@tangle-network/agent-eval 0.123.3 → 0.123.5

package/CHANGELOG.md

@@ -8,6 +8,8 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
 
 ### Added
 
+- `selfImprove({ budget: { candidateConcurrency } })` exposes the existing `runOptimization()` control for scoring candidate campaigns in parallel; it remains opt-in and defaults to one candidate campaign at a time.
+- `llmPolicyEditProposer()` and `projectPolicyEditHistory()` accept `scenarioOrder: 'input'` when controlled comparisons must preserve first-occurrence caller order; ranked evidence selection remains the default.
 - `callLlmJson()` accepts `jsonPayloadMode: 'exact'` when callers must reject fenced, prose-wrapped, or multi-root responses instead of extracting a JSON value.
 - `llmPolicyEditProposer({ redactCurrentSurfaceForModel })` can remove credentials and unrelated fields from the current surface sent to the model while applying validated edits to the complete original surface.
 - `CostLedger.listPending()` exposes immutable pending paid calls and distinguishes calls that are active, late after cancellation, or interrupted by a prior process so durable workflows can reconcile exact reservations before resuming.

package/dist/benchmarks/index.js

@@ -17,7 +17,7 @@ import {
   runBenchmarkAdapter,
   summarizeBenchmarkCampaign
 } from "../chunk-JKDNAOF5.js";
-import "../chunk-22VO7T2I.js";
+import "../chunk-LT4J7ULK.js";
 import "../chunk-SUN7QLPB.js";
 import "../chunk-D5JZ7UDZ.js";
 import "../chunk-MHPEGJHC.js";

package/dist/campaign/index.d.ts

@@ -5702,6 +5702,36 @@ interface HaloProposerOptions {
 /** Wrap the real halo-engine CLI as a SurfaceProposer (prompt-tier). */
 declare function haloProposer(opts: HaloProposerOptions): SurfaceProposer;
 
+/** One measured scenario row eligible for PolicyEdit author context. */
+interface PolicyEditAuthorScenarioRow {
+    scenarioId: string;
+    composite: number;
+}
+type PolicyEditAuthorScenarioOrder = 'ranked' | 'input';
+interface SelectPolicyEditAuthorRowsOptions {
+    /** Maximum returned rows. Must be a positive safe integer. */
+    limit: number;
+    /** Optional score to compare against, keyed by scenario ID. */
+    referenceByScenario?: ReadonlyMap<string, number>;
+    /** Ranked evidence selection by default; `input` preserves first-occurrence caller order. */
+    scenarioOrder?: PolicyEditAuthorScenarioOrder;
+}
+interface SerializedJsonBudget {
+    json: string;
+    actualChars: number;
+    maxChars: number;
+}
+/**
+ * Select a bounded, deterministic evidence slice for a PolicyEdit author.
+ *
+ * Rows are deduplicated by scenario ID, keeping the first measured row. Ranked
+ * selection (the default) interleaves hardest score, largest regression, and
+ * largest improvement. Input selection retains first-occurrence caller order.
+ */
+declare function selectPolicyEditAuthorRows<T extends PolicyEditAuthorScenarioRow>(rows: readonly T[], options: SelectPolicyEditAuthorRowsOptions): T[];
+/** Serialize once and fail before dispatch when author context exceeds its budget. */
+declare function assertPolicyEditAuthorContextBudget(value: unknown, maxChars: number): SerializedJsonBudget;
+
 declare const JSON_POLICY_EDIT_TARGET_SURFACES: readonly ["prompt", "tool-contract", "runtime-config", "memory", "agent-profile"];
 type JsonPolicyEditTargetSurface = (typeof JSON_POLICY_EDIT_TARGET_SURFACES)[number];
 declare const DEFAULT_POLICY_EDIT_HISTORY_LIMITS: Readonly<{
@@ -5744,8 +5774,10 @@ interface PolicyEditHistoryProjectionOptions {
     maxGenerations?: number;
     /** Number of candidates retained per generation. Default: 16. */
     maxCandidatesPerGeneration?: number;
-    /** Scored tasks retained per candidate/outcome after deterministic extreme selection. */
+    /** Maximum scored tasks retained per candidate/outcome. */
     maxScenariosPerCandidate?: number;
+    /** Ranked evidence selection by default; `input` preserves first-occurrence caller order. */
+    scenarioOrder?: PolicyEditAuthorScenarioOrder;
     /** Optional pseudonymizer applied before scenario IDs enter author text. */
     scenarioIdTransform?: (scenarioId: string) => string;
     /** Objectives used to compute forecast residuals from measured composite deltas. */
@@ -5841,8 +5873,10 @@ interface LlmPolicyEditProposerOptions {
     maxHistoryGenerations?: number;
     /** Candidates retained per admitted generation. Default: 16. */
     maxHistoryCandidatesPerGeneration?: number;
-    /** Scored tasks retained per candidate/outcome after deterministic extreme selection. */
+    /** Maximum scored tasks retained per candidate/outcome. */
     maxScenariosPerCandidate?: number;
+    /** Ranked evidence selection by default; `input` preserves first-occurrence caller order. */
+    scenarioOrder?: PolicyEditAuthorScenarioOrder;
     /** Evidence-bearing findings retained after deterministic severity/confidence ordering. */
     maxFindings?: number;
     /** Hard character limit over system + schema + serialized author context. */
@@ -5948,33 +5982,6 @@ interface PolicyEditProposerOptions {
  */
 declare function policyEditProposer(opts?: PolicyEditProposerOptions): SurfaceProposer;
 
