@tangle-network/agent-eval 0.216.1 → 0.217.1

package/dist/contract/index.d.ts

@@ -1,14 +1,14 @@
 import { a as CostLedgerHandle } from "../cost-ledger-Jk62OBfF.js";
-import { a as AnalystFinding, b as ProposalFinding, w as makeProposalFinding, x as ProposalFindingOrigin } from "../types-C1tnkEgM.js";
-import { A as createChatClient, T as CreateChatClientOpts, x as ChatClient } from "../types-C02qtnG8.js";
-import { $ as paretoSignificanceGate, G as BuildEvidenceVectorOptions, J as ParetoSignificanceGateOptions, K as EvidenceVector, Q as paretoPolicy, U as AxisEvidence, W as AxisVerdict, X as PromotionPolicy, Y as PromotionObjective, Z as buildEvidenceVector, q as ObjectiveSource } from "../statistical-heldout-CYpGaoDH.js";
-import { $i as RunImprovementLoopOptions, $r as compareOptimizationMethods, Bi as ReferenceEquivalenceJudgeInput, Ci as ComparisonCost, Dr as DefaultProductionGateOptions, Ea as FinalEvidenceUse, Er as DefaultProductionGateCheck, Gi as runReferenceEquivalenceJudge, Gr as SkillOptOptimizationMethodConfig, Hi as ReferenceEquivalenceJudgeResult, Ji as llmJudge, Jr as skillOptOptimizationMethod, Ki as LlmJudgeDimension, Kr as SkillOptRunnerCommand, Ma as campaignSplitDigest, Or as DefaultProductionRewardHackingOptions, Qi as transientDispatchFailure, Ri as REFERENCE_EQUIVALENCE_INPUT_LIMITS, Si as OptimizationPackageSource, Ta as FinalEvidencePolicy, Tr as heldOutGate, Ui as ReferenceEquivalenceScenario, Vi as ReferenceEquivalenceJudgeOptions, Wi as createReferenceEquivalenceJudge, Wr as composeGate, Xr as OptimizationMethodComparison, Yr as CompareOptimizationMethodsOptions, ai as GepaRunnerCommand, bi as OptimizationMethodResult, ci as OptimizerModelBudget, da as CampaignCellFailureReceipt, di as ExternalTextOptimizerContext, ea as RunImprovementLoopResult, ei as GepaAdaptiveEngineRun, fa as CampaignCellRetryPolicy, fi as ExternalTextOptimizerResult, gi as OptimizationMethod, ii as GepaOptimizationRecipe, kr as defaultProductionGate, la as RunEvalOptions, li as externalTextOptimizationMethod, ma as runCampaign, mi as ExternalTextEvaluationResponse, ni as GepaEngineRun, oi as gepaOptimizationMethod, pa as RunCampaignOptions, pi as ExternalOptimizationExample, qi as LlmJudgeOptions, qr as SkillOptTrainerConfig, ri as GepaOptimizationMethodConfig, si as OpenAICompatibleOptimizerModel, ta as runImprovementLoop, ti as GepaEngineOptions, ua as runEval, ui as ExternalTextOptimizationMethodConfig, vi as OptimizationMethodInput, wi as OptimizationTokenUsage, wr as HeldOutGateOptions, yi as OptimizationMethodProvenance, zi as REFERENCE_EQUIVALENCE_JUDGE_VERSION } from "../index-CyszSg4O.js";
-import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-Cq8mSbq8.js";
-import { $t as fsCampaignStorage, Zt as CampaignStorage, en as inMemoryCampaignStorage } from "../search-kernel-DyNEa6IU.js";
+import { S as makeProposalFinding, i as AnalystFinding, v as ProposalFinding, y as ProposalFindingOrigin } from "../types-CW8ewATZ.js";
+import { A as createChatClient, T as CreateChatClientOpts, x as ChatClient } from "../types-D-mgRzlh.js";
+import { G as ObjectiveSource, H as AxisVerdict, J as PromotionPolicy, K as ParetoSignificanceGateOptions, U as BuildEvidenceVectorOptions, V as AxisEvidence, W as EvidenceVector, X as paretoPolicy, Y as buildEvidenceVector, Z as paretoSignificanceGate, q as PromotionObjective } from "../statistical-heldout-C9AbgREN.js";
+import { Br as OptimizerModelBudget, Ci as transientDispatchFailure, Cr as SkillOptOptimizationMethodConfig, Dr as CompareOptimizationMethodsOptions, Ei as runImprovementLoop, Er as skillOptOptimizationMethod, Fi as runEval, Fr as GepaOptimizationMethodConfig, Gr as ExternalOptimizationExample, Hr as ExternalTextOptimizationMethodConfig, Ii as CampaignCellFailureReceipt, Ir as GepaOptimizationRecipe, Jr as OptimizationMethod, Kr as ExternalTextEvaluationResponse, Li as CampaignCellRetryPolicy, Lr as GepaRunnerCommand, Mr as GepaAdaptiveEngineRun, Nr as GepaEngineOptions, Or as OptimizationMethodComparison, Pi as RunEvalOptions, Pr as GepaEngineRun, Qr as OptimizationMethodResult, Ri as RunCampaignOptions, Rr as gepaOptimizationMethod, Sr as composeGate, Ti as RunImprovementLoopResult, Tr as SkillOptTrainerConfig, Ur as ExternalTextOptimizerContext, Vr as externalTextOptimizationMethod, Wr as ExternalTextOptimizerResult, Xi as FinalEvidenceUse, Xr as OptimizationMethodInput, Yi as FinalEvidencePolicy, Zr as OptimizationMethodProvenance, _i as ReferenceEquivalenceScenario, ar as DefaultProductionGateOptions, bi as llmJudge, ei as OptimizationPackageSource, gi as ReferenceEquivalenceJudgeResult, hi as ReferenceEquivalenceJudgeOptions, ir as DefaultProductionGateCheck, jr as compareOptimizationMethods, mi as ReferenceEquivalenceJudgeInput, ni as OptimizationTokenUsage, nr as HeldOutGateOptions, or as DefaultProductionRewardHackingOptions, rr as heldOutGate, sr as defaultProductionGate, ta as campaignSplitDigest, ti as ComparisonCost, vi as LlmJudgeDimension, wi as RunImprovementLoopOptions, wr as SkillOptRunnerCommand, yi as LlmJudgeOptions, zi as runCampaign, zr as OpenAICompatibleOptimizerModel } from "../index-ZuWx52WG.js";
+import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-D8GhJoHk.js";
+import { Kt as fsCampaignStorage, Wt as CampaignStorage, qt as inMemoryCampaignStorage } from "../search-kernel-DxqjBxA-.js";
 import { i as InMemoryOutcomeStore, n as FileSystemOutcomeStore, o as OutcomeStore, r as FileSystemOutcomeStoreOptions, t as DeploymentOutcome } from "../outcome-store-CNt4iZ67.js";
-import { r as HostedTenant } from "../client-BzL9niqi.js";
-import { _ as ReleaseSummary, a as summarizeExecution, c as ExecutionInsight, d as InsightReport, f as InterRaterInsight, g as Recommendation, h as OutcomeCorrelationInsight, i as SummarizeExecutionOptions, l as FailureClassTally, m as LiftInsight, n as analyzeRuns, o as CostProvenanceSummary, p as JudgeInsight, r as ExecutionReport, s as ExecutionErrorOutcomeCell, t as AnalyzeRunsOptions, u as FailureClusterInsight, v as ScalarDistribution, y as TokenUsageInsight } from "../analyze-runs-D68fd-su.js";
