@remnic/bench 9.69.35 → 9.69.36

package/dist/{chunk-VJ4YENME.js → chunk-PWFYSAUK.js}

@@ -308,6 +308,84 @@ function isWhitespace(char) {
   return char === " " || char === "	" || char === "\n" || char === "\r";
 }
 
+// src/integrity/canary-adapter.ts
+import process2 from "process";
+var CANARY_FIXED_RECALL = "__remnic_canary_response__";
+var CANARY_SCORE_FLOOR = 0.1;
+function parseCanaryFloor(value, label = "canary floor") {
+  if (typeof value !== "number" || !Number.isFinite(value) || value < 0) {
+    throw new Error(`Invalid ${label}: ${String(value)}`);
+  }
+  return value;
+}
+function resolveCanaryFloorFromEnv(raw = process2.env.REMNIC_BENCH_CANARY_FLOOR) {
+  if (raw === void 0) {
+    return CANARY_SCORE_FLOOR;
+  }
+  if (raw === "") {
+    throw new Error(`Invalid REMNIC_BENCH_CANARY_FLOOR: ${raw}`);
+  }
+  try {
+    return parseCanaryFloor(Number(raw), "REMNIC_BENCH_CANARY_FLOOR");
+  } catch {
+    throw new Error(`Invalid REMNIC_BENCH_CANARY_FLOOR: ${raw}`);
+  }
+}
+function resolveEffectiveCanaryFloor(preset = void 0, envRaw = process2.env.REMNIC_BENCH_CANARY_FLOOR) {
+  if (preset !== void 0) {
+    return parseCanaryFloor(preset);
+  }
+  return resolveCanaryFloorFromEnv(envRaw);
+}
+function createCanaryAdapter(options = {}) {
+  const response = options.response ?? CANARY_FIXED_RECALL;
+  const emptySearch = options.emptySearch ?? false;
+  return {
+    async store(_sessionId, _messages) {
+    },
+    async recall(_sessionId, _query, _budgetChars) {
+      return response;
+    },
+    async search(_query, _limit, _sessionId) {
+      if (emptySearch) {
+        return [];
+      }
+      return [
+        {
+          turnIndex: 0,
+          role: "assistant",
+          snippet: response,
+          sessionId: "__canary__",
+          score: 0
+        }
+      ];
+    },
+    async reset(_sessionId) {
+    },
+    async getStats(_sessionId) {
+      return {
+        totalMessages: 0,
+        totalSummaryNodes: 0,
+        maxDepth: 0
+      };
+    },
+    async destroy() {
+    }
+  };
+}
+function assertCanaryUnderFloor(benchmark, score, floor = CANARY_SCORE_FLOOR) {
+  const validatedFloor = parseCanaryFloor(floor);
+  if (!Number.isFinite(score)) {
+    return { benchmark, score, floor: validatedFloor, passed: false };
+  }
+  return {
+    benchmark,
+    score,
+    floor: validatedFloor,
+    passed: score <= validatedFloor
+  };
+}
+
 // src/run-identity.ts
 var BENCHMARK_RUN_ID_ENV = "REMNIC_BENCH_RUN_ID";
 var generatedBenchmarkRunId;
@@ -532,13 +610,18 @@ function replaceLoneSurrogates(value) {
   return out;
 }
 async function writeBenchmarkResult(result, outputDir) {
+  const canaryFloor = resolveEffectiveCanaryFloor(result.meta.canaryFloor);
   const outputRoot = path3.resolve(outputDir);
   await mkdir2(outputRoot, { recursive: true });
-  const safeBenchmark = sanitizeFilenameSegment(result.meta.benchmark);
-  const safeRemnicVersion = sanitizeFilenameSegment(result.meta.remnicVersion);
-  const timestamp = sanitizeFilenameSegment(result.meta.timestamp.replace(/[:.]/g, "-"));
+  const stamped = {
+    ...result,
+    meta: { ...result.meta, canaryFloor }
+  };
+  const safeBenchmark = sanitizeFilenameSegment(stamped.meta.benchmark);
+  const safeRemnicVersion = sanitizeFilenameSegment(stamped.meta.remnicVersion);
+  const timestamp = sanitizeFilenameSegment(stamped.meta.timestamp.replace(/[:.]/g, "-"));
   const filePath = resolveContainedPath(outputRoot, `${safeBenchmark}-v${safeRemnicVersion}-${timestamp}.json`);
-  const publicBaseResult = sanitizeBenchmarkResultForJson(redactBenchmarkResultSecrets(result));
+  const publicBaseResult = sanitizeBenchmarkResultForJson(redactBenchmarkResultSecrets(stamped));
   const leaderboardArtifacts = await writeLeaderboardArtifactsForResult(publicBaseResult, outputRoot).catch(
     (error) => [
       {
@@ -1834,7 +1917,8 @@ var BENCHMARK_INTEGRITY_META_SCHEMA = {
     qrelsSealedHash: { type: "string", pattern: "^[0-9a-f]{64}$" },
     judgePromptHash: { type: "string", pattern: "^[0-9a-f]{64}$" },
     datasetHash: { type: "string", pattern: "^[0-9a-f]{64}$" },
-    canaryScore: { type: "number" }
+    canaryScore: { type: "number" },
+    canaryFloor: { type: "number", minimum: 0 }
   }
 };
 function integrityMetaIsComplete(value) {
@@ -1851,6 +1935,9 @@ function integrityMetaIsComplete(value) {
   if (candidate.canaryScore !== void 0 && (typeof candidate.canaryScore !== "number" || !Number.isFinite(candidate.canaryScore))) {
     return false;
   }
+  if (candidate.canaryFloor !== void 0 && (typeof candidate.canaryFloor !== "number" || !Number.isFinite(candidate.canaryFloor) || candidate.canaryFloor < 0)) {
+    return false;
+  }
   return true;
 }
 function assertIntegrityMetaPresent(value) {
@@ -1876,6 +1963,9 @@ function assertIntegrityMetaPresent(value) {
   if (candidate.canaryScore !== void 0 && (typeof candidate.canaryScore !== "number" || !Number.isFinite(candidate.canaryScore))) {
     missing.push("canaryScore");
   }
+  if (candidate.canaryFloor !== void 0 && (typeof candidate.canaryFloor !== "number" || !Number.isFinite(candidate.canaryFloor) || candidate.canaryFloor < 0)) {
+    missing.push("canaryFloor");
+  }
   if (missing.length > 0) {
     throw new Error(
       `Result integrity metadata is incomplete or malformed: ${missing.join(", ")}`
@@ -2968,7 +3058,8 @@ function toPublishedBenchmarkFeedEntry(result, provenance) {
       qrelsSealedHash: result.meta.qrelsSealedHash,
       judgePromptHash: result.meta.judgePromptHash,
       datasetHash: result.meta.datasetHash,
-      ...result.meta.canaryScore !== void 0 ? { canaryScore: result.meta.canaryScore } : {}
+      ...result.meta.canaryScore !== void 0 ? { canaryScore: result.meta.canaryScore } : {},
+      ...result.meta.canaryFloor !== void 0 ? { canaryFloor: result.meta.canaryFloor } : {}
     }
   };
 }
@@ -16726,6 +16817,13 @@ export {
   buildAmaBenchLeaderboardRows,
   serializeJsonl,
   isSecretKey,
+  CANARY_FIXED_RECALL,
+  CANARY_SCORE_FLOOR,
+  parseCanaryFloor,
+  resolveCanaryFloorFromEnv,
+  resolveEffectiveCanaryFloor,
+  createCanaryAdapter,
+  assertCanaryUnderFloor,
   redactBenchmarkResultSecrets,
   sanitizeBenchmarkResultForJson,
   writeBenchmarkResult,

package/dist/index.d.ts

@@ -362,6 +362,13 @@ interface BenchmarkIntegrityMeta {
      * Omitted only during the canary's own run.
