@remnic/bench 9.6.32 → 9.6.33

package/dist/index.d.ts

@@ -2152,6 +2152,7 @@ declare function createCodexCliProvider(config: CodexCliProviderConfig, deps?: C
 declare function redactBenchmarkResultSecrets<T>(value: T): T;
 declare function writeBenchmarkResult(result: BenchmarkResult, outputDir: string): Promise<string>;
 declare function getRemnicVersion(): Promise<string>;
+declare function getGitSha(): string;
 
 interface DiscoverAllProvidersOptions {
     includeCodexCli?: boolean;
@@ -3322,6 +3323,148 @@ declare function sanitizeLoComoResultReference(path: string): string;
 declare function diagnoseLoComoRecallDelta(options: DiagnoseLoComoRecallDeltaOptions): LoComoRecallDeltaReport;
 declare function renderLoComoRecallDeltaMarkdown(report: LoComoRecallDeltaReport): string;
 
+interface LoCoMoQA {
+    question: string;
+    answer: string;
+    evidence: string[];
+    category: number;
+}
+interface LoCoMoConversation {
+    sample_id: string;
+    conversation: Record<string, unknown>;
+    qa: LoCoMoQA[];
+    event_summary?: unknown;
+    observation?: unknown;
+    session_summary?: unknown;
+}
+
+/**
+ * LoCoMo runner migrated into @remnic/bench for phase 1.
+ *
+ * As of issue #566 PR 2/7, the per-item lifecycle (reset → ingest →
+ * recall → answer → judge → score) lives in `../harness.ts`. This
+ * module only knows about dataset loading, session extraction, and
+ * how to translate a `LoCoMoConversation` into a `HarnessPlan`.
+ */
+
+interface LoCoMoContentDigest {
+    sha256: string;
+    charCount: number;
+    lineCount: number;
+}
+interface LoCoMoCompositionLineReceipt {
+    inputOrdinal: number;
+    input: LoCoMoContentDigest;
+    output: LoCoMoContentDigest;
+    stage: "direct" | "linked";
+    hop?: number;
+    visible: boolean;
+    outputStart: number;
+    outputEnd: number;
+    visibleStart: number;
+    visibleEnd: number;
+}
+interface LoCoMoRecallCompositionReceipt {
+    schemaVersion: 1;
+    mode: "focused" | "fallback";
+    multiHopRecallComposition: boolean;
+    input: LoCoMoContentDigest;
+    output: LoCoMoContentDigest;
+    selectedLines: LoCoMoCompositionLineReceipt[];
+}
+
+declare const LOCOMO_RETRIEVAL_TRACE_SCHEMA_VERSION: 1;
+declare const LOCOMO_RETRIEVAL_TRACE_SELECTION_VERSION: 1;
+declare const LOCOMO_RETRIEVAL_TRACE_BUDGET_VERSION: 1;
+type LoCoMoRetrievalTraceProfile = "baseline" | "real";
+type LoCoMoRetrievalTraceSelector = {
+    taskIds: readonly string[];
+    sampleSize?: never;
+    seed?: never;
+} | {
+    taskIds?: never;
+    sampleSize: number;
+    seed: number;
+};
+interface LoCoMoRetrievalTraceSelectionManifest {
+    algorithm: "explicit-task-ids" | "sha256-seeded-sample";
+    version: typeof LOCOMO_RETRIEVAL_TRACE_SELECTION_VERSION;
+    seed?: number;
+    candidateCount: number;
+    selectedCount: number;
+    selectedTaskIds: string[];
+    selectedTaskIdsSha256: string;
+}
+interface LoCoMoRetrievalTraceCoreCaptureReceipt {
+    budget: BenchRecallTraceCoreCapture["budget"];
+    filters: BenchRecallTraceCoreCapture["filters"];
+    results: Array<Pick<BenchRecallTraceCoreCapture["results"][number], "memoryIdRef" | "servedBy" | "scoreDecomposition" | "admittedBy" | "rejectedBy" | "disclosure" | "estimatedTokens">>;
+}
+interface LoCoMoRetrievalTraceSelectionReceipt extends Omit<BenchRecallTraceSelection, "summary"> {
+    summary?: Omit<NonNullable<BenchRecallTraceSelection["summary"]>, "id">;
+}
+interface LoCoMoRetrievalStructuralTrace extends Omit<BenchRecallTrace, "coreCapture" | "selections"> {
+    selections: LoCoMoRetrievalTraceSelectionReceipt[];
+    coreCapture?: LoCoMoRetrievalTraceCoreCaptureReceipt;
+}
+interface LoCoMoRetrievalSessionReceipt {
+    session: LoCoMoContentDigest;
+    trace: LoCoMoRetrievalStructuralTrace;
+}
+interface LoCoMoRetrievalTaskReceipt {
+    taskId: string;
+    question: LoCoMoContentDigest;
+    recallBudgetChars: number;
+    sessions: LoCoMoRetrievalSessionReceipt[];
+    composition: LoCoMoRecallCompositionReceipt;
+}
+interface LoCoMoRetrievalTraceReceipt {
+    schemaVersion: typeof LOCOMO_RETRIEVAL_TRACE_SCHEMA_VERSION;
+    benchmarkId: "locomo";
+    captureKind: "retrieval-only";
+    artifactHash: string;
+    sensitivity: {
+        classification: "restricted";
+        contentEncoding: "sha256+length";
+        containsGold: false;
+        containsRawContent: false;
+    };
+    provenance: {
+        gitSha: string;
+        remnicVersion: string;
+        runtimeProfile: LoCoMoRetrievalTraceProfile;
+        adapterMode: "direct";
+        replayExtractionMode: "skip";
+        providerFree: true;
+        dataset: {
+            id: "locomo-10";
+            sha256: string;
+        };
+        retrievalConfigSha256: string;
+        recallBudget: {
+            algorithm: "benchmarkRecallBudgetForSessionCount";
+            version: typeof LOCOMO_RETRIEVAL_TRACE_BUDGET_VERSION;
+        };
+    };
+    selection: LoCoMoRetrievalTraceSelectionManifest;
+    tasks: LoCoMoRetrievalTaskReceipt[];
+}
+interface CaptureLoCoMoRetrievalTraceOptions {
+    datasetDir: string;
+    runtimeProfile: LoCoMoRetrievalTraceProfile;
+    system: BenchMemoryAdapter;
+    retrievalConfig: Record<string, unknown>;
+    selector: LoCoMoRetrievalTraceSelector;
+    gitSha: string;
+    remnicVersion: string;
+    multiHopRecallComposition?: boolean;
+    providerFreeConfirmed: true;
+}
+declare function preflightLoCoMoRetrievalTraceCapture(options: Omit<CaptureLoCoMoRetrievalTraceOptions, "system">): Promise<void>;
+declare function buildProviderFreeLoCoMoRetrievalConfig(retrievalConfig: Record<string, unknown>): Record<string, unknown>;
+declare function captureLoCoMoRetrievalTrace(options: CaptureLoCoMoRetrievalTraceOptions): Promise<LoCoMoRetrievalTraceReceipt>;
+declare function serializeLoCoMoRetrievalTraceReceipt(receipt: LoCoMoRetrievalTraceReceipt): string;
+
 /**
  * Dataset-contamination guard.
  *
@@ -3507,21 +3650,6 @@ interface LongMemEvalItem {
     answer_session_ids: string[];
 }
 
-interface LoCoMoQA {
-    question: string;
-    answer: string;
-    evidence: string[];
-    category: number;
-}
-interface LoCoMoConversation {
-    sample_id: string;
-    conversation: Record<string, unknown>;
-    qa: LoCoMoQA[];
-    event_summary?: unknown;
-    observation?: unknown;
-    session_summary?: unknown;
-}
-
 /**
  * Shared dataset loader helpers for the published LongMemEval + LoCoMo
  * benchmark runners. Wraps the fs probe + JSON parse + fallback logic
@@ -3552,6 +3680,8 @@ interface LoadedDataset<T> {
     source: DatasetSource;
     /** Filename relative to `datasetDir` when source === "dataset". */
     filename?: string;
+    /** SHA-256 of the exact dataset file, or canonical bundled smoke fixture. */
+    sha256?: string;
     items: T[];
     /** Parse/read errors encountered while probing candidate filenames. */
     errors: string[];
@@ -5408,4 +5538,4 @@ declare function checkCodingGraphRegression(report: CodingGraphBenchReport, base
  */
 declare function buildBaselineFromReport(report: CodingGraphBenchReport, note: string): CodingGraphBaseline;
 
-export { AMA_BENCH_DIAGNOSTIC_VARIANTS, ASSISTANT_AGENT_CONFIG_KEY, ASSISTANT_JUDGE_CONFIG_KEY, ASSISTANT_MEETING_PREP_SCENARIOS, ASSISTANT_MEETING_PREP_SMOKE_SCENARIOS, ASSISTANT_MORNING_BRIEF_SCENARIOS, ASSISTANT_MORNING_BRIEF_SMOKE_SCENARIOS, ASSISTANT_NEXT_BEST_ACTION_SCENARIOS, ASSISTANT_NEXT_BEST_ACTION_SMOKE_SCENARIOS, ASSISTANT_RUBRIC_DIMENSIONS, ASSISTANT_RUBRIC_ID_KEY, ASSISTANT_SEEDS_CONFIG_KEY, ASSISTANT_SPOT_CHECK_DIR_KEY, ASSISTANT_SYNTHESIS_SCENARIOS, ASSISTANT_SYNTHESIS_SMOKE_SCENARIOS, type AblationConfigOverrides, type AbstentionRetrievalCase, type AggregateMetrics, type AmaBenchDiagnosticAdapterOptions, type AmaBenchDiagnosticAnswererMode, type AmaBenchDiagnosticBreakdown, type AmaBenchDiagnosticMatrixArtifact, type AmaBenchDiagnosticRecallMode, type AmaBenchDiagnosticRunContext, type AmaBenchDiagnosticTaskEvidence, type AmaBenchDiagnosticTaskRow, type AmaBenchDiagnosticVariant, type AmaBenchDiagnosticVariantSummary, type AnthropicProviderConfig, type AssistantAgent, type AssistantMemoryFact, type AssistantMemoryGraph, type AssistantRubricDimension, type AssistantRubricRequest, type AssistantRubricScores, type AssistantRunnerOptions, type AssistantScenario, type AssistantStance, type AttackRecallOptions, type AttackRetrievalHit, type AttackerMode, BENCHMARK_ARTIFACT_SCHEMA_VERSION, BENCHMARK_INTEGRITY_META_SCHEMA, BENCHMARK_REPRO_MANIFEST_FILENAME, BENCHMARK_REPRO_MANIFEST_SCHEMA_VERSION, BENCHMARK_RESULT_SCHEMA, BENCHMARK_SPLIT_TYPES, type BaselineRow, type BaselineScenario, type BeamDatasetPreview, type BenchConfig, type BenchJudge, type BenchJudgeResult, type BenchMemoryAdapter, type