@remnic/bench 9.6.22 → 9.6.24

package/README.md

@@ -91,7 +91,7 @@ status` to report ChatGPT authentication. A 473-credit safety reserve leaves
 then measure a quick task before choosing a workload bound:
 
 ```bash
-BUILD_WEEK_RUN_ROOT="$HOME/.remnic/bench/build-week-2026"
+export BUILD_WEEK_RUN_ROOT="$HOME/.remnic/bench/build-week-2026"
 export BUILD_WEEK_RESULTS_DIR="$BUILD_WEEK_RUN_ROOT/results"
 umask 077
 mkdir -p "$BUILD_WEEK_RUN_ROOT" "$BUILD_WEEK_RESULTS_DIR"
@@ -101,8 +101,10 @@ export REMNIC_BENCH_CODEX_CREDIT_BUDGET=2473
 export REMNIC_BENCH_CODEX_CREDIT_RESERVE=473
 export REMNIC_BENCH_CODEX_CREDIT_LEDGER="$BUILD_WEEK_RUN_ROOT/codex-credit-ledger.json"
 
-remnic bench run --quick longmemeval \
+remnic bench run longmemeval \
   --runtime-profile real \
+  --limit 1 \
+  --dataset-dir ./bench-datasets/longmemeval \
   --results-dir "$BUILD_WEEK_RESULTS_DIR" \
   --drain-timeout 600000 \
   --system-provider codex-cli --system-model gpt-5.6-luna \
@@ -114,6 +116,7 @@ remnic bench run --quick longmemeval \
 
 remnic bench run longmemeval \
   --runtime-profile real --limit <LEDGER_DERIVED_LIMIT> \
+  --dataset-dir ./bench-datasets/longmemeval \
   --results-dir "$BUILD_WEEK_RESULTS_DIR" \
   --drain-timeout 600000 \
   --system-provider codex-cli --system-model gpt-5.6-luna \
@@ -130,6 +133,11 @@ actual `turn.completed` JSONL usage. Stop dispatching at 2,000 spent; the
 473-credit reserve absorbs only a final in-flight call whose exact cost becomes
 known after completion. Missing exact terminal usage blocks the ledger pending
 manual account reconciliation.
+The measured probe is a full-mode run bounded to one staged LongMemEval item.
+Full mode fails before provider dispatch if that explicit dataset directory is
+missing or unreadable, so it cannot fall back to the bundled quick fixture.
+Both commands pin the same gitignored dataset source and do not fall back to
+the CLI-managed dataset store.
 Rates per one million tokens are Luna: 25 input, 2.5 cached input, 150 output;
 Terra: 62.5 input, 6.25 cached input, 375 output. A bounded result is a trial,
 not a full leaderboard artifact.

package/dist/index.d.ts

@@ -1521,12 +1521,29 @@ interface CodexCliNativeUsage {
 interface CodexCreditReceiptScope extends CodexCliNativeUsage {
     calls: number;
     credits: number;
+    unattributedReconciliationCount: number;
+    unattributedReconciledCredits: number;
     models: Array<CodexCliNativeUsage & {
         model: string;
         calls: number;
         credits: number;
     }>;
 }
+interface CodexCreditReconciliationReceipt {
+    schemaVersion: 1;
+    priorLedgerSha256: string;
+    ledgerSha256: string;
+    at: string;
+    attribution: "account-wide-unattributed";
+    affectedRunId: string;
+    originalBudgetCredits: number;
+    priorRecordedSpentCredits: number;
+    observedRemainingCredits: number;
+    unattributedCredits: number;
+    totalSpentCredits: number;
+    totalRemainingCredits: number;
+    affectedBlockedEventSha256: string;
+}
 interface CodexCreditReceipt {
     schemaVersion: 1;
     ledgerSha256: string;
@@ -1541,6 +1558,16 @@ interface CodexCreditReceipt {
         id: string;
     };
 }
+declare function reconcileCodexCreditLedger(args: {
+    ledgerPath: string;
+    priorLedgerSha256: string;
+    observedRemainingCredits: number;
+    originalBudgetBalanceConfirmed: true;
+    noCreditsAddedOrRefundedConfirmed: true;
+    accountWideUnattributedChargeAccepted: true;
+    affectedRunId: string;
+}): Promise<CodexCreditReconciliationReceipt>;
+declare function buildCodexCreditReceipt(ledgerPath: string, runId?: string): Promise<CodexCreditReceipt>;
 
