@tangle-network/agent-eval 0.115.2 → 0.116.0

package/dist/{gepa-dne9JDPL.d.ts → gepa-CQelRtuC.d.ts}

@@ -1,8 +1,8 @@
 import { D as DatasetScenario, c as Dataset } from './dataset-NENEzRgk.js';
 import { T as TraceStore } from './store-BsVi7ncX.js';
-import { S as Scenario, C as CampaignResult, G as GateResult, c as DispatchFn, b as JudgeConfig, L as LabeledScenarioStore, d as CampaignTraceWriter, e as GenerationRecord, M as MutableSurface, P as ParetoParent, f as SurfaceProposer, g as Gate } from './types-C5gJrOVT.js';
+import { S as Scenario, C as CampaignResult, G as GateResult, c as DispatchFn, b as JudgeConfig, L as LabeledScenarioStore, d as CampaignTraceWriter, e as GenerationRecord, M as MutableSurface, P as ParetoParent, f as SurfaceProposer, g as Gate } from './types-Ca_63YSD.js';
 import { C as CampaignStorage } from './storage-Dw_f7WMt.js';
-import { L as LlmClientOptions } from './llm-client-DyqEH4jH.js';
+import { L as LlmClientOptions } from './policy-edit-Clb2v6Oa.js';
 
 /**
  * Pareto frontier — multi-objective optimization over candidate runs.
@@ -332,10 +332,11 @@ declare function planCampaignRun<TScenario extends Scenario, TArtifact>(opts: Pl
 /**
  * `runOptimization` — the improvement loop body. Runs N generations: the
  * `SurfaceProposer` proposes K candidate surfaces per generation, each
- * candidate runs a campaign (the measurement), top-scoring promote to the
- * next generation. Proposer-agnostic — the same loop runs an evolutionary
- * population mutator (`evolutionaryProposer`) or any reflective / agentic
- * proposer; they differ only in how `propose()` picks candidates.
+ * candidate runs a campaign (the measurement), and only a candidate that beats
+ * the single global incumbent becomes the next generation's parent.
+ * Proposer-agnostic — the same loop runs an evolutionary population mutator
+ * (`evolutionaryProposer`) or any reflective / agentic proposer; they differ
+ * only in how `propose()` picks candidates.
  *
  * This is `runLoop`'s shape (plan → measure → decide) specialized to surface
  * improvement: `proposer.propose` = plan, `runCampaign` = the measurement
@@ -357,7 +358,8 @@ interface RunOptimizationBaseOptions<TScenario extends Scenario, TArtifact> exte
     proposer: SurfaceProposer;
     populationSize: number;
     maxGenerations: number;
-    /** How many top-scoring candidates carry to the next generation. Default 2. */
+    /** @deprecated The loop has one global incumbent and can promote only the
+     *  single candidate that beats it. Retained for source compatibility. */
     promoteTopK?: number;
     /** DEPTH knob forwarded to the proposer's `propose()` — max iterations the
      *  agentic generator may take per candidate. */
@@ -420,7 +422,7 @@ interface RunOptimizationResult<TArtifact, TScenario extends Scenario> {
     paretoFrontier: ParetoParent[];
 }
 /**
- * Improvement loop body: N generations of propose → campaign → rank, maintaining a Pareto frontier and promoting the top-scoring candidates to the next generation.
+ * Improvement loop body: N generations of propose → campaign → rank, maintaining a Pareto frontier and one global incumbent across generations.
  */
 declare function runOptimization<TScenario extends Scenario, TArtifact>(opts: RunOptimizationOptions<TScenario, TArtifact>): Promise<RunOptimizationResult<TArtifact, TScenario>>;
 

package/dist/hosted/index.d.ts

@@ -1,10 +1,14 @@
-import { M as MutableSurface, h as GateDecision } from '../types-C5gJrOVT.js';
-import { I as InsightReport } from '../insight-report-IwwvqZZv.js';
-import '../run-record-B7RTi_ix.js';
+import { M as MutableSurface, h as GateDecision } from '../types-Ca_63YSD.js';
+import { I as InsightReport } from '../insight-report-oMVxDTxl.js';
+import '../policy-edit-Clb2v6Oa.js';
+import '../run-record-CZmcpWPo.js';
 import '@tangle-network/agent-interface';
 import '../errors-oeQrLqXC.js';
 import '../schema-SGWcK9wa.js';
-import '../summary-report-BJ5aNwZ1.js';
+import '../store-9cAScOcb.js';
+import '../types-C7DGg5ex.js';
+import '@tangle-network/tcloud';
+import '../summary-report-DTNgQycC.js';
 import '../failure-cluster-C48PiReX.js';
 import '../store-BsVi7ncX.js';
 import '../judge-calibration-7C-IDmKr.js';

package/dist/{index-BTEpx9He.d.ts → index-DbCXJfZ1.d.ts}

@@ -1,5 +1,5 @@
-import { S as Scenario, D as DispatchContext, C as CampaignResult } from './types-C5gJrOVT.js';
-import { b as RunSplitTag } from './run-record-B7RTi_ix.js';
+import { S as Scenario, D as DispatchContext, C as CampaignResult } from './types-Ca_63YSD.js';
+import { a as RunSplitTag } from './run-record-CZmcpWPo.js';
 import { C as CampaignStorage } from './storage-Dw_f7WMt.js';
 
 /**

package/dist/index.d.ts