-/** One measured scenario row eligible for PolicyEdit author context. */
-interface PolicyEditAuthorScenarioRow {
-    scenarioId: string;
-    composite: number;
-}
-interface SelectPolicyEditAuthorRowsOptions {
-    /** Maximum returned rows. Must be a positive safe integer. */
-    limit: number;
-    /** Optional score to compare against, keyed by scenario ID. */
-    referenceByScenario?: ReadonlyMap<string, number>;
-}
-interface SerializedJsonBudget {
-    json: string;
-    actualChars: number;
-    maxChars: number;
-}
-/**
- * Select a bounded, deterministic evidence slice for a PolicyEdit author.
- *
- * Rows are deduplicated by scenario ID, keeping the first measured row. The
- * result then interleaves three ranked views: hardest score, largest regression,
- * and largest improvement. A row selected by multiple views appears once.
- */
-declare function selectPolicyEditAuthorRows<T extends PolicyEditAuthorScenarioRow>(rows: readonly T[], options: SelectPolicyEditAuthorRowsOptions): T[];
-/** Serialize once and fail before dispatch when author context exceeds its budget. */
-declare function assertPolicyEditAuthorContextBudget(value: unknown, maxChars: number): SerializedJsonBudget;
-
 /**
  * Typed Ax output for analyst findings.
  *
@@ -7678,4 +7685,4 @@ declare function verifyCodeSurface(surface: CodeSurface, worktreeDir?: string):
  *  identity against the checkout at `worktreeRef`. */
 declare function resolveWorktreePath(surface: CodeSurface, worktreeDir?: string): string;
 