      */
     canaryScore?: number;
+    /**
+     * Effective canary floor that gated `canaryScore` (custom
+     * `REMNIC_BENCH_CANARY_FLOOR` or the canonical default), persisted by
+     * `writeBenchmarkResult`. Readers apply it without the env so a custom
+     * floor survives restarts.
+     */
+    canaryFloor?: number;
 }
 declare const INTEGRITY_META_FIELDS: readonly ["splitType", "qrelsSealedHash", "judgePromptHash", "datasetHash"];
 declare const BENCHMARK_INTEGRITY_META_SCHEMA: {
@@ -387,6 +394,10 @@ declare const BENCHMARK_INTEGRITY_META_SCHEMA: {
         readonly canaryScore: {
             readonly type: "number";
         };
+        readonly canaryFloor: {
+            readonly type: "number";
+            readonly minimum: 0;
+        };
     };
 };
 declare function integrityMetaIsComplete(value: unknown): value is BenchmarkIntegrityMeta;
@@ -576,9 +587,10 @@ interface BenchmarkResult {
          */
         canaryScore?: number;
         /**
-         * Canary floor in force when this result was produced. Artifacts written
-         * under a custom floor persist it here so tooling badges against the
-         * gate that actually ran; absent means `CANARY_SCORE_FLOOR` applied.
+         * Effective canary floor (validated `REMNIC_BENCH_CANARY_FLOOR` or the
+         * canonical default) that gated `canaryScore`. Persisted by
+         * `writeBenchmarkResult` so readers judge the score against the
+         * producing run's floor without the environment variable.
          */
         canaryFloor?: number;
         /** "partial" if the benchmark was interrupted; absent or "complete" otherwise. */
@@ -1636,6 +1648,10 @@ declare const BENCHMARK_RESULT_SCHEMA: {
                 readonly canaryScore: {
                     readonly type: "number";
                 };
+                readonly canaryFloor: {
+                    readonly type: "number";
+                    readonly minimum: 0;
+                };
             };
         };
         readonly config: {
@@ -3914,6 +3930,8 @@ interface PublishedBenchmarkFeedEntry {
         judgePromptHash: string;
         datasetHash: string;
         canaryScore?: number;
+        /** Effective canary floor persisted with the result (see meta.canaryFloor). */
+        canaryFloor?: number;
     };
 }
 interface BuildBenchmarkPublishFeedOptions {
@@ -4352,6 +4370,23 @@ declare function serializeSealedQrels(artifact: SealedQrelsArtifact): string;
 declare const CANARY_FIXED_RECALL = "__remnic_canary_response__";
 /** The score floor the canary must NOT exceed for any benchmark. */
 declare const CANARY_SCORE_FLOOR = 0.1;
+/**
+ * Accept a finite canary floor >= 0. Reject NaN, infinities, negatives,
+ * and non-numbers. Callers treat absence separately so a malformed
+ * present value is never rewritten to the default.
+ */
+declare function parseCanaryFloor(value: unknown, label?: string): number;
+/**
+ * Effective canary floor for this process: a validated
+ * `REMNIC_BENCH_CANARY_FLOOR` override, else the canonical default.
+ * An explicitly empty override is invalid, not absent.
+ */
+declare function resolveCanaryFloorFromEnv(raw?: string | undefined): number;
+/**
+ * Producer/audit precedence: a present preset must be a valid floor;
+ * otherwise fall through to the env parser (absent env → 0.1).
+ */
+declare function resolveEffectiveCanaryFloor(preset?: unknown, envRaw?: string | undefined): number;
 interface CanaryAdapterOptions {
     /**
      * Override the response string used by `recall`. Useful for running two
@@ -9838,4 +9873,4 @@ declare function pickOne<T>(rng: SeededRandom, items: readonly T[]): T;
 /** Deterministic Fisher-Yates shuffle returning a new array. */
 declare function shuffled<T>(rng: SeededRandom, items: readonly T[]): T[];
 
-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 ActionIntentV1, ActionIntentV1Schema, type AggregateMetrics, type AmaBenchDiagnosticAdapterOptions, type AmaBenchDiagnosticAnswererMode, type AmaBenchDiagnosticBreakdown, type AmaBenchDiagnosticMatrixArtifact, type AmaBenchDiagnosticRecallMode, type AmaBenchDiagnosticRunContext, type AmaBenchDiagnosticTaskEvidence, type AmaBenchDiagnosticTaskRow, type AmaBenchDiagnosticVariant, type AmaBenchDiagnosticVariantSummary, type AnalyzeRepeatedFailureOptions, 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, type AttributeOptions, type AttributionClass, type AttributionEnvironment, type AttributionLabel, type AttributionMemory, type AttributionReport, BENCHMARK_ARTIFACT_SCHEMA_VERSION, BENCHMARK_INTEGRITY_META_SCHEMA, BENCHMARK_REPRO_MANIFEST_FILENAME, BENCHMARK_REPRO_MANIFEST_SCHEMA_VERSION, BENCHMARK_RESULT_SCHEMA, BENCHMARK_SPLIT_TYPES, BUILD_WEEK_EVIDENCE_RECEIPT_SCHEMA_VERSION, BUILD_WEEK_LIMITATIONS, type BaseTask, BaseTaskSchema, type BaselineRow, type BaselineScenario, type BeamDatasetPreview, type BenchAggregateMetric, type BenchAssistantTaskDetails, type BenchConfig, type BenchIntegritySplit, type BenchIntegritySummary, type BenchJudge, type BenchJudgeResult, type BenchMemoryAdapter, type BenchMetricHighlight, type BenchModelSource, type BenchPerSeedScore, 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 BenchResultFileWarning, type BenchResultSummary, type BenchResultSummaryPayload, type BenchRuntimeProfile, type BenchTaskScoreEntry, type BenchTaskSummary, type BenchTier, type BenchmarkArtifact, type BenchmarkArtifactEnvironment, type BenchmarkArtifactHardware, type BenchmarkArtifactJudgeCalibration, type BenchmarkArtifactPerTaskScore, type BenchmarkArtifactSystem, type BenchmarkArtifactTier, type BenchmarkCategory, type BenchmarkDefinition, type BenchmarkExecutionProvenance, type BenchmarkIntegrityMeta, type BenchmarkMeta, type BenchmarkMode, type BenchmarkReport, type BenchmarkReproManifest, type BenchmarkReproManifestDataset, type BenchmarkReproManifestFile, type BenchmarkReproManifestResult, type BenchmarkReproManifestSupplementalArtifact, type BenchmarkResult, type BenchmarkSplitType, type BenchmarkStatus, type BenchmarkSuiteResult, type BenchmarkTier, type BootstrapKappaOptions, type BootstrapKappaResult, type BuildBenchmarkArtifactInput, type BuildBenchmarkPublishFeedOptions, type BuildBenchmarkReproManifestOptions, type BuildBuildWeekEvidenceReceiptOptions, type BuildWeekEvidenceReceipt, type