@@ -1,12 +1,12 @@
-export { A as ActionExecutionPolicy, a as ActionPolicyDecision, C as ControlRunToRunRecordOptions, L as LlmJsonCall, b as LlmReviewerConfig, P as ProposeFn, c as ProposeInput, d as ProposeOutput, e as ProposeReviewConfig, f as ProposeReviewControlAction, g as ProposeReviewControlConfig, h as ProposeReviewControlResult, i as ProposeReviewControlState, j as ProposeReviewReport, k as ProposeReviewShot, R as Review, l as ReviewFn, m as ReviewInput, n as ReviewMemoryEntry, o as ReviewMemoryStore, p as RunEvidenceMetadata, V as Verification, q as VerifyFn, r as controlFailureClassFromVerification, s as controlRunToRunRecord, t as createLlmReviewer, u as evaluateActionPolicy, v as inMemoryReviewStore, w as jsonlReviewStore, x as runProposeReview, y as runProposeReviewAsControlLoop, z as scoreFromEvals } from './control-CcBiAEnn.js';
-import { R as RunRecord, b as RunSplitTag } from './run-record-B7RTi_ix.js';
-export { g as AGENT_PROFILE_KINDS, h as AgentInterfaceProfileLike, A as AgentProfileCell, f as AgentProfileCellInput, i as AgentProfileCellSchemaVersion, j as AgentProfileCellValidationError, k as AgentProfileDimensionValue, l as AgentProfileHarness, a as AgentProfileJson, m as AgentProfileKind, n as AgentProfileSource, o as AgentProfileSourceInput, J as JudgeScoresRecord, c as RunCostProvenance, e as RunJudgeMetadata, p as RunOutcome, q as RunRecordValidationError, d as RunTokenUsage, r as agentProfileCellHashMaterial, s as agentProfileCellKey, t as assertRunAgentProfileCell, u as buildAgentInterfaceProfileCell, v as buildAgentProfileCell, w as groupRunsByAgentProfileCell, x as isRunRecord, y as modelHasSnapshot, z as parseRunRecordSafe, B as requireAgentProfileCell, C as resolveRunCostProvenance, D as roundTripRunRecord, E as toAgentProfileJson, F as validateAgentProfileCell, G as validateRunRecord, H as verifyAgentProfileCell } from './run-record-B7RTi_ix.js';
-import { B as BehavioralMetrics, A as RunScore, a as RunTrace, E as RunScoreWeights } from './semantic-concept-judge-DxJmRkyJ.js';
-export { G as ConceptComplexity, H as ConceptFinding, J as ConceptSpec, L as ConceptWeightStrategy, C as CreateAnalystAiConfig, M as DEFAULT_COMPLEXITY_WEIGHTS, N as DEFAULT_RUN_SCORE_WEIGHTS, D as DEFAULT_TRACE_ANALYST_KINDS, c as DiffPolicy, F as FAILURE_MODE_KIND_SPEC, g as FindingSubject, h as FindingSubjectKind, j as FindingsDiff, k as FindingsStore, I as IMPROVEMENT_KIND_SPEC, l as KNOWLEDGE_GAP_KIND_SPEC, m as KNOWLEDGE_POISONING_KIND_SPEC, P as PersistedFinding, R as RunCritic, O as RunCriticOptions, Q as SEMANTIC_CONCEPT_JUDGE_VERSION, n as SKILL_USAGE_ANALYST, b as SemanticConceptJudgeInput, S as SemanticConceptJudgeOptions, T as SemanticConceptJudgeResult, o as SkillUsageAnalyst, U as SuboptimalCode, V as SuboptimalSignal, W as aggregateRunScore, X as clamp01, Y as computeTraceMetrics, t as createAnalystAi, Z as createSemanticConceptJudge, u as defaultIsMaterial, v as diffFindings, _ as runSemanticConceptJudge } from './semantic-concept-judge-DxJmRkyJ.js';
-import { m as ChatRequest, q as CreateChatClientOpts } from './kind-factory-DcNg13sZ.js';
-export { a as Analyst, b as AnalystContext, i as AnalystCost, A as AnalystFinding, j as AnalystInputKind, k as AnalystRequirements, f as AnalystRunEvent, e as AnalystRunInputs, d as AnalystRunResult, c as AnalystRunSummary, g as AnalystSeverity, l as ChatCallOpts, C as ChatClient, n as ChatResponse, o as ChatTransport, p as CliBridgeTransportOpts, r as CreateTraceAnalystKindOpts, D as DirectProviderTransportOpts, E as EvidenceRef, M as MockTransportOpts, s as RawAnalystFinding, u as RouterTransportOpts, S as SandboxSdkTransportOpts, v as TraceAnalystGolden, T as TraceAnalystKindSpec, w as computeFindingId, x as createChatClient, y as createTraceAnalystKind, z as makeFinding, F as renderPriorFindings } from './kind-factory-DcNg13sZ.js';
-export { a as AnalystHooks, A as AnalystRegistry, b as AnalystRegistryOptions, B as BudgetPolicy, D as DefaultAnalystRegistryOptions, R as RegistryRunOpts, c as buildDefaultAnalystRegistry } from './default-registry-DltpYR5u.js';
-export { F as FindingToPolicyEditOptions, P as POLICY_EDIT_AXES, a as POLICY_EDIT_TARGET_SURFACES, b as PolicyEdit, c as PolicyEditAdmission, d as PolicyEditAdmissionOptions, e as PolicyEditAxis, f as PolicyEditChange, g as PolicyEditExpectedGain, h as PolicyEditGainDirection, i as PolicyEditGainUnit, j as PolicyEditInit, k as PolicyEditRisk, l as PolicyEditSchemaVersion, m as PolicyEditSource, n as PolicyEditTarget, o as PolicyEditTargetSurface, p as PolicyEditValidationError, q as admitPolicyEdit, r as applyPolicyEditToSurface, s as computePolicyEditId, t as isPolicyEdit, u as makePolicyEdit, v as policyEditFromFinding, w as policyEditsFromFindings, x as scorePolicyEditReadiness, y as validatePolicyEdit } from './policy-edit-RLn8GWof.js';