-export { type AcceptedEdit, type AceProposerOptions, type AnalystArtifact, type AnalystScenario, type AnalyzeCrossSurfaceInteractionsInput, type AnalyzeOtlpTraceFileOptions, type ApplySkillPatchResult, type AxisEvidence, type AxisVerdict, type BuildAnalystSurfaceDispatchOptions, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type BuiltinOptimizationMethodConfig, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellResult, type CampaignCostMeter, type CampaignResult, type CampaignRunPlan, type CampaignRunPlanCell, type CampaignScenarioIdentity, type CampaignStorage, type CampaignTokenUsage, type CampaignTraceWriter, type CodeSurface, type CodeSurfaceVerification, type CompareOptimizationMethodsOptions, type ComparisonCost, type CompositeProposerOptions, type CostLedgerHandle, 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_POLICY_EDIT_HISTORY_LIMITS, type DefaultProductionGateOptions, type DimensionRegression, type DiscriminationScore, type DispatchContext, type DispatchFn, type EmitLoopProvenanceArgs, type EmitLoopProvenanceResult, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type EvidenceVector, type EvolutionaryProposerOptions, type FailureModeRecallJudgeOptions, type FapoAttributionSignals, type FapoFailureCluster, type FapoOptimizationLevel, type FapoOptimizationMethodConfig, type FapoProposerOptions, type FapoReviewInput, type FapoReviewIssue, type FapoReviewResult, type FapoScopeContract, FileSearchLedger, FsLabeledScenarioStore, type FsLabeledScenarioStoreOptions, type Gate, type GateContext, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaProposerConstraints, type GepaProposerOptions, type GitWorktreeAdapterOptions, type Governor, type GovernorContext, type GovernorOp, type HaloProposerOptions, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type HeuristicGovernorOptions, type JsonPolicyEditTargetSurface, type JsonPrimitive, type JsonValue, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, LabeledScenarioStoreError, type LabeledScenarioWrite, Lineage, type LineageEdge, type LineageGraph, type LineageNode, type LineageNodeInput, type LineageStore, LineageStoreConflictError, type LlmJudgeDimension, type LlmJudgeOptions, type LlmPolicyEditProposerOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceArgsFromResult, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceEvidence, type LoopProvenanceRecord, type MemoryCurationProposerOptions, type MutableSurface, type Mutator, type NeutralizationGateOptions, type ObjectiveSource, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodInput, type OptimizationMethodPairwise, type OptimizationMethodResult, type OptimizationMethodRunOptions, type OptimizationMethodScore, type OptimizationProposer, type OptimizerConfig, POLICY_EDIT_CANDIDATE_RECORD_SCHEMA, type PairedHoldout, type ParameterCandidate, type ParameterChange, type ParameterSweepProposerOptions, type ParetoParent, type ParetoSignificanceGateOptions, type PendingCostCallView, type PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PolicyEditAuthorScenarioRow, type PolicyEditCandidateRecord, type PolicyEditCandidateSummary, type PolicyEditFindingInput, type PolicyEditFindingSource, type PolicyEditHistoryCandidateContext, type PolicyEditHistoryGenerationContext, type PolicyEditHistoryProjectionOptions, type PolicyEditObjective, type PolicyEditOutcomeContext, type PolicyEditProposerOptions, type PowerPreflight, type PowerPreflightOptions, type PremeasuredOptimizationBaseline, type ProfileDispatchFn, ProfileMatrixError, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposalTrackContext, type ProposeContext, type ProposePatchesArgs, type ProposedCandidate, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RejectedEdit, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunLineageLoopOptions, type RunLineageLoopResult, type RunLineageLoopSeed, type RunLineageOptions, type RunLineageResult, type RunLineageSeed, type RunLineageStepResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunSkillOptOptions, type RunSkillOptResult, SEARCH_LEDGER_SCHEMA, type Scenario, type ScenarioAggregate, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type ScoredSurfaceOutcome, type SearchAccountingAudit, type SearchArtifactRef, type SearchAttemptAccounting, type SearchCandidateDecidedEvent, type SearchCandidateLineage, type SearchCandidateRegisteredEvent, type SearchCandidateSlot, type SearchCandidateSlotClosedEvent, type SearchCandidateSurface, type SearchCompletedEvent, type SearchCostAccounting, type SearchFailureReason, type SearchLedger, type SearchLedgerAppendResult, SearchLedgerConflictError, type SearchLedgerEntry, SearchLedgerError, type SearchLedgerEvent, type SearchLedgerHash, SearchLedgerIntegrityError, type SearchLedgerReplay, type SearchModelIdentity, type SearchOperationKind, type SearchOperationRecordedEvent, type SearchPlan, type SearchPlannedEvent, type SearchPlannedOperation, type SearchPlannedTask, type SearchSourceRef, type SearchSurfaceEffect, type SearchSurfaceEvidence, type SearchSurfaceKind, type SearchTaskAttemptedEvent, type SearchTaskOutcome, type SearchTokenAccounting, type SelectPolicyEditAuthorRowsOptions, type SequentialDecideFn, type