@tangle-network/agent-eval 0.222.1 → 1.0.0

package/dist/benchmarks/index.d.ts

@@ -1,8 +1,8 @@
 import { c as RunSplitTag } from "../run-record-D2BFMPGl.js";
-import "../index-Bdcpx2s8.js";
-import { a as CampaignResult } from "../types-e3HGaI7P.js";
-import { Wt as CampaignStorage } from "../search-kernel-B77eE7Ac.js";
-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-DUR0dqET.js";
+import "../index-B7v2QHlD.js";
+import { a as CampaignResult } from "../types-C54Uk_a1.js";
+import { Wt as CampaignStorage } from "../search-kernel-CBdwFZ05.js";
+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-0oMg7zee.js";
 interface BenchmarkMetricCalibrationOptions<TPayload = unknown, TArtifact = string> {
   adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>;
   item: BenchmarkDatasetItem<TPayload>;

package/dist/benchmarks/index.js

@@ -1,9 +1,9 @@
 import { t as __exportAll } from "../rolldown-runtime-BJohB4LM.js";
 import { o as summarizeNumberSeries } from "../descriptive-cPwITVfy.js";
-import { t as mean } from "../mean-CdDYG-us.js";
-import { Z as runCampaign } from "../llm-judge-cgbhTKQp.js";
+import { t as mean } from "../mean-BPuUNznQ.js";
+import { Z as runCampaign } from "../llm-judge-D6JtIqiU.js";
 import { M as fsCampaignStorage } from "../external-optimizer-subprocess-U616hV_l.js";
-import "../campaign-ddr1N3yD.js";
+import "../campaign-DdOgQfp8.js";
 import { join, resolve } from "node:path";
 import { readFile, readdir, stat } from "node:fs/promises";
 import { parse } from "smol-toml";

package/dist/campaign/index.d.ts

@@ -1,10 +1,10 @@
 import { o as CostLedgerHandle, r as CostBreakdown, v as PendingCostCallView } from "../cost-ledger-Czro9wmo.js";
-import { S as makeProposalFinding, v as ProposalFinding, y as ProposalFindingOrigin } from "../types-BEvjDziM.js";
+import { S as makeProposalFinding, v as ProposalFinding, y as ProposalFindingOrigin } from "../types-DzCO-S1m.js";
 import { L as SeriesDistribution } from "../schema-DkFJr3Y8.js";
-import { $ as paretoPolicy, G as AxisVerdict, H as PowerPreflightOptions, J as ObjectiveSource, K as BuildEvidenceVectorOptions, Q as buildEvidenceVector, U as powerPreflight, V as PowerPreflight, W as AxisEvidence, X as PromotionObjective, Y as ParetoSignificanceGateOptions, Z as PromotionPolicy, a as dimensionRegressions, c as SequentialDecideFn, d as SequentialObservation, et as paretoSignificanceGate, f as SequentialPairedGate, g as sequentialPairedGate, h as sequentialDecide, i as PairedHoldout, l as SequentialDecideOptions, m as SequentialStreamState, n as HeldoutSignificance, o as heldoutSignificance, p as SequentialPairedGateOptions, q as EvidenceVector, r as HeldoutSignificanceOptions, s as pairHoldout, t as DimensionRegression, u as SequentialGateDecision } from "../statistical-heldout-BYxHe3KO.js";
+import { $ as paretoPolicy, G as AxisVerdict, H as PowerPreflightOptions, J as ObjectiveSource, K as BuildEvidenceVectorOptions, Q as buildEvidenceVector, U as powerPreflight, V as PowerPreflight, W as AxisEvidence, X as PromotionObjective, Y as ParetoSignificanceGateOptions, Z as PromotionPolicy, a as dimensionRegressions, c as SequentialDecideFn, d as SequentialObservation, et as paretoSignificanceGate, f as SequentialPairedGate, g as sequentialPairedGate, h as sequentialDecide, i as PairedHoldout, l as SequentialDecideOptions, m as SequentialStreamState, n as HeldoutSignificance, o as heldoutSignificance, p as SequentialPairedGateOptions, q as EvidenceVector, r as HeldoutSignificanceOptions, s as pairHoldout, t as DimensionRegression, u as SequentialGateDecision } from "../statistical-heldout-Cei036nE.js";
 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-qiYRUo_N.js";
-import { $r as ExternalOptimizerObservationArtifact, $t as CrossSurfaceCandidate, A as RetainedDispatchOptions, Ai as CampaignRunPlanCell, An as BuildLoopProvenanceArgs, Ar as externalTextOptimizationMethod, At as importExternalEvaluations, B as RunProfileMatrixSegmentOptions, Bi as CampaignCoverage, Bn as emitLoopProvenance, Br as OptimizationMethodProvenance, Bt as LoadEvalFixtureScenariosOptions, C as searchModelIdentity, Ci as RunEvalOptions, Cn as CrossSurfaceSelectionPolicy, Cr as GepaEngineRun, Ct as RolloutCall, D as ScenarioSignal, Di as RunCampaignOptions, Dn as BuildTraceAnalystSurfaceDispatchOptions, Dr as gepaOptimizationMethod, Dt as GepaEvaluationImport, E as DiscriminationScore, Ei as CampaignCellRetryPolicy, En as measureCampaignCost, Er as GepaRunnerCommand, Et as ExternalSearchOptimizer, F as FinalizedProfileMatrixResult, Fi as CellScheduleSlot, Fn as LoopProvenanceCandidate, Fr as ExternalTextEvaluationResponse, Ft as EvalFixtureFile, G as ProfileSummary, Gi as campaignSplitDigestFromIdentities, Gn as HeldOutGateOptions, Gr as OptimizationTokenUsage, Gt as planEvalFixtureRun, H as finalizeProfileMatrix, Hi as campaignCoverage, Hr as OptimizationMethodRunOptions, Ht as discoverEvalFixtures, I as ProfileMatrixCoverage, Ii as buildCellSchedule, In as LoopProvenanceEvidence, Ir as decodeExternalTextCandidate, It as EvalFixtureLoadOptions, J as ScenarioRollup, Jn as DefaultProductionGateOptions, Jr as GepaCandidatePopulationCandidate, Jt as