+export { A as ActionExecutionPolicy, a as ActionPolicyDecision, C as ControlRunToRunRecordOptions, L as LlmJsonCall, b as LlmReviewerConfig, P as ProposeFn, c as ProposeInput, d as ProposeOutput, e as ProposeReviewConfig, f as ProposeReviewControlAction, g as ProposeReviewControlConfig, h as ProposeReviewControlResult, i as ProposeReviewControlState, j as ProposeReviewReport, k as ProposeReviewShot, R as Review, l as ReviewFn, m as ReviewInput, n as ReviewMemoryEntry, o as ReviewMemoryStore, p as RunEvidenceMetadata, V as Verification, q as VerifyFn, r as controlFailureClassFromVerification, s as controlRunToRunRecord, t as createLlmReviewer, u as evaluateActionPolicy, v as inMemoryReviewStore, w as jsonlReviewStore, x as runProposeReview, y as runProposeReviewAsControlLoop, z as scoreFromEvals } from './control-DbcDxouY.js';
+import { R as RunRecord, a as RunSplitTag } from './run-record-CZmcpWPo.js';
+export { g as AGENT_PROFILE_KINDS, h as AgentInterfaceProfileLike, A as AgentProfileCell, e as AgentProfileCellInput, i as AgentProfileCellSchemaVersion, j as AgentProfileCellValidationError, k as AgentProfileDimensionValue, l as AgentProfileHarness, f as AgentProfileJson, m as AgentProfileKind, n as AgentProfileSource, o as AgentProfileSourceInput, J as JudgeScoresRecord, b as RunCostProvenance, d as RunJudgeMetadata, p as RunOutcome, q as RunRecordValidationError, c as RunTokenUsage, r as agentProfileCellHashMaterial, s as agentProfileCellKey, t as assertRunAgentProfileCell, u as buildAgentInterfaceProfileCell, v as buildAgentProfileCell, w as groupRunsByAgentProfileCell, x as isRunRecord, y as modelHasSnapshot, z as parseRunRecordSafe, B as requireAgentProfileCell, C as resolveRunCostProvenance, D as roundTripRunRecord, E as toAgentProfileJson, F as validateAgentProfileCell, G as validateRunRecord, H as verifyAgentProfileCell } from './run-record-CZmcpWPo.js';
+import { B as BehavioralMetrics, A as RunScore, a as RunTrace, E as RunScoreWeights } from './semantic-concept-judge-CKjePUMh.js';
+export { G as ConceptComplexity, H as ConceptFinding, J as ConceptSpec, L as ConceptWeightStrategy, C as CreateAnalystAiConfig, M as DEFAULT_COMPLEXITY_WEIGHTS, N as DEFAULT_RUN_SCORE_WEIGHTS, D as DEFAULT_TRACE_ANALYST_KINDS, c as DiffPolicy, F as FAILURE_MODE_KIND_SPEC, g as FindingSubject, h as FindingSubjectKind, j as FindingsDiff, k as FindingsStore, I as IMPROVEMENT_KIND_SPEC, l as KNOWLEDGE_GAP_KIND_SPEC, m as KNOWLEDGE_POISONING_KIND_SPEC, P as PersistedFinding, R as RunCritic, O as RunCriticOptions, Q as SEMANTIC_CONCEPT_JUDGE_VERSION, n as SKILL_USAGE_ANALYST, b as SemanticConceptJudgeInput, S as SemanticConceptJudgeOptions, T as SemanticConceptJudgeResult, o as SkillUsageAnalyst, U as SuboptimalCode, V as SuboptimalSignal, W as aggregateRunScore, X as clamp01, Y as computeTraceMetrics, t as createAnalystAi, Z as createSemanticConceptJudge, u as defaultIsMaterial, v as diffFindings, _ as runSemanticConceptJudge } from './semantic-concept-judge-CKjePUMh.js';
+import { L as LlmClientOptions, m as ChatRequest, q as CreateChatClientOpts } from './policy-edit-Clb2v6Oa.js';
+export { A as Analyst, b as AnalystContext, h as AnalystCost, d as AnalystFinding, j as AnalystInputKind, k as AnalystRequirements, g as AnalystRunEvent, f as AnalystRunInputs, e as AnalystRunResult, c as AnalystRunSummary, i as AnalystSeverity, l as ChatCallOpts, C as ChatClient, n as ChatResponse, o as ChatTransport, p as CliBridgeTransportOpts, D as DirectProviderTransportOpts, E as EvidenceRef, F as FindingToPolicyEditOptions, a5 as LlmCallError, a6 as LlmCallRequest, a7 as LlmCallResult, a8 as LlmClient, a9 as LlmMessage, aa as LlmRouteAssertionError, a as LlmRouteRequirements, ab as LlmUsage, M as MockTransportOpts, r as POLICY_EDIT_AXES, s as POLICY_EDIT_CANDIDATE_RECORD_SCHEMA, t as POLICY_EDIT_TARGET_SURFACES, u as PolicyEdit, v as PolicyEditAdmission, w as PolicyEditAdmissionOptions, x as PolicyEditAxis, P as PolicyEditCandidateRecord, y as PolicyEditChange, z as PolicyEditExpectedGain, B as PolicyEditGainDirection, G as PolicyEditGainUnit, H as PolicyEditInit, I as PolicyEditRisk, J as PolicyEditSchemaVersion, K as PolicyEditSource, N as PolicyEditTarget, O as PolicyEditTargetSurface, Q as PolicyEditValidationError, R as RouterTransportOpts, S as SandboxSdkTransportOpts, T as admitPolicyEdit, U as applyPolicyEditToSurface, ac as assertLlmRoute, ad as backoffMs, ae as callLlm, af as callLlmJson, V as computeFindingId, W as computePolicyEditId, X as createChatClient, Y as isPolicyEdit, ag as isTransientLlmError, Z as makeFinding, _ as makePolicyEdit, $ as makePolicyEditCandidateRecord, a0 as policyEditFromFinding, a1 as policyEditsFromFindings, ah as probeLlm, a2 as scorePolicyEditReadiness, ai as stripFencedJson, a3 as validatePolicyEdit, a4 as validatePolicyEditCandidateRecord } from './policy-edit-Clb2v6Oa.js';