BuildWeekEvidenceReceiptProvider, type BuildWeekLimitationCode, 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 CodexCliNativeUsage, type CodexCliProviderConfig, CodexCreditAccountingError, type CodexCreditBudgetConfig, CodexCreditDispatchError, 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 ControlledGateDecision, type ControlledResponsesAgentDriverConfig, type ControlledResponsesCaps, type ControlledResponsesDisposition, ControlledResponsesDriver, type ControlledResponsesDriverConfig, type ControlledResponsesEpisodeInput, type ControlledResponsesEpisodeResult, type ControlledResponsesFault, type ControlledResponsesResponseEvent, type ControlledResponsesToolDefinition, type ControlledResponsesToolEvent, type ControlledResponsesTransport, type CustomBenchmarkScoring, type CustomBenchmarkSpec, type CustomBenchmarkTask, DATASET_SPLITS, 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, DRIFT_GEN_DEFAULTS, DRIFT_GEN_VERSION, type DatasetSource, type DatasetSplit, type DiagnoseLoComoProfileDeltaOptions, type DiagnoseLoComoRecallDeltaOptions, type DiscoveredModel, type DriftGenAuditRecord, type DriftGenCorpus, type DriftGenManifest, type DriftGenOptions, type DriftGenResult, type DriftSession, type DriftSessionTurn, type DriftValidationReport, type DriftValidationStats, EMPTY_CONTAMINATION_MANIFEST, type EffectSizeInterpretation, type EffectSizeSummary, type EvaluateTaskStateOptions, type ExplainResult, type ExtractedEntity, type ExtractedLink, type ExtractedPage, type ExtractionAttackOptions, type ExtractionAttackResult, type ExtractionAttackTarget, type FixtureGenerator, type FixtureOutput, type FixtureVariant, GATE_STATUSES, GENERAL_ANSWER_JUDGE_RUBRIC, type GateStatus, type GeneratedFile, type GeneratedRepo, type GoldEntity, type GoldEntityType, type GoldFact, type GoldFactKind, type GoldGraph, type GoldLink, type GoldMemoryAttribution, type GoldPage, type GoldProbe, type GoldProbeCategory, type H6BenchmarkDataset, H6BenchmarkDatasetSchema, type H6TrapId, H6_ACTION_INTENT_JSON_SCHEMA, H6_ARMS, H6_DATASET_JSON_SCHEMA, H6_DECISION_RULE, H6_FROZEN_INVENTORY_HASH, H6_FROZEN_SEED, H6_FROZEN_SPLITS, H6_SUPPORT_ARTIFACT_PATHS, H6_TASK_JSON_SCHEMA, H6_TRAP_FINGERPRINT_JSON_SCHEMA, H6_TRAP_IDS, HOST_FAULT_RETRY_LIMIT, type HarnessRng, INJECTION_SUITE_ARMS, INJECTION_SUITE_FAMILIES, INJECTION_SUITE_VERSION, INTEGRITY_CIPHER_ALGORITHM, INTEGRITY_HASH_ALGORITHM, INTEGRITY_META_FIELDS, INVENTED_DOMAINS, type IngestionBenchAdapter, type IngestionLog, type InjectionSuiteArm, type InjectionSuiteCliInput, type InjectionSuiteCliResult, type InjectionSuiteEpisodeRow, type InjectionSuiteFamily, type InjectionSuiteRowIdentity, 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_DELTA_SCHEMA_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 LoCoMoCategory, type LoCoMoRetrievalMechanism, type LoCoMoRetrievalMechanismSummary, type LoCoMoRetrievalSessionReceipt, type LoCoMoRetrievalStructuralTrace, type LoCoMoRetrievalTaskDelta, type LoCoMoRetrievalTaskReceipt, type LoCoMoRetrievalTraceCoreCaptureReceipt, type LoCoMoRetrievalTraceDeltaReport, type LoCoMoRetrievalTraceProfile, type LoCoMoRetrievalTraceReceipt, type LoCoMoRetrievalTraceSelectionManifest, type LoCoMoRetrievalTraceSelector, type LoCoMoStructuralMultisetDelta, 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, MAX_ROW_ATTEMPTS, 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 MaterializeOptions, type MaterializedRepo, 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 OllamaChatMessage, 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 PairedAnswerReplayCache, type PairedAnswerReplayEntry, type ParsedOllamaChatResponse, 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, REPEATED_FAILURE_ARMS, REPEATED_FAILURE_CONFIDENCE_LEVEL, REPEATED_FAILURE_INVALID_REASONS, REPEATED_FAILURE_STATISTICS_DRAWS, REQUIRED_FRONTMATTER_FIELDS, type RecallMetrics, type RecoveredMemory, type RegressionDetail, type RegressionGateResult$1 as RegressionGateResult, type RemnicAdapterOptions, type RepeatedFailureActionEvaluator, type RepeatedFailureArm, type RepeatedFailureCheckpointLoadResult, type RepeatedFailureCliCommandResult, type RepeatedFailureEffectAnalysis, type RepeatedFailureEpisode, type RepeatedFailureEpisodeDriver, type RepeatedFailureEpisodeEvidence, type RepeatedFailureEpisodeInput, type RepeatedFailureEpisodeRow, type RepeatedFailureExpectedDesign, type RepeatedFailureFactPairAudit, type RepeatedFailureFinalRepoEvidence, type RepeatedFailureFinalState, type RepeatedFailureGateEvent, type RepeatedFailureHolmResult, type RepeatedFailureInterval, type RepeatedFailureInvalidReason, type RepeatedFailureIsolationIdentity, type RepeatedFailureLocalToolHost, type RepeatedFailureNullableInterval, RepeatedFailureOllamaChatDriver, type RepeatedFailureOllamaChatDriverConfig, type RepeatedFailureProposedAction, type RepeatedFailureRowCheckpoint, type RepeatedFailureRowClaim, type RepeatedFailureRowIdentity, RepeatedFailureRowStore, type RepeatedFailureRowStoreOptions, type RepeatedFailureRunMetadata, type RepeatedFailureStatisticalAnalysis, type RepeatedFailureSuiteManifest, type RepeatedFailureSupportDecision, type RepeatedFailureTaskCut, type RepeatedFailureTimidityAnalysis, type RepeatedFailureTokenUsage, type RepeatedFailureTokenizer, type RepeatedFailureToolDefinition, type RepeatedFailureToolExecutionResult, type RepeatedFailureTrapAuditArtifact, type RepeatedFailureTrapAuditExpected, type RepeatedFailureTrapAuditMetrics, type RepeatedFailureTrapAuditRow, type RepeatedFailureTrapAuditRowIdentity, type RepeatedFailureTrapAuditThresholds, type RepeatedFailureTry, type ReplayRepeatedFailureStatisticsOptions, type ReportCardProvenanceContext, type ResolveBenchRuntimeProfileOptions, type ResolvedBenchRuntimeProfile, type ResolvedLocalLabProfile, type ResolvedLocalLabRole, type ResolvedRunBenchmarkOptions, type ResponsesApiOutputItem, type ResponsesApiRequest, type ResponsesApiResponse, type ResponsesApiUsage, type RetrievalMissStage, type RevisionShas, type RotatedChoices, type RunBenchmarkOptions, type RunJudgeCalibrationOptions, type RunProceduralAblationCliArgs, type RunProceduralAblationOptions, type RunRepeatedFailureCliCommandInput, type RunRepeatedFailureSuiteOptions, type RunRepeatedFailureSuiteResult, type RunSequentialPhasesOptions, type