CrossSurfaceAdditionDecision, K as RunProfileMatrixOptions, Ki as SearchLedgerError, Kn as heldOutGate, Kr as costFromLedgerSummary, Kt as analyzeCrossSurfaceInteractions, L as ProfileMatrixPlan, Li as cellCachePath, Ln as LoopProvenanceOptimizationMethod, Lr as OptimizationMethod, Lt as EvalFixtureRunPlan, M as createRetainedDispatch, Mi as planCampaignRun, Mn as EmitLoopProvenanceResult, Mr as ExternalTextOptimizerContext, Mt as NeutralizationGateOptions, N as CreateProfileMatrixPlanOptions, Ni as CacheIssueReason, Nn as LoopProvenanceArgsFromResult, Nr as ExternalTextOptimizerResult, Nt as neutralizationGate, O as scoreDiscrimination, Oi as runCampaign, On as TraceAnalystArtifact, Or as OpenAICompatibleOptimizerModel, Ot as GepaPopulationImport, P as FinalizeProfileMatrixOptions, Pi as CacheRead, Pn as LoopProvenanceBackend, Pr as ExternalOptimizationExample, Pt as EvalFixture, Qr as ExternalOptimizerExecutionSummary, Qt as CrossSurfaceBootstrapPolicy, R as ProfileMatrixRow, Ri as FinalEvidencePolicy, Rn as LoopProvenanceRecord, Rr as OptimizationMethodComposition, Rt as EvalFixtureScenario, S as campaignCellSearchResult, Sn as CrossSurfaceRelativeCost, Sr as GepaEngineOptions, St as RolloutArgumentDiffOptions, T as searchReceiptAccounting, Ti as CampaignCellFailureReceipt, Tn as CrossSurfaceTaskRow, Tr as GepaOptimizationRecipe, Tt as UngroundedLiteralReport, U as runProfileMatrixSegment, Ui as campaignScenarioIdentity, Un as campaignBreakdown, Ur as OptimizationPackageSource, Ut as loadEvalFixture, V as createProfileMatrixPlan, Vi as assertCampaignDesign, Vn as CampaignBreakdown, Vr as OptimizationMethodResult, Vt as PlanEvalFixtureRunOptions, W as ProfileDispatchFn, Wi as campaignSplitDigest, Wn as campaignMeanComposite, Wr as ComparisonCost, Wt as loadEvalFixtureScenarios, Xn as defaultProductionGate, Xr as GepaCandidateSelectionScore, Xt as CrossSurfaceAttemptCompleteness, Y as runProfileMatrix, Yn as DefaultProductionRewardHackingOptions, Yr as GepaCandidatePopulationSummary, Yt as CrossSurfaceAdditionRejectionReason, Zr as readGepaCandidatePopulationArtifact, Zt as CrossSurfaceBestSingleSelection, _ as surfaceDispatchRef, _i as RunOptimizationOptions, _n as CrossSurfacePairCompatibility, _r as OptimizationMethodComparison, _t as scopedOptimizationMethod, a as gitWorktreeAdapter, an as CrossSurfaceComponentEvidence, at as ScoreboardSummary, b as SingleRunLockOptions, bi as OpenAutoPrOptions, bn as CrossSurfacePairwiseEntry, br as compareOptimizationMethods, bt as FsLabeledScenarioStoreOptions, c as AutoevalsScoreLike, ci as LlmJudgeOptions, cn as CrossSurfaceEligibility, ct as renderScoreboardMarkdown, d as PhoenixEvaluatorLike, di as isTransientTransportFailure, dn as CrossSurfaceInteractionAwareSelection, dr as composeGate, dt as userStoryScoreboard, ei as ExternalOptimizerObservationSummary, en as CrossSurfaceCandidateComparison, et as PlaybackContext, f as assertCodeSurfaceIdentity, fi as transientDispatchFailure, fn as CrossSurfaceInteractionEffect, fr as SkillOptOptimizationMethodConfig, g as surfaceContentHash, gi as PremeasuredOptimizationBaseline, gn as CrossSurfaceNaiveStackSelection, gr as CompareOptimizationMethodsOptions, h as renderSurfaceDiff, hi as runImprovementLoop, hn as CrossSurfaceInteractionTask, hr as skillOptOptimizationMethod, i as WorktreeAdapter, ii as ReferenceEquivalenceJudgeOptions, in as CrossSurfaceComponent, it as ScoreboardRow, j as RetainedDispatchOutcome, ji as PlanCampaignRunOptions, jn as EmitLoopProvenanceArgs, jr as ExternalTextOptimizationMethodConfig, jt as importGepaPopulation, k as resolveRunDir, ki as CampaignRunPlan, kn as TraceAnalystScenario, kr as OptimizerModelBudget, kt as externalSurface, l as AutoevalsScorerLike, li as llmJudge, ln as CrossSurfaceEvidenceBreakdown, lt as scoreUserStory, m as componentSurfaceIdentityMaterial, mi as RunImprovementLoopResult, mn as CrossSurfaceInteractionReport, mr as SkillOptTrainerConfig, n as GitWorktreeAdapterOptions, ni as readExternalOptimizerObservationArtifact, nn as CrossSurfaceCandidateOutcome, nt as PlaybackStep, o as resolveWorktreePath, oi as ReferenceEquivalenceScenario, on as CrossSurfaceCompositionStep, ot as UserStory, p as codeSurfaceIdentityMaterial, pi as RunImprovementLoopOptions, pn as CrossSurfaceInteractionPath, pr as SkillOptRunnerCommand, q as RunProfileMatrixResult, qi as SearchLedgerIntegrityError, qn as DefaultProductionGateCheck, qr as GepaCandidatePopulationArtifact, qt as AnalyzeCrossSurfaceInteractionsInput, r as Worktree, rn as CrossSurfaceCandidateSummary, rt as ScoreboardRenderOptions, s as verifyCodeSurface, si as LlmJudgeDimension, sn as CrossSurfaceDistribution, st as UserStoryVerdict, t as CodeSurfaceVerification, ti as ExternalOptimizerSubmittedCandidate, tn as CrossSurfaceCandidateEvidence, tt as PlaybackDriver, u as PhoenixEvaluationResultLike, ui as TransientFailureOptions, un as CrossSurfaceIneligibilityReason, ut as scoreboardSummary, v as surfaceHash, vi as RunOptimizationResult, vn as CrossSurfacePairEvidence, vr as OptimizationMethodPairwise, vt as sequentialOptimizationMethod, w as searchProposalExecution, wi as runEval, wn as CrossSurfaceSelections, wr as GepaOptimizationMethodConfig, wt as ScoredRollout, x as acquireSingleRunLock, xi as OpenAutoPrResult, xn as CrossSurfaceRankedSingle, xr as GepaAdaptiveEngineRun, xt as RolloutArgumentDiff, y as SingleRunLock, yi as runOptimization, yn as CrossSurfacePairIncompatibilityReason, yr as OptimizationMethodScore, yt as neutralizeText, z as ProfileMatrixSegmentResult, zi as FinalEvidenceUse, zn as canonicalDigest, zr as OptimizationMethodInput, zt as EvalFixtureValidationMode } from "../index-Bdcpx2s8.js";