+export { a as AnalystHooks, A as AnalystRegistry, b as AnalystRegistryOptions, B as BudgetPolicy, D as DefaultAnalystRegistryOptions, R as RegistryRunOpts, c as buildDefaultAnalystRegistry } from './default-registry-DDfv22MQ.js';
+export { C as CreateTraceAnalystKindOpts, a as RawAnalystFinding, c as TraceAnalystGolden, T as TraceAnalystKindSpec, d as createTraceAnalystKind, r as renderPriorFindings } from './kind-factory-DWOvXjR_.js';
 import { TCloud } from '@tangle-network/tcloud';
 import { B as BenchmarkRunnerConfig, S as Scenario, c as BenchmarkReport, P as ProductClientConfig, C as CheckResult, T as TestResult, d as PersonaConfig, D as DriverResult, e as DriverState, b as JudgeFn, f as CollectedArtifacts, g as ScenarioResult, h as TurnMetrics, i as ScenarioFile, j as CompletionCriterion } from './types-C7DGg5ex.js';
 export { A as ArtifactCheck, k as ArtifactResult, E as EvalResult, F as FeedbackPattern, l as JudgeConfig, a as JudgeInput, m as JudgeRubric, J as JudgeScore, n as PersonaRigor, R as RouteMap, o as RubricDimension, p as Turn, q as TurnResult } from './types-C7DGg5ex.js';
@@ -16,32 +16,31 @@ import { F as FailureClass, T as ToolSpan, h as BudgetSpec, B as BudgetLedgerEnt
 export { A as Artifact, E as EventKind, i as FAILURE_CLASSES, G as GenericSpan, J as JudgeSpan, M as Message, c as RetrievalSpan, g as RunLayer, f as RunStatus, d as SandboxSpan, j as SpanBase, b as SpanKind, k as SpanStatus, l as TRACE_SCHEMA_VERSION, e as TraceEvent, m as isJudgeSpan, n as isLlmSpan, o as isRetrievalSpan, p as isSandboxSpan, q as isToolSpan } from './schema-SGWcK9wa.js';
 import { A as AgentEvalError, J as JudgeError, a as ConfigError } from './errors-oeQrLqXC.js';
 export { b as AgentEvalErrorCode, C as CaptureIntegrityError, N as NotFoundError, R as ReplayError, V as ValidationError, c as VerificationError } from './errors-oeQrLqXC.js';
-import { c as CorrectnessChecker } from './pre-registration-D8h7ZxNL.js';
-export { A as ArtifactCheckArtifact, e as ArtifactEventLike, f as ArtifactValidator, g as BackendIntegrityError, B as BackendIntegrityReport, h as ComparePairedArmsOptions, C as CompletionRequirement, a as CompletionVerdict, H as HypothesisManifest, i as HypothesisResult, j as LlmCorrectnessCheckerOpts, L as LlmJudgeDimension, d as LlmJudgeOptions, M as MatchedPair, k as PairArmsOptions, m as PairArmsResult, n as PairedArmRow, P as PairedArmsComparison, o as PairedCorrectness, p as PairedMetricDelta, q as ProducedProposal, b as ProducedState, r as ProposalEventLike, s as RequirementCheck, R as RuntimeEventLike, t as SatisfiedBy, S as SignedManifest, u as SignedManifestAlgo, T as TaskGold, v as ToolCallEventLike, V as ValidationContext, w as ValidationIssue, x as ValidationResult, y as assertRealBackend, z as byteLengthRange, D as canonicalize, E as comparePairedArms, F as completionVerdict, G as composeValidators, I as containsAll, J as createLlmCorrectnessChecker, K as createTokenRecallChecker, N as evaluateHypothesis, O as extractProducedState, Q as hashJson, U as jsonHasKeys, l as llmJudge, W as pairArms, X as parseCorrectnessResponse, Y as regexMatch, Z as signManifest, _ as summarizeBackendIntegrity, $ as verifyCompletion, a0 as verifyManifest } from './pre-registration-D8h7ZxNL.js';
+import { c as CorrectnessChecker } from './pre-registration--vU0mMtD.js';
+export { A as ArtifactCheckArtifact, e as ArtifactEventLike, f as ArtifactValidator, g as BackendIntegrityError, B as BackendIntegrityReport, h as ComparePairedArmsOptions, C as CompletionRequirement, a as CompletionVerdict, H as HypothesisManifest, i as HypothesisResult, j as LlmCorrectnessCheckerOpts, L as LlmJudgeDimension, d as LlmJudgeOptions, M as MatchedPair, k as PairArmsOptions, m as PairArmsResult, n as PairedArmRow, P as PairedArmsComparison, o as PairedCorrectness, p as PairedMetricDelta, q as ProducedProposal, b as ProducedState, r as ProposalEventLike, s as RequirementCheck, R as RuntimeEventLike, t as SatisfiedBy, S as SignedManifest, u as SignedManifestAlgo, T as TaskGold, v as ToolCallEventLike, V as ValidationContext, w as ValidationIssue, x as ValidationResult, y as assertRealBackend, z as byteLengthRange, D as canonicalize, E as comparePairedArms, F as completionVerdict, G as composeValidators, I as containsAll, J as createLlmCorrectnessChecker, K as createTokenRecallChecker, N as evaluateHypothesis, O as extractProducedState, Q as hashJson, U as jsonHasKeys, l as llmJudge, W as pairArms, X as parseCorrectnessResponse, Y as regexMatch, Z as signManifest, _ as summarizeBackendIntegrity, $ as verifyCompletion, a0 as verifyManifest } from './pre-registration--vU0mMtD.js';
