentity, k as SearchLedgerEvent, l as SearchCandidateSurface, m as SearchClaim, n as SEARCH_LEDGER_SCHEMA, o as SearchAudit, p as SearchCellStage, q as SearchSplits, r as SearchArtifactKind, s as SearchBudget, t as NodeEstimate, tt as SearchUnknown, u as SearchCellAllocatedEvent, v as SearchClaimTest, w as SearchEdgeRecordedEvent, x as SearchCostAccounting, y as SearchCloseReason, z as SearchOperationRecordedEvent } from "../search-ledger-types-p2xKfiDJ.js";
|
|
6
6
|
import { $ as SearchNode, A as LabeledScenarioWrite, B as ScoredSurfaceOutcome, C as JudgeDimension, D as LabeledScenarioSampleArgs, E as LabeledScenarioRecord, F as ProposeContext, G as labelTrustRank, H as SurfaceProposer, I as ProposedCandidate, J as renderSearchSummary, K as SearchProposerView, L as RedactionStatus, M as OptimizerConfig, N as ParetoParent, O as LabeledScenarioSource, P as ProposalTrackContext, Q as SearchDecisionRecord, R as Scenario, S as JudgeConfig, T as LabelTrust, U as TraceSpan, V as SessionScript, W as isProposedCandidate, X as SearchCell, Y as searchProposerView, Z as SearchCompletion, _ as GateDecision, a as CampaignResult, at as SearchUnitScore, b as GenerationRecord, c as CampaignTraceWriter, ct as searchTaskSetDigest, d as DispatchContext, et as SearchOperation, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, it as SearchStateView, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, lt as searchUnitScores, m as GateCheckStatus, n as CampaignArtifactWriter, nt as SearchSpend, o as CampaignScenarioIdentity, ot as searchCellId, p as Gate, q as SearchSummaryOptions, r as CampaignCellResult, rt as SearchState, s as CampaignTokenUsage, st as searchNodeId, t as CampaignAggregates, tt as SearchScoredCell, u as ComponentSurface, v as GateResult, w as JudgeScore, x as JudgeAggregate, y as GenerationCandidate, z as ScenarioAggregate } from "../types-DfvQOe91.js";
|
|
7
|
-
import { $ as
|
|
7
|
+
import { $ as ScoreboardRow, $i as RunImprovementLoopOptions, $n as TraceAnalystScenario, $r as GepaOptimizationRecipe, $t as importExternalEvaluations, A as ScenarioSignal, Aa as assertCampaignDesign, Ai as GepaCandidatePopulationSummary, An as CrossSurfaceComponentEvidence, B as FinalizedProfileMatrixResult, Bn as CrossSurfaceInteractionTask, Bt as FsLabeledScenarioStoreOptions, C as SingleRunLockOptions, Ca as CacheRead, Ci as compareOptimizationMethods, Cn as CrossSurfaceBootstrapPolicy, Cr as campaignMeanComposite, D as searchProposalExecution, Da as cellCachePath, Di as costFromLedgerSummary, Dn as CrossSurfaceCandidateOutcome, Dr as DefaultProductionGateOptions, E as searchModelIdentity, Ea as buildCellSchedule, Ei as combineComparisonCosts, En as CrossSurfaceCandidateEvidence, Er as DefaultProductionGateCheck, F as RetainedDispatchOptions, Fi as ExternalOptimizerObservationSummary, Fn as CrossSurfaceIneligibilityReason, G as RunProfileMatrixSegmentOptions, Gn as CrossSurfacePairwiseEntry, Gr as SkillOptOptimizationMethodConfig, Gt as ScoredRollout, H as ProfileMatrixPlan, Hn as CrossSurfacePairCompatibility, Ht as RolloutArgumentDiff, I as RetainedDispatchOutcome, Ii as ExternalOptimizerSubmittedCandidate, In as CrossSurfaceInteractionAwareSelection, It as scopedOptimizationMethod, J as runProfileMatrixSegment, Ji as llmJudge, Jn as CrossSurfaceSelectionPolicy, Jr as skillOptOptimizationMethod, Jt as rolloutArgumentDiff, K as createProfileMatrixPlan, Ki as LlmJudgeDimension, Kn as CrossSurfaceRankedSingle, Kr as SkillOptRunnerCommand, Kt as UngroundedLiteralReport, L as createRetainedDispatch, Li as readExternalOptimizerObservationArtifact, Ln as CrossSurfaceInteractionEffect, Lt as sequentialOptimizationMethod, M as selectDiscriminative, Ma as campaignScenarioIdentity, Mi as readGepaCandidatePopulationArtifact, Mn as CrossSurfaceDistribution, N as resolveRunDir, Na as campaignSplitDigest, Ni as ExternalOptimizerExecutionSummary, Nn as CrossSurfaceEligibility, O as searchReceiptAccounting, Oa as FinalEvidencePolicy, Oi as GepaCandidatePopulationArtifact, On as CrossSurfaceCandidateSummary, Or as DefaultProductionRewardHackingOptions, P as tangleTracesRoot, Pa as campaignSplitDigestFromIdentities, Pi as ExternalOptimizerObservationArtifact, Pn as CrossSurfaceEvidenceBreakdown, Q as ScoreboardRenderOptions, Qi as transientDispatchFailure, Qn as TraceAnalystArtifact, Qr as GepaOptimizationMethodConfig, Qt as externalSurface, R as CreateProfileMatrixPlanOptions, Rn as CrossSurfaceInteractionPath, Rt as neutralizeText, S as SingleRunLock, Sa as CacheIssueReason, Si as OptimizationTokenUsage, Sn as CrossSurfaceBestSingleSelection, Sr as campaignBreakdown, T as campaignCellSearchResult, Ta as CellScheduleSlot, Ti as ComparisonCost, Tn as CrossSurfaceCandidateComparison, Tr as heldOutGate, U as ProfileMatrixRow, Ui as ReferenceEquivalenceScenario, Un as CrossSurfacePairEvidence, Ut as RolloutArgumentDiffOptions, V as ProfileMatrixCoverage, Vi as ReferenceEquivalenceJudgeOptions, Vn as CrossSurfaceNaiveStackSelection, Vt as LabeledScenarioStoreError, W as ProfileMatrixSegmentResult, Wi as createReferenceEquivalenceJudge, Wn as CrossSurfacePairIncompatibilityReason, Wr as composeGate, Wt as RolloutCall, X as PlaybackDriver, Xi as isTransientTransportFailure, Xn as CrossSurfaceTaskRow, Xr as GepaEngineOptions, Xt as GepaEvaluationImport, Y as PlaybackContext, Yi as TransientFailureOptions, Yn as CrossSurfaceSelections, Yr as GepaAdaptiveEngineRun, Yt as ExternalSearchOptimizer, Z as PlaybackStep, Zi as quotaExhaustedUntil, Zn as BuildTraceAnalystSurfaceDispatchOptions, Zr as GepaEngineRun, Zt as GepaPopulationImport, _ as componentSurfaceIdentityMaterial, _a as runCampaign, _i as OptimizationMethodProvenance, _n as analyzeCrossSurfaceInteractions, _r as provenanceRecordPath, a as WorktreeAdapterError, aa as runOptimization, ai as ExternalTextOptimizationMethodConfig, an as EvalFixture, ar as LoopProvenanceArgsFromResult, at as scoreUserStory, b as surfaceDispatchRef, ba as PlanCampaignRunOptions, bi as OptimizationMethodScore, bn as CrossSurfaceAdditionRejectionReason, br as CampaignBreakdown, c as verifyCodeSurface, ca as SearchLedgerIntegrityError, ci as ExternalOptimizationExample, cn as EvalFixtureRunPlan, cr as LoopProvenanceEvidence, ct as ProfileDispatchFn, d as PhoenixEvaluationResultLike, da as openAutoPr, di as CompareOptimizationMethodsOptions, dn as LoadEvalFixtureScenariosOptions, dr as buildLoopProvenanceRecord, dt as RunProfileMatrixOptions, ea as RunImprovementLoopResult, ei as GepaRunnerCommand, en as importGepaPopulation, er as buildTraceAnalystSurfaceDispatch, et as ScoreboardSummary, f as PhoenixEvaluatorLike, fa as RunEvalOptions, fi as OptimizationMethod, fn as PlanEvalFixtureRunOptions, fr as campaignMeasurementDigest, ft as RunProfileMatrixResult, g as codeSurfaceIdentityMaterial, ga as RunCampaignOptions, gi as OptimizationMethodPairwise, gn as planEvalFixtureRun, gr as