-import { A as LabeledScenarioWrite, B as ScoredSurfaceOutcome, C as JudgeDimension, D as LabeledScenarioSampleArgs, E as LabeledScenarioRecord, F as ProposeContext, G as SearchProposerView, H as SurfaceProposer, I as ProposedCandidate, J as searchProposerView, K as SearchSummaryOptions, L as RedactionStatus, M as OptimizerConfig, N as ParetoParent, O as LabeledScenarioSource, P as ProposalTrackContext, R as Scenario, S as JudgeConfig, T as LabelTrust, U as TraceSpan, V as SessionScript, W as isProposedCandidate, X as SearchStateView, Y as SearchState, Z as searchUnitScores, _ as GateDecision, a as CampaignResult, at as SearchCompletion, b as GenerationRecord, c as CampaignTraceWriter, ct as SearchOperation, d as DispatchContext, dt as SearchUnitScore, f as DispatchFn, ft as searchCellId, g as GateContribution, h as GateContext, i as CampaignCostMeter, it as SearchCell, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, lt as SearchScoredCell, m as GateCheckStatus, mt as searchTaskSetDigest, n as CampaignArtifactWriter, o as CampaignScenarioIdentity, ot as SearchDecisionRecord, p as Gate, pt as searchNodeId, q as renderSearchSummary, r as CampaignCellResult, s as CampaignTokenUsage, st as SearchNode, t as CampaignAggregates, u as ComponentSurface, ut as SearchSpend, v as GateResult, w as JudgeScore, x as JudgeAggregate, y as GenerationCandidate, z as ScenarioAggregate } from "../types-e3HGaI7P.js";
-import { A as SearchRungDecision, At as SearchHistoryReceipt, B as beam, Bt as openSearchLedger, C as searchClaimDecision, Ct as surfaceNode, D as SearchAllocationView, Dt as SearchHistoryCoverage, E as AshaOptions, Et as SearchHistoryAuditSummary, F as SearchExpansion, Ft as FileSearchLedger, G as NodePosterior, Gt as createRunCostLedger, H as incumbent, Ht as replaySearchLedgerText, I as SearchPolicy, It as OpenSearchLedgerOptions, J as estimateNode, K as SEARCH_ESTIMATOR, Kt as fsCampaignStorage, L as SearchPolicyNode, Lt as SearchLedger, M as asha, Mt as createSearchHistoryReceipt, N as uniform, Nt as verifySearchHistoryArtifact, O as SearchAllocator, Ot as SearchHistoryCoverageRow, P as AideOptions, Pt as verifySearchHistoryReceipt, Q as searchPosterior, R as SearchPolicyView, Rt as SearchLedgerAppendResult, S as planSearchClaim, St as surfaceDiff, T as verifySearchClaim, Tt as SearchHistoryAdmissionOptions, U as incumbentWithOperatorBandit, Ut as validateSearchLedgerEvent, V as crowdedFrontierParent, Vt as parseSearchLedgerLine, W as EstimateNodeCellsInput, Wt as CampaignStorage, X as searchCellSetDigest, Y as estimateNodeFromCells, Z as searchEstimateMethod, _ as SEARCH_CLAIM_RULE, _t as SearchRecorderOptions, a as SearchArtifactCodec, b as SearchClaimVerification, bt as developmentClaim, c as SearchExecutor, d as SearchProposalRequest, dt as AllocateSearchCellInput, f as SearchProposalResult, ft as RecordSearchEdgeInput, g as PlanSearchClaimInput, gt as SearchRecorder, h as SearchRunResult, ht as SearchOpening, i as RunSearchOptions, j as SearchUnitExtension, jt as assertSearchHistoryMatchesState, k as SearchCellPlan, kt as SearchHistoryPolicy, l as SearchLane, m as SearchProposerPort, mt as SearchLedgerBinding, n as searchDivergence, o as SearchCellResult, p as SearchProposedChild, pt as RegisterSearchNodeInput, q as SearchPosterior, qt as inMemoryCampaignStorage, r as searchPolicyView, s as SearchCellWork, t as runSearch, u as SearchProposalBlob, v as SearchClaimDecision, vt as SearchRunIdentity, w as searchClaimReserveUsd, wt as CreateSearchHistoryReceiptInput, x as decideSearchClaim, xt as searchEdgeId, y as SearchClaimPlan, yt as SettleSearchCellInput, z as aide, zt as SearchLedgerTrustedHeadMode } from "../search-kernel-B77eE7Ac.js";
-import { B as ExternalOptimizerModelCallRequest, F as ExternalOptimizerEndpointFormat, G as ExternalOptimizerRunnerCommand, H as ExternalOptimizerModelExecutionObservation, I as ExternalOptimizerEvaluationObservation, L as ExternalOptimizerEvaluationRefusalReason, M as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, N as ExternalOptimizerCallbackLimits, P as ExternalOptimizerChatRequest, R as ExternalOptimizerModelBudget, U as ExternalOptimizerProcessLimits, V as ExternalOptimizerModelCallResult, j as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, q as ExternalTextCandidate, z as ExternalOptimizerModelCall } from "../agent-profile-CSEQRC9l.js";