 declare const BENCHMARK_REPRO_MANIFEST_FILENAME = "MANIFEST.json";
 declare const BENCHMARK_REPRO_MANIFEST_SCHEMA_VERSION = 1;
@@ -5224,4 +5251,4 @@ 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 BenchRecallOptions, type BenchRecallSupportAssessment, type BenchRecallSupportRequest, type BenchRecallSupportStatus, 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 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, 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, 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, getRemnicVersion, hashBenchmarkArtifact, hashBytes, hashCanonicalJson, 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, redactBenchmarkResultSecrets, renderBaselineMarkdown, renderBenchmarkResultExport, renderLoComoProfileDeltaMarkdown, renderLoComoRecallDeltaMarkdown, renderMemorySummaryForJudge, renderMemoryViewForAgent, resolveAssistantAgent, resolveAssistantRubricId, resolveAssistantSeeds, resolveAssistantSpotCheckDir, resolveBenchRuntimeProfile, resolveBenchmarkPhaseTimeoutMs, resolveBenchmarkProgressLogging, resolveBenchmarkResultReference, resolveBenchmarkRunCount, 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 BenchRecallOptions, type BenchRecallSupportAssessment, type BenchRecallSupportRequest, type BenchRecallSupportStatus, 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, 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, getRemnicVersion, hashBenchmarkArtifact, hashBytes, hashCanonicalJson, 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, 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 };

package/dist/index.js

@@ -5509,6 +5509,7 @@ import { mkdir as mkdir2, open, readFile as readFile3, rename as rename2, rmdir,
 import os from "os";
 import path3 from "path";
 var ONE_MILLION = 1e6;
+var CREDIT_EPSILON = 1e-9;
 var MAX_BOUNDED_CALL_CREDITS = 300;
 var SOL_MODEL = /^gpt-5\.6-sol$/i;
 var CREDIT_RATES = [
@@ -5534,6 +5535,123 @@ var CodexCreditDispatchError = class extends Error {
     this.name = "CodexCreditDispatchError";
   }
 };
+async function reconcileCodexCreditLedger(args) {
+  if (args.originalBudgetBalanceConfirmed !== true) {
+    throw new Error(
+      "Codex credit reconciliation requires confirmation that the observed balance belongs to the ledger's original budget"
+    );
+  }
+  if (args.noCreditsAddedOrRefundedConfirmed !== true) {
+    throw new Error(
+      "Codex credit reconciliation requires confirmation that no credits were added or refunded after the original budget was established"
+    );
+  }
+  if (args.accountWideUnattributedChargeAccepted !== true) {
+    throw new Error(
+      "Codex credit reconciliation requires acknowledgment that all unexplained account activity will be charged as account-wide unattributed spend"
+    );
+  }
+  if (!isSha256(args.priorLedgerSha256)) {
+    throw new Error("priorLedgerSha256 must be a lowercase SHA-256 digest");
+  }
+  if (!Number.isFinite(args.observedRemainingCredits) || args.observedRemainingCredits < 0) {
+    throw new Error("observedRemainingCredits must be a finite non-negative number");
+  }
+  const affectedRunId = parseRequiredRunId(args.affectedRunId, "affectedRunId");
+  const ledgerPath = path3.resolve(expandHomeRelativePath(args.ledgerPath));
+  const previous = completionQueue;
+  let release;
+  completionQueue = new Promise((resolve) => {
+    release = resolve;
+  });
+  await previous;
+  const lockPath = `${ledgerPath}.lock`;
+  let lock;
+  try {
+    await mkdir2(path3.dirname(lockPath), { recursive: true, mode: 448 });