loopProvenanceSpans, h as assertCodeSurfaceIdentity, ha as CampaignCellRetryPolicy, hi as OptimizationMethodInput, hn as loadEvalFixtureScenarios, hr as loopProvenanceArgsFromResult, i as WorktreeAdapter, ia as RunOptimizationResult, ii as externalTextOptimizationMethod, in as neutralizationGate, ir as EmitLoopProvenanceResult, it as renderScoreboardMarkdown, j as scoreDiscrimination, ja as assertCampaignSplitIdentity, ji as GepaCandidateSelectionScore, jn as CrossSurfaceCompositionStep, k as DiscriminationScore, ka as FinalEvidenceUse, ki as GepaCandidatePopulationCandidate, kn as CrossSurfaceComponent, kr as defaultProductionGate, l as AutoevalsScoreLike, la as OpenAutoPrOptions, li as ExternalTextEvaluationResponse, ln as EvalFixtureScenario, lr as LoopProvenanceOptimizationMethod, lt as ProfileMatrixError, m as phoenixEvaluatorJudge, ma as CampaignCellFailureReceipt, mi as OptimizationMethodComposition, mn as loadEvalFixture, mr as emitLoopProvenance, mt as runProfileMatrix, n as GitWorktreeAdapterOptions, na as PremeasuredOptimizationBaseline, ni as OpenAICompatibleOptimizerModel, nn as recordSkillOptSearch, nr as BuildLoopProvenanceArgs, nt as UserStoryVerdict, o as gitWorktreeAdapter, oa as SearchLedgerConflictError, oi as ExternalTextOptimizerContext, on as EvalFixtureFile, or as LoopProvenanceBackend, ot as scoreboardSummary, p as autoevalsScorerJudge, pa as runEval, pi as OptimizationMethodComparison, pn as discoverEvalFixtures, pr as canonicalDigest, pt as ScenarioRollup, q as finalizeProfileMatrix, qi as LlmJudgeOptions, qn as CrossSurfaceRelativeCost, qr as SkillOptTrainerConfig, qt as classifyUngroundedLiterals, r as Worktree, ra as RunOptimizationOptions, ri as OptimizerModelBudget, rn as NeutralizationGateOptions, rr as EmitLoopProvenanceArgs, rt as makePlaybackDispatch, s as resolveWorktreePath, sa as SearchLedgerError, si as ExternalTextOptimizerResult, sn as EvalFixtureLoadOptions, sr as LoopProvenanceCandidate, st as userStoryScoreboard, t as CodeSurfaceVerification, ta as runImprovementLoop, ti as gepaOptimizationMethod, tn as recordGepaSearch, tr as traceAnalystQualityJudge, tt as UserStory, u as AutoevalsScorerLike, ua as OpenAutoPrResult, ui as decodeExternalTextCandidate, un as EvalFixtureValidationMode, ur as LoopProvenanceRecord, ut as ProfileSummary, v as renderSurfaceDiff, va as CampaignRunPlan, vi as OptimizationMethodResult, vn as AnalyzeCrossSurfaceInteractionsInput, vr as provenanceSpansPath, w as acquireSingleRunLock, wa as readCachedCell, wi as optimizationTokenUsageFromSummary, wn as CrossSurfaceCandidate, wr as HeldOutGateOptions, x as surfaceHash, xa as planCampaignRun, xi as OptimizationPackageSource, xn as CrossSurfaceAttemptCompleteness, y as surfaceContentHash, ya as CampaignRunPlanCell, yi as OptimizationMethodRunOptions, yn as CrossSurfaceAdditionDecision, yr as verifyLoopProvenanceRecord, z as FinalizeProfileMatrixOptions, zn as CrossSurfaceInteractionReport, zt as FsLabeledScenarioStore } from "../index-giYkbitB.js";
|
|
8
8
|
import { A as SearchCellPlan, At as SearchHistoryCoverageRow, B as aide, Bt as FileSearchLedger, C as decideSearchClaim, Ct as searchEdgeId, D as AshaOptions, Dt as SearchHistoryAdmissionOptions, E as verifySearchClaim, Et as CreateSearchHistoryReceiptInput, F as AideOptions, Ft as assertSearchHistoryMatchesState, G as EstimateNodeCellsInput, Gt as openSearchLedger, H as crowdedFrontierParent, Ht as SearchLedger, I as SearchExpansion, It as createSearchHistoryReceipt, J as SearchPosterior, Jt as validateSearchLedgerEvent, K as NodePosterior, Kt as parseSearchLedgerLine, L as SearchPolicy, Lt as searchHistoryCoverageRow, M as SearchUnitExtension, Mt as SearchHistoryReceipt, N as asha, Nt as SearchHistoryRequiredError, O as SearchAllocationView, Ot as SearchHistoryAuditSummary, P as uniform, Pt as assertCompleteSearchHistory, Q as searchPosterior, Qt as inMemoryCampaignStorage, R as SearchPolicyNode, Rt as verifySearchHistoryArtifact, S as SearchClaimVerification, St as developmentClaim, T as searchClaimReserveUsd, Tt as surfaceNode, U as incumbent, Ut as SearchLedgerAppendResult, V as beam, Vt as OpenSearchLedgerOptions, W as incumbentWithOperatorBandit, Wt as SearchLedgerTrustedHeadMode, X as estimateNodeFromCells, Xt as createRunCostLedger, Y as estimateNode, Yt as CampaignStorage, Z as searchCellSetDigest, Zt as fsCampaignStorage, _ as searchPolicyView, _t as SearchOpening, a as SearchCellWork, b as SearchClaimDecision, bt as SearchRunIdentity, c as SearchProposalBlob, d as SearchProposedChild, f as SearchProposerPort, g as searchExpansionIndex, gt as SearchLedgerBinding, h as searchDivergence, ht as RegisterSearchNodeInput, i as SearchCellResult, j as SearchRungDecision, jt as SearchHistoryPolicy, k as SearchAllocator, kt as SearchHistoryCoverage, l as SearchProposalRequest, m as runSearch, mt as RecordSearchEdgeInput, n as SEARCH_KERNEL_SOURCE, o as SearchExecutor, p as SearchRunResult, pt as AllocateSearchCellInput, q as SEARCH_ESTIMATOR, qt as replaySearchLedgerText, r as SearchArtifactCodec, s as SearchLane, t as RunSearchOptions, u as SearchProposalResult, v as PlanSearchClaimInput, vt as SearchRecorder, w as planSearchClaim, wt as surfaceDiff, x as SearchClaimPlan, xt as SettleSearchCellInput, y as SEARCH_CLAIM_RULE, yt as SearchRecorderOptions, z as SearchPolicyView, zt as verifySearchHistoryReceipt } from "../search-kernel-C9CXvFR7.js";
|
|
9
9
|
import { B as ExternalOptimizerModelBudget, F as ExternalOptimizerCallbackLimits, G as ExternalOptimizerProcessLimits, H as ExternalOptimizerModelCallRequest, I as ExternalOptimizerChatRequest, L as ExternalOptimizerEndpointFormat, N as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, P as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, Q as resolveExternalOptimizerProcessLimits, R as ExternalOptimizerEvaluationObservation, U as ExternalOptimizerModelCallResult, V as ExternalOptimizerModelCall, W as ExternalOptimizerModelExecutionObservation, Y as ExternalTextCandidate, Z as resolveExternalOptimizerCallbackLimits, q as ExternalOptimizerRunnerCommand, z as ExternalOptimizerEvaluationRefusalReason } from "../agent-profile-DDfIjdyS.js";
|
|
10
|
-