+import { $r as ExternalOptimizerObservationArtifact, $t as CrossSurfaceCandidate, A as RetainedDispatchOptions, Ai as CampaignRunPlanCell, An as BuildLoopProvenanceArgs, Ar as externalTextOptimizationMethod, At as importExternalEvaluations, B as RunProfileMatrixSegmentOptions, Bi as CampaignCoverage, Bn as emitLoopProvenance, Br as OptimizationMethodProvenance, Bt as LoadEvalFixtureScenariosOptions, C as searchModelIdentity, Ci as RunEvalOptions, Cn as CrossSurfaceSelectionPolicy, Cr as GepaEngineRun, Ct as RolloutCall, D as ScenarioSignal, Di as RunCampaignOptions, Dn as BuildTraceAnalystSurfaceDispatchOptions, Dr as gepaOptimizationMethod, Dt as GepaEvaluationImport, E as DiscriminationScore, Ei as CampaignCellRetryPolicy, En as measureCampaignCost, Er as GepaRunnerCommand, Et as ExternalSearchOptimizer, F as FinalizedProfileMatrixResult, Fi as CellScheduleSlot, Fn as LoopProvenanceCandidate, Fr as ExternalTextEvaluationResponse, Ft as EvalFixtureFile, G as ProfileSummary, Gi as campaignSplitDigestFromIdentities, Gn as HeldOutGateOptions, Gr as OptimizationTokenUsage, Gt as planEvalFixtureRun, H as finalizeProfileMatrix, Hi as campaignCoverage, Hr as OptimizationMethodRunOptions, Ht as discoverEvalFixtures, I as ProfileMatrixCoverage, Ii as buildCellSchedule, In as LoopProvenanceEvidence, Ir as decodeExternalTextCandidate, It as EvalFixtureLoadOptions, J as ScenarioRollup, Jn as DefaultProductionGateOptions, Jr as GepaCandidatePopulationCandidate, Jt as CrossSurfaceAdditionDecision, K as RunProfileMatrixOptions, Ki as SearchLedgerError, Kn as heldOutGate, Kr as costFromLedgerSummary, Kt as analyzeCrossSurfaceInteractions, L as ProfileMatrixPlan, Li as cellCachePath, Ln as LoopProvenanceOptimizationMethod, Lr as OptimizationMethod, Lt as EvalFixtureRunPlan, M as createRetainedDispatch, Mi as planCampaignRun, Mn as EmitLoopProvenanceResult, Mr as ExternalTextOptimizerContext, Mt as NeutralizationGateOptions, N as CreateProfileMatrixPlanOptions, Ni as CacheIssueReason, Nn as LoopProvenanceArgsFromResult, Nr as ExternalTextOptimizerResult, Nt as neutralizationGate, O as scoreDiscrimination, Oi as runCampaign, On as TraceAnalystArtifact, Or as OpenAICompatibleOptimizerModel, Ot as GepaPopulationImport, P as FinalizeProfileMatrixOptions, Pi as CacheRead, Pn as LoopProvenanceBackend, Pr as ExternalOptimizationExample, Pt as EvalFixture, Qr as ExternalOptimizerExecutionSummary, Qt as CrossSurfaceBootstrapPolicy, R as ProfileMatrixRow, Ri as FinalEvidencePolicy, Rn as LoopProvenanceRecord, Rr as OptimizationMethodComposition, Rt as EvalFixtureScenario, S as campaignCellSearchResult, Sn as CrossSurfaceRelativeCost, Sr as GepaEngineOptions, St as RolloutArgumentDiffOptions, T as searchReceiptAccounting, Ti as CampaignCellFailureReceipt, Tn as CrossSurfaceTaskRow, Tr as GepaOptimizationRecipe, Tt as UngroundedLiteralReport, U as runProfileMatrixSegment, Ui as campaignScenarioIdentity, Un as campaignBreakdown, Ur as OptimizationPackageSource, Ut as loadEvalFixture, V as createProfileMatrixPlan, Vi as assertCampaignDesign, Vn as CampaignBreakdown, Vr as OptimizationMethodResult, Vt as PlanEvalFixtureRunOptions, W as ProfileDispatchFn, Wi as campaignSplitDigest, Wn as campaignMeanComposite, Wr as ComparisonCost, Wt as loadEvalFixtureScenarios, Xn as defaultProductionGate, Xr as GepaCandidateSelectionScore, Xt as CrossSurfaceAttemptCompleteness, Y as runProfileMatrix, Yn as DefaultProductionRewardHackingOptions, Yr as GepaCandidatePopulationSummary, Yt as CrossSurfaceAdditionRejectionReason, Zr as readGepaCandidatePopulationArtifact, Zt as CrossSurfaceBestSingleSelection, _ as surfaceDispatchRef, _i as RunOptimizationOptions, _n as CrossSurfacePairCompatibility, _r as OptimizationMethodComparison, _t as scopedOptimizationMethod, a as gitWorktreeAdapter, an as CrossSurfaceComponentEvidence, at as ScoreboardSummary, b as SingleRunLockOptions, bi as OpenAutoPrOptions, bn as CrossSurfacePairwiseEntry, br as compareOptimizationMethods, bt as FsLabeledScenarioStoreOptions, c as AutoevalsScoreLike, ci as LlmJudgeOptions, cn as CrossSurfaceEligibility, ct as renderScoreboardMarkdown, d as PhoenixEvaluatorLike, di as isTransientTransportFailure, dn as CrossSurfaceInteractionAwareSelection, dr as composeGate, dt as userStoryScoreboard, ei as ExternalOptimizerObservationSummary, en as CrossSurfaceCandidateComparison, et as PlaybackContext, f as assertCodeSurfaceIdentity, fi as transientDispatchFailure, fn as CrossSurfaceInteractionEffect, fr as SkillOptOptimizationMethodConfig, g as surfaceContentHash, gi as PremeasuredOptimizationBaseline, gn as CrossSurfaceNaiveStackSelection, gr as CompareOptimizationMethodsOptions, h as renderSurfaceDiff, hi as runImprovementLoop, hn as CrossSurfaceInteractionTask, hr as skillOptOptimizationMethod, i as WorktreeAdapter, ii as ReferenceEquivalenceJudgeOptions, in as CrossSurfaceComponent, it as ScoreboardRow, j as RetainedDispatchOutcome, ji as PlanCampaignRunOptions, jn as EmitLoopProvenanceArgs, jr as ExternalTextOptimizationMethodConfig, jt as importGepaPopulation, k as resolveRunDir, ki as CampaignRunPlan, kn as TraceAnalystScenario, kr as OptimizerModelBudget, kt as externalSurface, l as AutoevalsScorerLike, li as llmJudge, ln as CrossSurfaceEvidenceBreakdown, lt as scoreUserStory, m as componentSurfaceIdentityMaterial, mi as RunImprovementLoopResult, mn as CrossSurfaceInteractionReport, mr as SkillOptTrainerConfig, n as GitWorktreeAdapterOptions, ni as readExternalOptimizerObservationArtifact, nn