BenchModelSource, type BenchReasoningEffort, type BenchRecallLineageStatus, type BenchRecallOptions, type BenchRecallSupportAssessment, type BenchRecallSupportRequest, type BenchRecallSupportStatus, type BenchRecallTrace, type BenchRecallTraceCoreCapture, type BenchRecallTraceLcmCandidate, type BenchRecallTraceRange, type BenchRecallTraceSection, type BenchRecallTraceSelection, type BenchRecallWithTraceResult, type BenchResponder, type BenchResponse, type BenchRuntimeProfile, type BenchTier, type BenchmarkArtifact, type BenchmarkArtifactEnvironment, type BenchmarkArtifactHardware, type BenchmarkArtifactJudgeCalibration, type BenchmarkArtifactPerTaskScore, type BenchmarkArtifactSystem, type BenchmarkArtifactTier, type BenchmarkCategory, type BenchmarkDefinition, type BenchmarkIntegrityMeta, type BenchmarkMeta, type BenchmarkMode, type BenchmarkReport, type BenchmarkReproManifest, type BenchmarkReproManifestDataset, type BenchmarkReproManifestFile, type BenchmarkReproManifestResult, type BenchmarkResult, type BenchmarkSplitType, type BenchmarkStatus, type BenchmarkSuiteResult, type BenchmarkTier, type BootstrapKappaOptions, type BootstrapKappaResult, type BuildBenchmarkArtifactInput, type BuildBenchmarkPublishFeedOptions, type BuildBenchmarkReproManifestOptions, type BuiltInProvider, CALIBRATION_SLICE_SIZE, CANARY_FIXED_RECALL, CANARY_SCORE_FLOOR, DEFAULT_10K_FIXTURE as CODING_GRAPH_10K_FIXTURE, CODING_GRAPH_BENCH_SCHEMA_VERSION, DEFAULT_TOLERANCE_PERCENT as CODING_GRAPH_DEFAULT_TOLERANCE, MIN_ITERATIONS as CODING_GRAPH_MIN_ITERATIONS, DEFAULT_SMOKE_FIXTURE as CODING_GRAPH_SMOKE_FIXTURE, type CalibrationAnswer, type CalibrationVerdictPair, type CanaryAdapterOptions, type CanaryFloorCheck, type ClaudeCliProviderConfig, type CodexCliProviderConfig, type CodexCreditReceipt, type CodexCreditReceiptScope, type CodexCreditReconciliationReceipt, type CodingGraphBaseline, type CodingGraphBenchConfig, type CodingGraphBenchReport, type MachineFingerprint as CodingGraphMachineFingerprint, type CodingGraphMetricKey, type RegressionMetricDetail as CodingGraphRegressionDetail, type RegressionMetricKey as CodingGraphRegressionKey, type RegressionGateResult as CodingGraphRegressionResult, type CohenKappaResult, type ComparisonMetricDelta, type ComparisonResult, type CompletionOpts, type CompletionResult, type ConfidenceInterval, type ContaminationCheckResult, type ContaminationEntry, type ContaminationManifest, type CustomBenchmarkScoring, type CustomBenchmarkSpec, type CustomBenchmarkTask, DEFAULT_ABLATION_BENCHMARK, DEFAULT_ABLATION_BOOTSTRAP_SEED, DEFAULT_ASSISTANT_RUBRIC_ID, DEFAULT_BASELINE_SCENARIOS, DEFAULT_JUDGE_BINARIZATION_THRESHOLD, DEFAULT_KAPPA_BOOTSTRAP_SAMPLES, DEFAULT_KAPPA_CONFIDENCE_LEVEL, DEFAULT_OPENAI_RESPONSES_JUDGE_MODEL, type DatasetSource, type DiagnoseLoComoProfileDeltaOptions, type DiagnoseLoComoRecallDeltaOptions, type DiscoveredModel, EMPTY_CONTAMINATION_MANIFEST, type EffectSizeInterpretation, type EffectSizeSummary, type ExplainResult, type ExtractedEntity, type ExtractedLink, type ExtractedPage, type ExtractionAttackOptions, type ExtractionAttackResult, type ExtractionAttackTarget, type FixtureGenerator, type FixtureOutput, type FixtureVariant, GENERAL_ANSWER_JUDGE_RUBRIC, type GeneratedFile, type GeneratedRepo, type GoldEntity, type GoldEntityType, type GoldGraph, type GoldLink, type GoldPage, type HarnessRng, INTEGRITY_CIPHER_ALGORITHM, INTEGRITY_HASH_ALGORITHM, INTEGRITY_META_FIELDS, type IngestionBenchAdapter, type IngestionLog, JUDGE_CALIBRATION_KAPPA_THRESHOLD, JUDGE_CALIBRATION_PROTOCOL_VERSION, type JudgeCalibrationCheckpointProvenance, type JudgeCalibrationIdentities, type JudgeCalibrationResult, type JudgeCategory, type KappaConfidenceInterval, LOCAL_LAB_PROVIDER_KINDS, LOCOMO_DATASET_FILENAMES, LOCOMO_FULL_TASK_COUNT, LOCOMO_RECALL_DIFF_LINE_LIMIT, LOCOMO_RECALL_EXCERPT_CHARS, LONG_MEM_EVAL_DATASET_FILENAMES, type LeaderboardArtifactWrite, type LettaAdapterConfig, LettaMemCorrectAdapter, type LlmJudge, type LlmProvider, type LoComoCategoryDelta, type LoComoFinalContextRegression, type LoComoMetricDelta, type LoComoProfileArtifactEvidence, type LoComoProfileDeltaReport, type LoComoRawResultEvidence, type LoComoRecallCategoryDelta, type LoComoRecallContextSummary, type LoComoRecallDeltaReport, type LoComoRecallLineDelta, type LoComoRecallLineEvidence, type LoComoRecallMetricDelta, type LoComoRecallResultProvenance, type LoComoRecallTextDigest, type LoComoTaskRegression, type LoadDatasetOptions, type LoadSealedQrelsOptions, type LoadedDataset, type LoadedJudgeCalibrationState, type LocalLabManifest, type LocalLabManifestNotes, type LocalLabPhase, type LocalLabPhaseDescriptor, type LocalLabPhaseExecute, type LocalLabPhaseName, type LocalLabPhaseOutcome, LocalLabPreflightError, type LocalLabPreflightFailure, type LocalLabPreflightInput, type LocalLabPreflightOptions, type LocalLabPreflightResult, type LocalLabPreflightSuccess, type LocalLabProviderKind, type LocalLabRoleConfig, type LocalLlmProviderConfig, MEMCORRECT_CORRECTION_ACCEPTANCE_RUBRIC, MEMCORRECT_CORRECTION_ACCEPTANCE_RUBRIC_VERSION, MEMCORRECT_STALE_HARM_RUBRIC, MEMCORRECT_STALE_HARM_RUBRIC_VERSION, MEMORY_EVAL_DIMENSIONS, MEMORY_EVAL_PUBLIC_LINE, MIN_CALIBRATION_SOURCE_TASKS, MITIGATED_BASELINE_SCENARIOS, type McpArgumentSemantic, type McpBackendErrorCode, type McpBackendResult, type McpBenchMemoryAdapter, type McpConformanceResult, type McpHttpTransportConfig, type McpListedTool, type McpMemCorrectAdapter, type McpMemoryAdapterOptions, McpMemoryBackendError, type McpMemoryToolMapping, type McpMemoryTransportConfig, type McpStdioTransportConfig, type McpToolCallResult, type McpToolClient, type McpToolMappingEntry, type McpToolMappingValue, type McpToolOperation, type Mem0AdapterConfig, Mem0MemCorrectAdapter, type MemCorrectGeneratorOptions, type MemCorrectJudgeRequest, type MemCorrectJudgeResult, type MemCorrectSystemAdapter, type MemoryEvalCategory, type MemoryEvalDimension, type MemoryEvalDimensionId, type MemoryEvalMetric, type MemoryGraph, type MemoryStats, type MemorySystem, type Message, type MetricAggregate, type MicroMetric, MissingCredentialError, type MitigatedBaselineConfig, type MitigatedTargetConfig, type MultipleChoiceQuestion, OPENAI_RESPONSES_JUDGE_RUBRIC_VERSION, OTHER_NAMESPACE_MEMORIES, type OllamaProviderConfig, type OpenAiCompatibleProviderConfig, OpenAiResponsesJudgeError, type OpenAiResponsesJudgeErrorCode, type OpenAiResponsesJudgeTelemetry, OpenAiResponsesProvider, type OpenAiResponsesProviderConfig, type OpenAiResponsesVerdict, type OpenAiResponsesVerdictResult, PROCEDURAL_REAL_SCENARIOS, PROCEDURAL_REAL_SCENARIOS_SMOKE, PUBLISHED_BENCHMARK_ARTIFACT_IDS, type PersonalizationRetrievalCase, type PreflightDiscoveredModel, type ProceduralAblationArtifact, type ProceduralAblationPerCase, type ProceduralAblationScenario, type ProceduralRealScenario, type ProceduralRealScenarioCategory, type ProviderBaseConfig, type ProviderConfig, type ProviderDiscoveryResult, type ProviderFactoryConfig, type PublishSkipReason, type PublishSkipRecord, type PublishedBenchmarkFeed, type PublishedBenchmarkFeedEntry, type PublishedBenchmarkId, REQUIRED_FRONTMATTER_FIELDS, type RecallMetrics, type RecoveredMemory, type RegressionDetail, type RegressionGateResult$1 as RegressionGateResult, type RemnicAdapterOptions, type ReportCardProvenanceContext, type ResolveBenchRuntimeProfileOptions, type ResolvedBenchRuntimeProfile, type ResolvedLocalLabProfile, type ResolvedLocalLabRole, type ResolvedRunBenchmarkOptions, type RotatedChoices, type RunBenchmarkOptions, type RunJudgeCalibrationOptions, type RunProceduralAblationCliArgs, type RunProceduralAblationOptions, type RunSequentialPhasesOptions, SCHEMA_TIER_FIXTURE, SCHEMA_TIER_SMOKE_FIXTURE, SEALED_PROMPT_REGISTRY, SINGLE_FLAG_ABLATION_MATRIX, SYNTHETIC_MEMORIES, type SanitizedDiagnosticProvider, type SavedBaseline, type SchemaTierCorpus, type SchemaTierFixture, type SchemaTierName, type SchemaTierPage, type SchemaTierPageFrontmatter, type SealedArtifact, type SealedJudgeDecision, type SealedJudgeInput, type SealedQrelsArtifact, type SealedQrelsHandle, type SealedRubric, type SearchResult, type SeededMemory, type SeededRng, type SequentialPhaseHooks, type SingleFlagAblationCell, type SingleFlagAblationId, type SpotCheckLogger, type StatisticalReport, type StructuredJudge, StructuredJudgeError, type StructuredJudgeErrorCode, type StructuredJudgeProvider, type StructuredJudgeTelemetry, type StructuredJudgeVerdict, type StructuredJudgeVerdictResult, type StructuredVerdictRequest, type SyntheticEdge, type SyntheticEmailIngestionAdapterOptions, type SyntheticFileIR, type SyntheticRepoConfig, type SyntheticSymbol, type SyntheticTargetOptions, type TaskResult, type