SequentialDecideOptions, type SequentialDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SerializedJsonBudget, type SessionScript, type SingleRunLock, type SingleRunLockOptions, type SkillOptEpochRecord, type SkillOptEvidence, type SkillOptProposer, type SkillOptProposerOptions, type SkillPatch, type SkillPatchOp, SkillPatchParseError, type SkillPatchRejection, type SurfaceProposer, type SurfaceScore, type TraceAnalystPriorFindings, type TraceAnalystProposerOptions, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, WorktreeAdapterError, aceProposer, acquireSingleRunLock, analyzeCrossSurfaceInteractions, analyzeOtlpTraceFile, applySkillPatch, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertPolicyEditAuthorContextBudget, buildAnalystSurfaceDispatch, buildEvidenceVector, buildLoopProvenanceRecord, callbackGovernor, campaignBreakdown, campaignLineageStore, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, compareOptimizationMethods, composeGate, compositeProposer, costFromLedgerSummary, countSentenceEdits, createReferenceEquivalenceJudge, createRunCostLedger, defaultProductionGate, detectScale, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, evolutionaryProposer, extractFapoAttributionSignals, extractH2Sections, failureModeRecallJudge, fapoEscalationMethod, fapoProposer, fsCampaignStorage, fsLineageStore, gepaParetoMethod, gepaProposer, gepaReflectionMethod, gitWorktreeAdapter, haloProposer, heldOutGate, heldoutSignificance, heuristicGovernor, inMemoryCampaignStorage, isProposedCandidate, isTransientTransportFailure, labelTrustRank, lineageNodeId, llmJudge, llmPolicyEditProposer, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, memLineageStore, memoryCurationProposer, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, pairHoldout, parameterSweepProposer, paretoPolicy, paretoSignificanceGate, parseSkillPatchResponse, patchEditCount, planCampaignRun, planEvalFixtureRun, policyEditProposer, powerPreflight, projectPolicyEditHistory, provenanceRecordPath, provenanceSpansPath, renderScoreboardMarkdown, renderSurfaceDiff, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runLineage, runLineageLoop, runOptimization, runProfileMatrix, runSkillOpt, scoreDiscrimination, scoreUserStory, scoreboardSummary, selectDiscriminative, selectPolicyEditAuthorRows, sequentialDecide, sequentialPairedGate, skillOptMethod, skillOptProposer, surfaceContentHash, surfaceHash, tangleTracesRoot, traceAnalystProposer, userStoryScoreboard, validatePolicyEditCandidateRecord, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord };
+export { type AcceptedEdit, type AceProposerOptions, type AnalystArtifact, type AnalystScenario, type AnalyzeCrossSurfaceInteractionsInput, type AnalyzeOtlpTraceFileOptions, type ApplySkillPatchResult, type AxisEvidence, type AxisVerdict, type BuildAnalystSurfaceDispatchOptions, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type BuiltinOptimizationMethodConfig, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellResult, type CampaignCostMeter, type CampaignResult, type CampaignRunPlan, type CampaignRunPlanCell, type CampaignScenarioIdentity, type CampaignStorage, type CampaignTokenUsage, type CampaignTraceWriter, type CodeSurface, type CodeSurfaceVerification, type CompareOptimizationMethodsOptions, type ComparisonCost, type CompositeProposerOptions, type CostLedgerHandle, 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_POLICY_EDIT_HISTORY_LIMITS, type DefaultProductionGateOptions, type DimensionRegression, type DiscriminationScore, type DispatchContext, type DispatchFn, type EmitLoopProvenanceArgs, type EmitLoopProvenanceResult, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type EvidenceVector, type EvolutionaryProposerOptions, type FailureModeRecallJudgeOptions, type FapoAttributionSignals, type FapoFailureCluster, type FapoOptimizationLevel, type FapoOptimizationMethodConfig, type FapoProposerOptions, type FapoReviewInput, type FapoReviewIssue, type FapoReviewResult, type FapoScopeContract, FileSearchLedger, FsLabeledScenarioStore, type FsLabeledScenarioStoreOptions, type Gate, type GateContext, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaProposerConstraints, type GepaProposerOptions, type GitWorktreeAdapterOptions, type Governor, type GovernorContext, type GovernorOp, type HaloProposerOptions, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type HeuristicGovernorOptions, type JsonPolicyEditTargetSurface, type JsonPrimitive, type JsonValue, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, LabeledScenarioStoreError, type LabeledScenarioWrite, Lineage, type LineageEdge, type LineageGraph, type LineageNode, type LineageNodeInput, type LineageStore, LineageStoreConflictError, type LlmJudgeDimension, type LlmJudgeOptions, type LlmPolicyEditProposerOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceArgsFromResult, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceEvidence, type LoopProvenanceRecord, type MemoryCurationProposerOptions, type MutableSurface, type Mutator, type NeutralizationGateOptions, type ObjectiveSource, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodInput, type OptimizationMethodPairwise, type