-import { _ as SelfImproveMethodResult, a as DefinedAgentEval, c as SelfImproveMethodOptions, d as SelfImproveProposerOptions, f as SelfImproveProposerResult, g as SelfImproveMethodProvenance, h as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveOptions, m as SelfImproveRunError, n as AgentEvalEvaluateOptions, o as defineAgentEval, p as SelfImproveResult, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveProgressEvent } from "../define-agent-eval-WSck0CZJ.js";
-import { $ as FromRunRecordDirOptions, A as CodeAgentSessionActionKind, At as SealCandidateBenchmarkSuiteOptions, B as AgentTraceConversation, C as CodeAgentJsonlLine, Ct as CandidateExperimentRun, D as CodeAgentSessionMetrics, Dt as PairedMeasurementAdapter, E as CodeAgentSessionIntakeResult, Et as PairedMeasurement, F as CodeAgentSessionSource, G as AuthoringProvenance, H as AgentTraceIndex, I as CodeAgentSessionTerminalStatus, J as partitionRunsByAuthoringModel, K as PartitionByAuthoringModelResult, L as observeCodeAgentSession, M as CodeAgentSessionActionSurface, N as CodeAgentSessionExecutionReceipt, O as ParsedCodeAgentJsonl, Ot as PairedMeasurementEvaluation, P as CodeAgentSessionObservation, Q as evalReportingSuite, R as AgentTraceContributor, S as streamCodeAgentJsonlFile, St as CandidateExperimentExecutionInput, T as CodeAgentSessionIntakeOptions, Tt as EvaluatePairedMeasurementsOptions, U as AgentTraceRange, V as AgentTraceFile, W as AgentTraceRecord, X as EvalReportingSuiteOptions, Y as EvalReportingSuiteInput, Z as EvalReportingSuiteResult, _ as fromOpenCodeSession, _t as sealCandidateBenchmarkSuite, a as FromFeedbackTableOptions, at as CompareAgentProfileImprovementExperimentOptions, b as parseCodeAgentJsonl, bt as verifyCandidateExperiment, c as CodeAgentStoreObservation, ct as measuredComparisonFromAgentProfileImprovementExperiment, d as CodeAgentStoreScope, dt as sealAgentProfileImprovementSuite, et as FromRunRecordDirResult, f as CodeAgentStoreSession, ft as sealAgentProfileImprovementTask, g as fromKimiCodeSession, gt as runCandidateExperiment, h as fromCodexSession, ht as measuredComparisonFromCandidateExperiment, i as FeedbackTableRow, it as AgentProfileImprovementExperimentRun, j as CodeAgentSessionActionStatus, k as CodeAgentSessionAction, kt as RunCandidateExperimentOptions, l as CodeAgentStoreRejection, lt as runAgentProfileImprovementExperiment, m as fromClaudeCodeSession, mt as evaluatePairedMeasurements, n as fromOtelSpans, nt as fromRunRecordDir, o as FromFeedbackTableResult, ot as RunAgentProfileImprovementExperimentOptions, p as observeCodeAgentStore, pt as verifyAgentProfileImprovementExperimentComparison, q as parseAgentTrace, r as FeedbackTableMeta, rt as AgentProfileImprovementExperimentExecutionInput, s as fromFeedbackTable, st as SealAgentProfileImprovementSuiteOptions, t as FromOtelSpansOptions, tt as RunRecordRejection, u as CodeAgentStoreRejections, ut as sealAgentProfileImprovementExperiment, v as fromPiGraphSession, vt as sealCandidateBenchmarkTask, w as CodeAgentSessionDiagnostic, wt as CompareCandidateExperimentOptions, x as parseCodeAgentJsonlFile, xt as verifyCandidateExperimentComparison, y as fromPiSession, yt as sealCandidateExperiment, z as AgentTraceContributorType } from "../index-DPcngtwt.js";
-import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-DzHiGruW.js";
-export { type AgentEvalAgent, type AgentEvalEvaluateOptions, type AgentEvalImproveOptions, type AgentProfileImprovementExperimentExecutionInput, type AgentProfileImprovementExperimentRun, type AgentTraceContributor, type AgentTraceContributorType, type AgentTraceConversation, type AgentTraceFile, type AgentTraceIndex, type AgentTraceRange, type AgentTraceRecord, type AnalystFinding, type AnalyzeRunsOptions, type AuthoringProvenance, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type CampaignAggregates, type CampaignArtifactWriter, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCellRetryPolicy, type CampaignCostMeter, type CampaignResult, type CampaignStorage, type CampaignTraceWriter, type CandidateExperimentExecutionInput, type CandidateExperimentRun, type ChatClient, type CodeAgentJsonlLine, type CodeAgentSessionAction, type CodeAgentSessionActionKind, type CodeAgentSessionActionStatus, type CodeAgentSessionActionSurface, type CodeAgentSessionDiagnostic, type CodeAgentSessionExecutionReceipt, type CodeAgentSessionIntakeOptions, type CodeAgentSessionIntakeResult, type CodeAgentSessionMetrics, type CodeAgentSessionObservation, type CodeAgentSessionSource, type CodeAgentSessionTerminalStatus, type CodeAgentStoreObservation, type CodeAgentStoreRejection, type CodeAgentStoreRejections, type CodeAgentStoreScope, type CodeAgentStoreSession, type CodeSurface, type CompareAgentProfileImprovementExperimentOptions, type CompareCandidateExperimentOptions, type CompareOptimizationMethodsOptions, type ComparisonCost, type CostLedgerHandle, type CostProvenanceSummary, type CreateChatClientOpts, type DefaultAnalystRegistryOptions, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DefineAgentEvalOptions, type DefinedAgentEval, type DeploymentOutcome, type DispatchFn as Dispatch, type DispatchContext, type EvalReportingSuiteInput, type EvalReportingSuiteOptions, type EvalReportingSuiteResult, type EvaluatePairedMeasurementsOptions, type EvidenceVector, type ExecutionErrorOutcomeCell, type ExecutionInsight, type ExecutionReport, type ExternalOptimizationExample, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, type FailureClassTally, type FailureClusterInsight, type FeedbackTableMeta, type FeedbackTableRow, FileSystemOutcomeStore, type FileSystemOutcomeStoreOptions, type FinalEvidencePolicy, type FinalEvidenceUse, type FromFeedbackTableOptions, type FromFeedbackTableResult, type FromOtelSpansOptions, type FromRunRecordDirOptions, type FromRunRecordDirResult, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaEngineOptions, type