+    lock = await acquireLedgerLock(lockPath);
+    const contents = await readFile3(ledgerPath);
+    const currentSha256 = sha256(contents);
+    if (currentSha256 !== args.priorLedgerSha256) {
+      throw new Error(
+        `Codex credit ledger changed since operator observation: expected ${args.priorLedgerSha256}, found ${currentSha256}; obtain a fresh ledger hash and remaining balance`
+      );
+    }
+    const ledger = parseLedger(JSON.parse(contents.toString("utf8")));
+    if (!ledger.blockedReason) {
+      throw new Error("Codex credit ledger is not blocked; reconciliation is not permitted");
+    }
+    if (args.observedRemainingCredits > ledger.budgetCredits) {
+      throw new Error("observedRemainingCredits cannot exceed the ledger budget");
+    }
+    const rawUnattributedCredits = ledger.budgetCredits - args.observedRemainingCredits - ledger.spentCredits;
+    if (!Number.isFinite(rawUnattributedCredits) || rawUnattributedCredits < -CREDIT_EPSILON) {
+      throw new Error(
+        "observedRemainingCredits implies less spend than the ledger already records; reconciliation refused"
+      );
+    }
+    const unattributedCredits = Math.abs(rawUnattributedCredits) <= CREDIT_EPSILON ? 0 : rawUnattributedCredits;
+    const observedSpentCredits = ledger.spentCredits + unattributedCredits;
+    const at = (/* @__PURE__ */ new Date()).toISOString();
+    const reconciliation = {
+      at,
+      basis: "operator-observed-original-budget-balance",
+      attribution: "account-wide-unattributed",
+      priorLedgerSha256: currentSha256,
+      originalBudgetCredits: ledger.budgetCredits,
+      priorRecordedSpentCredits: ledger.spentCredits,
+      observedRemainingCredits: args.observedRemainingCredits,
+      credits: normalizeZero(unattributedCredits),
+      confirmations: {
+        observedBalanceBelongsToOriginalBudget: true,
+        noCreditsAddedOrRefunded: true,
+        accountWideUnattributedChargeAccepted: true
+      },
+      affectedBlockedEvent: {
+        runId: affectedRunId,
+        blockedReason: ledger.blockedReason
+      }
+    };
+    const nextLedger = {
+      ...ledger,
+      spentCredits: observedSpentCredits,
+      reconciliations: [...ledger.reconciliations ?? [], reconciliation],
+      blockedReason: void 0
+    };
+    await writeLedger(ledgerPath, nextLedger);
+    await writeLockState(lock, "settled");
+    const nextContents = await readFile3(ledgerPath);
+    return {
+      schemaVersion: 1,
+      priorLedgerSha256: currentSha256,
+      ledgerSha256: sha256(nextContents),
+      at,
+      attribution: "account-wide-unattributed",
+      affectedRunId,
+      originalBudgetCredits: ledger.budgetCredits,
+      priorRecordedSpentCredits: ledger.spentCredits,
+      observedRemainingCredits: args.observedRemainingCredits,
+      unattributedCredits: reconciliation.credits,
+      totalSpentCredits: observedSpentCredits,
+      totalRemainingCredits: args.observedRemainingCredits,
+      affectedBlockedEventSha256: sha256(
+        JSON.stringify(reconciliation.affectedBlockedEvent)
+      )
+    };
+  } finally {
+    try {
+      if (lock) {
+        try {
+          await lock.close();
+        } finally {
+          await removeOwnedLedgerLock(lockPath);
+        }
+      }
+    } finally {
+      release();
+    }
+  }
+}
 function resolveCodexCreditBudgetConfig(env = process.env, fallbackRunId) {
   const rawBudget = env.REMNIC_BENCH_CODEX_CREDIT_BUDGET?.trim();
   if (!rawBudget) return void 0;
@@ -5793,11 +5911,12 @@ async function buildCodexCreditReceipt(ledgerPath, runId) {