export { type AideOptions, type AllocateSearchCellInput, type AnalyzeCrossSurfaceInteractionsInput, type AshaOptions, type AutoevalsScoreLike, type AutoevalsScorerLike, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type BuildTraceAnalystSurfaceDispatchOptions, type CacheIssueReason, type CacheRead, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCellRetryPolicy, type CampaignCostMeter, type CampaignResult, type CampaignRunPlan, type CampaignRunPlanCell, type CampaignScenarioIdentity, type CampaignStorage, type CampaignTokenUsage, type CampaignTraceWriter, type CellScheduleSlot, type CodeSurface, type CodeSurfaceVerification, type CompareOptimizationMethodsOptions, type ComparisonCost, type ComponentSurface, type CostLedgerHandle, type CreateProfileMatrixPlanOptions, type CreateSearchHistoryReceiptInput, type CrossSurfaceAdditionDecision, type CrossSurfaceAdditionRejectionReason, type CrossSurfaceAttemptCompleteness, type CrossSurfaceBestSingleSelection, type CrossSurfaceBootstrapPolicy, type CrossSurfaceCandidate, type CrossSurfaceCandidateComparison, type CrossSurfaceCandidateEvidence, type CrossSurfaceCandidateOutcome, type CrossSurfaceCandidateSummary, type CrossSurfaceComponent, type CrossSurfaceComponentEvidence, type CrossSurfaceCompositionStep, type CrossSurfaceDistribution, type CrossSurfaceEligibility, type CrossSurfaceEvidenceBreakdown, type CrossSurfaceIneligibilityReason, type CrossSurfaceInteractionAwareSelection, type CrossSurfaceInteractionEffect, type CrossSurfaceInteractionPath, type CrossSurfaceInteractionReport, type CrossSurfaceInteractionTask, type CrossSurfaceNaiveStackSelection, type CrossSurfacePairCompatibility, type CrossSurfacePairEvidence, type CrossSurfacePairIncompatibilityReason, type CrossSurfacePairwiseEntry, type CrossSurfaceRankedSingle, type CrossSurfaceRelativeCost, type CrossSurfaceSelectionPolicy, type CrossSurfaceSelections, type CrossSurfaceTaskRow, DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DimensionRegression, type DiscriminationScore, type DispatchContext, type DispatchFn, type EmitLoopProvenanceArgs, type EmitLoopProvenanceResult, type EstimateNodeCellsInput, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type EvidenceVector, type ExternalOptimizationExample, type ExternalOptimizerCallbackLimits, type ExternalOptimizerChatRequest, type ExternalOptimizerEndpointFormat, type ExternalOptimizerEvaluationObservation, type ExternalOptimizerEvaluationRefusalReason, type ExternalOptimizerExecutionSummary, type ExternalOptimizerModelBudget, type ExternalOptimizerModelCall, type ExternalOptimizerModelCallRequest, type ExternalOptimizerModelCallResult, type ExternalOptimizerModelExecutionObservation, type ExternalOptimizerObservationArtifact, type ExternalOptimizerObservationSummary, type ExternalOptimizerProcessLimits, type ExternalOptimizerRunnerCommand, type ExternalOptimizerSubmittedCandidate, type ExternalSearchOptimizer, type ExternalTextCandidate, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, FileSearchLedger, type FinalEvidencePolicy, type FinalEvidenceUse, type FinalizeProfileMatrixOptions, type FinalizedProfileMatrixResult, FsLabeledScenarioStore, type FsLabeledScenarioStoreOptions, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaCandidatePopulationArtifact, type GepaCandidatePopulationCandidate, type GepaCandidatePopulationSummary, type GepaCandidateSelectionScore, type GepaEngineOptions, type GepaEngineRun, type GepaEvaluationImport, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaPopulationImport, type GepaRunnerCommand, type GitWorktreeAdapterOptions, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, LabeledScenarioStoreError, type LabeledScenarioWrite, type LlmJudgeDimension, type LlmJudgeOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceArgsFromResult, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceEvidence, type LoopProvenanceOptimizationMethod, type LoopProvenanceRecord, type MutableSurface, type NeutralizationGateOptions, type NodeEstimate, type NodePosterior, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodComposition, type OptimizationMethodInput, type OptimizationMethodPairwise, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationMethodRunOptions, type OptimizationMethodScore, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type PairedHoldout, type ParetoParent, type ParetoSignificanceGateOptions, type PendingCostCallView, type PhoenixEvaluationResultLike, type PhoenixEvaluatorLike, type PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlanSearchClaimInput, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PowerPreflight, type PowerPreflightOptions, type PremeasuredOptimizationBaseline, type ProfileDispatchFn, type ProfileMatrixCoverage, ProfileMatrixError, type ProfileMatrixPlan, type ProfileMatrixRow, type ProfileMatrixSegmentResult, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, type ProposalTrackContext, type ProposeContext, type ProposedCandidate, type RecordSearchEdgeInput, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RegisterSearchNodeInput, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunProfileMatrixSegmentOptions, type RunSearchOptions, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_KERNEL_SOURCE, SEARCH_LEDGER_SCHEMA, type Scenario, type ScenarioAggregate, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type ScoredSurfaceOutcome, type SearchAllocationView, type SearchAllocator, type SearchArtifactCodec, type SearchArtifactKind, type SearchArtifactRef, type SearchAttemptAccounting, type SearchAudit, type SearchBudget, type SearchCancelReason, type SearchCandidateSurface, type SearchCell, type SearchCellAllocatedEvent, type SearchCellCancelledEvent, type SearchCellPlan, type SearchCellResult, type SearchCellSettledEvent, type SearchCellStage, type SearchCellWork, type SearchClaim, type SearchClaimDecision, type SearchClaimEstimator, type SearchClaimFinalist, type SearchClaimPlan, type SearchClaimPower, type SearchClaimTest, type SearchClaimVerification, type SearchCloseReason, type SearchClosedEvent, type SearchCompletion, type SearchCostAccounting, type SearchDecisionRecord, type SearchEdgeAttribution, type SearchEdgeOperator, type SearchEdgeRecordedEvent, type SearchEstimateMethod, type SearchExecutionIdentity, type SearchExecutor, type SearchExpansion, type SearchFailureReason, type SearchHistoryAdmissionOptions, type SearchHistoryAuditSummary, type SearchHistoryCoverage, type SearchHistoryCoverageRow, type SearchHistoryPolicy, type SearchHistoryReceipt, SearchHistoryRequiredError, type SearchLane, type SearchLedger, type SearchLedgerAppendResult, type SearchLedgerBinding, SearchLedgerConflictError, type SearchLedgerEntry, SearchLedgerError, type SearchLedgerEvent, type SearchLedgerHash, SearchLedgerIntegrityError, type SearchLedgerTrustedHeadMode, type SearchModelIdentity, type SearchNode, type SearchNodeDecidedEvent, type SearchNodeDecision, type SearchNodeRef, type SearchNodeRegisteredEvent, type SearchNodeStatus, type SearchOpenedEvent, type SearchOpening, type SearchOperation, type SearchOperationKind, type SearchOperationRecordedEvent, type SearchOperationStartedEvent, type SearchPolicy, type SearchPolicyNode, type SearchPolicyView, type SearchPosterior, type SearchProposalBlob, type SearchProposalRequest, type SearchProposalResult, type SearchProposedChild, type SearchProposer, type SearchProposerKind, type SearchProposerPort, type SearchProposerView, SearchRecorder, type SearchRecorderOptions, type SearchReservation, type SearchRunIdentity, type SearchRunResult, type SearchRungDecision, type SearchScoredCell, type