GepaEngineRun, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaRunnerCommand, type HeldOutGateOptions, type HostedTenant, InMemoryOutcomeStore, type InsightReport, type InterRaterInsight, type JudgeConfig, type JudgeDimension, type JudgeInsight, type JudgeScore, type LiftInsight, type LlmJudgeDimension, type LlmJudgeOptions, type MutableSurface, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodInput, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type OutcomeCorrelationInsight, type OutcomeStore, type PairedMeasurement, type PairedMeasurementAdapter, type PairedMeasurementEvaluation, type ParetoSignificanceGateOptions, type ParsedCodeAgentJsonl, type PartitionByAuthoringModelResult, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, type Recommendation, type ReferenceEquivalenceJudgeInput, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceJudgeResult, type ReferenceEquivalenceScenario, type ReleaseSummary, type RunAgentProfileImprovementExperimentOptions, type RunCampaignOptions, type RunCandidateExperimentOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunRecordRejection, type ScalarDistribution, type Scenario, type SealAgentProfileImprovementSuiteOptions, type SealCandidateBenchmarkSuiteOptions, type SelfImproveBudget, type SelfImproveMethodOptions, type SelfImproveMethodProvenance, type SelfImproveMethodResult, type SelfImproveOptions, type SelfImproveProgressEvent, type SelfImproveProposerOptions, type SelfImproveProposerResult, type SelfImproveResult, SelfImproveRunError, type SessionScript, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SummarizeExecutionOptions, type SurfaceProposer, type TokenUsageInsight, analyzeRuns, buildDefaultAnalystRegistry, buildEvidenceVector, campaignSplitDigest, compareOptimizationMethods, composeGate, createChatClient, createReferenceEquivalenceJudge, defaultProductionGate, defineAgentEval, evalReportingSuite, evaluatePairedMeasurements, externalTextOptimizationMethod, fromClaudeCodeSession, fromCodexSession, fromFeedbackTable, fromKimiCodeSession, fromOpenCodeSession, fromOtelSpans, fromPiGraphSession, fromPiSession, fromRunRecordDir, fsCampaignStorage, gepaOptimizationMethod, heldOutGate, inMemoryCampaignStorage, llmJudge, makeProposalFinding, measuredComparisonFromAgentProfileImprovementExperiment, measuredComparisonFromCandidateExperiment, observeCodeAgentSession, observeCodeAgentStore, paretoPolicy, paretoSignificanceGate, parseAgentTrace, parseCodeAgentJsonl, parseCodeAgentJsonlFile, partitionRunsByAuthoringModel, runAgentProfileImprovementExperiment, runCampaign, runCandidateExperiment, runEval, runImprovementLoop, runReferenceEquivalenceJudge, sealAgentProfileImprovementExperiment, sealAgentProfileImprovementSuite, sealAgentProfileImprovementTask, sealCandidateBenchmarkSuite, sealCandidateBenchmarkTask, sealCandidateExperiment, selfImprove, skillOptOptimizationMethod, streamCodeAgentJsonlFile, summarizeExecution, transientDispatchFailure, verifyAgentProfileImprovementExperimentComparison, verifyCandidateExperiment, verifyCandidateExperimentComparison };
+import { n as HostedTenant } from "../client-hwo3cJjR.js";
+import { _ as ReleaseSummary, a as summarizeExecution, c as ExecutionInsight, d as InsightReport, f as InterRaterInsight, g as Recommendation, h as OutcomeCorrelationInsight, i as SummarizeExecutionOptions, l as FailureClassTally, m as LiftInsight, n as analyzeRuns, o as CostProvenanceSummary, p as JudgeInsight, r as ExecutionReport, s as ExecutionErrorOutcomeCell, t as AnalyzeRunsOptions, u as FailureClusterInsight, v as ScalarDistribution, y as TokenUsageInsight } from "../analyze-runs-DyeNcenW.js";
+import { _ as SelfImproveMethodResult, a as DefinedAgentEval, c as SelfImproveMethodOptions, d as SelfImproveProposerOptions, f as SelfImproveProposerResult, g as SelfImproveMethodProvenance, h as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveOptions, m as SelfImproveRunError, n as AgentEvalEvaluateOptions, o as defineAgentEval, p as SelfImproveResult, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveProgressEvent } from "../define-agent-eval-BIkH_gAR.js";
+import { $ as AgentProfileImprovementExperimentExecutionInput, A as CodeAgentSessionActionSurface, B as AgentTraceIndex, C as CodeAgentSessionIntakeOptions, Ct as PairedMeasurementAdapter, D as CodeAgentSessionAction, E as ParsedCodeAgentJsonl, Et as SealCandidateBenchmarkSuiteOptions, F as observeCodeAgentSession, G as parseAgentTrace, H as AgentTraceRecord, I as AgentTraceContributor, J as EvalReportingSuiteOptions, K as partitionRunsByAuthoringModel, L as AgentTraceContributorType, M as CodeAgentSessionObservation, N as CodeAgentSessionSource, O as CodeAgentSessionActionKind, P as CodeAgentSessionTerminalStatus, Q as RunRecordRejection, R as AgentTraceConversation, S as CodeAgentSessionDiagnostic, St as PairedMeasurement, T as CodeAgentSessionMetrics, Tt as RunCandidateExperimentOptions, U as AuthoringProvenance, V as AgentTraceRange, W as PartitionByAuthoringModelResult, X as FromRunRecordDirOptions, Y as EvalReportingSuiteResult, Z as FromRunRecordDirResult, _ as fromPiSession, _t as verifyCandidateExperimentComparison, a as FromFeedbackTableOptions, at as runAgentProfileImprovementExperiment, b as streamCodeAgentJsonlFile, bt as CompareCandidateExperimentOptions, c as CodeAgentStoreObservation, ct as sealAgentProfileImprovementTask, d as CodeAgentStoreScope, dt as measuredComparisonFromCandidateExperiment, et as AgentProfileImprovementExperimentRun, f as CodeAgentStoreSession, ft as runCandidateExperiment, g as fromOpenCodeSession, gt as verifyCandidateExperiment, h as fromKimiCodeSession, ht as sealCandidateExperiment, i as FeedbackTableRow, it as measuredComparisonFromAgentProfileImprovementExperiment, j as CodeAgentSessionExecutionReceipt, k as CodeAgentSessionActionStatus, l as CodeAgentStoreRejection, lt as verifyAgentProfileImprovementExperimentComparison, m as fromCodexSession, mt as sealCandidateBenchmarkTask, n as fromOtelSpans, nt as RunAgentProfileImprovementExperimentOptions, o as FromFeedbackTableResult, ot as sealAgentProfileImprovementExperiment, p as fromClaudeCodeSession, pt as sealCandidateBenchmarkSuite, q as EvalReportingSuiteInput, r as FeedbackTableMeta, rt as SealAgentProfileImprovementSuiteOptions, s as fromFeedbackTable, st as sealAgentProfileImprovementSuite, t as FromOtelSpansOptions, tt as CompareAgentProfileImprovementExperimentOptions, u as CodeAgentStoreRejections, ut as evaluatePairedMeasurements, v as parseCodeAgentJsonl, vt as CandidateExperimentExecutionInput, w as CodeAgentSessionIntakeResult, wt as PairedMeasurementEvaluation, x as CodeAgentJsonlLine, xt as EvaluatePairedMeasurementsOptions, y as parseCodeAgentJsonlFile, yt as CandidateExperimentRun, z as AgentTraceFile } from "../index-BiT6RtNH.js";