as CrossSurfaceCandidateOutcome, nt as PlaybackStep, o as resolveWorktreePath, oi as ReferenceEquivalenceScenario, on as CrossSurfaceCompositionStep, ot as UserStory, p as codeSurfaceIdentityMaterial, pi as RunImprovementLoopOptions, pn as CrossSurfaceInteractionPath, pr as SkillOptRunnerCommand, q as RunProfileMatrixResult, qi as SearchLedgerIntegrityError, qn as DefaultProductionGateCheck, qr as GepaCandidatePopulationArtifact, qt as AnalyzeCrossSurfaceInteractionsInput, r as Worktree, rn as CrossSurfaceCandidateSummary, rt as ScoreboardRenderOptions, s as verifyCodeSurface, si as LlmJudgeDimension, sn as CrossSurfaceDistribution, st as UserStoryVerdict, t as CodeSurfaceVerification, ti as ExternalOptimizerSubmittedCandidate, tn as CrossSurfaceCandidateEvidence, tt as PlaybackDriver, u as PhoenixEvaluationResultLike, ui as TransientFailureOptions, un as CrossSurfaceIneligibilityReason, ut as scoreboardSummary, v as surfaceHash, vi as RunOptimizationResult, vn as CrossSurfacePairEvidence, vr as OptimizationMethodPairwise, vt as sequentialOptimizationMethod, w as searchProposalExecution, wi as runEval, wn as CrossSurfaceSelections, wr as GepaOptimizationMethodConfig, wt as ScoredRollout, x as acquireSingleRunLock, xi as OpenAutoPrResult, xn as CrossSurfaceRankedSingle, xr as GepaAdaptiveEngineRun, xt as RolloutArgumentDiff, y as SingleRunLock, yi as runOptimization, yn as CrossSurfacePairIncompatibilityReason, yr as OptimizationMethodScore, yt as neutralizeText, z as ProfileMatrixSegmentResult, zi as FinalEvidenceUse, zn as canonicalDigest, zr as OptimizationMethodInput, zt as EvalFixtureValidationMode } from "../index-B7v2QHlD.js";
+import { A as LabeledScenarioWrite, B as ScoredSurfaceOutcome, C as JudgeDimension, D as LabeledScenarioSampleArgs, E as LabeledScenarioRecord, F as ProposeContext, G as SearchProposerView, H as SurfaceProposer, I as ProposedCandidate, J as searchProposerView, K as SearchSummaryOptions, L as RedactionStatus, M as OptimizerConfig, N as ParetoParent, O as LabeledScenarioSource, P as ProposalTrackContext, R as Scenario, S as JudgeConfig, T as LabelTrust, U as TraceSpan, V as SessionScript, W as isProposedCandidate, X as SearchStateView, Y as SearchState, Z as searchUnitScores, _ as GateDecision, a as CampaignResult, at as SearchCompletion, b as GenerationRecord, c as CampaignTraceWriter, ct as SearchOperation, d as DispatchContext, dt as SearchUnitScore, f as DispatchFn, ft as searchCellId, g as GateContribution, h as GateContext, i as CampaignCostMeter, it as SearchCell, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, lt as SearchScoredCell, m as GateCheckStatus, mt as searchTaskSetDigest, n as CampaignArtifactWriter, o as CampaignScenarioIdentity, ot as SearchDecisionRecord, p as Gate, pt as searchNodeId, q as renderSearchSummary, r as CampaignCellResult, s as CampaignTokenUsage, st as SearchNode, t as CampaignAggregates, u as ComponentSurface, ut as SearchSpend, v as GateResult, w as JudgeScore, x as JudgeAggregate, y as GenerationCandidate, z as ScenarioAggregate } from "../types-C54Uk_a1.js";
+import { A as SearchRungDecision, At as SearchHistoryReceipt, B as beam, Bt as openSearchLedger, C as searchClaimDecision, Ct as surfaceNode, D as SearchAllocationView, Dt as SearchHistoryCoverage, E as AshaOptions, Et as SearchHistoryAuditSummary, F as SearchExpansion, Ft as FileSearchLedger, G as NodePosterior, Gt as createRunCostLedger, H as incumbent, Ht as replaySearchLedgerText, I as SearchPolicy, It as OpenSearchLedgerOptions, J as estimateNode, K as SEARCH_ESTIMATOR, Kt as fsCampaignStorage, L as SearchPolicyNode, Lt as SearchLedger, M as asha, Mt as createSearchHistoryReceipt, N as uniform, Nt as verifySearchHistoryArtifact, O as SearchAllocator, Ot as SearchHistoryCoverageRow, P as AideOptions, Pt as verifySearchHistoryReceipt, Q as searchPosterior, R as SearchPolicyView, Rt as SearchLedgerAppendResult, S as planSearchClaim, St as surfaceDiff, T as verifySearchClaim, Tt as SearchHistoryAdmissionOptions, U as incumbentWithOperatorBandit, Ut as validateSearchLedgerEvent, V as crowdedFrontierParent, Vt as parseSearchLedgerLine, W as EstimateNodeCellsInput, Wt as CampaignStorage, X as searchCellSetDigest, Y as estimateNodeFromCells, Z as searchEstimateMethod, _ as SEARCH_CLAIM_RULE, _t as SearchRecorderOptions, a as SearchArtifactCodec, b as SearchClaimVerification, bt as developmentClaim, c as SearchExecutor, d as SearchProposalRequest, dt as AllocateSearchCellInput, f as SearchProposalResult, ft as RecordSearchEdgeInput, g as PlanSearchClaimInput, gt as SearchRecorder, h as SearchRunResult, ht as SearchOpening, i as RunSearchOptions, j as SearchUnitExtension, jt as assertSearchHistoryMatchesState, k as SearchCellPlan, kt as SearchHistoryPolicy, l as SearchLane, m as SearchProposerPort, mt as SearchLedgerBinding, n as searchDivergence, o as SearchCellResult, p as SearchProposedChild, pt as RegisterSearchNodeInput, q as SearchPosterior, qt as inMemoryCampaignStorage, r as searchPolicyView, s as SearchCellWork, t as runSearch, u as SearchProposalBlob, v as SearchClaimDecision, vt as SearchRunIdentity, w as searchClaimReserveUsd, wt as CreateSearchHistoryReceiptInput, x as decideSearchClaim, xt as searchEdgeId, y as SearchClaimPlan, yt as SettleSearchCellInput, z as aide, zt as SearchLedgerTrustedHeadMode } from "../search-kernel-CBdwFZ05.js";