   const contents = await readFile3(resolvedPath);
   const ledger = parseLedger(JSON.parse(contents.toString("utf8")));
   const normalizedRunId = parseOptionalRunId(runId);
-  const cumulative = summarizeLedgerEntries(ledger.entries);
+  const reconciliations = ledger.reconciliations ?? [];
+  const cumulative = summarizeLedgerEntries(ledger.entries, reconciliations);
   const runEntries = normalizedRunId ? ledger.entries.filter((entry) => entry.runId === normalizedRunId) : [];
   return {
     schemaVersion: 1,
-    ledgerSha256: createHash3("sha256").update(contents).digest("hex"),
+    ledgerSha256: sha256(contents),
     budgetCredits: ledger.budgetCredits,
     reserveCredits: ledger.reserveCredits,
     plannedSpendCeilingCredits: ledger.budgetCredits - ledger.reserveCredits,
@@ -5805,7 +5924,12 @@ async function buildCodexCreditReceipt(ledgerPath, runId) {
     totalRemainingCredits: ledger.budgetCredits - ledger.spentCredits,
     blocked: ledger.blockedReason !== void 0,
     cumulative,
-    ...normalizedRunId ? { run: { id: normalizedRunId, ...summarizeLedgerEntries(runEntries) } } : {}
+    ...normalizedRunId ? {
+      run: {
+        id: normalizedRunId,
+        ...summarizeLedgerEntries(runEntries)
+      }
+    } : {}
   };
 }
 function resolveRate(model) {
@@ -5852,11 +5976,36 @@ function parseLedger(parsed) {
     (sum, entry) => sum + (typeof entry?.credits === "number" ? entry.credits ?? 0 : 0),
     0
   ) : Number.NaN;
-  if (parsed.schemaVersion !== 1 || typeof parsed.budgetCredits !== "number" || !Number.isFinite(parsed.budgetCredits) || parsed.budgetCredits <= 0 || typeof parsed.reserveCredits !== "number" || !Number.isFinite(parsed.reserveCredits) || parsed.reserveCredits < 0 || parsed.reserveCredits >= parsed.budgetCredits || typeof parsed.spentCredits !== "number" || !Number.isFinite(parsed.spentCredits) || parsed.spentCredits < 0 || !Array.isArray(parsed.entries) || !parsed.entries.every(isLedgerEntry) || Math.abs(entryCredits - parsed.spentCredits) > 1e-9 || parsed.blockedReason !== void 0 && typeof parsed.blockedReason !== "string") {
+  const reconciliationCredits = parsed.reconciliations === void 0 ? 0 : Array.isArray(parsed.reconciliations) ? parsed.reconciliations.reduce(
+    (sum, reconciliation) => sum + (typeof reconciliation?.credits === "number" ? reconciliation.credits ?? 0 : 0),
+    0
+  ) : Number.NaN;
+  if (parsed.schemaVersion !== 1 || typeof parsed.budgetCredits !== "number" || !Number.isFinite(parsed.budgetCredits) || parsed.budgetCredits <= 0 || typeof parsed.reserveCredits !== "number" || !Number.isFinite(parsed.reserveCredits) || parsed.reserveCredits < 0 || parsed.reserveCredits >= parsed.budgetCredits || typeof parsed.spentCredits !== "number" || !Number.isFinite(parsed.spentCredits) || parsed.spentCredits < 0 || !Array.isArray(parsed.entries) || !parsed.entries.every(isLedgerEntry) || parsed.reconciliations !== void 0 && (!Array.isArray(parsed.reconciliations) || !parsed.reconciliations.every(
+    (reconciliation) => isLedgerReconciliationWithinBudget(reconciliation, parsed.budgetCredits)
+  )) || Math.abs(entryCredits + reconciliationCredits - parsed.spentCredits) > 1e-9 || parsed.blockedReason !== void 0 && typeof parsed.blockedReason !== "string") {
     throw new Error("ledger schema is invalid");
   }
   return parsed;
 }
+function isLedgerReconciliation(value) {
+  if (!value || typeof value !== "object") return false;
+  const candidate = value;
+  const forbiddenUsageFields = [
+    "runId",
+    "unknownEvent",
+    "model",
+    "inputTokens",
+    "cachedInputTokens",
+    "outputTokens",
+    "reasoningOutputTokens"
