@tangle-network/agent-eval 0.128.2 → 0.129.0

package/dist/benchmarks/index.d.ts

@@ -137,9 +137,8 @@ interface CostLedgerSummary {
  *     { baseUrl: 'https://router.tangle.tools/v1', apiKey: process.env.KEY },
  *   )
  *
- * This is THE llm-calling seam for agent-eval primitives that need structured
- * output (semantic concept judge, reviewer directives, critic scores). Primitives
- * that need free-form text use `callLlm` and parse output themselves.
+ * `createChatClient` wraps this implementation for provider-neutral package
+ * entry points. Direct callers can use `callLlm` or `callLlmJson`.
  */
 
 interface LlmUsage {
@@ -290,7 +289,7 @@ interface DispatchContext {
 /** The canonical judge verdict shape — one declaration, shared by campaign
  *  judges and the multishot judge runner (which re-exports this type).
  *
- *  Scale is PRODUCER-DEFINED: campaign convention is [0,1]; the legacy
+ *  Scale is PRODUCER-DEFINED: campaign convention is [0,1]; the
  *  multishot runner emits 0-10. Cross-scale comparison must go through
  *  `detectScale` (src/campaign/gates/statistical-heldout.ts, used by
  *  promotion-policy) — never renormalize a producer's values in place, as
@@ -470,8 +469,6 @@ interface CampaignAggregates {
     byScenario: Record<string, ScenarioAggregate>;
     /** Canonical campaign accounting, including worker and judge calls. */
     cost: CostLedgerSummary;
-    /** Compatibility alias of `cost.totalCostUsd`. */
-    totalCostUsd: number;
     /** Cells whose dispatch completed, including cells whose later judge failed. */
     cellsExecuted: number;
     cellsSkipped: number;
@@ -482,7 +479,7 @@ interface CampaignAggregates {
     cellsDispatchFailed?: number;
     /** Present on results that record failure stages. */
     cellsJudgeFailed?: number;
-    /** Legacy failures whose stage was not recorded. */
+    /** Failures whose stage could not be classified. */
     cellsUnclassifiedFailed?: number;
 }
 interface CampaignResult<TArtifact = unknown, TScenario extends Scenario = Scenario> {
@@ -734,7 +731,7 @@ interface CampaignStorage {
     write(path: string, content: string | Uint8Array): void;
     /** Append only when the current UTF-8 byte length matches `expectedBytes`.
      * Returns the new length, or undefined when another writer won. */
-    append?(path: string, content: string, expectedBytes: number): number | undefined;
+    append(path: string, content: string, expectedBytes: number): number | undefined;
 }
 
 interface BenchmarkRunOptions<TPayload = unknown, TArtifact = string> {

package/dist/benchmarks/index.js

@@ -16,24 +16,25 @@ import {
   routing_exports,
   runBenchmarkAdapter,
   summarizeBenchmarkCampaign
-} from "../chunk-XPRT64IE.js";
-import "../chunk-UB2LOJ6Q.js";
-import "../chunk-NKAGIDE2.js";
-import "../chunk-EZJEIH2R.js";
+} from "../chunk-IYCLP2N2.js";
+import "../chunk-QB6BDBP2.js";
+import "../chunk-2QU3YOPR.js";
+import "../chunk-C6LXANRU.js";
 import "../chunk-WGXIEX7P.js";
-import "../chunk-NYLOYM6N.js";
-import "../chunk-2MKQIFS4.js";
-import "../chunk-PBE2LOSS.js";
-import "../chunk-DPUHNQLN.js";
-import "../chunk-MHELPNRP.js";
-import "../chunk-WS3NZZQQ.js";
+import "../chunk-EG66UGL4.js";
+import "../chunk-E7QXT7SX.js";
+import "../chunk-SFLLL76A.js";
+import "../chunk-7FO3TNPI.js";
+import "../chunk-ZHTZ4EYI.js";
+import "../chunk-VCZ5FQYW.js";
 import "../chunk-VI2UW6B6.js";
 import "../chunk-5DTSBUL2.js";
 import "../chunk-GGE4NNQT.js";
 import "../chunk-P6FYH6K4.js";
 import "../chunk-PC4UYEBM.js";
-import "../chunk-2JX3CFMB.js";
+import "../chunk-56TAVBOK.js";
 import "../chunk-MA6HLL3S.js";
+import "../chunk-OIUOT4QD.js";
 import "../chunk-ONWEPEDO.js";
 import "../chunk-K4DBDHLK.js";
 import "../chunk-PZ5AY32C.js";