 import { T as TraceEmitter } from './emitter-BRchAAAx.js';
 export { R as RunCompleteHook, a as RunCompleteHookContext, S as SpanHandle, b as TraceEmitterOptions, l as llmSpanFromProvider } from './emitter-BRchAAAx.js';
-import { h as ReleaseConfidenceThresholds, f as ReleaseConfidenceScorecard } from './release-report-CCtzajxP.js';
-export { A as ActionableSideInfo, o as AsiSeverity, B as BootstrapOptions, a as BootstrapResult, J as JudgeReplayGateArgs, R as ReleaseConfidenceAxis, b as ReleaseConfidenceAxisName, c as ReleaseConfidenceInput, d as ReleaseConfidenceIssue, e as ReleaseConfidenceMetrics, g as ReleaseConfidenceStatus, i as ReleaseTraceEvidence, j as RenderReleaseReportOptions, V as Verdict, k as assertReleaseConfidence, l as bootstrapCi, m as evaluateReleaseConfidence, n as judgeReplayGate, r as renderReleaseReport } from './release-report-CCtzajxP.js';
+import { h as ReleaseConfidenceThresholds, f as ReleaseConfidenceScorecard } from './release-report-CamNDe90.js';
+export { A as ActionableSideInfo, o as AsiSeverity, B as BootstrapOptions, a as BootstrapResult, J as JudgeReplayGateArgs, R as ReleaseConfidenceAxis, b as ReleaseConfidenceAxisName, c as ReleaseConfidenceInput, d as ReleaseConfidenceIssue, e as ReleaseConfidenceMetrics, g as ReleaseConfidenceStatus, i as ReleaseTraceEvidence, j as RenderReleaseReportOptions, V as Verdict, k as assertReleaseConfidence, l as bootstrapCi, m as evaluateReleaseConfidence, n as judgeReplayGate, r as renderReleaseReport } from './release-report-CamNDe90.js';
 export { c as CliffsMagnitude, d as CorpusAgreementOptions, e as CorpusAgreementPerDimension, C as CorpusAgreementReport, f as CorpusScoreRecord, g as EProcess, h as EProcessOptions, E as EProcessState, i as EProcessStep, M as McNemarResult, P as PairedBootstrapOptions, a as PairedBootstrapResult, j as PairedSignTestResult, k as ProportionInterval, R as RiskDifferenceResult, S as SignTestAlternative, W as WeightedCompositeInput, l as WeightedCompositeResult, b as benjaminiHochberg, m as bonferroni, n as cliffsDelta, o as cohensD, q as confidenceInterval, r as corpusInterRaterAgreement, s as corpusInterRaterAgreementFromJudgeScores, t as eProcess, u as holm, v as interRaterReliability, x as interpretCliffs, y as mannWhitneyU, z as mcnemar, A as mcnemarPower, B as mcnemarRequiredN, D as mulberry32, F as normalizeScores, p as pairedBootstrap, G as pairedMde, H as pairedRiskDifference, I as pairedSignTest, J as pairedTTest, K as partialCredit, L as passAtK, N as pearsonR, O as ranks, Q as requiredSampleSize, T as spearmanR, U as weightedComposite, V as weightedMean, w as wilcoxonSignedRank, X as wilson } from './statistics-oUbOJe-S.js';
 import { OtelExporter, OtelExportConfig } from './traces.js';
 export { CaptureFetchContext, CaptureFetchOptions, DEFAULT_REDACTION_RULES, ExportableSpan, ExtractedUsage, FlattenOtlpOptions, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_COST_ATTR_KEYS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_MODEL_ATTR_KEYS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_OUTPUT_TOKEN_ATTR_KEYS, OPENINFERENCE_SPAN_KIND, OTEL_AGENT_EVAL_SCOPE, OtlpExport, OtlpFileTraceStore, OtlpFileTraceStoreOptions, OtlpFlatLine, OtlpResourceSpans, OtlpSpan, OtlpToRunRecordsOptions, OtlpTraceRunRecord, ProjectedOtlpSpan, REDACTION_VERSION, RedactionReport, RedactionRule, ReplayCache, ReplayCacheEntry, ReplayCacheMissError, ReplayCacheStats, ReplayFetchOptions, SPAN_KIND_ATTR_KEYS, SpanNotFoundError, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_SUBAGENT_DESCRIPTION, TraceAggregate, TraceAnalystHookOptions, TraceFileMissingError, TraceInsightContext, TraceInsightFinding, TraceInsightPanelRole, TraceInsightPromptInput, TraceInsightQualityGate, TraceInsightQuestion, TraceInsightReadiness, TraceInsightSuite, TraceInsightTask, TraceNotFoundError, TraceStoreSource, TraceStoreToOtlpOptions, TracesToOtlpResult, asNumber, asString, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, captureFetchToRawSink, convertTraceStoresToOtlp, createOtelExporter, createOtelTracingStore, createReplayFetch, defaultTraceInsightPanel, describeTraceInsightScope, domainEvidencePattern, exportRunAsOtlp, extractOtlpAttributes, extractUsage, extractUsageFromResponse, extractUsageFromSse, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, inferDomainKeywords, inferOtlpKind, iterateRawCalls, otelRunCompleteHook, otlpToRunRecords, otlpToTraceRunRecords, planTraceInsightQuestions, projectOtlpFlatLine, readOtlpStatus, redactString, redactValue, scoreTraceInsightReadiness, stringField, tokenizeDomainWords, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceSpanKindToOpenInferenceKind } from './traces.js';