+import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-DXgLmoDI.js";
+export { type AgentEvalAgent, type AgentEvalEvaluateOptions, type AgentEvalImproveOptions, type AgentProfileImprovementExperimentExecutionInput, type AgentProfileImprovementExperimentRun, type AgentTraceContributor, type AgentTraceContributorType, type AgentTraceConversation, type AgentTraceFile, type AgentTraceIndex, type AgentTraceRange, type AgentTraceRecord, type AnalystFinding, type AnalyzeRunsOptions, type AuthoringProvenance, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type CampaignAggregates, type CampaignArtifactWriter, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCellRetryPolicy, type CampaignCostMeter, type CampaignResult, type CampaignStorage, type CampaignTraceWriter, type CandidateExperimentExecutionInput, type CandidateExperimentRun, type ChatClient, type CodeAgentJsonlLine, type CodeAgentSessionAction, type CodeAgentSessionActionKind, type CodeAgentSessionActionStatus, type CodeAgentSessionActionSurface, type CodeAgentSessionDiagnostic, type CodeAgentSessionExecutionReceipt, type CodeAgentSessionIntakeOptions, type CodeAgentSessionIntakeResult, type CodeAgentSessionMetrics, type CodeAgentSessionObservation, type CodeAgentSessionSource, type CodeAgentSessionTerminalStatus, type CodeAgentStoreObservation, type CodeAgentStoreRejection, type CodeAgentStoreRejections, type CodeAgentStoreScope, type CodeAgentStoreSession, type CodeSurface, type CompareAgentProfileImprovementExperimentOptions, type CompareCandidateExperimentOptions, type CompareOptimizationMethodsOptions, type ComparisonCost, type CostLedgerHandle, type CostProvenanceSummary, type CreateChatClientOpts, type DefaultAnalystRegistryOptions, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DefineAgentEvalOptions, type DefinedAgentEval, type DeploymentOutcome, type DispatchFn as Dispatch, type DispatchContext, type EvalReportingSuiteInput, type EvalReportingSuiteOptions, type EvalReportingSuiteResult, type EvaluatePairedMeasurementsOptions, type EvidenceVector, type ExecutionErrorOutcomeCell, type ExecutionInsight, type ExecutionReport, type ExternalOptimizationExample, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, type FailureClassTally, type FailureClusterInsight, type FeedbackTableMeta, type FeedbackTableRow, FileSystemOutcomeStore, type FileSystemOutcomeStoreOptions, type FinalEvidencePolicy, type FinalEvidenceUse, type FromFeedbackTableOptions, type FromFeedbackTableResult, type FromOtelSpansOptions, type FromRunRecordDirOptions, type FromRunRecordDirResult, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaEngineOptions, type GepaEngineRun, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaRunnerCommand, type HeldOutGateOptions, type HostedTenant, InMemoryOutcomeStore, type InsightReport, type InterRaterInsight, type JudgeConfig, type JudgeDimension, type JudgeInsight, type JudgeScore, type LiftInsight, type LlmJudgeDimension, type LlmJudgeOptions, type MutableSurface, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodInput, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type OutcomeCorrelationInsight, type OutcomeStore, type PairedMeasurement, type PairedMeasurementAdapter, type PairedMeasurementEvaluation, type ParetoSignificanceGateOptions, type ParsedCodeAgentJsonl, type PartitionByAuthoringModelResult, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, type Recommendation, type ReferenceEquivalenceJudgeInput, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceJudgeResult, type ReferenceEquivalenceScenario, type ReleaseSummary, type RunAgentProfileImprovementExperimentOptions, type RunCampaignOptions, type RunCandidateExperimentOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunRecordRejection, type ScalarDistribution, type Scenario, type SealAgentProfileImprovementSuiteOptions, type SealCandidateBenchmarkSuiteOptions, type SelfImproveBudget, type SelfImproveMethodOptions, type SelfImproveMethodProvenance, type SelfImproveMethodResult, type SelfImproveOptions, type SelfImproveProgressEvent, type SelfImproveProposerOptions, type SelfImproveProposerResult, type SelfImproveResult, SelfImproveRunError, type SessionScript, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SummarizeExecutionOptions, type SurfaceProposer, type TokenUsageInsight, analyzeRuns, buildDefaultAnalystRegistry, buildEvidenceVector, campaignSplitDigest, compareOptimizationMethods, composeGate, createChatClient, defaultProductionGate, defineAgentEval, evaluatePairedMeasurements, externalTextOptimizationMethod, fromClaudeCodeSession, fromCodexSession, fromFeedbackTable, fromKimiCodeSession, fromOpenCodeSession, fromOtelSpans, fromPiSession, fsCampaignStorage, gepaOptimizationMethod, heldOutGate, inMemoryCampaignStorage, llmJudge, makeProposalFinding, measuredComparisonFromAgentProfileImprovementExperiment, measuredComparisonFromCandidateExperiment, observeCodeAgentSession, paretoPolicy, paretoSignificanceGate, parseAgentTrace, parseCodeAgentJsonl, parseCodeAgentJsonlFile, partitionRunsByAuthoringModel, runAgentProfileImprovementExperiment, runCampaign, runCandidateExperiment, runEval, runImprovementLoop, sealAgentProfileImprovementExperiment, sealAgentProfileImprovementSuite, sealAgentProfileImprovementTask, sealCandidateBenchmarkSuite, sealCandidateBenchmarkTask, sealCandidateExperiment, selfImprove, skillOptOptimizationMethod, streamCodeAgentJsonlFile, summarizeExecution, transientDispatchFailure, verifyAgentProfileImprovementExperimentComparison, verifyCandidateExperiment, verifyCandidateExperimentComparison };