RunTrapAuditOptions, SCHEMA_TIER_FIXTURE, SCHEMA_TIER_SMOKE_FIXTURE, SEALED_PROMPT_REGISTRY, SINGLE_FLAG_ABLATION_MATRIX, STATE_CLASSIFICATIONS, 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 SeededRandom, type SeededRng, type SequentialPhaseHooks, type SingleFlagAblationCell, type SingleFlagAblationId, type SpotCheckLogger, type StageObservation, type StageStatus, type StateClassification, type StateEvaluationResult, type StatisticalReport, type StrategyPatch, StrategyPatchSchema, type StructuredJudge, StructuredJudgeError, type StructuredJudgeErrorCode, type StructuredJudgeProvider, type StructuredJudgeTelemetry, type StructuredJudgeVerdict, type StructuredJudgeVerdictResult, type StructuredVerdictRequest, type SyntheticEdge, type SyntheticEmailIngestionAdapterOptions, type SyntheticFile, type SyntheticFileIR, SyntheticFileSchema, type SyntheticRepoConfig, type SyntheticSymbol, type SyntheticTargetOptions, TRAP_TAXONOMY, type TaskAttribution, type TaskAttributionGoldWitnessV1, type TaskAttributionRetrievalWitnessV1, type TaskAttributionWitness, type TaskAttributionWitnessRuntimeV1, type TaskAttributionWitnessV1, type TaskResult, type TaskTokenUsage, type TaskVariant, TaskVariantSchema, type TemporalRetrievalCase, type ThirdPartyAdapterConfig, type TierDetail, type TimelineEntry, type TokenUsage, type TrapFingerprintV1, TrapFingerprintV1Schema, type TrapTaxonomyItem, TrapTaxonomyItemSchema, type ValidationIssue, type ValidationReport, type WallMetric, type WriteBenchmarkArtifactResult, type ZepAdapterConfig, ZepMemCorrectAdapter, addContaminationEntry, aggregateTaskScores, analyzeRepeatedFailureRows, answerBenchmarkQuestion, applyPatchAndCommit, assertCanaryUnderFloor, assertIntegrityMetaPresent, assertPublishableIntegrity, assertSha256Hex, assertTrapDatasetPreflight, assistantMeetingPrepDefinition, assistantMorningBriefDefinition, assistantNextBestActionDefinition, assistantSynthesisDefinition, attributeGoldMemory, attributeRun, attributeTask, backlinkF1, binarizeJudgeScore, bootstrapCohensKappaConfidenceInterval, bootstrapMeanConfidenceInterval, buildAmaBenchDiagnosticMatrixArtifact, buildAmaBenchDiagnosticVariantSummary, buildAmaBenchLeaderboardRows, buildBaselineFromReport, buildBenchmarkArtifact, buildBenchmarkArtifactFilename, buildBenchmarkPublishFeed, buildBenchmarkReproManifest, buildBenchmarkRunSeeds, buildBuildWeekEvidenceReceipt, buildCodexCreditReceipt, buildDriftCorpus, buildJudgePayload, buildOracleTrajectoryRecall, buildProviderFreeLoCoMoRetrievalConfig, buildRepeatedFailureRowKey, buildSchemaTierFixture, buildSchemaTierSmokeFixture, calculateCodexBudgetUnits, calculateJaccardSimilarity, calendarFixture, canonicalJsonStringify, captureBenchmarkExecutionProvenance, captureLoCoMoRetrievalTrace, captureMachineFingerprint, chatFixture, checkCodingGraphRegression, checkDatasetContamination, checkRegression, clampScore, cohensD, compareResults, computeBenchmarkReproDatasetInventoryHash, computeBenchmarkReproManifestArtifactHash, computeCohensKappa, computeH6InventoryHash, computeH6SupportArtifactHashes, computeRevisionShas, computeSealHash, computeTrapAuditArtifactHash, computeTrapAuditMetrics, containsAnswer, createSeededRng$1 as createAdamSeededRng, createAmaBenchDiagnosticAdapter, createAnthropicProvider, createCanaryAdapter, createClaudeCliProvider, createCodexCliProvider, createSeededRng as createCodingGraphSeededRng, createControlledResponsesAgentDriver, createDeterministicSpotCheckLogger, createGatewayResponder, createLightweightAdapter, createLiteLlmProvider, createLocalLlmProvider, createMcpDemoMemCorrectAdapter, createMcpDemoMemoryAdapter, createMcpMemCorrectAdapter, createMcpMemoryAdapter, createMitigatedTarget, createOllamaProvider, createOpenAiCompatibleProvider, createOpenAiResponsesBenchJudge, createOpenAiResponsesProvider, createSeededRandom$1 as createProceduralAblationSeededRandom, createProvider, createProviderBackedAmaBenchRecommendedJudge, createProviderBackedJudge, createProviderBackedResponder, createProviderBackedStructuredJudge, createRemnicAdapter, createRepeatedFailureOllamaChatDriver, createResponderFromProvider, createSeededRandom, createSeededRng$2 as createSeededRng, createSpotCheckFileLogger, createStructuredBenchJudge, createStructuredJudgeFromProvider, createSyntheticEmailIngestionAdapter, createSyntheticTarget, createTimeoutGuardedAdapter, decideRepeatedFailureContent, decideRepeatedFailureStudy, decideRepeatedFailureTiming, defaultBenchmarkBaselineDir, defaultBenchmarkPublishPath, deleteBenchmarkResults, diagnoseLoCoMoRetrievalTraceDelta, diagnoseLoComoProfileDelta, diagnoseLoComoRecallDelta, discoverAllProviders, discoveryEndpointFor, emailFixture, entityRecall, evaluateTaskState, exactMatch, executeLocalRow, extractMetrics as extractCodingGraphMetrics, extractContentWords, extractMarkdownSectionsByTitle, f1Score, fixtureToAblationScenarios, formatHandoffNote, formatMissingDatasetError, generateDriftCorpus, generateFamilyVariants, generateH6BenchmarkDataset, generateReport, generateSuiteVariants, generateSyntheticRepo, getAblationCell, getBenchmark, getBenchmarkLowerIsBetter, getGitSha, getMemoryEvalDimension, getProviderBackedJudgePromptIdentity, getRemnicVersion, getTrapTaxonomyItem, hashBenchmarkArtifact, hashBytes, hashCanonicalJson, hashOrderedQuestionIds, hashString, holmAdjust, injectionSuiteResumeContractHash, integrityMetaIsComplete, interpretEffectSize, isAmaBenchUnknownLikeAnswer, isContaminationEntry, isContaminationManifest, isRepeatedFailureTimidityEquivalent, isSafeSyntheticPath, isSealedQrelsArtifact, isSha256Hex, isStructuredJudgeProvider, isTaskFailed, judgeMemCorrectCorrectionAcceptance, judgeMemCorrectStaleMemoryHarm, lexicalSimilarity, linkMatches, listBenchmarkBaselines, listBenchmarkResults, listBenchmarks, listMemoryEvalBenchmarkIds, listMemoryEvalDimensions, llmJudgeScore, llmJudgeScoreDetailed, loadAblationFixture, loadBaseline, loadBeamDatasetPreview, loadBenchmarkArtifact, loadBenchmarkBaseline, loadBenchmarkReportCardProvenance, loadBenchmarkResult, loadBenchmarkResultSummaries, loadCommittedH6BenchmarkDataset, loadCustomBenchmarkFile, loadJudgeCalibrationState, loadLoCoMo10, loadLocalLabManifest, loadLongMemEvalS, loadSealKeyFromEnv, loadSealedQrels, loadSealedRubric, matchEntity, materializeTaskRepo, mergeContaminationManifests