+import { B as ExternalOptimizerModelCallRequest, F as ExternalOptimizerEndpointFormat, G as ExternalOptimizerRunnerCommand, H as ExternalOptimizerModelExecutionObservation, I as ExternalOptimizerEvaluationObservation, L as ExternalOptimizerEvaluationRefusalReason, M as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, N as ExternalOptimizerCallbackLimits, P as ExternalOptimizerChatRequest, R as ExternalOptimizerModelBudget, U as ExternalOptimizerProcessLimits, V as ExternalOptimizerModelCallResult, j as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, q as ExternalTextCandidate, z as ExternalOptimizerModelCall } from "../agent-profile-CoD1Dotf.js";
 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 CampaignCoverage, 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 CostBreakdown, 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, 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, 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, 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_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, type SearchLane, type SearchLedger, type SearchLedgerAppendResult, type SearchLedgerBinding, 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 SequentialGateDecision, 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, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCodeSurfaceIdentity, assertSearchHistoryMatchesState, beam, buildCellSchedule, buildEvidenceVector, campaignBreakdown, campaignCellSearchResult, campaignCoverage, campaignMeanComposite, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, codeSurfaceIdentityMaterial, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, makeProposalFinding, measureCampaignCost, neutralizationGate, neutralizeText, openSearchLedger, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveRunDir, resolveWorktreePath, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimDecision, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchEstimateMethod, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };

package/dist/campaign/index.js

@@ -1,16 +1,16 @@
-import { a as heldoutSignificance, i as dimensionRegressions, o as pairHoldout } from "../statistical-heldout-TP6ndL0B.js";
-import { $ as resolveRunDir, A as campaignCellSearchResult, C as surfaceDiff, D as createSearchHistoryReceipt, E as assertSearchHistoryMatchesState, F as uniform, M as searchProposalExecution, N as searchReceiptAccounting, O as verifySearchHistoryArtifact, P as asha, Q as planCampaignRun, S as searchEdgeId, X as runEval, Z as runCampaign, _ as crowdedFrontierParent, b as SearchRecorder, c as readGepaCandidatePopulationArtifact, et as buildCellSchedule, f as runImprovementLoop, g as beam, h as aide, j as searchModelIdentity, k as verifySearchHistoryReceipt, m as isProposedCandidate, n as isTransientTransportFailure, p as runOptimization, r as transientDispatchFailure, t as llmJudge, tt as cellCachePath, u as costFromLedgerSummary, v as incumbent, w as surfaceNode, x as developmentClaim, y as incumbentWithOperatorBandit, z as defaultProductionGate } from "../llm-judge-cgbhTKQp.js";
-import { A as campaignSplitDigest, O as campaignCoverage, T as assertCampaignDesign, f as campaignBreakdown, j as campaignSplitDigestFromIdentities, k as campaignScenarioIdentity, l as canonicalDigest, p as campaignMeanComposite, u as emitLoopProvenance, v as powerPreflight } from "../campaign-evidence-DN8rWuBj.js";
+import { a as heldoutSignificance, i as dimensionRegressions, o as pairHoldout } from "../statistical-heldout-C9hz_AF9.js";
+import { $ as resolveRunDir, A as campaignCellSearchResult, C as surfaceDiff, D as createSearchHistoryReceipt, E as assertSearchHistoryMatchesState, F as uniform, M as searchProposalExecution, N as searchReceiptAccounting, O as verifySearchHistoryArtifact, P as asha, Q as planCampaignRun, S as searchEdgeId, X as runEval, Z as runCampaign, _ as crowdedFrontierParent, b as SearchRecorder, c as readGepaCandidatePopulationArtifact, et as buildCellSchedule, f as runImprovementLoop, g as beam, h as aide, j as searchModelIdentity, k as verifySearchHistoryReceipt, m as isProposedCandidate, n as isTransientTransportFailure, p as runOptimization, r as transientDispatchFailure, t as llmJudge, tt as cellCachePath, u as costFromLedgerSummary, v as incumbent, w as surfaceNode, x as developmentClaim, y as incumbentWithOperatorBandit, z as defaultProductionGate } from "../llm-judge-D6JtIqiU.js";
+import { A as campaignSplitDigest, O as campaignCoverage, T as assertCampaignDesign, f as campaignBreakdown, j as campaignSplitDigestFromIdentities, k as campaignScenarioIdentity, l as canonicalDigest, p as campaignMeanComposite, u as emitLoopProvenance, v as powerPreflight } from "../campaign-evidence-MjCjkb5a.js";
 import { a as SearchLedgerError, o as SearchLedgerIntegrityError } from "../search-ledger-file-DTBD6X-y.js";