package/dist/contract/index.js

@@ -1,29 +1,30 @@
-import { a as hashCanonical } from "../canonical-B2RqLofv.js";
-import { a as parseRunRecordSafe } from "../run-record-Boc5IUTp.js";
-import { i as pairedBootstrap } from "../mean-DHoMiudD.js";
-import { t as decidePairedPromotion } from "../paired-promotion-decision-DsAPYGCm.js";
-import { a as heldoutSignificance } from "../statistical-heldout-42NlCNDe.js";
-import { r as summarizeExecution, t as analyzeRuns } from "../analyze-runs-C84irm4i.js";
-import { G as defaultProductionGate, a as transientDispatchFailure, m as runImprovementLoop, nt as runEval, rt as runCampaign, t as llmJudge } from "../llm-judge-CG12SQJq.js";
+import { a as hashCanonical } from "../canonical-Dq2hOQ8L.js";
+import "../run-record-DhEJgoPz.js";
+import { r as summarizeExecution, t as analyzeRuns } from "../analyze-runs-DqPJaJ4X.js";
+import { i as pairedBootstrap } from "../mean-BbA_JCzW.js";
+import { t as decidePairedPromotion } from "../paired-promotion-decision-D0SHb-ZL.js";
+import { a as heldoutSignificance } from "../statistical-heldout-Bksv92Th.js";
+import { B as defaultProductionGate, Q as runCampaign, Z as runEval, i as transientDispatchFailure, o as clamp01, p as runImprovementLoop, t as llmJudge } from "../llm-judge-C8jMTNNG.js";
 import { i as isModelPriced, n as estimateCost } from "../metrics-BKaPBtIG.js";
-import { t as CostLedger } from "../cost-ledger-BP8V4kUQ.js";
-import { G as campaignSplitDigest } from "../campaign-evidence-Clrn4hBV.js";
-import { h as mapConcurrentRange } from "../journal-DfRz9kps.js";
-import { F as inMemoryCampaignStorage, P as fsCampaignStorage } from "../external-optimizer-subprocess-BqW0hHcI.js";
-import { o as makeProposalFinding } from "../types-IXD2HJzu.js";
-import { n as SelfImproveRunError, r as selfImprove, t as defineAgentEval } from "../define-agent-eval-CNTPzIkp.js";
-import { t as buildDefaultAnalystRegistry } from "../default-registry-COhk9g36.js";
-import { t as extractUsage } from "../extract-usage-BrnzRJ6H.js";
-import { t as createChatClient } from "../chat-client-l-MVDlDk.js";
+import { t as CostLedger } from "../cost-ledger-D0352Z4d.js";
+import { z as campaignSplitDigest } from "../campaign-evidence-CQYw8_0H.js";
+import { o as mapConcurrentRange } from "../journal-CR4QikYp.js";
+import { M as fsCampaignStorage, N as inMemoryCampaignStorage } from "../external-optimizer-subprocess-XIHBOSKt.js";
+import { a as makeProposalFinding } from "../types-DqFMNTgK.js";
+import { n as isRecord } from "../json-BhdsJaOQ.js";
+import { n as SelfImproveRunError, r as selfImprove, t as defineAgentEval } from "../define-agent-eval-BDxNlP3D.js";
+import { t as buildDefaultAnalystRegistry } from "../default-registry-ClYjq3QE.js";
+import { t as extractUsage } from "../extract-usage-BZLU01kR.js";
+import { t as createChatClient } from "../chat-client-ldqxYYTN.js";
 import { n as InMemoryOutcomeStore, t as FileSystemOutcomeStore } from "../outcome-store-DUcMj-3D.js";
-import { _ as runReferenceEquivalenceJudge, d as externalTextOptimizationMethod, g as createReferenceEquivalenceJudge, h as REFERENCE_EQUIVALENCE_JUDGE_VERSION, l as heldOutGate, m as REFERENCE_EQUIVALENCE_INPUT_LIMITS, n as compareOptimizationMethods, r as gepaOptimizationMethod, t as skillOptOptimizationMethod, u as composeGate } from "../skillopt-optimization-method-BiwA8nfX.js";
-import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-KO57gbyV.js";
-import { t as fromOtelSpans } from "../otel-spans-B9tUXRHF.js";
+import { c as composeGate, l as externalTextOptimizationMethod, n as compareOptimizationMethods, r as gepaOptimizationMethod, s as heldOutGate, t as skillOptOptimizationMethod } from "../skillopt-optimization-method-kyL94-ri.js";
+import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-6A8osGr_.js";
+import { t as fromOtelSpans } from "../otel-spans-CdN909k1.js";
 import { createHash } from "node:crypto";
 import { agentCandidateBenchmarkSuiteSchema, agentCandidateBenchmarkTaskSchema, agentCandidateBundleSchema, agentCandidateEvaluationPolicySchema, agentCandidateExperimentSchema, agentImprovementMeasuredComparisonSchema, agentProfileImprovementExperimentSchema, agentProfileImprovementMeasuredComparisonSchema, agentProfileImprovementRunCellSchema, agentProfileImprovementRunReceiptSchema, agentProfileImprovementSuiteInputsSchema, agentProfileImprovementSuiteSchema, agentProfileImprovementTaskSchema, candidateExecutionEvidenceSchema, canonicalCandidateDigest, canonicalCandidateJson, numbersApproximatelyEqual, omitTopLevelDigest } from "@tangle-network/agent-interface";
 import { createReadStream } from "node:fs";
-import { dirname, isAbsolute, join, relative, resolve, sep } from "node:path";
-import { mkdir, readFile, readdir, stat, writeFile } from "node:fs/promises";
+import "node:path";
+import "node:fs/promises";
 import { readSessionInput, summarizeSessionInput } from "@tangle-network/harness-sessions";
 //#region src/contract/fixed-spend.ts
 function addFixedSpend(left, right) {
@@ -1128,182 +1129,6 @@ const profileReceiptAdapter = {
 	passed: (receipt) => receipt.grading.passed
 };
 //#endregion
-//#region src/contract/intake/run-record-dir.ts
-/**
-* # `intake/run-record-dir` — load a directory or file of `RunRecord`s.
-*
-* The on-disk counterpart to the in-memory intake adapters: point it at a
-* single `.json` (array) / `.jsonl` (one record per line) file or at a
-* directory of such files, and it returns the substrate-canonical
-* `RunRecord[]` ready for `analyzeRuns({ runs })`.