TaskTokenUsage, type TemporalRetrievalCase, type ThirdPartyAdapterConfig, type TierDetail, type TimelineEntry, type TokenUsage, type WallMetric, type WriteBenchmarkArtifactResult, type ZepAdapterConfig, ZepMemCorrectAdapter, addContaminationEntry, aggregateTaskScores, answerBenchmarkQuestion, assertCanaryUnderFloor, assertIntegrityMetaPresent, assertPublishableIntegrity, assertSha256Hex, assistantMeetingPrepDefinition, assistantMorningBriefDefinition, assistantNextBestActionDefinition, assistantSynthesisDefinition, backlinkF1, binarizeJudgeScore, bootstrapCohensKappaConfidenceInterval, bootstrapMeanConfidenceInterval, buildAmaBenchDiagnosticMatrixArtifact, buildAmaBenchDiagnosticVariantSummary, buildAmaBenchLeaderboardRows, buildBaselineFromReport, buildBenchmarkArtifact, buildBenchmarkArtifactFilename, buildBenchmarkPublishFeed, buildBenchmarkReproManifest, buildBenchmarkRunSeeds, buildCodexCreditReceipt, buildJudgePayload, buildOracleTrajectoryRecall, buildSchemaTierFixture, buildSchemaTierSmokeFixture, calendarFixture, canonicalJsonStringify, captureMachineFingerprint, chatFixture, checkCodingGraphRegression, checkDatasetContamination, checkRegression, clampScore, cohensD, compareResults, computeCohensKappa, computeSealHash, containsAnswer, createSeededRng$1 as createAdamSeededRng, createAmaBenchDiagnosticAdapter, createAnthropicProvider, createCanaryAdapter, createClaudeCliProvider, createCodexCliProvider, createSeededRng as createCodingGraphSeededRng, createDeterministicSpotCheckLogger, createGatewayResponder, createLightweightAdapter, createLiteLlmProvider, createLocalLlmProvider, createMcpDemoMemCorrectAdapter, createMcpDemoMemoryAdapter, createMcpMemCorrectAdapter, createMcpMemoryAdapter, createMitigatedTarget, createOllamaProvider, createOpenAiCompatibleProvider, createOpenAiResponsesBenchJudge, createOpenAiResponsesProvider, createSeededRandom as createProceduralAblationSeededRandom, createProvider, createProviderBackedAmaBenchRecommendedJudge, createProviderBackedJudge, createProviderBackedResponder, createProviderBackedStructuredJudge, createRemnicAdapter, createResponderFromProvider, createSeededRng$2 as createSeededRng, createSpotCheckFileLogger, createStructuredBenchJudge, createStructuredJudgeFromProvider, createSyntheticEmailIngestionAdapter, createSyntheticTarget, createTimeoutGuardedAdapter, defaultBenchmarkBaselineDir, defaultBenchmarkPublishPath, deleteBenchmarkResults, diagnoseLoComoProfileDelta, diagnoseLoComoRecallDelta, discoverAllProviders, discoveryEndpointFor, emailFixture, entityRecall, exactMatch, extractMetrics as extractCodingGraphMetrics, extractMarkdownSectionsByTitle, f1Score, fixtureToAblationScenarios, formatHandoffNote, formatMissingDatasetError, generateReport, generateSyntheticRepo, getAblationCell, getBenchmark, getBenchmarkLowerIsBetter, getMemoryEvalDimension, getProviderBackedJudgePromptIdentity, getRemnicVersion, hashBenchmarkArtifact, hashBytes, hashCanonicalJson, hashOrderedQuestionIds, hashString, integrityMetaIsComplete, interpretEffectSize, isAmaBenchUnknownLikeAnswer, isContaminationEntry, isContaminationManifest, isSealedQrelsArtifact, isSha256Hex, isStructuredJudgeProvider, judgeMemCorrectCorrectionAcceptance, judgeMemCorrectStaleMemoryHarm, linkMatches, listBenchmarkBaselines, listBenchmarkResults, listBenchmarks, listMemoryEvalBenchmarkIds, listMemoryEvalDimensions, llmJudgeScore, llmJudgeScoreDetailed, loadAblationFixture, loadBaseline, loadBeamDatasetPreview, loadBenchmarkArtifact, loadBenchmarkBaseline, loadBenchmarkReportCardProvenance, loadBenchmarkResult, loadCustomBenchmarkFile, loadJudgeCalibrationState, loadLoCoMo10, loadLocalLabManifest, loadLongMemEvalS, loadSealKeyFromEnv, loadSealedQrels, loadSealedRubric, matchEntity, mergeContaminationManifests, openSeal, orchestrateBenchmarkRuns, pairedDeltaConfidenceInterval, parseBenchmarkArtifact, parseCustomBenchmark, parseLocalLabManifest, parseRubricResponse, parseSealedQrels, pickStableQualifiedName, precisionAtK, preflightLocalLabRole, projectFolderFixture, recallAtK, reconcileCodexCreditLedger, redactBenchmarkResultSecrets, renderBaselineMarkdown, renderBenchmarkResultExport, renderLoComoProfileDeltaMarkdown, renderLoComoRecallDeltaMarkdown, renderMemorySummaryForJudge, renderMemoryViewForAgent, resolveAssistantAgent, resolveAssistantRubricId, resolveAssistantSeeds, resolveAssistantSpotCheckDir, resolveBenchRuntimeProfile, resolveBenchmarkPhaseTimeoutMs, resolveBenchmarkProgressLogging, resolveBenchmarkResultReference, resolveBenchmarkRunCount, resolveLocalLabJudgeProviderConfig, resolveLocalLabProfile, resolveLocalLabRole, resolveStructuredJudge, rotateDistractors, rougeL, runAssistantBenchmark, runAssistantMeetingPrepBenchmark, runAssistantMorningBriefBenchmark, runAssistantNextBestActionBenchmark, runAssistantSynthesisBenchmark, runBaseline, runBenchSuite, runBenchmark, runCodingGraphBenchmark, runCustomBenchmarkFile, runExplain, runExtractionAttack, runJudgeCalibration, runMitigatedBaseline, runProceduralAblation, runProceduralAblationCli, runSealedJudge, runSequentialPhases, safeHexEqual, sanitizeLoComoResultReference, saveBaseline, saveBenchmarkBaseline, schemaCompleteness, sealPayload, selectAmaBenchDiagnosticVariants, selectCalibrationSlice, selectFixtureVariant, serializeBenchmarkArtifact, serializeJsonl, serializeSealedQrels, shuffleTasks, timed, verifyRubricDigest, writeBenchmarkArtifact, writeBenchmarkPublishFeed, writeBenchmarkReproManifest, writeBenchmarkResult, writeJudgeCalibrationState, writeLeaderboardArtifactsForResult, zeroScores };
+export { AMA_BENCH_DIAGNOSTIC_VARIANTS, ASSISTANT_AGENT_CONFIG_KEY, ASSISTANT_JUDGE_CONFIG_KEY, ASSISTANT_MEETING_PREP_SCENARIOS, ASSISTANT_MEETING_PREP_SMOKE_SCENARIOS, ASSISTANT_MORNING_BRIEF_SCENARIOS, ASSISTANT_MORNING_BRIEF_SMOKE_SCENARIOS, ASSISTANT_NEXT_BEST_ACTION_SCENARIOS, ASSISTANT_NEXT_BEST_ACTION_SMOKE_SCENARIOS, ASSISTANT_RUBRIC_DIMENSIONS, ASSISTANT_RUBRIC_ID_KEY, ASSISTANT_SEEDS_CONFIG_KEY, ASSISTANT_SPOT_CHECK_DIR_KEY, ASSISTANT_SYNTHESIS_SCENARIOS, ASSISTANT_SYNTHESIS_SMOKE_SCENARIOS, type AblationConfigOverrides, type AbstentionRetrievalCase, type AggregateMetrics, type AmaBenchDiagnosticAdapterOptions, type AmaBenchDiagnosticAnswererMode, type AmaBenchDiagnosticBreakdown, type AmaBenchDiagnosticMatrixArtifact, type AmaBenchDiagnosticRecallMode, type AmaBenchDiagnosticRunContext, type AmaBenchDiagnosticTaskEvidence, type AmaBenchDiagnosticTaskRow, type AmaBenchDiagnosticVariant, type AmaBenchDiagnosticVariantSummary, type AnthropicProviderConfig, type AssistantAgent, type AssistantMemoryFact, type AssistantMemoryGraph, type AssistantRubricDimension, type AssistantRubricRequest, type AssistantRubricScores, type AssistantRunnerOptions, type AssistantScenario, type AssistantStance, type AttackRecallOptions, type AttackRetrievalHit, type AttackerMode, BENCHMARK_ARTIFACT_SCHEMA_VERSION, BENCHMARK_INTEGRITY_META_SCHEMA, BENCHMARK_REPRO_MANIFEST_FILENAME, BENCHMARK_REPRO_MANIFEST_SCHEMA_VERSION, BENCHMARK_RESULT_SCHEMA, BENCHMARK_SPLIT_TYPES, type BaselineRow, type BaselineScenario, type BeamDatasetPreview, type BenchConfig, type BenchJudge, type BenchJudgeResult, type BenchMemoryAdapter, type BenchModelSource, type BenchReasoningEffort, type BenchRecallLineageStatus, type BenchRecallOptions, type BenchRecallSupportAssessment, type BenchRecallSupportRequest, type BenchRecallSupportStatus, type BenchRecallTrace, type BenchRecallTraceCoreCapture, type BenchRecallTraceLcmCandidate, type BenchRecallTraceRange, type BenchRecallTraceSection, type BenchRecallTraceSelection, type BenchRecallWithTraceResult, type BenchResponder, type BenchResponse, type BenchRuntimeProfile, type BenchTier, type BenchmarkArtifact, type BenchmarkArtifactEnvironment, type BenchmarkArtifactHardware, type BenchmarkArtifactJudgeCalibration, type BenchmarkArtifactPerTaskScore, type BenchmarkArtifactSystem, type BenchmarkArtifactTier, type BenchmarkCategory, type BenchmarkDefinition, type BenchmarkIntegrityMeta, type BenchmarkMeta, type BenchmarkMode, type BenchmarkReport, type BenchmarkReproManifest, type BenchmarkReproManifestDataset, type BenchmarkReproManifestFile, type BenchmarkReproManifestResult, type BenchmarkResult, type BenchmarkSplitType, type BenchmarkStatus, type BenchmarkSuiteResult, type BenchmarkTier, type BootstrapKappaOptions, type BootstrapKappaResult, type BuildBenchmarkArtifactInput, type BuildBenchmarkPublishFeedOptions, type BuildBenchmarkReproManifestOptions, type BuiltInProvider, CALIBRATION_SLICE_SIZE, CANARY_FIXED_RECALL, CANARY_SCORE_FLOOR, DEFAULT_10K_FIXTURE as CODING_GRAPH_10K_FIXTURE, CODING_GRAPH_BENCH_SCHEMA_VERSION, DEFAULT_TOLERANCE_PERCENT as CODING_GRAPH_DEFAULT_TOLERANCE, MIN_ITERATIONS as CODING_GRAPH_MIN_ITERATIONS, DEFAULT_SMOKE_FIXTURE