, openSeal, orchestrateBenchmarkRuns, pairedDeltaConfidenceInterval, parseBenchmarkArtifact, parseCodexJsonlUsage, parseCustomBenchmark, parseLocalLabManifest, parseRepeatedFailureEpisodeRow, parseRubricResponse, parseSealedQrels, pickOne, pickStableQualifiedName, planInjectionSuiteRows, precisionAtK, preflightLoCoMoRetrievalTraceCapture, preflightLocalLabRole, projectFolderFixture, randomInt, recallAtK, reconcileCodexCreditLedger, redactBenchmarkResultSecrets, relativeRiskReduction, renderAttributionReportTable, renderBaselineMarkdown, renderBenchmarkResultExport, renderLoComoProfileDeltaMarkdown, renderLoComoRecallDeltaMarkdown, renderMemorySummaryForJudge, renderMemoryViewForAgent, replayRepeatedFailureStatistics, resolveAssistantAgent, resolveAssistantRubricId, resolveAssistantSeeds, resolveAssistantSpotCheckDir, resolveBenchRuntimeProfile, resolveBenchmarkPhaseTimeoutMs, resolveBenchmarkProgressLogging, resolveBenchmarkResultReference, resolveBenchmarkRunCount, resolveBenchmarkRunId, resolveCodexCreditBudgetConfig, resolveCommittedH6FixtureDirectory, resolveLocalLabJudgeProviderConfig, resolveLocalLabProfile, resolveLocalLabRole, resolveStructuredJudge, rotateDistractors, rougeL, runAssistantBenchmark, runAssistantMeetingPrepBenchmark, runAssistantMorningBriefBenchmark, runAssistantNextBestActionBenchmark, runAssistantSynthesisBenchmark, runAttributeCliCommand, runBaseline, runBenchSuite, runBenchmark, runCodingGraphBenchmark, runCustomBenchmarkFile, runDriftGenCliCommand, runExplain, runExtractionAttack, runInjectionSuiteCliCommand, runJudgeCalibration, runMitigatedBaseline, runProceduralAblation, runProceduralAblationCli, runRepeatedFailureCliCommand, runRepeatedFailurePaperReportCliCommand, runRepeatedFailureSuite, runSealedJudge, runSequentialPhases, runTrapAudit, runTrapAuditCliCommand, runWithinCodexCreditBudget, safeHexEqual, sanitizeBenchmarkResultForJson, sanitizeLoComoResultReference, saveBaseline, saveBenchmarkBaseline, schemaCompleteness, sealPayload, selectAmaBenchDiagnosticVariants, selectCalibrationSlice, selectFixtureVariant, serializeAttributionReport, serializeBenchmarkArtifact, serializeBuildWeekEvidenceReceipt, serializeH6FixtureJson, serializeJsonl, serializeLoCoMoRetrievalTraceDelta, serializeLoCoMoRetrievalTraceReceipt, serializeSealedQrels, shuffleTasks, shuffled, summarizeBenchmarkResult, timed, tokenizeContent, unresolvedHelperImports, validateDriftCorpus, validateH6Dataset, validateH6FixtureBundle, validateH6StateDefiningIndependence, validateOllamaChatEndpoint, verifyMatchingTrapAudit, verifyRubricDigest, verifyTrapAuditArtifact, writeBenchmarkArtifact, writeBenchmarkPublishFeed, writeBenchmarkReproManifest, writeBenchmarkResult, writeBuildWeekEvidenceReceipt, writeH6FixtureBundle, writeJudgeCalibrationState, writeLeaderboardArtifactsForResult, writeRepeatedFailurePaperArtifacts, writeRepeatedFailureRunMetadata, writeRepeatedFailureStatistics, 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 ActionIntentV1, ActionIntentV1Schema, type AggregateMetrics, type AmaBenchDiagnosticAdapterOptions, type AmaBenchDiagnosticAnswererMode, type AmaBenchDiagnosticBreakdown, type AmaBenchDiagnosticMatrixArtifact, type AmaBenchDiagnosticRecallMode, type AmaBenchDiagnosticRunContext, type AmaBenchDiagnosticTaskEvidence, type AmaBenchDiagnosticTaskRow, type AmaBenchDiagnosticVariant, type AmaBenchDiagnosticVariantSummary, type AnalyzeRepeatedFailureOptions, 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, type AttributeOptions, type AttributionClass, type AttributionEnvironment, type AttributionLabel, type AttributionMemory, type AttributionReport, BENCHMARK_ARTIFACT_SCHEMA_VERSION, BENCHMARK_INTEGRITY_META_SCHEMA, BENCHMARK_REPRO_MANIFEST_FILENAME, BENCHMARK_REPRO_MANIFEST_SCHEMA_VERSION, BENCHMARK_RESULT_SCHEMA, BENCHMARK_SPLIT_TYPES, BUILD_WEEK_EVIDENCE_RECEIPT_SCHEMA_VERSION, BUILD_WEEK_LIMITATIONS, type BaseTask, BaseTaskSchema, type BaselineRow, type BaselineScenario, type BeamDatasetPreview, type BenchAggregateMetric, type BenchAssistantTaskDetails, type BenchConfig, type BenchIntegritySplit, type BenchIntegritySummary, type BenchJudge, type BenchJudgeResult, type BenchMemoryAdapter, type BenchMetricHighlight, type BenchModelSource, type BenchPerSeedScore, 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 BenchResultFileWarning, type BenchResultSummary, type BenchResultSummaryPayload, type BenchRuntimeProfile, type BenchTaskScoreEntry, type BenchTaskSummary, type BenchTier, type BenchmarkArtifact, type BenchmarkArtifactEnvironment, type BenchmarkArtifactHardware, type BenchmarkArtifactJudgeCalibration, type BenchmarkArtifactPerTaskScore, type BenchmarkArtifactSystem, type BenchmarkArtifactTier, type BenchmarkCategory, type BenchmarkDefinition, type BenchmarkExecutionProvenance, type BenchmarkIntegrityMeta, type BenchmarkMeta, type BenchmarkMode, type BenchmarkReport, type BenchmarkReproManifest, type BenchmarkReproManifestDataset, type BenchmarkReproManifestFile, type BenchmarkReproManifestResult, type BenchmarkReproManifestSupplementalArtifact, type BenchmarkResult, type BenchmarkSplitType, type BenchmarkStatus, type BenchmarkSuiteResult, type BenchmarkTier, type BootstrapKappaOptions, type BootstrapKappaResult, type BuildBenchmarkArtifactInput, type BuildBenchmarkPublishFeedOptions, type BuildBenchmarkReproManifestOptions, type BuildBuildWeekEvidenceReceiptOptions, type BuildWeekEvidenceReceipt, type BuildWeekEvidenceReceiptProvider, type BuildWeekLimitationCode, 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 CodexCliNativeUsage, type CodexCliProviderConfig, CodexCreditAccountingError, type CodexCreditBudgetConfig, CodexCreditDispatchError, 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 ControlledGateDecision, type ControlledResponsesAgentDriverConfig, type ControlledResponsesCaps, type ControlledResponsesDisposition, ControlledResponsesDriver, type ControlledResponsesDriverConfig, type ControlledResponsesEpisodeInput, type ControlledResponsesEpisodeResult, type ControlledResponsesFault, type ControlledResponsesResponseEvent, type ControlledResponsesToolDefinition, type ControlledResponsesToolEvent, type ControlledResponsesTransport, type CustomBenchmarkScoring, type CustomBenchmarkSpec, type CustomBenchmarkTask, DATASET_SPLITS, 