package/dist/builder-eval/index.js

@@ -5,7 +5,7 @@ import {
 import {
   pearsonR,
   spearmanR
-} from "../chunk-MHELPNRP.js";
+} from "../chunk-ZHTZ4EYI.js";
 import {
   judgeSpans
 } from "../chunk-ZET2UAYW.js";

package/dist/campaign/index.d.ts

@@ -249,9 +249,8 @@ type CostLedgerHandle = Pick<CostLedger, Exclude<keyof CostLedger, 'listPending'
  *     { baseUrl: 'https://router.tangle.tools/v1', apiKey: process.env.KEY },
  *   )
  *
- * This is THE llm-calling seam for agent-eval primitives that need structured
- * output (semantic concept judge, reviewer directives, critic scores). Primitives
- * that need free-form text use `callLlm` and parse output themselves.
+ * `createChatClient` wraps this implementation for provider-neutral package
+ * entry points. Direct callers can use `callLlm` or `callLlmJson`.
  */
 
 interface LlmMessage {
@@ -335,42 +334,27 @@ interface LlmCallResult {
 type LlmCallMetadata = Pick<LlmCallResult, 'usage' | 'costUsd' | 'model' | 'durationMs'>;
 
 /**
- * ChatClient — the single LLM abstraction analysts call.
- *
- * agent-eval already ships an `LlmClient` (OpenAI-compatible, retry,
- * graceful JSON-schema degrade) and judges that talk to `TCloud`. Two
- * mixed patterns force every analyst author to pick a transport, which
- * couples analyst code to runtime concerns (cli-bridge vs router vs
- * sandbox-sdk) it shouldn't know about.
- *
- * `ChatClient` is one interface every analyst takes via `AnalystContext.chat`.
- * The operator decides at the registry boundary which transport binds
- * to it. Analyst code stays transport-agnostic; swapping production
- * (sandbox-sdk) for local dev (cli-bridge) or tests (mock) is a one-
- * line factory call.
- *
- * Designed to coexist: existing `LlmClient` callers and existing
- * `TCloud`-based judges keep working untouched. New analyst code uses
- * `ChatClient`. When old call sites migrate, they pick up budgeting,
- * cancellation, and unified telemetry for free.
+ * Provider-neutral chat contract for every model call made by agent-eval.
+ *
+ * Callers choose the transport at the package boundary with `createChatClient`.
+ * Evaluation code receives canonical requests and results without importing a
+ * provider SDK.
  */
 