as CODING_GRAPH_SMOKE_FIXTURE, type CalibrationAnswer, type CalibrationVerdictPair, type CanaryAdapterOptions, type CanaryFloorCheck, type CaptureLoCoMoRetrievalTraceOptions, type ClaudeCliProviderConfig, type CodexCliProviderConfig, type CodexCreditReceipt, type CodexCreditReceiptScope, type CodexCreditReconciliationReceipt, type CodingGraphBaseline, type CodingGraphBenchConfig, type CodingGraphBenchReport, type MachineFingerprint as CodingGraphMachineFingerprint, type CodingGraphMetricKey, type RegressionMetricDetail as CodingGraphRegressionDetail, type RegressionMetricKey as CodingGraphRegressionKey, type RegressionGateResult as CodingGraphRegressionResult, type CohenKappaResult, type ComparisonMetricDelta, type ComparisonResult, type CompletionOpts, type CompletionResult, type ConfidenceInterval, type ContaminationCheckResult, type ContaminationEntry, type ContaminationManifest, type CustomBenchmarkScoring, type CustomBenchmarkSpec, type CustomBenchmarkTask, DEFAULT_ABLATION_BENCHMARK, DEFAULT_ABLATION_BOOTSTRAP_SEED, DEFAULT_ASSISTANT_RUBRIC_ID, DEFAULT_BASELINE_SCENARIOS, DEFAULT_JUDGE_BINARIZATION_THRESHOLD, DEFAULT_KAPPA_BOOTSTRAP_SAMPLES, DEFAULT_KAPPA_CONFIDENCE_LEVEL, DEFAULT_OPENAI_RESPONSES_JUDGE_MODEL, type DatasetSource, type DiagnoseLoComoProfileDeltaOptions, type DiagnoseLoComoRecallDeltaOptions, type DiscoveredModel, EMPTY_CONTAMINATION_MANIFEST, type EffectSizeInterpretation, type EffectSizeSummary, type ExplainResult, type ExtractedEntity, type ExtractedLink, type ExtractedPage, type ExtractionAttackOptions, type ExtractionAttackResult, type ExtractionAttackTarget, type FixtureGenerator, type FixtureOutput, type FixtureVariant, GENERAL_ANSWER_JUDGE_RUBRIC, type GeneratedFile, type GeneratedRepo, type GoldEntity, type GoldEntityType, type GoldGraph, type GoldLink, type GoldPage, type HarnessRng, INTEGRITY_CIPHER_ALGORITHM, INTEGRITY_HASH_ALGORITHM, INTEGRITY_META_FIELDS, type IngestionBenchAdapter, type IngestionLog, JUDGE_CALIBRATION_KAPPA_THRESHOLD, JUDGE_CALIBRATION_PROTOCOL_VERSION, type JudgeCalibrationCheckpointProvenance, type JudgeCalibrationIdentities, type JudgeCalibrationResult, type JudgeCategory, type KappaConfidenceInterval, LOCAL_LAB_PROVIDER_KINDS, LOCOMO_DATASET_FILENAMES, LOCOMO_FULL_TASK_COUNT, LOCOMO_RECALL_DIFF_LINE_LIMIT, LOCOMO_RECALL_EXCERPT_CHARS, LOCOMO_RETRIEVAL_TRACE_BUDGET_VERSION, LOCOMO_RETRIEVAL_TRACE_SCHEMA_VERSION, LOCOMO_RETRIEVAL_TRACE_SELECTION_VERSION, LONG_MEM_EVAL_DATASET_FILENAMES, type LeaderboardArtifactWrite, type LettaAdapterConfig, LettaMemCorrectAdapter, type LlmJudge, type LlmProvider, type LoCoMoRetrievalSessionReceipt, type LoCoMoRetrievalStructuralTrace, type LoCoMoRetrievalTaskReceipt, type LoCoMoRetrievalTraceCoreCaptureReceipt, type LoCoMoRetrievalTraceProfile, type LoCoMoRetrievalTraceReceipt, type LoCoMoRetrievalTraceSelectionManifest, type LoCoMoRetrievalTraceSelector, type LoComoCategoryDelta, type LoComoFinalContextRegression, type LoComoMetricDelta, type LoComoProfileArtifactEvidence, type LoComoProfileDeltaReport, type LoComoRawResultEvidence, type LoComoRecallCategoryDelta, type LoComoRecallContextSummary, type LoComoRecallDeltaReport, type LoComoRecallLineDelta, type LoComoRecallLineEvidence, type LoComoRecallMetricDelta, type LoComoRecallResultProvenance, type LoComoRecallTextDigest, type LoComoTaskRegression, type LoadDatasetOptions, type LoadSealedQrelsOptions, type LoadedDataset, type LoadedJudgeCalibrationState, type LocalLabManifest, type LocalLabManifestNotes, type LocalLabPhase, type LocalLabPhaseDescriptor, type LocalLabPhaseExecute, type LocalLabPhaseName, type LocalLabPhaseOutcome, LocalLabPreflightError, type LocalLabPreflightFailure, type LocalLabPreflightInput, type LocalLabPreflightOptions, type LocalLabPreflightResult, type LocalLabPreflightSuccess, type LocalLabProviderKind, type LocalLabRoleConfig, type LocalLlmProviderConfig, MEMCORRECT_CORRECTION_ACCEPTANCE_RUBRIC, MEMCORRECT_CORRECTION_ACCEPTANCE_RUBRIC_VERSION, MEMCORRECT_STALE_HARM_RUBRIC, MEMCORRECT_STALE_HARM_RUBRIC_VERSION, MEMORY_EVAL_DIMENSIONS, MEMORY_EVAL_PUBLIC_LINE, MIN_CALIBRATION_SOURCE_TASKS, MITIGATED_BASELINE_SCENARIOS, type McpArgumentSemantic, type McpBackendErrorCode, type McpBackendResult, type McpBenchMemoryAdapter, type McpConformanceResult, type McpHttpTransportConfig, type McpListedTool, type McpMemCorrectAdapter, type McpMemoryAdapterOptions, McpMemoryBackendError, type McpMemoryToolMapping, type McpMemoryTransportConfig, type McpStdioTransportConfig, type McpToolCallResult, type McpToolClient, type McpToolMappingEntry, type McpToolMappingValue, type McpToolOperation, type Mem0AdapterConfig, Mem0MemCorrectAdapter, type MemCorrectGeneratorOptions, type MemCorrectJudgeRequest, type MemCorrectJudgeResult, type MemCorrectSystemAdapter, type MemoryEvalCategory, type MemoryEvalDimension, type MemoryEvalDimensionId, type MemoryEvalMetric, type MemoryGraph, type MemoryStats, type MemorySystem, type Message, type MetricAggregate, type MicroMetric, MissingCredentialError, type MitigatedBaselineConfig, type MitigatedTargetConfig, type MultipleChoiceQuestion, OPENAI_RESPONSES_JUDGE_RUBRIC_VERSION, OTHER_NAMESPACE_MEMORIES, type OllamaProviderConfig, type OpenAiCompatibleProviderConfig, OpenAiResponsesJudgeError, type OpenAiResponsesJudgeErrorCode, type OpenAiResponsesJudgeTelemetry, OpenAiResponsesProvider, type OpenAiResponsesProviderConfig, type OpenAiResponsesVerdict, type OpenAiResponsesVerdictResult, PROCEDURAL_REAL_SCENARIOS, PROCEDURAL_REAL_SCENARIOS_SMOKE, PUBLISHED_BENCHMARK_ARTIFACT_IDS, type PersonalizationRetrievalCase, type PreflightDiscoveredModel, type ProceduralAblationArtifact, type ProceduralAblationPerCase, type ProceduralAblationScenario, type ProceduralRealScenario, type ProceduralRealScenarioCategory, type ProviderBaseConfig, type ProviderConfig, type ProviderDiscoveryResult, type ProviderFactoryConfig, type PublishSkipReason, type PublishSkipRecord, type PublishedBenchmarkFeed, type PublishedBenchmarkFeedEntry, type PublishedBenchmarkId, REQUIRED_FRONTMATTER_FIELDS, type RecallMetrics, type RecoveredMemory, type RegressionDetail, type RegressionGateResult$1 as RegressionGateResult, type RemnicAdapterOptions, type ReportCardProvenanceContext, type ResolveBenchRuntimeProfileOptions, type ResolvedBenchRuntimeProfile, type ResolvedLocalLabProfile, type ResolvedLocalLabRole, type ResolvedRunBenchmarkOptions, type RotatedChoices, type RunBenchmarkOptions, type RunJudgeCalibrationOptions, type RunProceduralAblationCliArgs, type RunProceduralAblationOptions, type RunSequentialPhasesOptions, SCHEMA_TIER_FIXTURE, SCHEMA_TIER_SMOKE_FIXTURE, SEALED_PROMPT_REGISTRY, SINGLE_FLAG_ABLATION_MATRIX, SYNTHETIC_MEMORIES, type SanitizedDiagnosticProvider, type SavedBaseline, type SchemaTierCorpus, type SchemaTierFixture, type SchemaTierName, type SchemaTierPage, type SchemaTierPageFrontmatter, type SealedArtifact, type SealedJudgeDecision, type SealedJudgeInput, type SealedQrelsArtifact, type SealedQrelsHandle, type SealedRubric, type SearchResult, type SeededMemory, type SeededRng, type SequentialPhaseHooks, type SingleFlagAblationCell, type SingleFlagAblationId, type SpotCheckLogger, type StatisticalReport, type StructuredJudge, StructuredJudgeError, type StructuredJudgeErrorCode, type StructuredJudgeProvider, type StructuredJudgeTelemetry, type StructuredJudgeVerdict, type StructuredJudgeVerdictResult, type StructuredVerdictRequest, type SyntheticEdge, type SyntheticEmailIngestionAdapterOptions, type SyntheticFileIR, type SyntheticRepoConfig, type SyntheticSymbol, type SyntheticTargetOptions, type TaskResult, type TaskTokenUsage, type TemporalRetrievalCase, type ThirdPartyAdapterConfig, type TierDetail, type TimelineEntry, type TokenUsage, type WallMetric, type WriteBenchmarkArtifactResult, type ZepAdapterConfig, ZepMemCorrectAdapter, addContaminationEntry, aggregateTaskScores, answerBenchmarkQuestion, assertCanaryUnderFloor, assertIntegrityMetaPresent, assertPublishableIntegrity, assertSha256Hex, assistantMeetingPrepDefinition, assistantMorningBriefDefinition, assistantNextBestActionDefinition, assistantSynthesisDefinition, backlinkF1, binarizeJudgeScore, bootstrapCohensKappaConfidenceInterval, bootstrapMeanConfidenceInterval, buildAmaBenchDiagnosticMatrixArtifact, buildAmaBenchDiagnosticVariantSummary, buildAmaBenchLeaderboardRows, buildBaselineFromReport, buildBenchmarkArtifact, buildBenchmarkArtifactFilename, buildBenchmarkPublishFeed, buildBenchmarkReproManifest, buildBenchmarkRunSeeds, buildCodexCreditReceipt, buildJudgePayload, buildOracleTrajectoryRecall, buildProviderFreeLoCoMoRetrievalConfig, buildSchemaTierFixture, buildSchemaTierSmokeFixture, calendarFixture, canonicalJsonStringify, captureLoCoMoRetrievalTrace, captureMachineFingerprint, chatFixture, checkCodingGraphRegression, checkDatasetContamination, checkRegression, clampScore, cohensD, compareResults, computeCohensKappa, computeSealHash, containsAnswer, createSeededRng$1 