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, DRIFT_GEN_DEFAULTS, DRIFT_GEN_VERSION, type DatasetSource, type DatasetSplit, type DiagnoseLoComoProfileDeltaOptions, type DiagnoseLoComoRecallDeltaOptions, type DiscoveredModel, type DriftGenAuditRecord, type DriftGenCorpus, type DriftGenManifest, type DriftGenOptions, type DriftGenResult, type DriftSession, type DriftSessionTurn, type DriftValidationReport, type DriftValidationStats, EMPTY_CONTAMINATION_MANIFEST, type EffectSizeInterpretation, type EffectSizeSummary, type EvaluateTaskStateOptions, type ExplainResult, type ExtractedEntity, type ExtractedLink, type ExtractedPage, type ExtractionAttackOptions, type ExtractionAttackResult, type ExtractionAttackTarget, type FixtureGenerator, type FixtureOutput, type FixtureVariant, GATE_STATUSES, GENERAL_ANSWER_JUDGE_RUBRIC, type GateStatus, type GeneratedFile, type GeneratedRepo, type GoldEntity, type GoldEntityType, type GoldFact, type GoldFactKind, type GoldGraph, type GoldLink, type GoldMemoryAttribution, type GoldPage, type GoldProbe, type GoldProbeCategory, type H6BenchmarkDataset, H6BenchmarkDatasetSchema, type H6TrapId, H6_ACTION_INTENT_JSON_SCHEMA, H6_ARMS, H6_DATASET_JSON_SCHEMA, H6_DECISION_RULE, H6_FROZEN_INVENTORY_HASH, H6_FROZEN_SEED, H6_FROZEN_SPLITS, H6_SUPPORT_ARTIFACT_PATHS, H6_TASK_JSON_SCHEMA, H6_TRAP_FINGERPRINT_JSON_SCHEMA, H6_TRAP_IDS, HOST_FAULT_RETRY_LIMIT, type HarnessRng, INJECTION_SUITE_ARMS, INJECTION_SUITE_FAMILIES, INJECTION_SUITE_VERSION, INTEGRITY_CIPHER_ALGORITHM, INTEGRITY_HASH_ALGORITHM, INTEGRITY_META_FIELDS, INVENTED_DOMAINS, type IngestionBenchAdapter, type IngestionLog, type InjectionSuiteArm, type InjectionSuiteCliInput, type InjectionSuiteCliResult, type InjectionSuiteEpisodeRow, type InjectionSuiteFamily, type InjectionSuiteRowIdentity, 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_DELTA_SCHEMA_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 LoCoMoCategory, type LoCoMoRetrievalMechanism, type LoCoMoRetrievalMechanismSummary, type LoCoMoRetrievalSessionReceipt, type LoCoMoRetrievalStructuralTrace, type LoCoMoRetrievalTaskDelta, type LoCoMoRetrievalTaskReceipt, type LoCoMoRetrievalTraceCoreCaptureReceipt, type LoCoMoRetrievalTraceDeltaReport, type LoCoMoRetrievalTraceProfile, type LoCoMoRetrievalTraceReceipt, type LoCoMoRetrievalTraceSelectionManifest, type LoCoMoRetrievalTraceSelector, type LoCoMoStructuralMultisetDelta, 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, MAX_ROW_ATTEMPTS, 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 MaterializeOptions, type MaterializedRepo, 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 OllamaChatMessage, 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 PairedAnswerReplayCache, type PairedAnswerReplayEntry, type ParsedOllamaChatResponse, 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, REPEATED_FAILURE_ARMS, REPEATED_FAILURE_CONFIDENCE_LEVEL, REPEATED_FAILURE_INVALID_REASONS, REPEATED_FAILURE_STATISTICS_DRAWS, REQUIRED_FRONTMATTER_FIELDS, type RecallMetrics, type RecoveredMemory, type RegressionDetail, type RegressionGateResult$1 as RegressionGateResult, type RemnicAdapterOptions, type RepeatedFailureActionEvaluator, type RepeatedFailureArm, type RepeatedFailureCheckpointLoadResult, type RepeatedFailureCliCommandResult, type RepeatedFailureEffectAnalysis, type RepeatedFailureEpisode, type RepeatedFailureEpisodeDriver, type RepeatedFailureEpisodeEvidence, type RepeatedFailureEpisodeInput, type RepeatedFailureEpisodeRow, type RepeatedFailureExpectedDesign, type RepeatedFailureFactPairAudit, type RepeatedFailureFinalRepoEvidence, type RepeatedFailureFinalState, type RepeatedFailureGateEvent, type RepeatedFailureHolmResult, type RepeatedFailureInterval, type RepeatedFailureInvalidReason, type RepeatedFailureIsolationIdentity, type RepeatedFailureLocalToolHost, type RepeatedFailureNullableInterval, RepeatedFailureOllamaChatDriver, type RepeatedFailureOllamaChatDriverConfig, type RepeatedFailureProposedAction, type RepeatedFailureRowCheckpoint, type RepeatedFailureRowClaim, type RepeatedFailureRowIdentity, RepeatedFailureRowStore, type RepeatedFailureRowStoreOptions, type RepeatedFailureRunMetadata, type RepeatedFailureStatisticalAnalysis, type RepeatedFailureSuiteManifest, type RepeatedFailureSupportDecision, type RepeatedFailureTaskCut, type RepeatedFailureTimidityAnalysis, type RepeatedFailureTokenUsage, type RepeatedFailureTokenizer, type RepeatedFailureToolDefinition, type RepeatedFailureToolExecutionResult, type RepeatedFailureTrapAuditArtifact, type RepeatedFailureTrapAuditExpected, type RepeatedFailureTrapAuditMetrics, type RepeatedFailureTrapAuditRow, type RepeatedFailureTrapAuditRowIdentity, type RepeatedFailureTrapAuditThresholds, type RepeatedFailureTry, type ReplayRepeatedFailureStatisticsOptions, type ReportCardProvenanceContext, type ResolveBenchRuntimeProfileOptions, type ResolvedBenchRuntimeProfile, type ResolvedLocalLabProfile, type ResolvedLocalLabRole, type ResolvedRunBenchmarkOptions, type ResponsesApiOutputItem, type ResponsesApiRequest, type ResponsesApiResponse, type ResponsesApiUsage, type RetrievalMissStage, type RevisionShas, type RotatedChoices, type RunBenchmarkOptions, type RunJudgeCalibrationOptions, type RunProceduralAblationCliArgs, type RunProceduralAblationOptions, type RunRepeatedFailureCliCommandInput, type RunRepeatedFailureSuiteOptions, type RunRepeatedFailureSuiteResult, type RunSequentialPhasesOptions, type RunTrapAuditOptions, SCHEMA_TIER_FIXTURE, SCHEMA_TIER_SMOKE_FIXTURE, SEALED_PROMPT_REGISTRY, SINGLE_FLAG_ABLATION_MATRIX, STATE_CLASSIFICATIONS, 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 SeededRandom, type SeededRng, type SequentialPhaseHooks, type SingleFlagAblationCell, type SingleFlagAblationId, type SpotCheckLogger, type StageObservation, type StageStatus, type StateClassification, type StateEvaluationResult, type StatisticalReport, type StrategyPatch, StrategyPatchSchema, type StructuredJudge, StructuredJudgeError, type StructuredJudgeErrorCode, type StructuredJudgeProvider, type StructuredJudgeTelemetry, type StructuredJudgeVerdict, type StructuredJudgeVerdictResult, type StructuredVerdictRequest, type SyntheticEdge, type SyntheticEmailIngestionAdapterOptions, type SyntheticFile, type SyntheticFileIR, SyntheticFileSchema, type SyntheticRepoConfig, type