+  ];
+  return forbiddenUsageFields.every((field) => !Object.prototype.hasOwnProperty.call(candidate, field)) && isIsoTimestamp(candidate.at) && candidate.basis === "operator-observed-original-budget-balance" && candidate.attribution === "account-wide-unattributed" && isSha256(candidate.priorLedgerSha256) && typeof candidate.originalBudgetCredits === "number" && Number.isFinite(candidate.originalBudgetCredits) && candidate.originalBudgetCredits > 0 && typeof candidate.priorRecordedSpentCredits === "number" && Number.isFinite(candidate.priorRecordedSpentCredits) && candidate.priorRecordedSpentCredits >= 0 && typeof candidate.observedRemainingCredits === "number" && Number.isFinite(candidate.observedRemainingCredits) && candidate.observedRemainingCredits >= 0 && typeof candidate.credits === "number" && Number.isFinite(candidate.credits) && candidate.credits >= 0 && candidate.confirmations?.observedBalanceBelongsToOriginalBudget === true && candidate.confirmations.noCreditsAddedOrRefunded === true && candidate.confirmations.accountWideUnattributedChargeAccepted === true && isValidStoredRunId(candidate.affectedBlockedEvent?.runId) && typeof candidate.affectedBlockedEvent?.blockedReason === "string" && candidate.affectedBlockedEvent.blockedReason.length > 0;
+}
+function isLedgerReconciliationWithinBudget(value, budget) {
+  return typeof budget === "number" && isLedgerReconciliation(value) && value.originalBudgetCredits === budget && value.observedRemainingCredits <= budget && value.credits <= budget && Math.abs(
+    value.priorRecordedSpentCredits + value.credits + value.observedRemainingCredits - budget
+  ) <= 1e-9;
+}
 function isLedgerEntry(entry) {
   if (!entry || typeof entry !== "object") return false;
   const candidate = entry;
@@ -5869,7 +6018,7 @@ function isEntryCreditConsistent(entry) {
     return false;
   }
 }
-function summarizeLedgerEntries(entries) {
+function summarizeLedgerEntries(entries, reconciliations = []) {
   const byModel = /* @__PURE__ */ new Map();
   for (const entry of entries) {
     const modelEntries = byModel.get(entry.model) ?? [];
@@ -5879,7 +6028,12 @@ function summarizeLedgerEntries(entries) {
   const totals = summarizeUsage(entries);
   return {
     calls: entries.length,
-    credits: entries.reduce((sum, entry) => sum + entry.credits, 0),
+    credits: entries.reduce((sum, entry) => sum + entry.credits, 0) + reconciliations.reduce((sum, reconciliation) => sum + reconciliation.credits, 0),
+    unattributedReconciliationCount: reconciliations.length,
+    unattributedReconciledCredits: reconciliations.reduce(
+      (sum, reconciliation) => sum + reconciliation.credits,
+      0
+    ),
     ...totals,
     models: [...byModel.entries()].sort(([left], [right]) => left.localeCompare(right)).map(([model, modelEntries]) => ({
       model,
@@ -5958,6 +6112,30 @@ function hasControlCharacters(value) {
   }
   return false;
 }
+function parseRequiredRunId(value, name) {
+  const runId = value?.trim();
+  if (!runId || !isValidStoredRunId(runId)) {
+    throw new Error(`${name} must be 1 to 128 trimmed characters without control characters`);
+  }
+  return runId;
+}
+function isSha256(value) {
+  return typeof value === "string" && /^[a-f0-9]{64}$/.test(value);
+}
+function sha256(value) {
+  return createHash3("sha256").update(value).digest("hex");
+}
+function normalizeZero(value) {
+  return Object.is(value, -0) ? 0 : value;
+}
+function isIsoTimestamp(value) {
+  if (typeof value !== "string") return false;
+  try {
+    return new Date(value).toISOString() === value;
+  } catch {
+    return false;