as createAdamSeededRng, createAmaBenchDiagnosticAdapter, createAnthropicProvider, createCanaryAdapter, createClaudeCliProvider, createCodexCliProvider, createSeededRng as createCodingGraphSeededRng, createDeterministicSpotCheckLogger, createGatewayResponder, createLightweightAdapter, createLiteLlmProvider, createLocalLlmProvider, createMcpDemoMemCorrectAdapter, createMcpDemoMemoryAdapter, createMcpMemCorrectAdapter, createMcpMemoryAdapter, createMitigatedTarget, createOllamaProvider, createOpenAiCompatibleProvider, createOpenAiResponsesBenchJudge, createOpenAiResponsesProvider, createSeededRandom as createProceduralAblationSeededRandom, createProvider, createProviderBackedAmaBenchRecommendedJudge, createProviderBackedJudge, createProviderBackedResponder, createProviderBackedStructuredJudge, createRemnicAdapter, createResponderFromProvider, createSeededRng$2 as createSeededRng, createSpotCheckFileLogger, createStructuredBenchJudge, createStructuredJudgeFromProvider, createSyntheticEmailIngestionAdapter, createSyntheticTarget, createTimeoutGuardedAdapter, defaultBenchmarkBaselineDir, defaultBenchmarkPublishPath, deleteBenchmarkResults, diagnoseLoComoProfileDelta, diagnoseLoComoRecallDelta, discoverAllProviders, discoveryEndpointFor, emailFixture, entityRecall, exactMatch, extractMetrics as extractCodingGraphMetrics, extractMarkdownSectionsByTitle, f1Score, fixtureToAblationScenarios, formatHandoffNote, formatMissingDatasetError, generateReport, generateSyntheticRepo, getAblationCell, getBenchmark, getBenchmarkLowerIsBetter, getGitSha, getMemoryEvalDimension, getProviderBackedJudgePromptIdentity, getRemnicVersion, hashBenchmarkArtifact, hashBytes, hashCanonicalJson, hashOrderedQuestionIds, hashString, integrityMetaIsComplete, interpretEffectSize, isAmaBenchUnknownLikeAnswer, isContaminationEntry, isContaminationManifest, isSealedQrelsArtifact, isSha256Hex, isStructuredJudgeProvider, judgeMemCorrectCorrectionAcceptance, judgeMemCorrectStaleMemoryHarm, linkMatches, listBenchmarkBaselines, listBenchmarkResults, listBenchmarks, listMemoryEvalBenchmarkIds, listMemoryEvalDimensions, llmJudgeScore, llmJudgeScoreDetailed, loadAblationFixture, loadBaseline, loadBeamDatasetPreview, loadBenchmarkArtifact, loadBenchmarkBaseline, loadBenchmarkReportCardProvenance, loadBenchmarkResult, loadCustomBenchmarkFile, loadJudgeCalibrationState, loadLoCoMo10, loadLocalLabManifest, loadLongMemEvalS, loadSealKeyFromEnv, loadSealedQrels, loadSealedRubric, matchEntity, mergeContaminationManifests, openSeal, orchestrateBenchmarkRuns, pairedDeltaConfidenceInterval, parseBenchmarkArtifact, parseCustomBenchmark, parseLocalLabManifest, parseRubricResponse, parseSealedQrels, pickStableQualifiedName, precisionAtK, preflightLoCoMoRetrievalTraceCapture, preflightLocalLabRole, projectFolderFixture, recallAtK, reconcileCodexCreditLedger, redactBenchmarkResultSecrets, renderBaselineMarkdown, renderBenchmarkResultExport, renderLoComoProfileDeltaMarkdown, renderLoComoRecallDeltaMarkdown, renderMemorySummaryForJudge, renderMemoryViewForAgent, resolveAssistantAgent, resolveAssistantRubricId, resolveAssistantSeeds, resolveAssistantSpotCheckDir, resolveBenchRuntimeProfile, resolveBenchmarkPhaseTimeoutMs, resolveBenchmarkProgressLogging, resolveBenchmarkResultReference, resolveBenchmarkRunCount, resolveLocalLabJudgeProviderConfig, resolveLocalLabProfile, resolveLocalLabRole, resolveStructuredJudge, rotateDistractors, rougeL, runAssistantBenchmark, runAssistantMeetingPrepBenchmark, runAssistantMorningBriefBenchmark, runAssistantNextBestActionBenchmark, runAssistantSynthesisBenchmark, runBaseline, runBenchSuite, runBenchmark, runCodingGraphBenchmark, runCustomBenchmarkFile, runExplain, runExtractionAttack, runJudgeCalibration, runMitigatedBaseline, runProceduralAblation, runProceduralAblationCli, runSealedJudge, runSequentialPhases, safeHexEqual, sanitizeLoComoResultReference, saveBaseline, saveBenchmarkBaseline, schemaCompleteness, sealPayload, selectAmaBenchDiagnosticVariants, selectCalibrationSlice, selectFixtureVariant, serializeBenchmarkArtifact, serializeJsonl, serializeLoCoMoRetrievalTraceReceipt, serializeSealedQrels, shuffleTasks, timed, verifyRubricDigest, writeBenchmarkArtifact, writeBenchmarkPublishFeed, writeBenchmarkReproManifest, writeBenchmarkResult, writeJudgeCalibrationState, writeLeaderboardArtifactsForResult, zeroScores };

package/dist/index.js

@@ -19847,26 +19847,6 @@ import { collectTemporalLexicalCues } from "@remnic/core";
 import { readFile as readFile15 } from "fs/promises";
 import path17 from "path";
 
-// src/benchmarks/published/longmemeval/fixture.ts
-var LONG_MEM_EVAL_SMOKE_FIXTURE = [
-  {
-    question_id: 1,
-    question_type: "single-session-user",
-    question: "What city does the user live in?",
-    answer: "Paris",
-    question_date: "2025-01-01",
-    haystack_dates: ["2024-12-01"],
-    haystack_session_ids: ["session-1"],
-    haystack_sessions: [
-      [
-        { role: "user", content: "I moved to Paris last year." },
-        { role: "assistant", content: "Paris sounds great." }
-      ]
-    ],
-    answer_session_ids: ["session-1"]
-  }
-];
-
 // src/benchmarks/published/locomo/fixture.ts
 var LOCOMO_SMOKE_FIXTURE = [
   {
@@ -19916,6 +19896,26 @@ var LOCOMO_SMOKE_FIXTURE = [
   }
 ];
 
+// src/benchmarks/published/longmemeval/fixture.ts
+var LONG_MEM_EVAL_SMOKE_FIXTURE = [
+  {
+    question_id: 1,
+    question_type: "single-session-user",
+    question: "What city does the user live in?",
+    answer: "Paris",
+    question_date: "2025-01-01",
+    haystack_dates: ["2024-12-01"],
+    haystack_session_ids: ["session-1"],
+    haystack_sessions: [
+      [
+        { role: "user", content: "I moved to Paris last year." },
+        { role: "assistant", content: "Paris sounds great." }
+      ]
+    ],
+    answer_session_ids: ["session-1"]
+  }
+];
+
 // src/benchmarks/published/dataset-loader.ts
 var LONG_MEM_EVAL_DATASET_FILENAMES = Object.freeze([
   "longmemeval_oracle.json",
@@ -19963,6 +19963,7 @@ async function loadDataset4(options) {
         return {
           source: "dataset",
           filename,
+          sha256: hashString(raw),
           items: applyLimit4(parsed, limit),
           errors
         };
@@ -19978,6 +19979,7 @@ async function loadDataset4(options) {
   }
   return {
     source: "smoke",
+    sha256: hashCanonicalJson(options.smokeFixture),
     items: applyLimit4([...options.smokeFixture], limit),
     errors
   };
@@ -21263,11 +21265,12 @@ var locomoDefinition = {
   }
 };
 async function runLoCoMoBenchmark(options) {
-  const conversations = await loadDataset6(
+  const loaded = await loadLoCoMoDataset(
     options.mode,
     options.datasetDir,
     options.limit
   );
+  const conversations = loaded.items;
   const trialLimit = resolveTrialLimit(options.benchmarkOptions?.trialLimit);
   const multiHopRecallComposition = resolveLoCoMoBooleanOption(
     options.benchmarkOptions?.multiHopRecallComposition,
@@ -21276,7 +21279,7 @@ async function runLoCoMoBenchmark(options) {
   );
   const plans = applyTrialLimit(
     conversations.map(
-      (conversation) => buildPlan2(conversation, multiHopRecallComposition)
+      (conversation) => buildLoCoMoPlan(conversation, multiHopRecallComposition)
     ),
     trialLimit
   );
@@ -21347,7 +21350,7 @@ function applyTrialLimit(plans, trialLimit) {
   }
   return limitedPlans;
 }
-function buildPlan2(conversation, multiHopRecallComposition) {
+function buildLoCoMoPlan(conversation, multiHopRecallComposition) {
   const sessions = extractSessions(conversation.conversation);
   const speakerA = typeof conversation.conversation.speaker_a === "string" ? conversation.conversation.speaker_a : "Speaker A";
   const ingestSessions = [];
@@ -21383,9 +21386,9 @@ function buildTrial(conversationId, qa, questionIndex, sessionIds, multiHopRecal
     expected: qa.answer,
     recallSessionIds: sessionIds,
     answerFormat: "short-with-specifics",
-    recallTextTransform: ({ question, recalledText }) => prioritizeLoCoMoRecallText({
+    recallTextTransform: ({ question, recalledText }) => transformLoCoMoRecallText({
       question,
-      recalledText: sanitizeLoCoMoRecallText({ question, recalledText }),
+      recalledText,
       multiHopRecallComposition
     }),
     answerFallback: ({ question, recalledText }) => answerLoCoMoFromRecall(question, recalledText),
@@ -21586,10 +21589,20 @@ function sanitizeLoCoMoRecallText(args) {
     (id) => queryVisibleIds.has(id) ? id : ""
   );
 }