 /**
- * Unified chat interface. Mirrors LlmCallRequest/Result so the OpenAI-
- * compatible mental model stays. Two methods: a one-shot `chat()` and
- * an `streamChat()` for future agentic loops (not yet exposed).
+ * Unified chat interface using the package's canonical LLM request and result.
  */
 interface ChatClient {
-    /** Display name of the bound transport — included in telemetry. */
+    /** Display name of the bound transport, included in telemetry. */
     readonly transport: ChatTransport;
-    /** Default model when caller omits — operators bind this per environment. */
+    /** Default model when the caller omits one. */
     readonly defaultModel?: string;
     /** Total provider attempts this transport can make for one chat call. */
     readonly maximumAttempts?: number;
     /** Implementations must enforce `req.maxTokens` when it is present. */
     chat(req: ChatRequest, opts?: ChatCallOpts): Promise<ChatResponse>;
 }
-type ChatTransport = 'router' | 'sandbox-sdk' | 'cli-bridge' | 'direct-provider' | 'mock';
+type ChatTransport = 'router' | 'sandbox-sdk' | 'cli-bridge' | 'direct-provider' | 'custom' | 'mock';
 interface ChatRequest extends Omit<LlmCallRequest, 'model'> {
     /** Optional — falls back to ChatClient.defaultModel. */
     model?: string;
@@ -738,7 +722,7 @@ interface JudgeConfig<TArtifact, TScenario extends Scenario = Scenario> {
 /** The canonical judge verdict shape — one declaration, shared by campaign
  *  judges and the multishot judge runner (which re-exports this type).
  *
- *  Scale is PRODUCER-DEFINED: campaign convention is [0,1]; the legacy
+ *  Scale is PRODUCER-DEFINED: campaign convention is [0,1]; the
  *  multishot runner emits 0-10. Cross-scale comparison must go through
  *  `detectScale` (src/campaign/gates/statistical-heldout.ts, used by
  *  promotion-policy) — never renormalize a producer's values in place, as
@@ -961,9 +945,6 @@ interface SurfaceProposer<TFindings = unknown> {
         reason?: string;
     };
 }
-/** Optional vocabulary alias. The loop is the optimizer; this object is the
- * proposer inside that loop. */
-type OptimizationProposer<TFindings = unknown> = SurfaceProposer<TFindings>;
 interface OptimizerConfigBase {
     populationSize: number;
     maxGenerations: number;
@@ -1243,8 +1224,6 @@ interface CampaignAggregates {
     byScenario: Record<string, ScenarioAggregate>;
     /** Canonical campaign accounting, including worker and judge calls. */
     cost: CostLedgerSummary;
-    /** Compatibility alias of `cost.totalCostUsd`. */
-    totalCostUsd: number;
     /** Cells whose dispatch completed, including cells whose later judge failed. */
     cellsExecuted: number;
     cellsSkipped: number;
@@ -1255,7 +1234,7 @@ interface CampaignAggregates {
     cellsDispatchFailed?: number;
     /** Present on results that record failure stages. */
     cellsJudgeFailed?: number;
-    /** Legacy failures whose stage was not recorded. */
+    /** Failures whose stage could not be classified. */
     cellsUnclassifiedFailed?: number;
 }
 interface CampaignResult<TArtifact = unknown, TScenario extends Scenario = Scenario> {
@@ -2313,7 +2292,7 @@ interface CampaignStorage {
     write(path: string, content: string | Uint8Array): void;
     /** Append only when the current UTF-8 byte length matches `expectedBytes`.
      * Returns the new length, or undefined when another writer won. */
-    append?(path: string, content: string, expectedBytes: number): number | undefined;
+    append(path: string, content: string, expectedBytes: number): number | undefined;
 }
 /** Node-filesystem storage — the default. Lazily requires `node:fs` so the
  *  module imports cleanly in non-Node runtimes (where the caller passes
@@ -2455,9 +2434,9 @@ interface RunCampaignOptions<TScenario extends Scenario, TArtifact> {
     }) => string | undefined;
 }
 /** Durable `<cell>/failure-receipt.json` written before a failed cell can
- * trigger campaign-wide cancellation. The cell keeps its dispatch-only usage
- * fields for compatibility; `cost` covers every settled agent and judge call
- * attributed to this exact run attempt. */
+ * trigger campaign-wide cancellation. The cell records dispatch measurements;
+ * `cost` covers every settled agent and judge call attributed to this exact run
+ * attempt. */
 interface CampaignCellFailureReceipt<TArtifact = unknown> {
     schemaVersion: 1;
     runAttemptId: string;
@@ -3110,6 +3089,18 @@ interface VerifiableRewardExtractionOptions {
      * doesn't report one. Default `0.7`.
      */
     judgeConfidenceFloor?: number;
+    /**
+     * Whether the anti-Goodhart realness gate applies. Default `true`, and the
+     * default is the one every training path must keep.
+     *
+     * Set `false` ONLY for detection and analysis. `rl/reward-hacking.ts` does,
+     * for the same reason it reads `observedScore` for its proxy: it measures the
+     * DIVERGENCE between the judge signal and the deterministic one, and a
+     * deterministic reward that another gate already forced to 0 manufactures
+     * exactly that divergence on exactly the gamed population. The detector would
+     * then be re-reporting a verdict it was supposed to reach independently.
+     */
+    applyRealnessGate?: boolean;
 }
 