SearchSourceRef, type SearchSpend, type SearchSplit, type SearchSplitTasks, type SearchSplits, SearchState, SearchStateView, type SearchSummaryOptions, type SearchSurfaceEffect, type SearchSurfaceEvidence, type SearchSurfaceKind, type SearchTask, type SearchTaskOutcome, type SearchTokenAccounting, type SearchTraceRef, type SearchUnitExtension, type SearchUnitScore, type SearchUnknown, type SequentialDecideFn, type SequentialDecideOptions, type SequentialDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SequentialStreamState, type SeriesDistribution, type SessionScript, type SettleSearchCellInput, type SingleRunLock, type SingleRunLockOptions, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SurfaceProposer, type TraceAnalystArtifact, type TraceAnalystScenario, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, WorktreeAdapterError, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertCompleteSearchHistory, assertSearchHistoryMatchesState, autoevalsScorerJudge, beam, buildCellSchedule, buildEvidenceVector, buildLoopProvenanceRecord, buildTraceAnalystSurfaceDispatch, campaignBreakdown, campaignCellSearchResult, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, combineComparisonCosts, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createReferenceEquivalenceJudge, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, detectScale, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, labelTrustRank, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, makeProposalFinding, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, optimizationTokenUsageFromSummary, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, phoenixEvaluatorJudge, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, provenanceRecordPath, provenanceSpansPath, quotaExhaustedUntil, readCachedCell, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, recordGepaSearch, recordSkillOptSearch, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveExternalOptimizerCallbackLimits, resolveExternalOptimizerProcessLimits, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchExpansionIndex, searchHistoryCoverageRow, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, selectDiscriminative, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, tangleTracesRoot, traceAnalystQualityJudge, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };
|
|
10
|
+
export { type AideOptions, type AllocateSearchCellInput, type AnalyzeCrossSurfaceInteractionsInput, type AshaOptions, type AutoevalsScoreLike, type AutoevalsScorerLike, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type BuildTraceAnalystSurfaceDispatchOptions, type CacheIssueReason, type CacheRead, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCellRetryPolicy, type CampaignCostMeter, type CampaignResult, type CampaignRunPlan, type CampaignRunPlanCell, type CampaignScenarioIdentity, type CampaignStorage, type CampaignTokenUsage, type CampaignTraceWriter, type CellScheduleSlot, type CodeSurface, type CodeSurfaceVerification, type CompareOptimizationMethodsOptions, type ComparisonCost, type ComponentSurface, type CostLedgerHandle, type CreateProfileMatrixPlanOptions, type CreateSearchHistoryReceiptInput, type CrossSurfaceAdditionDecision, type CrossSurfaceAdditionRejectionReason, type CrossSurfaceAttemptCompleteness, type CrossSurfaceBestSingleSelection, type CrossSurfaceBootstrapPolicy, type CrossSurfaceCandidate, type CrossSurfaceCandidateComparison, type CrossSurfaceCandidateEvidence, type CrossSurfaceCandidateOutcome, type CrossSurfaceCandidateSummary, type CrossSurfaceComponent, type CrossSurfaceComponentEvidence, type CrossSurfaceCompositionStep, type CrossSurfaceDistribution, type CrossSurfaceEligibility, type CrossSurfaceEvidenceBreakdown, type CrossSurfaceIneligibilityReason, type CrossSurfaceInteractionAwareSelection, type CrossSurfaceInteractionEffect, type CrossSurfaceInteractionPath, type CrossSurfaceInteractionReport, type CrossSurfaceInteractionTask, type CrossSurfaceNaiveStackSelection, type CrossSurfacePairCompatibility, type CrossSurfacePairEvidence, type CrossSurfacePairIncompatibilityReason, type CrossSurfacePairwiseEntry, type CrossSurfaceRankedSingle, type CrossSurfaceRelativeCost, type CrossSurfaceSelectionPolicy, type CrossSurfaceSelections, type CrossSurfaceTaskRow, DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DimensionRegression, type DiscriminationScore, type DispatchContext, type DispatchFn, type EmitLoopProvenanceArgs, type EmitLoopProvenanceResult, type EstimateNodeCellsInput, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type EvidenceVector, type ExternalOptimizationExample, type ExternalOptimizerCallbackLimits, type ExternalOptimizerChatRequest, type ExternalOptimizerEndpointFormat, type ExternalOptimizerEvaluationObservation, type ExternalOptimizerEvaluationRefusalReason, type ExternalOptimizerExecutionSummary, type ExternalOptimizerModelBudget, type ExternalOptimizerModelCall, type ExternalOptimizerModelCallRequest, type ExternalOptimizerModelCallResult, type ExternalOptimizerModelExecutionObservation, type ExternalOptimizerObservationArtifact, type ExternalOptimizerObservationSummary, type ExternalOptimizerProcessLimits, type ExternalOptimizerRunnerCommand, type ExternalOptimizerSubmittedCandidate, type ExternalSearchOptimizer, type ExternalTextCandidate, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, FileSearchLedger, type FinalEvidencePolicy, type FinalEvidenceUse, type FinalizeProfileMatrixOptions, type FinalizedProfileMatrixResult, FsLabeledScenarioStore, type FsLabeledScenarioStoreOptions, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaCandidatePopulationArtifact, type GepaCandidatePopulationCandidate, type GepaCandidatePopulationSummary, type GepaCandidateSelectionScore, type GepaEngineOptions, type GepaEngineRun, type GepaEvaluationImport, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaPopulationImport, type GepaRunnerCommand, type GitWorktreeAdapterOptions, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, LabeledScenarioStoreError, type LabeledScenarioWrite, type LlmJudgeDimension, type LlmJudgeOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceArgsFromResult, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceEvidence, type LoopProvenanceOptimizationMethod, type LoopProvenanceRecord, type MutableSurface, type NeutralizationGateOptions, type NodeEstimate, type NodePosterior, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodComposition, type OptimizationMethodInput, type OptimizationMethodPairwise, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationMethodRunOptions, type OptimizationMethodScore, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type PairedHoldout, type ParetoParent, type ParetoSignificanceGateOptions, type PendingCostCallView, type PhoenixEvaluationResultLike, type PhoenixEvaluatorLike, type PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlanSearchClaimInput, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PowerPreflight, type PowerPreflightOptions, type PremeasuredOptimizationBaseline, type ProfileDispatchFn, type ProfileMatrixCoverage, ProfileMatrixError, type ProfileMatrixPlan, type ProfileMatrixRow, type ProfileMatrixSegmentResult, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, type ProposalTrackContext, type ProposeContext, type ProposedCandidate, type RecordSearchEdgeInput, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RegisterSearchNodeInput, type RetainedDispatchOptions, type RetainedDispatchOutcome, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunProfileMatrixSegmentOptions, type RunSearchOptions, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_KERNEL_SOURCE, SEARCH_LEDGER_SCHEMA, type Scenario, type ScenarioAggregate, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type ScoredSurfaceOutcome, type SearchAllocationView, type SearchAllocator, type SearchArtifactCodec, type SearchArtifactKind, type SearchArtifactRef, type SearchAttemptAccounting, type SearchAudit, type SearchBudget, type SearchCancelReason, type SearchCandidateSurface, type SearchCell, type SearchCellAllocatedEvent, type SearchCellCancelledEvent, type SearchCellPlan, type SearchCellResult, type SearchCellSettledEvent, type SearchCellStage, type SearchCellWork, type SearchClaim, type SearchClaimDecision, type SearchClaimEstimator, type SearchClaimFinalist, type SearchClaimPlan, type SearchClaimPower, type SearchClaimTest, type SearchClaimVerification, type SearchCloseReason, type SearchClosedEvent, type SearchCompletion, type SearchCostAccounting, type SearchDecisionRecord, type SearchEdgeAttribution, type SearchEdgeOperator, type SearchEdgeRecordedEvent, type SearchEstimateMethod, type SearchExecutionIdentity, type SearchExecutor, type SearchExpansion, type SearchFailureReason, type SearchHistoryAdmissionOptions, type SearchHistoryAuditSummary, type SearchHistoryCoverage, type SearchHistoryCoverageRow, type SearchHistoryPolicy, type SearchHistoryReceipt, SearchHistoryRequiredError, type SearchLane, type SearchLedger, type SearchLedgerAppendResult, type SearchLedgerBinding, SearchLedgerConflictError, type SearchLedgerEntry, SearchLedgerError, type SearchLedgerEvent, type SearchLedgerHash, SearchLedgerIntegrityError, type SearchLedgerTrustedHeadMode, type SearchModelIdentity, type SearchNode, type SearchNodeDecidedEvent, type SearchNodeDecision, type SearchNodeRef, type SearchNodeRegisteredEvent, type SearchNodeStatus, type SearchOpenedEvent, type SearchOpening, type SearchOperation, type SearchOperationKind, type SearchOperationRecordedEvent, type SearchOperationStartedEvent, type SearchPolicy, type SearchPolicyNode, type SearchPolicyView, type SearchPosterior, type SearchProposalBlob, type SearchProposalRequest, type SearchProposalResult, type SearchProposedChild, type SearchProposer, type SearchProposerKind, type SearchProposerPort, type SearchProposerView, SearchRecorder, type SearchRecorderOptions, type SearchReservation, type SearchRunIdentity, type SearchRunResult, type SearchRungDecision, type SearchScoredCell, type SearchSourceRef, type SearchSpend, type SearchSplit, type SearchSplitTasks, type SearchSplits, SearchState, SearchStateView, type SearchSummaryOptions, type SearchSurfaceEffect, type SearchSurfaceEvidence, type SearchSurfaceKind, type SearchTask, type SearchTaskOutcome, type SearchTokenAccounting, type SearchTraceRef, type SearchUnitExtension, type SearchUnitScore, type SearchUnknown, type SequentialDecideFn, type SequentialDecideOptions, type SequentialDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SequentialStreamState, type SeriesDistribution, type SessionScript, type SettleSearchCellInput, type SingleRunLock, type SingleRunLockOptions, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SurfaceProposer, type TraceAnalystArtifact, type TraceAnalystScenario, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, WorktreeAdapterError, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertCompleteSearchHistory, assertSearchHistoryMatchesState, autoevalsScorerJudge, beam, buildCellSchedule, buildEvidenceVector, buildLoopProvenanceRecord, buildTraceAnalystSurfaceDispatch, campaignBreakdown, campaignCellSearchResult, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, combineComparisonCosts, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createReferenceEquivalenceJudge, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, detectScale, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, labelTrustRank, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, makeProposalFinding, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, optimizationTokenUsageFromSummary, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, phoenixEvaluatorJudge, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, provenanceRecordPath, provenanceSpansPath, quotaExhaustedUntil, readCachedCell, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, recordGepaSearch, recordSkillOptSearch, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveExternalOptimizerCallbackLimits, resolveExternalOptimizerProcessLimits, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchExpansionIndex, searchHistoryCoverageRow, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, selectDiscriminative, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, tangleTracesRoot, traceAnalystQualityJudge, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };
|
package/dist/campaign/index.js
CHANGED
|
@@ -9,7 +9,7 @@ import { t as acquireSingleRunLock } from "../single-run-lock-fVu3MZWH.js";
|
|
|
9
9
|
import { n as searchProposerView, t as renderSearchSummary } from "../search-summary-C1nAHgkB.js";
|
|
10
10
|
import { i as makeProposalFinding } from "../types-CW5b90JH.js";
|
|
11
11
|
import { a as importGepaPopulation, c as heldOutGate, d as decodeExternalTextCandidate, f as readExternalOptimizerObservationArtifact, h as createReferenceEquivalenceJudge, i as importExternalEvaluations, l as composeGate, n as gepaOptimizationMethod, o as recordGepaSearch, r as externalSurface, s as recordSkillOptSearch, t as skillOptOptimizationMethod, u as externalTextOptimizationMethod } from "../skillopt-optimization-method-Bls9t-G_.js";
|
|
12
|
-
import { A as
|
|
12
|
+
import { A as loadEvalFixtureScenarios, C as FsLabeledScenarioStore, D as neutralizationGate, E as rolloutArgumentDiff, M as analyzeCrossSurfaceInteractions, N as buildTraceAnalystSurfaceDispatch, O as discoverEvalFixtures, P as traceAnalystQualityJudge, S as neutralizeText, T as classifyUngroundedLiterals, _ as scoreUserStory, a as autoevalsScorerJudge, b as scopedOptimizationMethod, c as selectDiscriminative, d as finalizeProfileMatrix, f as runProfileMatrixSegment, g as renderScoreboardMarkdown, h as makePlaybackDispatch, i as verifyCodeSurface, j as planEvalFixtureRun, k as loadEvalFixture, l as createRetainedDispatch, m as runProfileMatrix, n as gitWorktreeAdapter, o as phoenixEvaluatorJudge, p as ProfileMatrixError, r as resolveWorktreePath, s as scoreDiscrimination, t as WorktreeAdapterError, u as createProfileMatrixPlan, v as scoreboardSummary, w as LabeledScenarioStoreError, x as sequentialOptimizationMethod, y as userStoryScoreboard } from "../campaign-XXmOU0R0.js";