-function prioritizeLoCoMoRecallText(args) {
-  const lines = dedupePreserveOrder(
-    args.recalledText.replaceAll("\r\n", "\n").replaceAll("\r", "\n").split("\n").map((line) => line.trim()).filter((line) => line.length > 0)
-  );
+function transformLoCoMoRecallText(args) {
+  const sanitized = sanitizeLoCoMoRecallText(args);
+  return prioritizeLoCoMoRecallTextWithTrace({
+    ...args,
+    recalledText: sanitized
+  }).text;
+}
+function prioritizeLoCoMoRecallTextWithTrace(args) {
+  const inputLines = args.recalledText.replaceAll("\r\n", "\n").replaceAll("\r", "\n").split("\n").map((line) => line.trim()).filter((line) => line.length > 0);
+  const lines = dedupePreserveOrder(inputLines);
+  const inputOrdinalByLine = /* @__PURE__ */ new Map();
+  inputLines.forEach((line, index) => {
+    if (!inputOrdinalByLine.has(line)) inputOrdinalByLine.set(line, index);
+  });
   const questionTokens = expandLoCoMoQuestionTokens(
     tokenizeForLoCoMo(args.question)
   );
@@ -21622,25 +21635,90 @@ function prioritizeLoCoMoRecallText(args) {
     )
   ];
   if (direct.length === 0 && linkedHops.length === 0) {
-    return truncateLoCoMoContext(
+    const { text: text2 } = truncateLoCoMoContext(
       args.recalledText,
       LOCOMO_FALLBACK_CONTEXT_MAX_CHARS
     );
+    return {
+      text: text2,
+      receipt: {
+        schemaVersion: 1,
+        mode: "fallback",
+        multiHopRecallComposition: args.multiHopRecallComposition,
+        input: digestLoCoMoContent(args.recalledText),
+        output: digestLoCoMoContent(text2),
+        selectedLines: []
+      }
+    };
   }
-  const sections = [
-    "## LoCoMo Question-Focused Evidence",
-    ...direct.map((entry) => truncateLoCoMoLine(entry.line))
-  ];
+  const sections = ["## LoCoMo Question-Focused Evidence"];
+  const selectedRanges = [];
+  const appendSelectedLine = (input, stage, hop) => {
+    const output = truncateLoCoMoLine(input);
+    const inputOrdinal = inputOrdinalByLine.get(input);
+    if (inputOrdinal === void 0) {
+      throw new Error("LoCoMo composition selected a line outside its normalized input.");
+    }
+    const outputStart = sections.join("\n").length + 1;
+    sections.push(output);
+    selectedRanges.push({
+      input,
+      output,
+      inputOrdinal,
+      stage,
+      ...hop === void 0 ? {} : { hop },
+      outputStart,
+      outputEnd: outputStart + output.length
+    });
+  };
+  for (const entry of direct) appendSelectedLine(entry.line, "direct");
   for (const hop of linkedHops) {
-    sections.push(
-      `## LoCoMo Linked Evidence (hop ${hop.hop})`,
-      ...hop.lines.map(truncateLoCoMoLine)
-    );
+    sections.push(`## LoCoMo Linked Evidence (hop ${hop.hop})`);
+    for (const line of hop.lines) appendSelectedLine(line, "linked", hop.hop);
   }
-  return truncateLoCoMoContext(
+  const truncation = truncateLoCoMoContext(
     sections.join("\n"),
     LOCOMO_FOCUSED_CONTEXT_MAX_CHARS
   );
+  const { text, safePrefixEnd } = truncation;
+  const selectedLines = selectedRanges.map((entry) => {
+    const visibleStart = Math.min(entry.outputStart, safePrefixEnd);
+    const visibleEnd = Math.min(entry.outputEnd, safePrefixEnd);
+    const visible = visibleEnd - visibleStart === entry.output.length;
+    return buildCompositionLineReceipt(entry, visible, visibleStart, visibleEnd);
+  });
+  return {
+    text,
+    receipt: {
+      schemaVersion: 1,
+      mode: "focused",
+      multiHopRecallComposition: args.multiHopRecallComposition,
+      input: digestLoCoMoContent(args.recalledText),
+      output: digestLoCoMoContent(text),
+      selectedLines
+    }
+  };
+}
+function buildCompositionLineReceipt(entry, visible, visibleStart, visibleEnd) {
+  return {
+    inputOrdinal: entry.inputOrdinal,
+    input: digestLoCoMoContent(entry.input),
+    output: digestLoCoMoContent(entry.output),
+    stage: entry.stage,
+    ...entry.hop === void 0 ? {} : { hop: entry.hop },
+    visible,
+    outputStart: entry.outputStart,
+    outputEnd: entry.outputEnd,
+    visibleStart,
+    visibleEnd
+  };
+}
+function digestLoCoMoContent(value) {
+  return {
+    sha256: hashString(value),
+    charCount: value.length,
+    lineCount: value.length === 0 ? 0 : value.split("\n").length
+  };
 }
 function composeLoCoMoLinkedEvidence(args) {
   if (args.direct.length === 0 || args.remainingLineBudget <= 0) {
@@ -21898,13 +21976,16 @@ function truncateLoCoMoLine(line) {
 }
 function truncateLoCoMoContext(text, maxChars) {
   if (text.length <= maxChars) {
-    return text;
+    return { text, safePrefixEnd: text.length };
   }
   const truncated = text.slice(0, maxChars);
   const lastNewline = truncated.lastIndexOf("\n");
   const safePrefix = lastNewline > 0 ? truncated.slice(0, lastNewline) : truncated;
-  return `${safePrefix}
-[LoCoMo context truncated to ${maxChars} characters]`;
+  return {
+    text: `${safePrefix}
+[LoCoMo context truncated to ${maxChars} characters]`,
+    safePrefixEnd: safePrefix.length
+  };
 }
 function countHiddenEvidenceIdsInRecall(evidence, question, recalledText) {
   const queryVisibleIds = collectDialogueIds(question);
@@ -21925,7 +22006,7 @@ function collectDialogueIds(text) {
 function escapeRegExp2(value) {
   return value.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
 }
-async function loadDataset6(mode, datasetDir, limit) {
+async function loadLoCoMoDataset(mode, datasetDir, limit) {
   const loaded = await loadLoCoMo10({
     mode,
     datasetDir,
@@ -21957,7 +22038,15 @@ async function loadDataset6(mode, datasetDir, limit) {
       "[remnic-bench] LoCoMo falling back to smoke fixture: " + loaded.errors.join(" | ")
     );
   }
-  return loaded.items;
+  if (!loaded.sha256) {
+    throw new Error("LoCoMo dataset loader did not provide a content hash.");
+  }
+  return {
+    source: loaded.source,
+    ...loaded.filename === void 0 ? {} : { filename: loaded.filename },
+    sha256: loaded.sha256,
+    items: loaded.items
+  };
 }
 function parseDataset2(raw, filename) {
   const parsed = JSON.parse(raw);
@@ -22241,7 +22330,7 @@ async function loadBeamDatasetPreview(options) {
   }
   let dataset;
   try {
-    dataset = await loadDataset7(
+    dataset = await loadDataset6(
       options.mode === "quick" ? "quick" : "full",
       options.datasetDir,
       options.limit
@@ -22280,7 +22369,7 @@ async function loadBeamDatasetPreview(options) {
   };
 }
 async function runBeamBenchmark(options) {
-  const dataset = await loadDataset7(options.mode, options.datasetDir, options.limit);
+  const dataset = await loadDataset6(options.mode, options.datasetDir, options.limit);
   const tasks = [];
   const taskFilter = normalizeBeamTaskFilter(
     options.benchmarkOptions?.taskFilter
@@ -22467,7 +22556,7 @@ async function runBeamBenchmark(options) {
     }
   };
 }
-async function loadDataset7(mode, datasetDir, limit) {
+async function loadDataset6(mode, datasetDir, limit) {
   const normalizedLimit = normalizeLimit5(limit);
   const ensureDatasetEntries = (entryCount) => {
     if (entryCount === 0) {
@@ -23653,7 +23742,7 @@ var personaMemDefinition = {
   }
 };
 async function runPersonaMemBenchmark(options) {
-  const samples = await loadDataset8(options.mode, options.datasetDir, options.limit);
+  const samples = await loadDataset7(options.mode, options.datasetDir, options.limit);
   const tasks = [];
   const totalTasks = samples.length;
   for (let sampleIndex = 0; sampleIndex < samples.length; sampleIndex += 1) {
@@ -23833,7 +23922,7 @@ async function runPersonaMemBenchmark(options) {
     }
   };
 }
-async function loadDataset8(mode, datasetDir, limit) {
+async function loadDataset7(mode, datasetDir, limit) {
   const normalizedLimit = normalizeLimit6(limit);
   const ensureDatasetSamples = (samples) => {
     if (samples.length === 0) {
@@ -24441,7 +24530,7 @@ var memBenchDefinition = {
   }
 };
 async function runMemBenchBenchmark(options) {
-  const dataset = await loadDataset9(options.mode, options.datasetDir, options.limit);
+  const dataset = await loadDataset8(options.mode, options.datasetDir, options.limit);
   const tasks = [];
   const totalTasks = dataset.length;
   for (const testCase of dataset) {
@@ -24615,7 +24704,7 @@ async function runMemBenchBenchmark(options) {
     }
   };
 }
-async function loadDataset9(mode, datasetDir, limit) {
+async function loadDataset8(mode, datasetDir, limit) {
   const normalizedLimit = normalizeLimit7(limit);
   const ensureDatasetCases = (cases) => {
     if (cases.length === 0) {
@@ -25646,7 +25735,7 @@ var memoryAgentBenchDefinition = {
   }
 };
 async function runMemoryAgentBenchBenchmark(options) {
-  const rawDataset = await loadDataset10(options.mode, options.datasetDir, options.limit);