-import { a as AnalyzeTracesInput, A as AnalyzeTracesOptions, b as AnalyzeTracesResult } from './analyst-C8HHvfJp.js';
-export { c as AnalyzeTracesTurnSnapshot, d as analyzeTraces } from './analyst-C8HHvfJp.js';
-import { a as TraceAnalystSpan } from './store-C1YxJDEK.js';
-export { D as DEFAULT_TRACE_ANALYST_BUDGETS, b as DatasetOverview, E as ErrorCluster, Q as QueryTracesPage, S as SearchSpanResult, c as SearchTraceResult, d as SpanMatchRecord, e as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, T as TraceAnalysisStore, f as TraceAnalystByteBudgets, g as TraceAnalystFilters, h as TraceAnalystSpanKind, i as TraceAnalystSpanStatus, j as TraceAnalystTraceSummary, V as ViewSpansResult, k as ViewTraceOversized, l as ViewTraceResult } from './store-C1YxJDEK.js';
-import { b as JudgeConfig, J as JudgeScore, S as Scenario$1, g as Gate } from './types-C5gJrOVT.js';
-import { A as AnalyzeRunsOptions } from './analyze-runs-BYHg6Irm.js';
-import { q as Objective, s as ParetoResult, h as GepaProposerConstraints, a as RunImprovementLoopResult } from './gepa-dne9JDPL.js';
-export { t as DEFAULT_RED_TEAM_CORPUS, D as Direction, e as RedTeamCase, u as RedTeamCategory, v as RedTeamFinding, w as RedTeamPayload, x as RedTeamReport, y as crowdingDistance, z as dominates, A as paretoFrontier, B as paretoFrontierWithCrowding, E as redTeamDataset, F as redTeamReport, H as scalarScore, I as scoreRedTeamOutput, J as toolNamesForRun } from './gepa-dne9JDPL.js';
+import { a as AnalyzeTracesInput, A as AnalyzeTracesOptions, b as AnalyzeTracesResult } from './analyst-CFBc14Wc.js';
+export { c as AnalyzeTracesTurnSnapshot, d as analyzeTraces } from './analyst-CFBc14Wc.js';
+import { a as TraceAnalystSpan } from './store-9cAScOcb.js';
+export { D as DEFAULT_TRACE_ANALYST_BUDGETS, b as DatasetOverview, E as ErrorCluster, F as FileSystemRawProviderSink, c as FileSystemRawProviderSinkOptions, I as InMemoryRawProviderSink, d as InMemoryRawProviderSinkOptions, N as NoopRawProviderSink, P as ProviderRedactor, Q as QueryTracesPage, e as RawProviderDirection, f as RawProviderEvent, R as RawProviderSink, g as RawProviderSinkFilter, S as SearchSpanResult, h as SearchTraceResult, i as SpanMatchRecord, j as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, T as TraceAnalysisStore, k as TraceAnalystByteBudgets, l as TraceAnalystFilters, m as TraceAnalystSpanKind, n as TraceAnalystSpanStatus, o as TraceAnalystTraceSummary, V as ViewSpansResult, p as ViewTraceOversized, q as ViewTraceResult, r as defaultProviderRedactor, s as providerFromBaseUrl } from './store-9cAScOcb.js';
+import { b as JudgeConfig, J as JudgeScore, S as Scenario$1, g as Gate } from './types-Ca_63YSD.js';
+import { A as AnalyzeRunsOptions } from './analyze-runs-0rz_m29H.js';
+import { q as Objective, s as ParetoResult, h as GepaProposerConstraints, a as RunImprovementLoopResult } from './gepa-CQelRtuC.js';
+export { t as DEFAULT_RED_TEAM_CORPUS, D as Direction, e as RedTeamCase, u as RedTeamCategory, v as RedTeamFinding, w as RedTeamPayload, x as RedTeamReport, y as crowdingDistance, z as dominates, A as paretoFrontier, B as paretoFrontierWithCrowding, E as redTeamDataset, F as redTeamReport, H as scalarScore, I as scoreRedTeamOutput, J as toolNamesForRun } from './gepa-CQelRtuC.js';
 import { S as SandboxDriver, H as HarnessConfig, a as SandboxHarnessResult } from './test-graded-scenario-mzYBKspu.js';
 export { D as DockerSandboxDriver, c as SandboxHarness, d as SandboxResult, e as SubprocessSandboxDriver, f as SubprocessSandboxDriverOptions, g as TestGradedRunOptions, b as TestGradedRunResult, T as TestGradedScenario, h as TestOutputParser, i as composeParsers, j as jestTestParser, p as pytestTestParser, r as runTestGradedScenario, v as vitestTestParser } from './test-graded-scenario-mzYBKspu.js';
-export { b as RunIntegrityError, R as RunIntegrityExpectations, c as RunIntegrityIssue, d as RunIntegrityIssueCode, a as RunIntegrityReport, e as assertRunCaptured, t as throwIfRunIncomplete } from './integrity-qemeBAyx.js';
+export { b as RunIntegrityError, R as RunIntegrityExpectations, c as RunIntegrityIssue, d as RunIntegrityIssueCode, a as RunIntegrityReport, e as assertRunCaptured, t as throwIfRunIncomplete } from './integrity-C6PZ73iC.js';
 export { a as aggregateLlm, b as argHash, g as groupBy, j as judgeSpans, l as llmSpans, r as runFailureClass, c as runsForScenario, t as toolSpans } from './query-Ck190MOd.js';