 import { M as fsCampaignStorage, N as inMemoryCampaignStorage, j as createRunCostLedger, l as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, u as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS } from "../external-optimizer-subprocess-U616hV_l.js";
 import { a as componentSurfaceIdentityMaterial, c as surfaceDispatchRef, i as codeSurfaceIdentityMaterial, l as surfaceHash, o as renderSurfaceDiff, r as assertCodeSurfaceIdentity, s as surfaceContentHash } from "../json-recovery-DK4MDm5r.js";
 import { a as makeProposalFinding } from "../types-BK0blWhe.js";
-import { C as SEARCH_ESTIMATOR, D as searchEstimateMethod, E as searchCellSetDigest, O as searchPosterior, T as estimateNodeFromCells, _ as planSearchClaim, a as searchDivergence, b as verifySearchClaim, g as decideSearchClaim, h as SEARCH_CLAIM_RULE, i as runSearch, o as searchPolicyView, v as searchClaimDecision, w as estimateNode, y as searchClaimReserveUsd } from "../pareto-B9udAt2i.js";
+import { C as SEARCH_ESTIMATOR, D as searchEstimateMethod, E as searchCellSetDigest, O as searchPosterior, T as estimateNodeFromCells, _ as planSearchClaim, a as searchDivergence, b as verifySearchClaim, g as decideSearchClaim, h as SEARCH_CLAIM_RULE, i as runSearch, o as searchPolicyView, v as searchClaimDecision, w as estimateNode, y as searchClaimReserveUsd } from "../pareto-7_LEmYGG.js";
 import { a as validateSearchLedgerEvent, c as SearchStateView, d as searchNodeId, f as searchTaskSetDigest, i as replaySearchLedgerText, l as searchUnitScores, n as openSearchLedger, o as SEARCH_LEDGER_SCHEMA, r as parseSearchLedgerLine, s as SearchState, t as FileSearchLedger, u as searchCellId } from "../search-ledger-BsVi-Ngv.js";
 import { t as acquireSingleRunLock } from "../single-run-lock-BGlW3OSk.js";
-import { n as searchProposerView, t as renderSearchSummary } from "../search-summary-Ct7SrmVc.js";
-import { C as measureCampaignCost, S as analyzeCrossSurfaceInteractions, _ as neutralizationGate, a as createRetainedDispatch, b as loadEvalFixtureScenarios, c as runProfileMatrixSegment, d as scoreUserStory, f as scoreboardSummary, g as neutralizeText, h as sequentialOptimizationMethod, i as scoreDiscrimination, l as runProfileMatrix, m as scopedOptimizationMethod, n as resolveWorktreePath, o as createProfileMatrixPlan, p as userStoryScoreboard, r as verifyCodeSurface, s as finalizeProfileMatrix, t as gitWorktreeAdapter, u as renderScoreboardMarkdown, v as discoverEvalFixtures, x as planEvalFixtureRun, y as loadEvalFixture } from "../campaign-ddr1N3yD.js";
-import { a as importExternalEvaluations, c as composeGate, d as readExternalOptimizerObservationArtifact, i as externalSurface, l as externalTextOptimizationMethod, n as compareOptimizationMethods, o as importGepaPopulation, r as gepaOptimizationMethod, s as heldOutGate, t as skillOptOptimizationMethod, u as decodeExternalTextCandidate } from "../skillopt-optimization-method-qB1M554K.js";
-import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-CBYeD4zi.js";
-import { n as sequentialPairedGate, t as sequentialDecide } from "../sequential-B1I0tRfL.js";
+import { n as searchProposerView, t as renderSearchSummary } from "../search-summary-Bt3uuK53.js";
+import { C as measureCampaignCost, S as analyzeCrossSurfaceInteractions, _ as neutralizationGate, a as createRetainedDispatch, b as loadEvalFixtureScenarios, c as runProfileMatrixSegment, d as scoreUserStory, f as scoreboardSummary, g as neutralizeText, h as sequentialOptimizationMethod, i as scoreDiscrimination, l as runProfileMatrix, m as scopedOptimizationMethod, n as resolveWorktreePath, o as createProfileMatrixPlan, p as userStoryScoreboard, r as verifyCodeSurface, s as finalizeProfileMatrix, t as gitWorktreeAdapter, u as renderScoreboardMarkdown, v as discoverEvalFixtures, x as planEvalFixtureRun, y as loadEvalFixture } from "../campaign-DdOgQfp8.js";
+import { a as importExternalEvaluations, c as composeGate, d as readExternalOptimizerObservationArtifact, i as externalSurface, l as externalTextOptimizationMethod, n as compareOptimizationMethods, o as importGepaPopulation, r as gepaOptimizationMethod, s as heldOutGate, t as skillOptOptimizationMethod, u as decodeExternalTextCandidate } from "../skillopt-optimization-method-BKFbJEyz.js";
+import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-D7B4_jEZ.js";
+import { n as sequentialPairedGate, t as sequentialDecide } from "../sequential-De7BTAlV.js";
 export { DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, FileSearchLedger, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_LEDGER_SCHEMA, SearchLedgerError, SearchLedgerIntegrityError, SearchRecorder, SearchState, SearchStateView, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCodeSurfaceIdentity, assertSearchHistoryMatchesState, beam, buildCellSchedule, buildEvidenceVector, campaignBreakdown, campaignCellSearchResult, campaignCoverage, campaignMeanComposite, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, codeSurfaceIdentityMaterial, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, makeProposalFinding, measureCampaignCost, neutralizationGate, neutralizeText, openSearchLedger, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveRunDir, resolveWorktreePath, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimDecision, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchEstimateMethod, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };

package/dist/{campaign-ddr1N3yD.js → campaign-DdOgQfp8.js}

@@ -3,24 +3,25 @@ import { a as hashCanonical, i as compareCodeUnits, r as canonicalString } from
 import { a as agentProfileModelId, i as agentProfileId, r as agentProfileHash, s as harnessAxisOf } from "./agent-profile-Dngk3LZ0.js";
 import { i as buildAgentProfileCell } from "./agent-profile-cell-CT5JYueM.js";
 import { c as validateRunRecord, i as modelHasSnapshot, n as UNKNOWN_MODEL, o as runTaskScore } from "./run-record-DEjx43cI.js";
-import { t as comparePairedArms } from "./paired-arms-D-54hO-k.js";
-import { i as pairedBootstrap, t as mean } from "./mean-CdDYG-us.js";
-import { o as pairHoldout } from "./statistical-heldout-TP6ndL0B.js";
+import { t as comparePairedArms } from "./paired-arms-D5H_58e8.js";
+import { r as pairedBootstrap } from "./paired-tests-C2zUqgUW.js";
+import { n as meanOrNull, t as mean } from "./mean-BPuUNznQ.js";
+import { o as pairHoldout } from "./statistical-heldout-C9hz_AF9.js";
 import { d as mapConcurrent } from "./journal-C1uLJi0w.js";
 import { n as contentHash, t as canonicalJson } from "./verdict-cache-BVH-fl18.js";
-import { $ as resolveRunDir, Q as planCampaignRun, Z as runCampaign, l as combineComparisonCosts, nt as computeManifestHash, o as executeOptimizationMethod, tt as cellCachePath } from "./llm-judge-cgbhTKQp.js";
-import { n as campaignCellToRunRecord, r as projectCampaignCellQuality, t as campaignCellCostProvenance } from "./run-record-CZy4-TnZ.js";
-import { C as assertRealBackend, T as assertCampaignDesign, k as campaignScenarioIdentity, w as summarizeBackendIntegrity } from "./campaign-evidence-DN8rWuBj.js";
+import { $ as resolveRunDir, Q as planCampaignRun, Z as runCampaign, l as combineComparisonCosts, nt as computeManifestHash, o as executeOptimizationMethod, tt as cellCachePath } from "./llm-judge-D6JtIqiU.js";
+import { n as campaignCellToRunRecord, r as projectCampaignCellQuality, t as campaignCellCostProvenance } from "./run-record-BUm-Jggo.js";
+import { C as assertRealBackend, T as assertCampaignDesign, k as campaignScenarioIdentity, w as summarizeBackendIntegrity } from "./campaign-evidence-MjCjkb5a.js";
 import { M as fsCampaignStorage } from "./external-optimizer-subprocess-U616hV_l.js";
 import { r as assertCodeSurfaceIdentity, s as surfaceContentHash } from "./json-recovery-DK4MDm5r.js";
 import "./types-BK0blWhe.js";
-import "./pareto-B9udAt2i.js";
+import "./pareto-7_LEmYGG.js";
 import "./search-ledger-BsVi-Ngv.js";
-import "./search-summary-Ct7SrmVc.js";
+import "./search-summary-Bt3uuK53.js";
 import { r as verifyCompletion } from "./completion-verifier-rqbMh5-4.js";
 import "./external-optimizer-process-BRVmaDFe.js";
-import "./skillopt-optimization-method-qB1M554K.js";
-import "./sequential-B1I0tRfL.js";
+import "./skillopt-optimization-method-BKFbJEyz.js";
+import "./sequential-De7BTAlV.js";
 import { createHash } from "node:crypto";
 import { z } from "zod";
 import { closeSync, constants, existsSync, fstatSync, lstatSync, mkdirSync, mkdtempSync, openSync, readFileSync, readSync, readdirSync, readlinkSync, realpathSync, rmSync, statSync } from "node:fs";
@@ -531,9 +532,6 @@ function distribution(values) {
 		total
 	};
 }
-function meanOrNull$1(values) {
-	return values.length === 0 ? null : values.reduce((sum, value) => sum + value, 0) / values.length;
-}
 function unique(values) {
 	return [...new Set(values)];
 }
@@ -637,8 +635,8 @@ function interactionEffect(context, baseline, singles, composition) {
 		perTask,
 		n: complete.length,
 		nMissing: perTask.length - complete.length,
-		meanPassInteraction: meanOrNull$1(passInteractions),
-		meanScoreInteraction: meanOrNull$1(scoreInteractions),
+		meanPassInteraction: meanOrNull(passInteractions),
+		meanScoreInteraction: meanOrNull(scoreInteractions),
 		passBootstrap: bootstrapInteraction(passInteractions, context),
 		scoreBootstrap: bootstrapInteraction(scoreInteractions, context)
 	};
@@ -1689,9 +1687,6 @@ async function runProfileMatrix(opts) {
 function scoreOf(r) {
 	return runTaskScore(r);
 }
-function meanOrNull(values) {
-	return values.length === 0 ? null : mean(values);
-}
 function rollup(records, keyOf) {
 	const groups = /* @__PURE__ */ new Map();
 	for (const r of records) {
@@ -3101,4 +3096,4 @@ function resolveWorktreePath(surface, worktreeDir) {
 }
 export { measureCampaignCost as C, analyzeCrossSurfaceInteractions as S, neutralizationGate as _, createRetainedDispatch as a, loadEvalFixtureScenarios as b, runProfileMatrixSegment as c, scoreUserStory as d, scoreboardSummary as f, neutralizeText as g, sequentialOptimizationMethod as h, scoreDiscrimination as i, runProfileMatrix as l, scopedOptimizationMethod as m, resolveWorktreePath as n, createProfileMatrixPlan as o, userStoryScoreboard as p, verifyCodeSurface as r, finalizeProfileMatrix as s, gitWorktreeAdapter as t, renderScoreboardMarkdown as u, discoverEvalFixtures as v, planEvalFixtureRun as x, loadEvalFixture as y };
 
-//# sourceMappingURL=campaign-ddr1N3yD.js.map
+//# sourceMappingURL=campaign-DdOgQfp8.js.map