+  const rawDataset = await loadDataset9(options.mode, options.datasetDir, options.limit);
   const trialLimit = resolveTrialLimit2(options.benchmarkOptions?.trialLimit);
   const benchmarkOptions = trialLimit === void 0 ? options.benchmarkOptions : { ...options.benchmarkOptions ?? {}, trialLimit };
   const dataset = applyTrialLimit2(rawDataset, trialLimit);
@@ -26619,7 +26708,7 @@ function decodeUrlComponentSafely(value) {
     return value;
   }
 }
-async function loadDataset10(mode, datasetDir, limit) {
+async function loadDataset9(mode, datasetDir, limit) {
   const normalizedLimit = normalizeLimit8(limit);
   const ensureDatasetItems = (items) => {
     if (items.length === 0) {
@@ -39194,6 +39283,349 @@ function formatSignedScore2(value) {
   return `${value >= 0 ? "+" : ""}${formatScore2(value)}`;
 }
 
+// src/benchmarks/published/locomo/retrieval-trace-runner.ts
+var LOCOMO_RETRIEVAL_TRACE_SCHEMA_VERSION = 1;
+var LOCOMO_RETRIEVAL_TRACE_SELECTION_VERSION = 1;
+var LOCOMO_RETRIEVAL_TRACE_BUDGET_VERSION = 1;
+async function preflightLoCoMoRetrievalTraceCapture(options) {
+  assertCaptureOptions(options);
+  assertProviderFreeRetrievalConfig(options.retrievalConfig);
+  const loaded = await loadLoCoMoDataset("full", options.datasetDir);
+  const multiHopRecallComposition = options.multiHopRecallComposition ?? true;
+  const plans = loaded.items.map((conversation) => buildLoCoMoPlan(conversation, multiHopRecallComposition));
+  const selectable = plans.flatMap(
+    (plan, planIndex) => plan.trials.map((trial) => ({ taskId: trial.taskId, planIndex }))
+  );
+  selectLoCoMoRetrievalTraceTasks(selectable, options.selector);
+}
+function buildProviderFreeLoCoMoRetrievalConfig(retrievalConfig) {
+  const sanitized = sanitizeProviderFreeRetrievalConfig(retrievalConfig);
+  return assertProviderFreeRetrievalConfig({
+    ...sanitized,
+    localLlmEnabled: false,
+    localLlmFastEnabled: false,
+    recallPlannerEnabled: false,
+    embeddingFallbackEnabled: false,
+    hostEmbeddingProviderEnabled: false,
+    openaiApiKey: false,
+    modelSource: "plugin"
+  });
+}
+async function captureLoCoMoRetrievalTrace(options) {
+  assertCaptureOptions(options);
+  const retrievalConfig = assertProviderFreeRetrievalConfig(options.retrievalConfig);
+  const recallWithTrace = options.system.recallWithTrace?.bind(options.system);
+  if (!recallWithTrace) {
+    throw new Error("LoCoMo retrieval trace capture requires system.recallWithTrace().");
+  }
+  const loaded = await loadLoCoMoDataset("full", options.datasetDir);
+  const multiHopRecallComposition = options.multiHopRecallComposition ?? true;
+  const plans = loaded.items.map((conversation) => buildLoCoMoPlan(conversation, multiHopRecallComposition));
+  const selectable = plans.flatMap(
+    (plan, planIndex) => plan.trials.map(
+      (trial) => ({
+        taskId: trial.taskId,
+        question: trial.question,
+        recallSessionIds: [...trial.recallSessionIds],
+        planIndex
+      })
+    )
+  );
+  const selection = selectLoCoMoRetrievalTraceTasks(selectable, options.selector);
+  const selectedIds = new Set(selection.selectedTaskIds);
+  const tasks = [];
+  for (let planIndex = 0; planIndex < plans.length; planIndex += 1) {
+    const selected = selectable.filter((task) => task.planIndex === planIndex && selectedIds.has(task.taskId));
+    if (selected.length === 0) continue;
+    const plan = plans[planIndex];
+    if (!plan) throw new Error(`Missing LoCoMo plan at index ${planIndex}.`);
+    await options.system.reset();
+    for (const session of plan.ingestSessions) {
+      if (session.messages.length > 0) {
+        await options.system.store(session.sessionId, session.messages);
+      }
+    }
+    await options.system.drain?.();
+    for (const selectedTask of selected) {
+      const recallBudgetChars = benchmarkRecallBudgetForSessionCount(selectedTask.recallSessionIds.length);
+      const recalled = await Promise.all(
+        selectedTask.recallSessionIds.map(async (sessionId) => {
+          const result = await recallWithTrace(sessionId, selectedTask.question, recallBudgetChars);
+          return {
+            text: result.text,
+            receipt: {
+              session: digestContent(sessionId),
+              trace: sanitizeStructuralTrace(result.trace)
+            }
+          };
+        })
+      );
+      const rawRecalledText = recalled.map((entry) => entry.text).filter(Boolean).join("\n\n");
+      const sanitized = sanitizeLoCoMoRecallText({
+        question: selectedTask.question,
+        recalledText: rawRecalledText
+      });
+      const composition = prioritizeLoCoMoRecallTextWithTrace({
+        question: selectedTask.question,
+        recalledText: sanitized,
+        multiHopRecallComposition
+      });
+      tasks.push({
+        taskId: selectedTask.taskId,
+        question: digestContent(selectedTask.question),
+        recallBudgetChars,
+        sessions: recalled.map((entry) => entry.receipt),
+        composition: composition.receipt
+      });
+    }
+  }
+  const withoutHash = {
+    schemaVersion: LOCOMO_RETRIEVAL_TRACE_SCHEMA_VERSION,
+    benchmarkId: "locomo",
+    captureKind: "retrieval-only",
+    sensitivity: {
+      classification: "restricted",
+      contentEncoding: "sha256+length",
+      containsGold: false,
+      containsRawContent: false
+    },
+    provenance: {
+      gitSha: options.gitSha,
+      remnicVersion: options.remnicVersion,
+      runtimeProfile: options.runtimeProfile,
+      adapterMode: "direct",
+      replayExtractionMode: "skip",
+      providerFree: true,
+      dataset: { id: "locomo-10", sha256: loaded.sha256 },
+      retrievalConfigSha256: hashCanonicalJson(retrievalConfig),
+      recallBudget: {
+        algorithm: "benchmarkRecallBudgetForSessionCount",
+        version: LOCOMO_RETRIEVAL_TRACE_BUDGET_VERSION
+      }
+    },
+    selection,
+    tasks
+  };
+  return {
+    ...withoutHash,
+    artifactHash: hashCanonicalJson(withoutHash)
+  };
+}
+function selectLoCoMoRetrievalTraceTasks(tasks, selector) {
+  const allIds = tasks.map((task) => task.taskId);
+  if (new Set(allIds).size !== allIds.length) {
+    throw new Error("LoCoMo retrieval trace task ids must be unique.");
+  }
+  let selected;
+  let algorithm;
+  let seed;
+  const hasTaskIds = "taskIds" in selector && selector.taskIds !== void 0;
+  const hasSampleSize = "sampleSize" in selector && selector.sampleSize !== void 0;
+  if (Number(hasTaskIds) + Number(hasSampleSize) !== 1) {
+    throw new Error("Choose exactly one LoCoMo retrieval trace selector.");
+  }
+  if (hasTaskIds && "seed" in selector && selector.seed !== void 0) {
+    throw new Error("LoCoMo retrieval trace seed is valid only for seeded sampling.");
+  }
+  if (hasTaskIds) {
+    algorithm = "explicit-task-ids";
+    const requestedTaskIds = selector.taskIds;
+    if (!requestedTaskIds) throw new Error("LoCoMo explicit task ids are required.");
+    const requested = [...requestedTaskIds];
+    if (requested.length === 0) {
+      throw new Error("LoCoMo retrieval trace explicit task selection cannot be empty.");
+    }
+    if (new Set(requested).size !== requested.length) {
+      throw new Error("LoCoMo retrieval trace explicit task ids must not contain duplicates.");
+    }
+    const available = new Set(allIds);
+    const unknown = requested.filter((taskId) => !available.has(taskId));
+    if (unknown.length > 0) {
+      throw new Error(`Unknown LoCoMo retrieval trace task id: ${unknown[0]}`);
+    }
+    const requestedSet = new Set(requested);
+    selected = allIds.filter((taskId) => requestedSet.has(taskId));
+  } else {
+    algorithm = "sha256-seeded-sample";
+    const sampleSize = selector.sampleSize;
+    seed = selector.seed;
+    if (sampleSize === void 0 || seed === void 0) {
+      throw new Error("LoCoMo seeded sampling requires sampleSize and seed.");
+    }
+    if (!Number.isSafeInteger(sampleSize) || sampleSize <= 0 || sampleSize > allIds.length) {
+      throw new Error(`LoCoMo retrieval trace sampleSize must be an integer from 1 to ${allIds.length}.`);
+    }
+    if (!Number.isSafeInteger(seed) || seed < 0) {
+      throw new Error("LoCoMo retrieval trace seed must be a non-negative safe integer.");
+    }
+    const sampled = [...allIds].sort((left, right) => {
+      const leftHash = hashString(`${seed}\0${left}`);
+      const rightHash = hashString(`${seed}\0${right}`);
+      return leftHash.localeCompare(rightHash) || left.localeCompare(right);
+    }).slice(0, sampleSize);
+    const sampledSet = new Set(sampled);
+    selected = allIds.filter((taskId) => sampledSet.has(taskId));
+  }
+  return {
+    algorithm,
+    version: LOCOMO_RETRIEVAL_TRACE_SELECTION_VERSION,