OptimizationMethodResult, type OptimizationMethodRunOptions, type OptimizationMethodScore, type OptimizationProposer, type OptimizerConfig, POLICY_EDIT_CANDIDATE_RECORD_SCHEMA, type PairedHoldout, type ParameterCandidate, type ParameterChange, type ParameterSweepProposerOptions, type ParetoParent, type ParetoSignificanceGateOptions, type PendingCostCallView, type PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PolicyEditAuthorScenarioOrder, type PolicyEditAuthorScenarioRow, type PolicyEditCandidateRecord, type PolicyEditCandidateSummary, type PolicyEditFindingInput, type PolicyEditFindingSource, type PolicyEditHistoryCandidateContext, type PolicyEditHistoryGenerationContext, type PolicyEditHistoryProjectionOptions, type PolicyEditObjective, type PolicyEditOutcomeContext, type PolicyEditProposerOptions, type PowerPreflight, type PowerPreflightOptions, type PremeasuredOptimizationBaseline, type ProfileDispatchFn, ProfileMatrixError, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposalTrackContext, type ProposeContext, type ProposePatchesArgs, type ProposedCandidate, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RejectedEdit, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunLineageLoopOptions, type RunLineageLoopResult, type RunLineageLoopSeed, type RunLineageOptions, type RunLineageResult, type RunLineageSeed, type RunLineageStepResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunSkillOptOptions, type RunSkillOptResult, SEARCH_LEDGER_SCHEMA, type Scenario, type ScenarioAggregate, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type ScoredSurfaceOutcome, type SearchAccountingAudit, type SearchArtifactRef, type SearchAttemptAccounting, type SearchCandidateDecidedEvent, type SearchCandidateLineage, type SearchCandidateRegisteredEvent, type SearchCandidateSlot, type SearchCandidateSlotClosedEvent, type SearchCandidateSurface, type SearchCompletedEvent, type SearchCostAccounting, type SearchFailureReason, type SearchLedger, type SearchLedgerAppendResult, SearchLedgerConflictError, type SearchLedgerEntry, SearchLedgerError, type SearchLedgerEvent, type SearchLedgerHash, SearchLedgerIntegrityError, type SearchLedgerReplay, type SearchModelIdentity, type SearchOperationKind, type SearchOperationRecordedEvent, type SearchPlan, type SearchPlannedEvent, type SearchPlannedOperation, type SearchPlannedTask, type SearchSourceRef, type SearchSurfaceEffect, type SearchSurfaceEvidence, type SearchSurfaceKind, type SearchTaskAttemptedEvent, type SearchTaskOutcome, type SearchTokenAccounting, type SelectPolicyEditAuthorRowsOptions, type SequentialDecideFn, type SequentialDecideOptions, type SequentialDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SerializedJsonBudget, type SessionScript, type SingleRunLock, type SingleRunLockOptions, type SkillOptEpochRecord, type SkillOptEvidence, type SkillOptProposer, type SkillOptProposerOptions, type SkillPatch, type SkillPatchOp, SkillPatchParseError, type SkillPatchRejection, type SurfaceProposer, type SurfaceScore, type TraceAnalystPriorFindings, type TraceAnalystProposerOptions, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, WorktreeAdapterError, aceProposer, acquireSingleRunLock, analyzeCrossSurfaceInteractions, analyzeOtlpTraceFile, applySkillPatch, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertPolicyEditAuthorContextBudget, buildAnalystSurfaceDispatch, buildEvidenceVector, buildLoopProvenanceRecord, callbackGovernor, campaignBreakdown, campaignLineageStore, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, compareOptimizationMethods, composeGate, compositeProposer, costFromLedgerSummary, countSentenceEdits, createReferenceEquivalenceJudge, createRunCostLedger, defaultProductionGate, detectScale, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, evolutionaryProposer, extractFapoAttributionSignals, extractH2Sections, failureModeRecallJudge, fapoEscalationMethod, fapoProposer, fsCampaignStorage, fsLineageStore, gepaParetoMethod, gepaProposer, gepaReflectionMethod, gitWorktreeAdapter, haloProposer, heldOutGate, heldoutSignificance, heuristicGovernor, inMemoryCampaignStorage, isProposedCandidate, isTransientTransportFailure, labelTrustRank, lineageNodeId, llmJudge, llmPolicyEditProposer, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, memLineageStore, memoryCurationProposer, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, pairHoldout, parameterSweepProposer, paretoPolicy, paretoSignificanceGate, parseSkillPatchResponse, patchEditCount, planCampaignRun, planEvalFixtureRun, policyEditProposer, powerPreflight, projectPolicyEditHistory, provenanceRecordPath, provenanceSpansPath, renderScoreboardMarkdown, renderSurfaceDiff, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runLineage, runLineageLoop, runOptimization, runProfileMatrix, runSkillOpt, scoreDiscrimination, scoreUserStory, scoreboardSummary, selectDiscriminative, selectPolicyEditAuthorRows, sequentialDecide, sequentialPairedGate, skillOptMethod, skillOptProposer, surfaceContentHash, surfaceHash, tangleTracesRoot, traceAnalystProposer, userStoryScoreboard, validatePolicyEditCandidateRecord, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord };

package/dist/campaign/index.js

@@ -70,7 +70,7 @@ import {
   userStoryScoreboard,
   validateSearchLedgerEvent,
   verifyCodeSurface
-} from "../chunk-22VO7T2I.js";
+} from "../chunk-LT4J7ULK.js";
 import {
   assertCodeSurfaceIdentity,
   buildEvidenceVector,

package/dist/{chunk-22VO7T2I.js → chunk-LT4J7ULK.js}

@@ -5637,6 +5637,10 @@ function coerceCandidateSurface(surface) {
 // src/campaign/proposers/policy-edit-author-context.ts
 function selectPolicyEditAuthorRows(rows, options) {
   assertPositiveSafeInteger2(options.limit, "limit");
+  const scenarioOrder = options.scenarioOrder ?? "ranked";
+  if (scenarioOrder !== "ranked" && scenarioOrder !== "input") {
+    throw new Error("selectPolicyEditAuthorRows: scenarioOrder must be 'ranked' or 'input'");
+  }
   const unique2 = /* @__PURE__ */ new Map();
   for (const row of rows) {
     if (!row.scenarioId || row.scenarioId.trim() !== row.scenarioId) {
@@ -5659,6 +5663,9 @@ function selectPolicyEditAuthorRows(rows, options) {
       delta: reference === void 0 ? null : row.composite - reference
     });
   }
+  if (scenarioOrder === "input") {
+    return [...unique2.values()].slice(0, options.limit).map(({ row }) => row);
+  }
   const hardest = [...unique2.values()].sort(
     (a, b) => a.row.composite - b.row.composite || compareScenarioId(a.row, b.row)
   );
@@ -5937,6 +5944,7 @@ function llmPolicyEditProposer(opts) {
     ...opts.maxHistoryGenerations === void 0 ? {} : { maxGenerations: opts.maxHistoryGenerations },
     ...opts.maxHistoryCandidatesPerGeneration === void 0 ? {} : { maxCandidatesPerGeneration: opts.maxHistoryCandidatesPerGeneration },
     ...opts.maxScenariosPerCandidate === void 0 ? {} : { maxScenariosPerCandidate: opts.maxScenariosPerCandidate },
+    ...opts.scenarioOrder === void 0 ? {} : { scenarioOrder: opts.scenarioOrder },
     ...opts.scenarioIdTransform === void 0 ? {} : { scenarioIdTransform: opts.scenarioIdTransform }
   });
   const maxFindings = positiveLimit(
@@ -5990,12 +5998,14 @@ function llmPolicyEditProposer(opts) {
         baselineOutcome: projectOutcome(
           ctx.baselineOutcome,
           scenarioIds,
-          historyLimits.maxScenariosPerCandidate
+          historyLimits.maxScenariosPerCandidate,
+          historyLimits.scenarioOrder
         ),
         incumbentOutcome: projectOutcome(
           ctx.incumbentOutcome,
           scenarioIds,
           historyLimits.maxScenariosPerCandidate,
+          historyLimits.scenarioOrder,
           ctx.baselineOutcome
         ),
         history: projectPolicyEditHistoryWithProjector(
@@ -6105,6 +6115,7 @@ function projectPolicyEditHistoryWithProjector(history, limits, scenarioIds, obj
           candidate,
           scenarioIds,
           limits.maxScenariosPerCandidate,
+          limits.scenarioOrder,
           candidate.parentSurfaceHash ? candidateByHash.get(candidate.parentSurfaceHash)?.scenarios : void 0,
           objectiveByKey