 /**
@@ -6387,4 +6378,4 @@ declare function verifyCodeSurface(surface: CodeSurface, worktreeDir?: string):
  *  identity against the checkout at `worktreeRef`. */
 declare function resolveWorktreePath(surface: CodeSurface, worktreeDir?: string): string;
 
-export { type AnalystArtifact, type AnalystScenario, type AnalyzeCrossSurfaceInteractionsInput, type AxisEvidence, type AxisVerdict, type BuildAnalystSurfaceDispatchOptions, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellFailureReceipt, 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 ComponentSurface, 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, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, 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 ExternalOptimizationExample, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, type FailureModeRecallJudgeOptions, FileSearchLedger, FsLabeledScenarioStore, type FsLabeledScenarioStoreOptions, 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 GitWorktreeAdapterOptions, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, LabeledScenarioStoreError, type LabeledScenarioWrite, type LlmJudgeDimension, type LlmJudgeOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceArgsFromResult, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceEvidence, type LoopProvenanceOptimizationMethod, type LoopProvenanceRecord, type MutableSurface, type NeutralizationGateOptions, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodInput, type OptimizationMethodPairwise, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationMethodRunOptions, type OptimizationMethodScore, type OptimizationPackageSource, type OptimizationProposer, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type PairedHoldout, type ParetoParent, type ParetoSignificanceGateOptions, type PendingCostCallView, type PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PowerPreflight, type PowerPreflightOptions, type PremeasuredOptimizationBaseline, type ProfileDispatchFn, ProfileMatrixError, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposalTrackContext, type ProposeContext, type ProposedCandidate, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, 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 SequentialDecideFn, type SequentialDecideOptions, type SequentialDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SessionScript, type SingleRunLock, type SingleRunLockOptions, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SurfaceProposer, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, WorktreeAdapterError, acquireSingleRunLock, analyzeCrossSurfaceInteractions, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertComponentSurface, buildAnalystSurfaceDispatch, buildEvidenceVector, buildLoopProvenanceRecord, campaignBreakdown, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, compareOptimizationMethods, compareRankKeys, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createReferenceEquivalenceJudge, createRunCostLedger, defaultProductionGate, detectScale, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, externalTextOptimizationMethod, failureModeRecallJudge, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, inMemoryCampaignStorage, isProposedCandidate, isTransientTransportFailure, labelTrustRank, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, optimizationTokenUsageFromSummary, pairHoldout, paretoPolicy, paretoSignificanceGate, planCampaignRun, planEvalFixtureRun, powerPreflight, provenanceRecordPath, provenanceSpansPath, renderScoreboardMarkdown, renderSurfaceDiff, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, scoreDiscrimination, scoreUserStory, scoreboardSummary, selectDiscriminative, sequentialDecide, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceHash, tangleTracesRoot, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord };
+export { type AnalystArtifact, type AnalystScenario, type AnalyzeCrossSurfaceInteractionsInput, type AxisEvidence, type AxisVerdict, type BuildAnalystSurfaceDispatchOptions, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellFailureReceipt, 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 ComponentSurface, 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, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, 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 ExternalOptimizationExample, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, type FailureModeRecallJudgeOptions, FileSearchLedger, FsLabeledScenarioStore, type FsLabeledScenarioStoreOptions, 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 GitWorktreeAdapterOptions, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, LabeledScenarioStoreError, type LabeledScenarioWrite, type LlmJudgeDimension, type LlmJudgeOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceArgsFromResult, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceEvidence, type LoopProvenanceOptimizationMethod, type LoopProvenanceRecord, type MutableSurface, type NeutralizationGateOptions, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationMethod, type OptimizationMethodComparison, 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 PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PowerPreflight, type PowerPreflightOptions, type PremeasuredOptimizationBaseline, type ProfileDispatchFn, ProfileMatrixError, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposalTrackContext, type ProposeContext, type ProposedCandidate, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, 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 SequentialDecideFn, type SequentialDecideOptions, type SequentialDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SessionScript, type SingleRunLock, type SingleRunLockOptions, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SurfaceProposer, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, WorktreeAdapterError, acquireSingleRunLock, analyzeCrossSurfaceInteractions, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertComponentSurface, buildAnalystSurfaceDispatch, buildEvidenceVector, buildLoopProvenanceRecord, campaignBreakdown, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, compareOptimizationMethods, compareRankKeys, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createReferenceEquivalenceJudge, createRunCostLedger, defaultProductionGate, detectScale, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, externalTextOptimizationMethod, failureModeRecallJudge, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, inMemoryCampaignStorage, isProposedCandidate, isTransientTransportFailure, labelTrustRank, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, optimizationTokenUsageFromSummary, pairHoldout, paretoPolicy, paretoSignificanceGate, planCampaignRun, planEvalFixtureRun, powerPreflight, provenanceRecordPath, provenanceSpansPath, renderScoreboardMarkdown, renderSurfaceDiff, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, scoreDiscrimination, scoreUserStory, scoreboardSummary, selectDiscriminative, sequentialDecide, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceHash, tangleTracesRoot, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord };

package/dist/campaign/index.js

@@ -32,7 +32,7 @@ import {
   userStoryScoreboard,
   validateSearchLedgerEvent,
   verifyCodeSurface
-} from "../chunk-UB2LOJ6Q.js";
+} from "../chunk-QB6BDBP2.js";
 import {
   acquireSingleRunLock,
   assertCodeSurfaceIdentity,
@@ -79,7 +79,7 @@ import {
   surfaceContentHash,
   surfaceHash,
   verifyLoopProvenanceRecord
-} from "../chunk-NKAGIDE2.js";
+} from "../chunk-2QU3YOPR.js";
 import {
   SearchLedgerConflictError,
   SearchLedgerError,
@@ -96,21 +96,22 @@ import {
   resolveRunDir,
   runCampaign,
   tangleTracesRoot
-} from "../chunk-EZJEIH2R.js";
+} from "../chunk-C6LXANRU.js";
 import "../chunk-WGXIEX7P.js";
-import "../chunk-NYLOYM6N.js";
-import "../chunk-2MKQIFS4.js";
-import "../chunk-PBE2LOSS.js";
-import "../chunk-DPUHNQLN.js";
-import "../chunk-MHELPNRP.js";
-import "../chunk-WS3NZZQQ.js";
+import "../chunk-EG66UGL4.js";
+import "../chunk-E7QXT7SX.js";
+import "../chunk-SFLLL76A.js";
+import "../chunk-7FO3TNPI.js";
+import "../chunk-ZHTZ4EYI.js";
+import "../chunk-VCZ5FQYW.js";
 import "../chunk-VI2UW6B6.js";
 import "../chunk-5DTSBUL2.js";
 import "../chunk-GGE4NNQT.js";
 import "../chunk-P6FYH6K4.js";
 import "../chunk-PC4UYEBM.js";
-import "../chunk-2JX3CFMB.js";
+import "../chunk-56TAVBOK.js";
 import "../chunk-MA6HLL3S.js";
+import "../chunk-OIUOT4QD.js";
 import "../chunk-ONWEPEDO.js";
 import "../chunk-K4DBDHLK.js";
 import "../chunk-PZ5AY32C.js";