-export { F as FileSystemRawProviderSink, a as FileSystemRawProviderSinkOptions, I as InMemoryRawProviderSink, b as InMemoryRawProviderSinkOptions, N as NoopRawProviderSink, P as ProviderRedactor, c as RawProviderDirection, d as RawProviderEvent, R as RawProviderSink, e as RawProviderSinkFilter, f as defaultProviderRedactor, p as providerFromBaseUrl } from './raw-provider-sink-C46HDghv.js';
 import { T as TraceStore, R as RunFilter } from './store-BsVi7ncX.js';
 export { E as EventFilter, F as FileSystemTraceStore, a as FileSystemTraceStoreOptions, I as InMemoryTraceStore, S as SpanFilter } from './store-BsVi7ncX.js';
 export { D as DEFAULT_FAILURE_RULES, b as FailureClassification, c as FailureContext, d as FailureRule, e as classifyFailure } from './failure-cluster-C48PiReX.js';
-export { P as ProjectRuntimeTrajectoryEvidenceOptions, a as RuntimeTrajectoryEvidenceProjection, b as RuntimeTrajectoryEvidenceSummary, c as RuntimeTrajectoryHookEvent, R as RuntimeTrajectoryRecord, d as RuntimeTrajectoryRunRecord, p as parseRuntimeTrajectoryHookEvent, e as projectRuntimeTrajectoryEvidence } from './runtime-trajectory-Dws7Kpgi.js';
+export { P as ProjectRuntimeTrajectoryEvidenceOptions, a as RuntimeTrajectoryEvidenceProjection, b as RuntimeTrajectoryEvidenceSummary, c as RuntimeTrajectoryHookEvent, R as RuntimeTrajectoryRecord, d as RuntimeTrajectoryRunRecord, p as parseRuntimeTrajectoryHookEvent, e as projectRuntimeTrajectoryEvidence } from './runtime-trajectory-CC0jx9ql.js';
 import { a as BaselineReport, b as Trajectory, T as TrajectoryStep } from './baseline-DsNteOgR.js';
 export { B as BaselineOptions, M as MetricSamples, d as MetricVerdict, e as ToolStats, f as ToolUseMetrics, g as ToolUseOptions, h as buildTrajectory, i as compareToBaseline, c as computeToolUseMetrics, j as iqr, w as welchsTTest } from './baseline-DsNteOgR.js';
 import { HarnessType, AgentProfile } from '@tangle-network/agent-interface';
@@ -55,15 +54,13 @@ export { d as DatasetDifficulty, b as DatasetManifest, e as DatasetProvenance, a
 export { a as CalibrationResult, c as CandidateScore, C as ContinuousAgreement, d as ContinuousAgreementOptions, b as ContinuousCalibrationResult, G as GoldenItem, P as PositionalBiasResult, S as SelfPreferenceResult, V as VerbosityBiasResult, e as calibrateJudge, f as calibrateJudgeContinuous, g as continuousAgreement, p as positionalBias, s as selfPreference, v as verbosityBias } from './judge-calibration-7C-IDmKr.js';
 import { L as Layer, S as Severity, b as LayerResult, c as VerifyContext } from './multi-layer-verifier-BsqKuLyN.js';
 export { F as Finding, d as LayerStatus, M as MultiLayerVerifier, a as VerificationReport, V as VerifyOptions, g as gradeSemanticStatus } from './multi-layer-verifier-BsqKuLyN.js';
-import { L as LlmClientOptions } from './llm-client-DyqEH4jH.js';
-export { d as LlmCallError, b as LlmCallRequest, c as LlmCallResult, e as LlmClient, f as LlmMessage, g as LlmRouteAssertionError, a as LlmRouteRequirements, h as LlmUsage, i as assertLlmRoute, j as backoffMs, k as callLlm, l as callLlmJson, m as isTransientLlmError, p as probeLlm, s as stripFencedJson } from './llm-client-DyqEH4jH.js';
-export { B as BENCHMARK_SPLIT_SEED, a as BenchmarkAdapter, b as BenchmarkDatasetItem, c as BenchmarkEvaluation, d as BenchmarkFamily, e as BenchmarkResponder, f as BenchmarkScenario, g as BenchmarkSource, h as BenchmarkTaskKind, i as benchmarkDeterministicSplit, j as benchmarks } from './index-BTEpx9He.js';
-export { C as CallbackResearcher, d as CallbackResearcherOptions, e as CampaignFactoryParams, f as CampaignIntegrityPolicy, g as CampaignRunContext, h as CampaignRunOutcome, i as CampaignRunner, j as CampaignScenario, k as CampaignVariant, c as EvalCampaignOptions, b as EvalCampaignResult, E as ExperimentPlan, a as ExperimentResult, l as FailedRun, F as FailureMode, N as NoopResearcher, R as Researcher, S as SteeringChange, r as runEvalCampaign } from './researcher-Dq-EtpbE.js';
-export { G as GainDistributionBin, a as GainDistributionFigureSpec, b as GainDistributionOptions, m as GateDecision, n as GateEvidence, H as HeldOutGate, o as HeldOutGateConfig, q as HeldOutGateRejectionCode, P as ParetoFigureSpec, c as ParetoPoint, R as RESEARCH_REPORT_HARD_PAIR_FLOOR, d as ResearchReport, e as ResearchReportCandidate, f as ResearchReportDecision, g as ResearchReportMethodology, h as ResearchReportOptions, i as ResearchReportRecommendation, S as SummaryTable, j as SummaryTableOptions, k as SummaryTableRow, l as gainHistogram, p as paretoChart, r as researchReport, s as summaryTable } from './summary-report-BJ5aNwZ1.js';
+export { B as BENCHMARK_SPLIT_SEED, a as BenchmarkAdapter, b as BenchmarkDatasetItem, c as BenchmarkEvaluation, d as BenchmarkFamily, e as BenchmarkResponder, f as BenchmarkScenario, g as BenchmarkSource, h as BenchmarkTaskKind, i as benchmarkDeterministicSplit, j as benchmarks } from './index-DbCXJfZ1.js';