+    ...seed === void 0 ? {} : { seed },
+    candidateCount: allIds.length,
+    selectedCount: selected.length,
+    selectedTaskIds: selected,
+    selectedTaskIdsSha256: hashCanonicalJson(selected)
+  };
+}
+function serializeLoCoMoRetrievalTraceReceipt(receipt) {
+  return `${canonicalJsonStringify(receipt, 2)}
+`;
+}
+function sanitizeStructuralTrace(trace2) {
+  return {
+    schemaVersion: trace2.schemaVersion,
+    sensitivity: { ...trace2.sensitivity },
+    sections: trace2.sections.map((section) => ({ ...section })),
+    selections: trace2.selections.map(({ summary, ...selection }) => ({
+      ...selection,
+      ...selection.archiveRowIds === void 0 ? {} : { archiveRowIds: [...selection.archiveRowIds] },
+      ...summary === void 0 ? {} : { summary: { depth: summary.depth, msgStart: summary.msgStart, msgEnd: summary.msgEnd } }
+    })),
+    lcmCandidates: trace2.lcmCandidates.map((candidate) => ({ ...candidate })),
+    ...trace2.coreCapture === void 0 ? {} : {
+      coreCapture: {
+        budget: { ...trace2.coreCapture.budget },
+        filters: trace2.coreCapture.filters.map((filter) => ({ ...filter })),
+        results: trace2.coreCapture.results.map((result) => {
+          assertMemoryIdRef(result.memoryIdRef);
+          const score = result.scoreDecomposition;
+          return {
+            memoryIdRef: {
+              sha256: result.memoryIdRef.sha256,
+              length: result.memoryIdRef.length
+            },
+            servedBy: result.servedBy,
+            scoreDecomposition: {
+              ...score.vector === void 0 ? {} : { vector: score.vector },
+              ...score.bm25 === void 0 ? {} : { bm25: score.bm25 },
+              ...score.importance === void 0 ? {} : { importance: score.importance },
+              ...score.mmrPenalty === void 0 ? {} : { mmrPenalty: score.mmrPenalty },
+              ...score.tierPrior === void 0 ? {} : { tierPrior: score.tierPrior },
+              ...score.reinforcementBoost === void 0 ? {} : { reinforcementBoost: score.reinforcementBoost },
+              final: score.final
+            },
+            admittedBy: [...result.admittedBy],
+            ...result.rejectedBy === void 0 ? {} : { rejectedBy: result.rejectedBy },
+            ...result.disclosure === void 0 ? {} : { disclosure: result.disclosure },
+            ...result.estimatedTokens === void 0 ? {} : { estimatedTokens: result.estimatedTokens }
+          };
+        })
+      }
+    },
+    budget: { ...trace2.budget }
+  };
+}
+function digestContent(value) {
+  return {
+    sha256: hashString(value),
+    charCount: value.length,
+    lineCount: value.length === 0 ? 0 : value.split("\n").length
+  };
+}
+function assertCaptureOptions(options) {
+  if (!options.datasetDir.trim()) {
+    throw new Error("LoCoMo retrieval trace capture requires datasetDir.");
+  }
+  if (options.runtimeProfile !== "baseline" && options.runtimeProfile !== "real") {
+    throw new Error('LoCoMo retrieval trace runtimeProfile must be "baseline" or "real".');
+  }
+  if (!options.gitSha.trim() || !options.remnicVersion.trim() || options.gitSha === "unknown" || options.remnicVersion === "unknown") {
+    throw new Error("LoCoMo retrieval trace provenance requires gitSha and remnicVersion.");
+  }
+  if (options.providerFreeConfirmed !== true) {
+    throw new Error("LoCoMo retrieval trace capture requires explicit provider-free confirmation.");
+  }
+}
+function assertProviderFreeRetrievalConfig(value) {
+  const config = assertJsonConfig(value);
+  for (const key of [
+    "localLlmEnabled",
+    "localLlmFastEnabled",
+    "recallPlannerEnabled",
+    "embeddingFallbackEnabled",
+    "hostEmbeddingProviderEnabled",
+    "openaiApiKey"
+  ]) {
+    if (config[key] !== false) {
+      throw new Error(`retrievalConfig.${key} must be false for provider-free capture.`);
+    }
+  }
+  if (config.modelSource !== "plugin") {
+    throw new Error('retrievalConfig.modelSource must be "plugin" for provider-free capture.');
+  }
+  return config;
+}
+function assertMemoryIdRef(value) {
+  if (!value || typeof value !== "object" || !/^[0-9a-f]{64}$/u.test(value.sha256) || !Number.isSafeInteger(value.length) || value.length <= 0) {
+    throw new Error("LoCoMo retrieval trace requires a valid content-free memoryIdRef.");
+  }
+}
+function assertJsonConfig(value, path40 = "retrievalConfig") {
+  if (value === null || typeof value === "string" || typeof value === "boolean") return value;
+  if (typeof value === "number") {
+    if (!Number.isFinite(value)) throw new Error(`${path40} must contain only finite JSON numbers.`);
+    return value;
+  }
+  if (Array.isArray(value)) {
+    return value.map((entry, index) => assertJsonConfig(entry, `${path40}[${index}]`));
+  }
+  if (!value || typeof value !== "object") {
+    throw new Error(`${path40} must be JSON-serializable and provider-free.`);
+  }
+  const output = {};
+  for (const key of Object.keys(value).sort()) {
+    const child = value[key];
+    if (key === "openaiApiKey") {
+      if (child !== false) {
+        throw new Error(`${path40}.${key} must be exactly false for provider-free capture.`);
+      }
+      output[key] = false;
+      continue;
+    }
+    if (isSecretKey(key)) {
+      throw new Error(`${path40}.${key} contains secret-bearing configuration.`);
+    }
+    if (child === void 0) continue;
+    if (/^(?:gatewayConfig|gatewayAgentId|fastGatewayAgentId|internalProvider|llmProvider|llmModel)$/iu.test(key) || key === "modelSource" && child !== "plugin") {
+      throw new Error(`${path40}.${key} is provider-capable configuration.`);
+    }
+    output[key] = assertJsonConfig(child, `${path40}.${key}`);
+  }
+  return output;
+}
+function sanitizeProviderFreeRetrievalConfig(value, path40 = "retrievalConfig") {
+  if (value === null || typeof value === "string" || typeof value === "boolean") return value;
+  if (typeof value === "number") {
+    if (!Number.isFinite(value)) throw new Error(`${path40} must contain only finite JSON numbers.`);
+    return value;
+  }
+  if (Array.isArray(value)) {
+    return value.map((entry, index) => sanitizeProviderFreeRetrievalConfig(entry, `${path40}[${index}]`));
+  }
+  if (!value || typeof value !== "object") {
+    throw new Error(`${path40} must be JSON-serializable.`);
+  }
+  const output = {};
+  for (const key of Object.keys(value).sort()) {
+    const child = value[key];
+    if (child === void 0) continue;
+    if (isSecretKey(key) || /^(?:gatewayConfig|gatewayAgentId|fastGatewayAgentId|internalProvider|llmProvider|llmModel)$/iu.test(key) || key === "modelSource") {
+      continue;
+    }
+    output[key] = sanitizeProviderFreeRetrievalConfig(child, `${path40}.${key}`);
+  }
+  return output;
+}
+
 // src/integrity/sealed-qrels.ts
 import { readFile as readFile20 } from "fs/promises";
 function isSealedQrelsArtifact(value) {
@@ -43064,6 +43496,9 @@ export {
   LOCOMO_FULL_TASK_COUNT,
   LOCOMO_RECALL_DIFF_LINE_LIMIT,
   LOCOMO_RECALL_EXCERPT_CHARS,
+  LOCOMO_RETRIEVAL_TRACE_BUDGET_VERSION,
+  LOCOMO_RETRIEVAL_TRACE_SCHEMA_VERSION,
+  LOCOMO_RETRIEVAL_TRACE_SELECTION_VERSION,
   LONG_MEM_EVAL_DATASET_FILENAMES,
   LettaMemCorrectAdapter,
   LocalLabPreflightError,
@@ -43120,10 +43555,12 @@ export {
   buildCodexCreditReceipt,
   buildJudgePayload,
   buildOracleTrajectoryRecall,
+  buildProviderFreeLoCoMoRetrievalConfig,
   buildSchemaTierFixture,
   buildSchemaTierSmokeFixture,
   calendarFixture,
   canonicalJsonStringify,
+  captureLoCoMoRetrievalTrace,
   captureMachineFingerprint,
   chatFixture,
   checkCodingGraphRegression,
@@ -43192,6 +43629,7 @@ export {
   getAblationCell,
   getBenchmark,
   getBenchmarkLowerIsBetter,
+  getGitSha,
   getMemoryEvalDimension,
   getProviderBackedJudgePromptIdentity,
   getRemnicVersion,
@@ -43245,6 +43683,7 @@ export {
   parseSealedQrels,
   pickStableQualifiedName,
   precisionAtK,
+  preflightLoCoMoRetrievalTraceCapture,
   preflightLocalLabRole,
   projectFolderFixture,
   recallAtK,
@@ -43300,6 +43739,7 @@ export {
   selectFixtureVariant,
   serializeBenchmarkArtifact,
   serializeJsonl,
+  serializeLoCoMoRetrievalTraceReceipt,
   serializeSealedQrels,
   shuffleTasks,
   timed,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@remnic/bench",
-  "version": "9.6.32",
+  "version": "9.6.33",
   "description": "Retrieval latency ladder benchmarks + CI regression gates for @remnic/core",
   "type": "module",
   "main": "./dist/index.js",
@@ -39,8 +39,8 @@
     "hyparquet": "^1.25.7",
     "yaml": "^2.4.2",
     "zod": "^3.24.0",
-    "@remnic/coding-graph": "^9.6.32",
-    "@remnic/core": "^9.6.32"
+    "@remnic/coding-graph": "^9.6.33",
+    "@remnic/core": "^9.6.33"
   },
   "devDependencies": {
     "tsup": "^8.5.1",