|
|
13
13
|
import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-BI1RxzwD.js";
|
|
14
14
|
import { n as sequentialPairedGate, t as sequentialDecide } from "../sequential-DWn_np6P.js";
|
|
15
|
-
export { DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, FileSearchLedger, FsLabeledScenarioStore, LabeledScenarioStoreError, ProfileMatrixError, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_KERNEL_SOURCE, SEARCH_LEDGER_SCHEMA, SearchHistoryRequiredError, SearchLedgerConflictError, SearchLedgerError, SearchLedgerIntegrityError, SearchRecorder, SearchState, SearchStateView, WorktreeAdapterError, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertCompleteSearchHistory, assertSearchHistoryMatchesState, autoevalsScorerJudge, beam, buildCellSchedule, buildEvidenceVector, buildLoopProvenanceRecord, buildTraceAnalystSurfaceDispatch, campaignBreakdown, campaignCellSearchResult, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, combineComparisonCosts, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createReferenceEquivalenceJudge, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, detectScale, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, labelTrustRank, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, makeProposalFinding, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, optimizationTokenUsageFromSummary, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, phoenixEvaluatorJudge, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, provenanceRecordPath, provenanceSpansPath, quotaExhaustedUntil, readCachedCell, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, recordGepaSearch, recordSkillOptSearch, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveExternalOptimizerCallbackLimits, resolveExternalOptimizerProcessLimits, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchExpansionIndex, searchHistoryCoverageRow, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, selectDiscriminative, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, tangleTracesRoot, traceAnalystQualityJudge, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };
|
|
15
|
+
export { DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, FileSearchLedger, FsLabeledScenarioStore, LabeledScenarioStoreError, ProfileMatrixError, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_KERNEL_SOURCE, SEARCH_LEDGER_SCHEMA, SearchHistoryRequiredError, SearchLedgerConflictError, SearchLedgerError, SearchLedgerIntegrityError, SearchRecorder, SearchState, SearchStateView, WorktreeAdapterError, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertCompleteSearchHistory, assertSearchHistoryMatchesState, autoevalsScorerJudge, beam, buildCellSchedule, buildEvidenceVector, buildLoopProvenanceRecord, buildTraceAnalystSurfaceDispatch, campaignBreakdown, campaignCellSearchResult, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, combineComparisonCosts, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createReferenceEquivalenceJudge, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, detectScale, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, labelTrustRank, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, makeProposalFinding, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, optimizationTokenUsageFromSummary, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, phoenixEvaluatorJudge, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, provenanceRecordPath, provenanceSpansPath, quotaExhaustedUntil, readCachedCell, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, recordGepaSearch, recordSkillOptSearch, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveExternalOptimizerCallbackLimits, resolveExternalOptimizerProcessLimits, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchExpansionIndex, searchHistoryCoverageRow, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, selectDiscriminative, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, tangleTracesRoot, traceAnalystQualityJudge, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { s as ValidationError, t as AgentEvalError } from "./errors-Dngq5h35.js";
|
|
2
|
-
import { i as compareCodeUnits } from "./canonical-BsZVgBhW.js";
|
|
2
|
+
import { a as hashCanonical, i as compareCodeUnits, r as canonicalString } from "./canonical-BsZVgBhW.js";
|
|
3
3
|
import { c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, s as agentProfileHash, t as extractProducedState } from "./produced-state-CR2tAPLf.js";
|
|
4
4
|
import { s as buildAgentProfileCell } from "./agent-profile-cell-DcFW4vOc.js";
|
|
5
5
|
import { r as pairedBootstrap } from "./paired-tests-Ccr16gcW.js";
|
|
@@ -22,6 +22,7 @@ import "./skillopt-optimization-method-Bls9t-G_.js";
|
|
|
22
22
|
import "./sequential-DWn_np6P.js";
|
|
23
23
|
import { createHash } from "node:crypto";
|
|
24
24
|
import { closeSync, constants, existsSync, fstatSync, lstatSync, mkdirSync, mkdtempSync, openSync, readFileSync, readSync, readdirSync, readlinkSync, realpathSync, rmSync, statSync, writeFileSync } from "node:fs";
|
|
25
|
+
import { z } from "zod";
|
|
25
26
|
import { basename, dirname, isAbsolute, join, relative, resolve, sep } from "node:path";
|
|
26
27
|
import { devNull, tmpdir } from "node:os";
|
|
27
28
|
import { execFileSync } from "node:child_process";
|
|
@@ -2366,6 +2367,238 @@ function byteLength(value) {
|
|
|
2366
2367
|
return new TextEncoder().encode(value).byteLength;
|
|
2367
2368
|
}
|
|
2368
2369
|
//#endregion
|
|
2370
|
+
//#region src/campaign/retained-dispatch.ts
|
|
2371
|
+
const digest = z.string().regex(/^sha256:[a-f0-9]{64}$/);
|
|
2372
|
+
const headerSchema = z.object({
|
|
2373
|
+
kind: z.literal("retained-dispatch-v1"),
|
|
2374
|
+
scope: digest,
|
|
2375
|
+
limits: z.array(z.tuple([z.string().min(1), z.number().int().nonnegative()]))
|
|
2376
|
+
}).strict();
|
|
2377
|
+
const eventSchema = z.discriminatedUnion("kind", [z.object({
|
|
2378
|
+
kind: z.literal("started"),
|
|
2379
|
+
id: digest,
|
|
2380
|
+
lane: z.string().min(1)
|
|
2381
|
+
}).strict(), z.object({
|
|
2382
|
+
kind: z.literal("settled"),
|
|
2383
|
+
id: digest,
|
|
2384
|
+
receipt: digest
|
|
2385
|
+
}).strict()]);
|
|
2386
|
+
const resultSchema = z.object({
|
|
2387
|
+
kind: z.literal("retained-dispatch-result-v1"),
|
|
2388
|
+
id: digest,
|
|
2389
|
+
receipt: digest,
|
|
2390
|
+
value: z.unknown()
|
|
2391
|
+
}).strict();
|
|
2392
|
+
/**
|
|
2393
|
+
* Persist dispatch intent before an external call and its validated result afterward.