+export { C as CallbackResearcher, d as CallbackResearcherOptions, e as CampaignFactoryParams, f as CampaignIntegrityPolicy, g as CampaignRunContext, h as CampaignRunOutcome, i as CampaignRunner, j as CampaignScenario, k as CampaignVariant, c as EvalCampaignOptions, b as EvalCampaignResult, E as ExperimentPlan, a as ExperimentResult, l as FailedRun, F as FailureMode, N as NoopResearcher, R as Researcher, S as SteeringChange, r as runEvalCampaign } from './researcher-Dwbo_Fxx.js';
+export { G as GainDistributionBin, a as GainDistributionFigureSpec, b as GainDistributionOptions, m as GateDecision, n as GateEvidence, H as HeldOutGate, o as HeldOutGateConfig, q as HeldOutGateRejectionCode, P as ParetoFigureSpec, c as ParetoPoint, R as RESEARCH_REPORT_HARD_PAIR_FLOOR, d as ResearchReport, e as ResearchReportCandidate, f as ResearchReportDecision, g as ResearchReportMethodology, h as ResearchReportOptions, i as ResearchReportRecommendation, S as SummaryTable, j as SummaryTableOptions, k as SummaryTableRow, l as gainHistogram, p as paretoChart, r as researchReport, s as summaryTable } from './summary-report-DTNgQycC.js';
 export { I as InterimReleaseConfidence, a as InterimReleaseConfidenceInput, P as PairedEvalueOptions, b as PairedEvalueSequence, c as PairedEvalueStep, S as SequentialDecision, e as evaluateInterimReleaseConfidence, p as pairedEvalueSequence } from './sequential-5iSVfzl2.js';
 import '@ax-llm/ax';
 import 'zod';
-import './insight-report-IwwvqZZv.js';
+import './insight-report-oMVxDTxl.js';
 import './storage-Dw_f7WMt.js';
 
 /**

package/dist/index.js

@@ -61,7 +61,7 @@ import {
   pairArms,
   parseCorrectnessResponse,
   verifyCompletion
-} from "./chunk-5NVBGKPH.js";
+} from "./chunk-GSW3OBHK.js";
 import {
   DEFAULT_MUTATION_PRIMITIVES,
   DEFAULT_RED_TEAM_CORPUS,
@@ -85,7 +85,7 @@ import {
   scoreRedTeamOutput,
   surfaceContentHash,
   toolNamesForRun
-} from "./chunk-N6MTC3GK.js";
+} from "./chunk-3274WNK7.js";
 import {
   MODEL_PRICING,
   MetricsCollector,
@@ -119,29 +119,17 @@ import {
   defaultIsMaterial,
   diffFindings,
   runSemanticConceptJudge
-} from "./chunk-I2HNIE6N.js";
+} from "./chunk-NBSS5NDZ.js";
 import {
   buildDefaultAnalystRegistry,
   computeTraceMetrics
-} from "./chunk-LVTGFSHF.js";
+} from "./chunk-7GKEAIAD.js";
 import {
   DEFAULT_RUN_SCORE_WEIGHTS,
   Mutex,
-  POLICY_EDIT_AXES,
-  POLICY_EDIT_TARGET_SURFACES,
-  PolicyEditValidationError,
-  admitPolicyEdit,
   aggregateRunScore,
-  applyPolicyEditToSurface,
-  clamp01,
-  computePolicyEditId,
-  isPolicyEdit,
-  makePolicyEdit,
-  policyEditFromFinding,
-  policyEditsFromFindings,
-  scorePolicyEditReadiness,
-  validatePolicyEdit
-} from "./chunk-AN5UYSVD.js";
+  clamp01
+} from "./chunk-MPHTT5HE.js";
 import {
   AnalystRegistry,
   DEFAULT_TRACE_ANALYST_KINDS,
@@ -149,11 +137,26 @@ import {
   IMPROVEMENT_KIND_SPEC,
   KNOWLEDGE_GAP_KIND_SPEC,
   KNOWLEDGE_POISONING_KIND_SPEC,
+  POLICY_EDIT_AXES,
+  POLICY_EDIT_CANDIDATE_RECORD_SCHEMA,
+  POLICY_EDIT_TARGET_SURFACES,
+  PolicyEditValidationError,
+  admitPolicyEdit,
+  applyPolicyEditToSurface,
   computeFindingId,
+  computePolicyEditId,
   createTraceAnalystKind,
+  isPolicyEdit,
   makeFinding,
-  renderPriorFindings
-} from "./chunk-DWLIGZBX.js";
+  makePolicyEdit,
+  makePolicyEditCandidateRecord,
+  policyEditFromFinding,
+  policyEditsFromFindings,
+  renderPriorFindings,
+  scorePolicyEditReadiness,
+  validatePolicyEdit,
+  validatePolicyEditCandidateRecord
+} from "./chunk-CIUOICJT.js";
 import {
   allCriticalPassed,
   controlFailureClassFromVerification,
@@ -184,7 +187,7 @@ import {
 } from "./chunk-MOXWMGPC.js";
 import {
   runEvalCampaign
-} from "./chunk-QG5F6463.js";
+} from "./chunk-ONM6PEAE.js";
 import "./chunk-ARU2PZFM.js";
 import {
   evaluateInterimReleaseConfidence,
@@ -381,7 +384,7 @@ import {
   isTransientLlmError,
   probeLlm,
   stripFencedJson
-} from "./chunk-FUCQVFMU.js";
+} from "./chunk-GY4SYVPJ.js";
 import {
   FileSystemRawProviderSink,
   InMemoryRawProviderSink,
@@ -11523,6 +11526,7 @@ export {
   OTEL_AGENT_EVAL_SCOPE,
   OtlpFileTraceStore,
   POLICY_EDIT_AXES,
+  POLICY_EDIT_CANDIDATE_RECORD_SCHEMA,
   POLICY_EDIT_TARGET_SURFACES,
   PairwiseSteeringOptimizer,
   PolicyEditValidationError,
@@ -11832,6 +11836,7 @@ export {
   makeEvalTools,
   makeFinding,
   makePolicyEdit,
+  makePolicyEditCandidateRecord,
   mannWhitneyU,
   matchGoldens,
   matchSpan,
@@ -12009,6 +12014,7 @@ export {
   userQuestionsForKnowledgeGaps,
   validateAgentProfileCell,
   validatePolicyEdit,
+  validatePolicyEditCandidateRecord,
   validateProductBenchmarkManifest,
   validateProductBenchmarkRecord,
   validateProductBenchmarkRun,