         )
@@ -6130,10 +6141,11 @@ function selectHistoryCandidates(record, limit) {
   }
   return selected;
 }
-function projectHistoryCandidate(candidate, scenarioIds, maxScenarios, parentScenarios, objectiveByKey) {
+function projectHistoryCandidate(candidate, scenarioIds, maxScenarios, scenarioOrder, parentScenarios, objectiveByKey) {
   const referenceByScenario = parentScenarios ? new Map(parentScenarios.map((scenario) => [scenario.scenarioId, scenario.composite])) : void 0;
   const selectedScenarios = selectPolicyEditAuthorRows(candidate.scenarios, {
     limit: maxScenarios,
+    scenarioOrder,
     ...referenceByScenario ? { referenceByScenario } : {}
   });
   const validatedRecord = candidate.candidateRecord ? validatePolicyEditCandidateRecord(candidate.candidateRecord) : void 0;
@@ -6167,11 +6179,12 @@ function projectHistoryCandidate(candidate, scenarioIds, maxScenarios, parentSce
     forecastCalibration: forecastCalibration(candidate, validatedRecord, objectiveByKey)
   };
 }
-function projectOutcome(outcome, scenarioIds, maxScenarios, reference = void 0) {
+function projectOutcome(outcome, scenarioIds, maxScenarios, scenarioOrder, reference = void 0) {
   if (!outcome) return null;
   const referenceByScenario = reference ? new Map(reference.scenarios.map((scenario) => [scenario.scenarioId, scenario.composite])) : void 0;
   const selectedScenarios = selectPolicyEditAuthorRows(outcome.scenarios, {
     limit: maxScenarios,
+    scenarioOrder,
     ...referenceByScenario ? { referenceByScenario } : {}
   });
   return {
@@ -6424,6 +6437,7 @@ function validateHistoryLimits(options) {
   const maxGenerations = options.maxGenerations ?? DEFAULT_POLICY_EDIT_HISTORY_LIMITS.generations;
   const maxCandidatesPerGeneration = options.maxCandidatesPerGeneration ?? DEFAULT_POLICY_EDIT_HISTORY_LIMITS.candidatesPerGeneration;
   const maxScenariosPerCandidate = options.maxScenariosPerCandidate ?? DEFAULT_POLICY_EDIT_HISTORY_LIMITS.scenariosPerCandidate;
+  const scenarioOrder = options.scenarioOrder ?? "ranked";
   if (!Number.isSafeInteger(maxGenerations) || maxGenerations <= 0) {
     throw new Error("llmPolicyEditProposer: maxHistoryGenerations must be a positive safe integer");
   }
@@ -6437,10 +6451,14 @@ function validateHistoryLimits(options) {
       "llmPolicyEditProposer: maxScenariosPerCandidate must be a positive safe integer"
     );
   }
+  if (scenarioOrder !== "ranked" && scenarioOrder !== "input") {
+    throw new Error("llmPolicyEditProposer: scenarioOrder must be 'ranked' or 'input'");
+  }
   return {
     maxGenerations,
     maxCandidatesPerGeneration,
     maxScenariosPerCandidate,
+    scenarioOrder,
     scenarioIdTransform: options.scenarioIdTransform ?? ((scenarioId) => scenarioId)
   };
 }
@@ -8718,4 +8736,4 @@ export {
   verifyCodeSurface,
   resolveWorktreePath
 };
-//# sourceMappingURL=chunk-22VO7T2I.js.map
+//# sourceMappingURL=chunk-LT4J7ULK.js.map