|
|
2394
|
+
* Replays settled calls; never repeats an uncertain call or refunds its allowance.
|
|
2395
|
+
* This wraps a campaign's dispatch; runEval still owns scheduling, judging and reports.
|
|
2396
|
+
* Uses CampaignStorage's existing compare-and-append, not another database or scheduler.
|
|
2397
|
+
*/
|
|
2398
|
+
function createRetainedDispatch(options) {
|
|
2399
|
+
if (!options.runDir.trim()) throw new ValidationError("retained dispatch requires a run directory");
|
|
2400
|
+
const storage = options.storage ?? fsCampaignStorage();
|
|
2401
|
+
const parse = options.parse;
|
|
2402
|
+
const header = headerSchema.parse({
|
|
2403
|
+
kind: "retained-dispatch-v1",
|
|
2404
|
+
scope: hashCanonical(options.scope),
|
|
2405
|
+
limits: Object.entries(options.limits).sort(([a], [b]) => compareCodeUnits(a, b))
|
|
2406
|
+
});
|
|
2407
|
+
if (!header.limits.length) throw new ValidationError("retained dispatch requires named allowances");
|
|
2408
|
+
const limits = new Map(header.limits);
|
|
2409
|
+
const counts = /* @__PURE__ */ new Map();
|
|
2410
|
+
const calls = /* @__PURE__ */ new Map();
|
|
2411
|
+
const active = /* @__PURE__ */ new Map();
|
|
2412
|
+
const indexPath = join(options.runDir, "dispatches.jsonl");
|
|
2413
|
+
const receiptsDir = join(options.runDir, "dispatch-results");
|
|
2414
|
+
const resultPath = (id) => join(receiptsDir, `${id.slice(7)}.json`);
|
|
2415
|
+
storage.ensureDir(options.runDir);
|
|
2416
|
+
storage.ensureDir(receiptsDir);
|
|
2417
|
+
let retainedText = "";
|
|
2418
|
+
let revision = 0;
|
|
2419
|
+
let opened = false;
|
|
2420
|
+
const fail = (message) => {
|
|
2421
|
+
throw new ValidationError(`retained dispatch: ${message}`);
|
|
2422
|
+
};
|
|
2423
|
+
const apply = (event) => {
|
|
2424
|
+
const existing = calls.get(event.id);
|
|
2425
|
+
if (event.kind === "started") {
|
|
2426
|
+
const limit = limits.get(event.lane);
|
|
2427
|
+
const count = counts.get(event.lane) ?? 0;
|
|
2428
|
+
if (existing || limit === void 0 || count >= limit) fail("invalid or over-budget dispatch history");
|
|
2429
|
+
calls.set(event.id, { lane: event.lane });
|
|
2430
|
+
counts.set(event.lane, count + 1);
|
|
2431
|
+
} else {
|
|
2432
|
+
if (!existing || existing.receipt) return fail("settlement has no unique pending dispatch");
|
|
2433
|
+
existing.receipt = event.receipt;
|
|
2434
|
+
}
|
|
2435
|
+
};
|
|
2436
|
+
const refresh = () => {
|
|
2437
|
+
const stored = storage.read(indexPath);
|
|
2438
|
+
if (stored === void 0 && storage.exists(indexPath)) fail("cannot read existing history");
|
|
2439
|
+
const text = stored ?? "";
|
|
2440
|
+
if (!text.startsWith(retainedText) || text && !text.endsWith("\n")) fail("history was truncated, replaced or torn");
|
|
2441
|
+
for (const line of text.slice(retainedText.length).split("\n").filter(Boolean)) {
|
|
2442
|
+
const value = JSON.parse(line);
|
|
2443
|
+
if (!opened) {
|
|
2444
|
+
if (hashCanonical(headerSchema.parse(value)) !== hashCanonical(header)) fail("resume changed its scope or allowances");
|
|
2445
|
+
opened = true;
|
|
2446
|
+
} else apply(eventSchema.parse(value));
|
|
2447
|
+
}
|
|
2448
|
+
retainedText = text;
|
|
2449
|
+
revision = Buffer.byteLength(text);
|
|
2450
|
+
};
|
|
2451
|
+
const append = (event) => {
|
|
2452
|
+
const line = `${canonicalString(event)}\n`;
|
|
2453
|
+
const next = storage.append(indexPath, line, revision);
|
|
2454
|
+
if (next === void 0) {
|
|
2455
|
+
refresh();
|
|
2456
|
+
return false;
|
|
2457
|
+
}
|
|
2458
|
+
if (next !== revision + Buffer.byteLength(line)) fail("storage returned an invalid append revision");
|
|
2459
|
+
retainedText += line;
|
|
2460
|
+
revision = next;
|
|
2461
|
+
if (event.kind === "retained-dispatch-v1") opened = true;
|
|
2462
|
+
else apply(event);
|
|
2463
|
+
return true;
|
|
2464
|
+
};
|
|
2465
|
+
refresh();
|
|
2466
|
+
if (!opened && !append(header) && !opened) fail("history is busy; retry with the same run directory");
|
|
2467
|
+
const readResult = (id) => {
|
|
2468
|
+
const content = storage.read(resultPath(id));
|
|
2469
|
+
if (content === void 0) return void 0;
|
|
2470
|
+
const result = resultSchema.parse(JSON.parse(content));
|
|
2471
|
+
if (result.id !== id) return fail("retained result dispatch id mismatch");
|
|
2472
|
+
if (hashCanonical(result.value) !== result.receipt) return fail("retained result digest mismatch");
|
|
2473
|
+
return {
|
|
2474
|
+
value: parse(result.value),
|
|
2475
|
+
receipt: result.receipt
|
|
2476
|
+
};
|
|
2477
|
+
};
|
|
2478
|
+
const settle = (id, receipt) => {
|
|
2479
|
+
while (true) {
|
|
2480
|
+
const existing = calls.get(id);
|
|
2481
|
+
if (existing?.receipt) {
|
|
2482
|
+
if (existing.receipt !== receipt) fail("settled result digest mismatch");
|
|
2483
|
+
return;
|
|
2484
|
+
}
|
|
2485
|
+
if (!existing) fail("settlement has no pending dispatch");
|
|
2486
|
+
const previousRevision = revision;
|
|
2487
|
+
if (append({
|
|
2488
|
+
kind: "settled",
|
|
2489
|
+
id,
|
|
2490
|
+
receipt
|
|
2491
|
+
})) return;
|
|
2492
|
+
if (revision === previousRevision) fail("settlement is busy; retained result awaits reconciliation");
|
|
2493
|
+
}
|
|
2494
|
+
};
|
|
2495
|
+
const load = (id, receipt) => {
|
|
2496
|
+
const result = readResult(id);
|
|
2497
|
+
if (!result) return fail("settled result is missing");
|
|
2498
|
+
if (result.receipt !== receipt) return fail("settled result digest mismatch");