-*
-* Validation is at the boundary: each parsed object goes through
-* `parseRunRecordSafe`. By default an invalid record fails loud with its
-* file + index; pass `onInvalid: 'collect'` to keep the valid records and
-* receive the rejects as structured diagnostics instead.
-*/
-const ANALYSIS_ARTIFACT$1 = "analysis.json";
-function defaultInclude(fileName) {
-	if (fileName === ANALYSIS_ARTIFACT$1) return false;
-	return fileName.endsWith(".json") || fileName.endsWith(".jsonl");
-}
-/**
-* Resolve a file or directory path into validated `RunRecord[]`.
-*
-* A `.json` file must parse to a top-level array; a `.jsonl` file is one
-* record per non-empty line. Directories are read shallowly by default
-* (set `recursive` to descend); the `analysis.json` output artifact is
-* always excluded.
-*/
-async function fromRunRecordDir(path, options = {}) {
-	const onInvalid = options.onInvalid ?? "throw";
-	const include = options.include ?? defaultInclude;
-	const filePaths = (await stat(path)).isDirectory() ? await collectFiles(path, include, options.recursive ?? false) : [path];
-	const runs = [];
-	const rejected = [];
-	for (const file of filePaths) {
-		const raw = await parseRecordFile(file);
-		for (const { index, value } of raw) {
-			const parsed = parseRunRecordSafe(value);
-			if (parsed.ok) {
-				runs.push(parsed.value);
-				continue;
-			}
-			const rejection = {
-				file,
-				index,
-				reason: parsed.error.message
-			};
-			if (onInvalid === "throw") throw new Error(`fromRunRecordDir: invalid RunRecord in '${file}' at index ${index}: ${parsed.error.message}`);
-			rejected.push(rejection);
-		}
-	}
-	return {
-		runs,
-		rejected,
-		files: filePaths
-	};
-}
-/** Read a single `.json` / `.jsonl` file into `{ index, value }` pairs. A
-*  malformed JSONL line throws with its line number rather than being skipped —
-*  silent line-dropping is how corpora quietly shrink. */
-async function parseRecordFile(file) {
-	const trimmed = (await readFile(file, "utf8")).trim();
-	if (trimmed.length === 0) return [];
-	if (trimmed.startsWith("[")) {
-		const parsed = JSON.parse(trimmed);
-		if (!Array.isArray(parsed)) throw new Error(`fromRunRecordDir: file '${file}' did not parse to an array`);
-		return parsed.map((value, index) => ({
-			index,
-			value
-		}));
-	}
-	const out = [];
-	const lines = trimmed.split("\n");
-	for (let i = 0; i < lines.length; i++) {
-		const line = lines[i].trim();
-		if (line.length === 0) continue;
-		try {
-			out.push({
-				index: i,
-				value: JSON.parse(line)
-			});
-		} catch (err) {
-			throw new Error(`fromRunRecordDir: file '${file}' line ${i + 1} is not valid JSON: ${err instanceof Error ? err.message : String(err)}`);
-		}
-	}
-	return out;
-}
-/** Sorted file list under a directory, filtered by `include`. Sorted so the
-*  resulting `RunRecord` order — and any downstream fingerprint — is stable
-*  across filesystems. */
-async function collectFiles(dir, include, recursive) {
-	const entries = await readdir(dir, { withFileTypes: true });
-	const files = [];
-	const subdirs = [];
-	for (const entry of entries) {
-		if (entry.isDirectory()) {
-			if (recursive) subdirs.push(join(dir, entry.name));
-			continue;
-		}
-		if (include(entry.name)) files.push(join(dir, entry.name));
-	}
-	files.sort();
-	subdirs.sort();
-	for (const sub of subdirs) files.push(...await collectFiles(sub, include, recursive));
-	return files;
-}
-//#endregion
-//#region src/contract/eval-reporting-suite.ts
-/**
-* # `evalReportingSuite` — one call from runs (or a run dir) to `analysis.json`.
-*
-* A thin wrapper over the analysis primitive (`analyzeRuns`) and the on-disk
-* intake adapter (`fromRunRecordDir`). It does NOT reimplement any statistics,
-* distributions, or clustering — it resolves the input into validated
-* `RunRecord[]`, calls `analyzeRuns` with the options you'd pass it directly,
-* wraps the result in a small provenance envelope, and (optionally) writes a
-* single `analysis.json` artifact.
-*
-* ```ts
-* // From a directory of run files, write ./runs/analysis.json:
-* const suite = await evalReportingSuite('./runs', { write: true })
-* // From records already in memory, no write:
-* const suite = await evalReportingSuite(records, { analyze: { decisionThreshold: 0.03 } })
-* suite.report // the InsightReport — distributions, paired lift, findings rollup
-* ```
-*/
-const ANALYSIS_ARTIFACT = "analysis.json";
-/**
-* Resolve runs (or a run dir/file), run `analyzeRuns`, and optionally persist a
-* single `analysis.json`. The only analysis logic lives in `analyzeRuns`; this
-* function is composition + I/O.