+  }
+}
 
 // src/results-store.ts
 import { mkdir as mkdir3, readdir as readdir3, readFile as readFile4, unlink as unlink3, writeFile as writeFile3 } from "fs/promises";
@@ -32106,9 +32284,9 @@ function loadSealedRubric(id = DEFAULT_ASSISTANT_RUBRIC_ID, options = {}) {
   if (typeof prompt !== "string" || prompt.length === 0) {
     throw new Error(`sealed rubric not found in registry: ${id}`);
   }
-  const sha2562 = createHash9("sha256").update(prompt, "utf8").digest("hex");
+  const sha2563 = createHash9("sha256").update(prompt, "utf8").digest("hex");
   const version = parseVersionFromId(id);
-  return { id, version, prompt, sha256: sha2562 };
+  return { id, version, prompt, sha256: sha2563 };
 }
 function verifyRubricDigest(expectedSha256, options = {}) {
   const rubric = loadSealedRubric(options.id, { registry: options.registry });
@@ -38325,8 +38503,8 @@ function buildRegression(task, metric, excerptChars, maxDiffLines) {
     baselineScore,
     realScore,
     delta: realScore - baselineScore,
-    questionSha256: sha256(normalizeText3(task.baseline.question)),
-    expectedAnswerSha256: sha256(normalizeText3(task.baseline.expected)),
+    questionSha256: sha2562(normalizeText3(task.baseline.question)),
+    expectedAnswerSha256: sha2562(normalizeText3(task.baseline.expected)),
     baseline: {
       answer: textDigest(task.baseline.actual, excerptChars),
       recall: recallSummary(baselineRecall, task.baseline.expected)
@@ -38360,7 +38538,7 @@ function buildProvenance(evidence, profile, joined, side) {
     judgeProvider: judge.provider,
     judgeModel: judge.model,
     seeds: [...evidence.result.meta.seeds],
-    taskPayloadSha256: sha256(stableJson(payload))
+    taskPayloadSha256: sha2562(stableJson(payload))
   };
 }
 function summarizeMetric(tasks, metric) {
@@ -38400,7 +38578,7 @@ function recallSummary(text, expected) {
   const normalized = normalizeText3(text);
   const lines = normalized.split("\n").map((line) => line.trim()).filter(Boolean);
   return {
-    sha256: sha256(normalized),
+    sha256: sha2562(normalized),
     charCount: normalized.length,
     lineCount: lines.length,
     headings: [...new Set(lines.filter(isHeading))],
@@ -38411,7 +38589,7 @@ function recallSummary(text, expected) {
 function textDigest(text, excerptChars) {
   const normalized = normalizeText3(text);
   return {
-    sha256: sha256(normalized),
+    sha256: sha2562(normalized),
     charCount: normalized.length,
     excerpt: normalized.slice(0, excerptChars)
   };
@@ -38523,7 +38701,7 @@ function isHeading(line) {
 function normalizeText3(value) {
   return value.replaceAll("\r\n", "\n").replaceAll("\r", "\n");
 }
-function sha256(value) {
+function sha2562(value) {
   return createHash13("sha256").update(value).digest("hex");
 }
 function stableJson(value) {
@@ -42293,6 +42471,7 @@ export {
   buildBenchmarkPublishFeed,
   buildBenchmarkReproManifest,
   buildBenchmarkRunSeeds,
+  buildCodexCreditReceipt,
   buildJudgePayload,
   buildOracleTrajectoryRecall,
   buildSchemaTierFixture,
@@ -42421,6 +42600,7 @@ export {
   preflightLocalLabRole,
   projectFolderFixture,
   recallAtK,
+  reconcileCodexCreditLedger,
   redactBenchmarkResultSecrets,
   renderBaselineMarkdown,
   renderBenchmarkResultExport,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@remnic/bench",
-  "version": "9.6.22",
+  "version": "9.6.24",
   "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": "^9.6.22",
-    "@remnic/core": "^9.6.22"
+    "@remnic/coding-graph": "^9.6.24",
+    "@remnic/core": "^9.6.24"
   },
   "devDependencies": {
     "tsup": "^8.5.1",