|
|
2499
|
+
return {
|
|
2500
|
+
succeeded: true,
|
|
2501
|
+
value: result.value,
|
|
2502
|
+
dispatchId: id,
|
|
2503
|
+
receiptDigest: receipt,
|
|
2504
|
+
replayed: true
|
|
2505
|
+
};
|
|
2506
|
+
};
|
|
2507
|
+
return {
|
|
2508
|
+
/** The input binds all behavior-affecting request material, not just a caller-selected nonce. */
|
|
2509
|
+
run(request) {
|
|
2510
|
+
const id = hashCanonical([
|
|
2511
|
+
header.scope,
|
|
2512
|
+
request.lane,
|
|
2513
|
+
request.input
|
|
2514
|
+
]);
|
|
2515
|
+
const pending = active.get(id);
|
|
2516
|
+
if (pending) return pending;
|
|
2517
|
+
const work = async () => {
|
|
2518
|
+
if (!limits.has(request.lane)) return fail("unknown allowance lane");
|
|
2519
|
+
if (calls.has(id) && !calls.get(id).receipt) refresh();
|
|
2520
|
+
while (true) {
|
|
2521
|
+
const existing = calls.get(id);
|
|
2522
|
+
if (existing) {
|
|
2523
|
+
if (existing.receipt) return load(id, existing.receipt);
|
|
2524
|
+
const result = readResult(id);
|
|
2525
|
+
if (!result) return {
|
|
2526
|
+
succeeded: false,
|
|
2527
|
+
reason: "outcome_unknown",
|
|
2528
|
+
dispatchId: id
|
|
2529
|
+
};
|
|
2530
|
+
settle(id, result.receipt);
|
|
2531
|
+
return {
|
|
2532
|
+
succeeded: true,
|
|
2533
|
+
value: result.value,
|
|
2534
|
+
dispatchId: id,
|
|
2535
|
+
receiptDigest: result.receipt,
|
|
2536
|
+
replayed: true
|
|
2537
|
+
};
|
|
2538
|
+
}
|
|
2539
|
+
request.signal?.throwIfAborted();
|
|
2540
|
+
if ((counts.get(request.lane) ?? 0) >= limits.get(request.lane)) return {
|
|
2541
|
+
succeeded: false,
|
|
2542
|
+
reason: "allowance_exhausted",
|
|
2543
|
+
dispatchId: id
|
|
2544
|
+
};
|
|
2545
|
+
const previousRevision = revision;
|
|
2546
|
+
if (append({
|
|
2547
|
+
kind: "started",
|
|
2548
|
+
id,
|
|
2549
|
+
lane: request.lane
|
|
2550
|
+
})) break;
|
|
2551
|
+
if (revision === previousRevision) return fail("history is busy; retry with the same input");
|
|
2552
|
+
}
|
|
2553
|
+
let value;
|
|
2554
|
+
try {
|
|
2555
|
+
request.signal?.throwIfAborted();
|
|
2556
|
+
value = parse(await request.dispatch());
|
|
2557
|
+
} catch (error) {
|
|
2558
|
+
return {
|
|
2559
|
+
succeeded: false,
|
|
2560
|
+
reason: "outcome_unknown",
|
|
2561
|
+
dispatchId: id,
|
|
2562
|
+
diagnostic: error instanceof Error ? error.message : String(error)
|
|
2563
|
+
};
|
|
2564
|
+
}
|
|
2565
|
+
const encoded = canonicalString(value);
|
|
2566
|
+
const decoded = JSON.parse(encoded);
|
|
2567
|
+
const retainedValue = parse(decoded);
|
|
2568
|
+
if (canonicalString(retainedValue) !== encoded) return fail("result decoder is not stable across serialization");
|
|
2569
|
+
const receipt = hashCanonical(decoded);
|
|
2570
|
+
const retainedResult = canonicalString({
|
|
2571
|
+
kind: "retained-dispatch-result-v1",
|
|
2572
|
+
id,
|
|
2573
|
+
receipt,
|
|
2574
|
+
value: decoded
|
|
2575
|
+
});
|
|
2576
|
+
const file = resultPath(id);
|
|
2577
|
+
if (storage.append(file, retainedResult, 0) === void 0 && storage.read(file) !== retainedResult) fail("cannot retain dispatch result");
|
|
2578
|
+
settle(id, receipt);
|
|
2579
|
+
return {
|
|
2580
|
+
succeeded: true,
|
|
2581
|
+
value: retainedValue,
|
|
2582
|
+
dispatchId: id,
|
|
2583
|
+
receiptDigest: receipt,
|
|
2584
|
+
replayed: false
|
|
2585
|
+
};
|
|
2586
|
+
};
|
|
2587
|
+
const promise = work();
|
|
2588
|
+
active.set(id, promise);
|
|
2589
|
+
promise.finally(() => {
|
|
2590
|
+
active.delete(id);
|
|
2591
|
+
}).catch(() => {});
|
|
2592
|
+
return promise;
|
|
2593
|
+
},
|
|
2594
|
+
/** Counts include uncertain dispatches, which keep their reservation. No pricing is inferred. */
|
|
2595
|
+
committed() {
|
|
2596
|
+
refresh();
|
|
2597
|
+
return new Map(header.limits.map(([lane]) => [lane, counts.get(lane) ?? 0]));
|
|
2598
|
+
}
|
|
2599
|
+
};
|
|
2600
|
+
}
|
|
2601
|
+
//#endregion
|
|
2369
2602
|
//#region src/campaign/scenario-selection.ts
|
|
2370
2603
|
const DEFAULT_SATURATION_CEILING = .999;
|
|
2371
2604
|
/** Variance below this is treated as "every candidate scored the same". */
|
|
@@ -3091,6 +3324,6 @@ function resolveWorktreePath(surface, worktreeDir) {
|
|
|
3091
3324
|
return verifyCodeSurface(surface, worktreeDir).path;
|
|
3092
3325
|
}
|
|
3093
3326
|
//#endregion
|
|
3094
|
-
export {
|
|
3327
|
+
export { loadEvalFixtureScenarios as A, FsLabeledScenarioStore as C, neutralizationGate as D, rolloutArgumentDiff as E, analyzeCrossSurfaceInteractions as M, buildTraceAnalystSurfaceDispatch as N, discoverEvalFixtures as O, traceAnalystQualityJudge as P, neutralizeText as S, classifyUngroundedLiterals as T, scoreUserStory as _, autoevalsScorerJudge as a, scopedOptimizationMethod as b, selectDiscriminative as c, finalizeProfileMatrix as d, runProfileMatrixSegment as f, renderScoreboardMarkdown as g, makePlaybackDispatch as h, verifyCodeSurface as i, planEvalFixtureRun as j, loadEvalFixture as k, createRetainedDispatch as l, runProfileMatrix as m, gitWorktreeAdapter as n, phoenixEvaluatorJudge as o, ProfileMatrixError as p, resolveWorktreePath as r, scoreDiscrimination as s, WorktreeAdapterError as t, createProfileMatrixPlan as u, scoreboardSummary as v, LabeledScenarioStoreError as w, sequentialOptimizationMethod as x, userStoryScoreboard as y };
|
|
3095
3328
|
|
|
3096
|
-
//# sourceMappingURL=campaign-
|
|
3329
|
+
//# sourceMappingURL=campaign-XXmOU0R0.js.map
|