-*/
-async function evalReportingSuite(input, options = {}) {
-	const fromPath = typeof input === "string";
-	let runs;
-	let files = [];
-	let rejected = [];
-	if (fromPath) {
-		const loaded = await fromRunRecordDir(input, options.load);
-		runs = loaded.runs;
-		files = loaded.files;
-		rejected = loaded.rejected;
-	} else runs = input;
-	if (runs.length === 0) throw new Error(fromPath ? `evalReportingSuite: no RunRecords found at '${input}'` : "evalReportingSuite: no RunRecords to analyze");
-	const result = {
-		report: await analyzeRuns({
-			...options.analyze,
-			runs
-		}),
-		provenance: {
-			generatedAt: (/* @__PURE__ */ new Date()).toISOString(),
-			runCount: runs.length,
-			sourcePath: fromPath ? input : null,
-			files,
-			rejected
-		},
-		writtenTo: null
-	};
-	const target = resolveWriteTarget(options.write, fromPath ? input : null);
-	if (target) {
-		await mkdir(dirname(target), { recursive: true });
-		await writeFile(target, `${JSON.stringify(result, null, 2)}\n`, "utf8");
-		result.writtenTo = target;
-	}
-	return result;
-}
-/** Resolve where (if anywhere) to write `analysis.json`. Returns null when
-*  writing is disabled. Throws on `write: true` with in-memory input — there is
-*  no directory to anchor the artifact to, and silently inventing `cwd` would
-*  scatter files. */
-function resolveWriteTarget(write, sourcePath) {
-	if (!write) return null;
-	if (typeof write === "string") return write.endsWith("/") || !write.endsWith(".json") && !write.endsWith(".jsonl") ? join(write, ANALYSIS_ARTIFACT) : write;
-	if (sourcePath === null) throw new Error("evalReportingSuite: write:true needs a source path to anchor analysis.json — pass an explicit output path when analyzing in-memory records");
-	return sourcePath.endsWith(".json") || sourcePath.endsWith(".jsonl") ? join(dirname(sourcePath), ANALYSIS_ARTIFACT) : join(sourcePath, ANALYSIS_ARTIFACT);
-}
-//#endregion
 //#region src/contract/intake/agent-trace.ts
 function rangeLines(r) {
 	return Math.max(0, r.end_line - r.start_line + 1);
@@ -1385,9 +1210,6 @@ function partitionRunsByAuthoringModel(runs, index) {
 }
 //#endregion
 //#region src/contract/intake/code-agent-fields.ts
-function isRecord(value) {
-	return value !== null && typeof value === "object" && !Array.isArray(value);
-}
 function record(value) {
 	return isRecord(value) ? value : null;
 }
@@ -2200,7 +2022,8 @@ function emptyMetrics(entries) {
 }
 function penalizeErrors(base, metrics) {
 	const operations = Math.max(1, metrics.toolCalls + metrics.patchAttempts);
-	return clamp01(base * (1 - .5 * Math.min(1, (metrics.toolErrors + metrics.patchFailures) / operations)));
+	const errorRate = Math.min(1, (metrics.toolErrors + metrics.patchFailures) / operations);
+	return clamp01(base * (1 - .5 * errorRate));
 }
 function emptyTokenTotals() {
 	return {
@@ -2291,10 +2114,6 @@ function resolveSessionCost(explicit, metrics, model) {
 		}
 	};
 }
-function clamp01(value) {
-	if (!Number.isFinite(value)) return 0;
-	return Math.max(0, Math.min(1, value));
-}
 //#endregion
 //#region src/contract/intake/code-agent-codex-metrics.ts
 function codexMetrics(entries) {
@@ -2740,10 +2559,6 @@ function fromKimiCodeSession(options) {
 function fromPiSession(options) {
 	return fromCodeAgentSession("pi", options);
 }
-/** The graph IR a Pi graph run exports (`nodes[].ir`), which is not a Pi session. */
-function fromPiGraphSession(options) {
-	return fromCodeAgentSession("pi-graph", options);
-}
 function fromCodeAgentSession(source, options) {
 	const entries = options.entries.filter(isRecord);
 	if (entries.length === 0) return {
@@ -2974,197 +2789,6 @@ function finiteScore(value) {
 	return value !== void 0 && Number.isFinite(value) ? clamp01(value) : void 0;
 }
 //#endregion
-//#region src/contract/intake/code-agent-store.ts
-const SESSION_SUFFIX = ".jsonl";
-const REJECTION_SAMPLE_LIMIT = 8;
-/**
-* Read every session this run captured under `scope.root` and project them.
-*
-* A store directory that does not exist, or that holds no session the scope attributes, returns
-* `status: 'unavailable'`. That is not an error: a run whose harness store was never synced back
-* has no evidence, and saying so is the correct answer.
-*/
-async function observeCodeAgentStore(scope) {
-	if (!isAbsolute(scope.root)) throw new TypeError(`observeCodeAgentStore: scope.root must be an absolute path to this run's own store, received ${JSON.stringify(scope.root)}`);
-	const workspaceRoot = scope.workspaceRoot === void 0 ? void 0 : resolve(scope.workspaceRoot);
-	const paths = await listSessionFiles(resolve(scope.root));
-	const sessions = [];
-	const byReason = /* @__PURE__ */ new Map();
-	const sample = [];
-	const reject = (path, reason) => {
-		byReason.set(reason, (byReason.get(reason) ?? 0) + 1);
-		if (sample.length < REJECTION_SAMPLE_LIMIT) sample.push({
-			path,
-			reason
-		});
-	};
-	for (const path of paths) {
-		let entries;
-		try {
-			entries = (await parseCodeAgentJsonlFile(path)).entries;
-		} catch {
-			reject(path, "unreadable");
-			continue;
-		}
-		if (entries.length === 0) {
-			reject(path, "no-entries");
-			continue;
-		}
-		const workspace = sessionWorkspace(entries);
-		if (workspaceRoot !== void 0) {
-			if (workspace === void 0) {
-				reject(path, "no-workspace-recorded");
-				continue;
-			}
-			if (!isSameOrBelow(workspaceRoot, workspace)) {
-				reject(path, "outside-workspace");
-				continue;
-			}
-		}
-		const startedAtMs = sessionStartedAtMs(entries);
-		if (scope.window !== void 0) {
-			if (startedAtMs === void 0) {
-				reject(path, "no-start-recorded");
-				continue;
-			}
-			if (startedAtMs < scope.window.startedAtMs) {
-				reject(path, "before-run");
-				continue;
-			}
-			if (scope.window.finishedAtMs !== void 0 && startedAtMs > scope.window.finishedAtMs) {
-				reject(path, "after-run");
-				continue;
-			}
-		}
-		const observation = observeCodeAgentSession({
-			source: scope.source,
-			entries,
-			sourcePath: path
-		});
-		sessions.push({
-			path,
-			sessionId: observation.sessionId,
-			...workspace === void 0 ? {} : { workspace },
-			...startedAtMs === void 0 ? {} : { startedAtMs },
-			observation
-		});
-	}
-	const rejections = {
-		total: [...byReason.values()].reduce((sum, count) => sum + count, 0),
-		byReason: Object.fromEntries(byReason),
-		sample
-	};
-	if (sessions.length === 0) return {
-		status: "unavailable",
-		scope,
-		filesScanned: paths.length,
-		reason: unavailableReason(paths.length, rejections),
-		rejections
-	};
-	const bySurface = /* @__PURE__ */ new Map();
-	let actions = 0;
-	let subagentActions = 0;
-	let sessionsWithSubagentActions = 0;
-	for (const session of sessions) {
-		let sessionSubagents = 0;
-		for (const action of session.observation.actions) {
-			actions += 1;
-			bySurface.set(action.surface, (bySurface.get(action.surface) ?? 0) + 1);
-			if (action.surface === "subagent") sessionSubagents += 1;
-		}
-		subagentActions += sessionSubagents;
-		if (sessionSubagents > 0) sessionsWithSubagentActions += 1;
-	}
-	return {
-		status: "observed",
-		scope,
-		filesScanned: paths.length,
-		sessions,
-		actions,
-		subagentActions,
-		sessionsWithSubagentActions,
-		bySurface: Object.fromEntries(bySurface),
-		rejections
-	};
-}
-function unavailableReason(filesScanned, rejections) {
-	if (filesScanned === 0) return "the run captured no session files in this store";
-	return `no session in this store belongs to the run: ${filesScanned} file(s) scanned, all rejected (${Object.entries(rejections.byReason).map(([reason, count]) => `${count} ${reason}`).sort().join(", ")})`;
-}
-async function listSessionFiles(root) {
-	const found = [];
-	const walk = async (dir) => {
-		let entries;
-		try {
-			entries = await readdir(dir, { withFileTypes: true });
-		} catch {
-			return;
-		}
-		for (const entry of entries) {
-			const path = join(dir, entry.name);
-			if (entry.isDirectory()) await walk(path);
-			else if (entry.isFile() && entry.name.endsWith(SESSION_SUFFIX)) found.push(path);
-		}
-	};
-	await walk(root);
-	if (found.length === 0 && await isSessionFile(root)) found.push(root);
-	return found.sort();
-}
-async function isSessionFile(path) {
-	if (!path.endsWith(SESSION_SUFFIX)) return false;
-	try {
-		return (await stat(path)).isFile();
-	} catch {
-		return false;
-	}
-}
-/**
-* The working directory the session RECORDED, from the first entry that states one.
-*
-* codex writes it on `session_meta.payload.cwd`; claude-code writes `cwd` on every line. Neither is
-* inferred from the file's own location: where a captured file was copied to says nothing about
-* where the agent ran.
-*/
-function sessionWorkspace(entries) {
-	for (const entry of entries) {
-		const record = plainRecord(entry);
-		if (record === void 0) continue;
-		const direct = stringOf(record.cwd);
-		if (direct !== void 0) return direct;
-		const nested = stringOf(plainRecord(record.payload)?.cwd);
-		if (nested !== void 0) return nested;
-	}
-}
-/** When the session itself says it started, from the first entry carrying a parseable timestamp. */
-function sessionStartedAtMs(entries) {
-	for (const entry of entries) {
-		const record = plainRecord(entry);
-		if (record === void 0) continue;
-		const payload = plainRecord(record.payload);
-		for (const candidate of [payload?.timestamp, record.timestamp]) {
-			const text = stringOf(candidate);
-			if (text === void 0) continue;
-			const parsed = Date.parse(text);
-			if (Number.isFinite(parsed)) return parsed;
-		}
-		const epochMs = numberOf(record.timestampMs) ?? numberOf(payload?.timestampMs);
-		if (epochMs !== void 0) return epochMs;
-	}
-}
-function isSameOrBelow(root, candidate) {
-	const rel = relative(root, resolve(candidate));
-	return rel === "" || rel !== ".." && !rel.startsWith(`..${sep}`) && !isAbsolute(rel);
-}
-function plainRecord(value) {
-	return value !== null && typeof value === "object" && !Array.isArray(value) ? value : void 0;
-}
-function stringOf(value) {
-	return typeof value === "string" && value.length > 0 ? value : void 0;
-}
-function numberOf(value) {
-	return typeof value === "number" && Number.isFinite(value) ? value : void 0;
-}
-//#endregion
 //#region src/contract/intake/feedback-table.ts
 function fromFeedbackTable(opts) {
 	const { ratings, meta = [], scale, emitRaterScores = true } = opts;
@@ -3244,6 +2868,6 @@ function fromFeedbackTable(opts) {
 	};
 }
 //#endregion
-export { FileSystemOutcomeStore, InMemoryOutcomeStore, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, SelfImproveRunError, analyzeRuns, buildDefaultAnalystRegistry, buildEvidenceVector, campaignSplitDigest, compareOptimizationMethods, composeGate, createChatClient, createReferenceEquivalenceJudge, defaultProductionGate, defineAgentEval, evalReportingSuite, evaluatePairedMeasurements, externalTextOptimizationMethod, fromClaudeCodeSession, fromCodexSession, fromFeedbackTable, fromKimiCodeSession, fromOpenCodeSession, fromOtelSpans, fromPiGraphSession, fromPiSession, fromRunRecordDir, fsCampaignStorage, gepaOptimizationMethod, heldOutGate, inMemoryCampaignStorage, llmJudge, makeProposalFinding, measuredComparisonFromAgentProfileImprovementExperiment, measuredComparisonFromCandidateExperiment, observeCodeAgentSession, observeCodeAgentStore, paretoPolicy, paretoSignificanceGate, parseAgentTrace, parseCodeAgentJsonl, parseCodeAgentJsonlFile, partitionRunsByAuthoringModel, runAgentProfileImprovementExperiment, runCampaign, runCandidateExperiment, runEval, runImprovementLoop, runReferenceEquivalenceJudge, sealAgentProfileImprovementExperiment, sealAgentProfileImprovementSuite, sealAgentProfileImprovementTask, sealCandidateBenchmarkSuite, sealCandidateBenchmarkTask, sealCandidateExperiment, selfImprove, skillOptOptimizationMethod, streamCodeAgentJsonlFile, summarizeExecution, transientDispatchFailure, verifyAgentProfileImprovementExperimentComparison, verifyCandidateExperiment, verifyCandidateExperimentComparison };
+export { FileSystemOutcomeStore, InMemoryOutcomeStore, SelfImproveRunError, analyzeRuns, buildDefaultAnalystRegistry, buildEvidenceVector, campaignSplitDigest, compareOptimizationMethods, composeGate, createChatClient, defaultProductionGate, defineAgentEval, evaluatePairedMeasurements, externalTextOptimizationMethod, fromClaudeCodeSession, fromCodexSession, fromFeedbackTable, fromKimiCodeSession, fromOpenCodeSession, fromOtelSpans, fromPiSession, fsCampaignStorage, gepaOptimizationMethod, heldOutGate, inMemoryCampaignStorage, llmJudge, makeProposalFinding, measuredComparisonFromAgentProfileImprovementExperiment, measuredComparisonFromCandidateExperiment, observeCodeAgentSession, paretoPolicy, paretoSignificanceGate, parseAgentTrace, parseCodeAgentJsonl, parseCodeAgentJsonlFile, partitionRunsByAuthoringModel, runAgentProfileImprovementExperiment, runCampaign, runCandidateExperiment, runEval, runImprovementLoop, sealAgentProfileImprovementExperiment, sealAgentProfileImprovementSuite, sealAgentProfileImprovementTask, sealCandidateBenchmarkSuite, sealCandidateBenchmarkTask, sealCandidateExperiment, selfImprove, skillOptOptimizationMethod, streamCodeAgentJsonlFile, summarizeExecution, transientDispatchFailure, verifyAgentProfileImprovementExperimentComparison, verifyCandidateExperiment, verifyCandidateExperimentComparison };
 
 //# sourceMappingURL=index.js.map