SyntheticSymbol, type SyntheticTargetOptions, TRAP_TAXONOMY, type TaskAttribution, type TaskAttributionGoldWitnessV1, type TaskAttributionRetrievalWitnessV1, type TaskAttributionWitness, type TaskAttributionWitnessRuntimeV1, type TaskAttributionWitnessV1, type TaskResult, type TaskTokenUsage, type TaskVariant, TaskVariantSchema, type TemporalRetrievalCase, type ThirdPartyAdapterConfig, type TierDetail, type TimelineEntry, type TokenUsage, type TrapFingerprintV1, TrapFingerprintV1Schema, type TrapTaxonomyItem, TrapTaxonomyItemSchema, type ValidationIssue, type ValidationReport, type WallMetric, type WriteBenchmarkArtifactResult, type ZepAdapterConfig, ZepMemCorrectAdapter, addContaminationEntry, aggregateTaskScores, analyzeRepeatedFailureRows, answerBenchmarkQuestion, applyPatchAndCommit, assertCanaryUnderFloor, assertIntegrityMetaPresent, assertPublishableIntegrity, assertSha256Hex, assertTrapDatasetPreflight, assistantMeetingPrepDefinition, assistantMorningBriefDefinition, assistantNextBestActionDefinition, assistantSynthesisDefinition, attributeGoldMemory, attributeRun, attributeTask, backlinkF1, binarizeJudgeScore, bootstrapCohensKappaConfidenceInterval, bootstrapMeanConfidenceInterval, buildAmaBenchDiagnosticMatrixArtifact, buildAmaBenchDiagnosticVariantSummary, buildAmaBenchLeaderboardRows, buildBaselineFromReport, buildBenchmarkArtifact, buildBenchmarkArtifactFilename, buildBenchmarkPublishFeed, buildBenchmarkReproManifest, buildBenchmarkRunSeeds, buildBuildWeekEvidenceReceipt, buildCodexCreditReceipt, buildDriftCorpus, buildJudgePayload, buildOracleTrajectoryRecall, buildProviderFreeLoCoMoRetrievalConfig, buildRepeatedFailureRowKey, buildSchemaTierFixture, buildSchemaTierSmokeFixture, calculateCodexBudgetUnits, calculateJaccardSimilarity, calendarFixture, canonicalJsonStringify, captureBenchmarkExecutionProvenance, captureLoCoMoRetrievalTrace, captureMachineFingerprint, chatFixture, checkCodingGraphRegression, checkDatasetContamination, checkRegression, clampScore, cohensD, compareResults, computeBenchmarkReproDatasetInventoryHash, computeBenchmarkReproManifestArtifactHash, computeCohensKappa, computeH6InventoryHash, computeH6SupportArtifactHashes, computeRevisionShas, computeSealHash, computeTrapAuditArtifactHash, computeTrapAuditMetrics, containsAnswer, createSeededRng$1 as createAdamSeededRng, createAmaBenchDiagnosticAdapter, createAnthropicProvider, createCanaryAdapter, createClaudeCliProvider, createCodexCliProvider, createSeededRng as createCodingGraphSeededRng, createControlledResponsesAgentDriver, createDeterministicSpotCheckLogger, createGatewayResponder, createLightweightAdapter, createLiteLlmProvider, createLocalLlmProvider, createMcpDemoMemCorrectAdapter, createMcpDemoMemoryAdapter, createMcpMemCorrectAdapter, createMcpMemoryAdapter, createMitigatedTarget, createOllamaProvider, createOpenAiCompatibleProvider, createOpenAiResponsesBenchJudge, createOpenAiResponsesProvider, createSeededRandom$1 as createProceduralAblationSeededRandom, createProvider, createProviderBackedAmaBenchRecommendedJudge, createProviderBackedJudge, createProviderBackedResponder, createProviderBackedStructuredJudge, createRemnicAdapter, createRepeatedFailureOllamaChatDriver, createResponderFromProvider, createSeededRandom, createSeededRng$2 as createSeededRng, createSpotCheckFileLogger, createStructuredBenchJudge, createStructuredJudgeFromProvider, createSyntheticEmailIngestionAdapter, createSyntheticTarget, createTimeoutGuardedAdapter, decideRepeatedFailureContent, decideRepeatedFailureStudy, decideRepeatedFailureTiming, defaultBenchmarkBaselineDir, defaultBenchmarkPublishPath, deleteBenchmarkResults, diagnoseLoCoMoRetrievalTraceDelta, diagnoseLoComoProfileDelta, diagnoseLoComoRecallDelta, discoverAllProviders, discoveryEndpointFor, emailFixture, entityRecall, evaluateTaskState, exactMatch, executeLocalRow, extractMetrics as extractCodingGraphMetrics, extractContentWords, extractMarkdownSectionsByTitle, f1Score, fixtureToAblationScenarios, formatHandoffNote, formatMissingDatasetError, generateDriftCorpus, generateFamilyVariants, generateH6BenchmarkDataset, generateReport, generateSuiteVariants, generateSyntheticRepo, getAblationCell, getBenchmark, getBenchmarkLowerIsBetter, getGitSha, getMemoryEvalDimension, getProviderBackedJudgePromptIdentity, getRemnicVersion, getTrapTaxonomyItem, hashBenchmarkArtifact, hashBytes, hashCanonicalJson, hashOrderedQuestionIds, hashString, holmAdjust, injectionSuiteResumeContractHash, integrityMetaIsComplete, interpretEffectSize, isAmaBenchUnknownLikeAnswer, isContaminationEntry, isContaminationManifest, isRepeatedFailureTimidityEquivalent, isSafeSyntheticPath, isSealedQrelsArtifact, isSha256Hex, isStructuredJudgeProvider, isTaskFailed, judgeMemCorrectCorrectionAcceptance, judgeMemCorrectStaleMemoryHarm, lexicalSimilarity, linkMatches, listBenchmarkBaselines, listBenchmarkResults, listBenchmarks, listMemoryEvalBenchmarkIds, listMemoryEvalDimensions, llmJudgeScore, llmJudgeScoreDetailed, loadAblationFixture, loadBaseline, loadBeamDatasetPreview, loadBenchmarkArtifact, loadBenchmarkBaseline, loadBenchmarkReportCardProvenance, loadBenchmarkResult, loadBenchmarkResultSummaries, loadCommittedH6BenchmarkDataset, loadCustomBenchmarkFile, loadJudgeCalibrationState, loadLoCoMo10, loadLocalLabManifest, loadLongMemEvalS, loadSealKeyFromEnv, loadSealedQrels, loadSealedRubric, matchEntity, materializeTaskRepo, mergeContaminationManifests, openSeal, orchestrateBenchmarkRuns, pairedDeltaConfidenceInterval, parseBenchmarkArtifact, parseCanaryFloor, parseCodexJsonlUsage, parseCustomBenchmark, parseLocalLabManifest, parseRepeatedFailureEpisodeRow, parseRubricResponse, parseSealedQrels, pickOne, pickStableQualifiedName, planInjectionSuiteRows, precisionAtK, preflightLoCoMoRetrievalTraceCapture, preflightLocalLabRole, projectFolderFixture, randomInt, recallAtK, reconcileCodexCreditLedger, redactBenchmarkResultSecrets, relativeRiskReduction, renderAttributionReportTable, renderBaselineMarkdown, renderBenchmarkResultExport, renderLoComoProfileDeltaMarkdown, renderLoComoRecallDeltaMarkdown, renderMemorySummaryForJudge, renderMemoryViewForAgent, replayRepeatedFailureStatistics, resolveAssistantAgent, resolveAssistantRubricId, resolveAssistantSeeds, resolveAssistantSpotCheckDir, resolveBenchRuntimeProfile, resolveBenchmarkPhaseTimeoutMs, resolveBenchmarkProgressLogging, resolveBenchmarkResultReference, resolveBenchmarkRunCount, resolveBenchmarkRunId, resolveCanaryFloorFromEnv, resolveCodexCreditBudgetConfig, resolveCommittedH6FixtureDirectory, resolveEffectiveCanaryFloor, resolveLocalLabJudgeProviderConfig, resolveLocalLabProfile, resolveLocalLabRole, resolveStructuredJudge, rotateDistractors, rougeL, runAssistantBenchmark, runAssistantMeetingPrepBenchmark, runAssistantMorningBriefBenchmark, runAssistantNextBestActionBenchmark, runAssistantSynthesisBenchmark, runAttributeCliCommand, runBaseline, runBenchSuite, runBenchmark, runCodingGraphBenchmark, runCustomBenchmarkFile, runDriftGenCliCommand, runExplain, runExtractionAttack, runInjectionSuiteCliCommand, runJudgeCalibration, runMitigatedBaseline, runProceduralAblation, runProceduralAblationCli, runRepeatedFailureCliCommand, runRepeatedFailurePaperReportCliCommand, runRepeatedFailureSuite, runSealedJudge, runSequentialPhases, runTrapAudit, runTrapAuditCliCommand, runWithinCodexCreditBudget, safeHexEqual, sanitizeBenchmarkResultForJson, sanitizeLoComoResultReference, saveBaseline, saveBenchmarkBaseline, schemaCompleteness, sealPayload, selectAmaBenchDiagnosticVariants, selectCalibrationSlice, selectFixtureVariant, serializeAttributionReport, serializeBenchmarkArtifact, serializeBuildWeekEvidenceReceipt, serializeH6FixtureJson, serializeJsonl, serializeLoCoMoRetrievalTraceDelta, serializeLoCoMoRetrievalTraceReceipt, serializeSealedQrels, shuffleTasks, shuffled, summarizeBenchmarkResult, timed, tokenizeContent, unresolvedHelperImports, validateDriftCorpus, validateH6Dataset, validateH6FixtureBundle, validateH6StateDefiningIndependence, validateOllamaChatEndpoint, verifyMatchingTrapAudit, verifyRubricDigest, verifyTrapAuditArtifact, writeBenchmarkArtifact, writeBenchmarkPublishFeed, writeBenchmarkReproManifest, writeBenchmarkResult, writeBuildWeekEvidenceReceipt, writeH6FixtureBundle, writeJudgeCalibrationState, writeLeaderboardArtifactsForResult, writeRepeatedFailurePaperArtifacts, writeRepeatedFailureRunMetadata, writeRepeatedFailureStatistics, zeroScores };

package/dist/index.js

@@ -7,6 +7,8 @@ import {
   BaseTaskSchema,
   BenchmarkRunBlockReason,
   BenchmarkRunBlockedError,
+  CANARY_FIXED_RECALL,
+  CANARY_SCORE_FLOOR,
   CodexCreditAccountingError,
   CodexCreditDispatchError,
   ControlledResponsesDriver,
@@ -54,6 +56,7 @@ import {
   aggregateTaskScores,
   analyzeRepeatedFailureRows,
   applyPatchAndCommit,
+  assertCanaryUnderFloor,
   assertIntegrityMetaPresent,
   assertNoSymlinkComponents,
   assertPublishableIntegrity,
@@ -82,6 +85,7 @@ import {
   computeTrapAuditMetrics,
   containsAnswer,
   countFactTokens,
+  createCanaryAdapter,
   createControlledResponsesAgentDriver,
   createRepeatedFailureOllamaChatDriver,
   createSeededRandom,
@@ -128,6 +132,7 @@ import {
   materializeTaskRepo,
   mergeContaminationManifests,
   openSeal,
+  parseCanaryFloor,
   parseCodexJsonlUsage,
   parseDesign,
   parseEpisodesJsonl,
@@ -145,9 +150,11 @@ import {
   replayRepeatedFailureStatistics,
   resolveBenchmarkResultReference,
   resolveBenchmarkRunId,
+  resolveCanaryFloorFromEnv,
   resolveCodexCreditBudgetConfig,
   resolveCommittedH6FixtureDirectory,
   resolveContainedPath,
+  resolveEffectiveCanaryFloor,
   retryFetch,
   rougeL,
   runRepeatedFailureCliCommand,
@@ -180,7 +187,7 @@ import {
   writeLeaderboardArtifactsForResult,
   writeRepeatedFailureRunMetadata,
   writeRepeatedFailureStatistics
-} from "./chunk-VJ4YENME.js";
+} from "./chunk-PWFYSAUK.js";
 
 // src/build-week-evidence-receipt.ts
 import { createHash as createHash2 } from "crypto";
@@ -7570,7 +7577,8 @@ var BENCHMARK_RESULT_SCHEMA = {
         qrelsSealedHash: { type: "string", pattern: "^[0-9a-f]{64}$" },
         judgePromptHash: { type: "string", pattern: "^[0-9a-f]{64}$" },
         datasetHash: { type: "string", pattern: "^[0-9a-f]{64}$" },
-        canaryScore: { type: "number" }
+        canaryScore: { type: "number" },
+        canaryFloor: { type: "number", minimum: 0 }
       }
     },
     config: {
@@ -41871,60 +41879,6 @@ function serializeSealedQrels(artifact) {
   return canonicalJsonStringify(normalized);
 }
 
-// src/integrity/canary-adapter.ts
-var CANARY_FIXED_RECALL = "__remnic_canary_response__";
-var CANARY_SCORE_FLOOR = 0.1;
-function createCanaryAdapter(options = {}) {
-  const response = options.response ?? CANARY_FIXED_RECALL;
-  const emptySearch = options.emptySearch ?? false;
-  return {
-    async store(_sessionId, _messages) {
-    },
-    async recall(_sessionId, _query, _budgetChars) {
-      return response;
-    },
-    async search(_query, _limit, _sessionId) {
-      if (emptySearch) {
-        return [];
-      }
-      return [
-        {
-          turnIndex: 0,
-          role: "assistant",
-          snippet: response,
-          sessionId: "__canary__",
-          score: 0
-        }
-      ];
-    },
-    async reset(_sessionId) {
-    },
-    async getStats(_sessionId) {
-      return {
-        totalMessages: 0,
-        totalSummaryNodes: 0,
-        maxDepth: 0
-      };
-    },
-    async destroy() {
-    }
-  };
-}
-function assertCanaryUnderFloor(benchmark, score, floor = CANARY_SCORE_FLOOR) {
-  if (!Number.isFinite(floor) || floor < 0) {
-    throw new Error(`Canary floor must be a non-negative finite number; got ${floor}.`);
-  }
-  if (!Number.isFinite(score)) {
-    return { benchmark, score, floor, passed: false };
-  }
-  return {
-    benchmark,
-    score,
-    floor,
-    passed: score <= floor
-  };
-}
-
 // src/integrity/randomize.ts
 function createSeededRng(seed) {
   if (!Number.isFinite(seed)) {
@@ -47725,7 +47679,7 @@ async function writeRepeatedFailurePaperArtifacts(options) {
 }
 async function runRepeatedFailurePaperReportCliCommand(options) {
   try {
-    const { replayRepeatedFailureStatistics: replayRepeatedFailureStatistics2 } = await import("./repeated-failure-suite-runner-5BL2FBIP.js");
+    const { replayRepeatedFailureStatistics: replayRepeatedFailureStatistics2 } = await import("./repeated-failure-suite-runner-ONGURPYC.js");
     const replay = await replayRepeatedFailureStatistics2(options);
     if (replay.exitCode !== 0) return replay;
     const result = await writeRepeatedFailurePaperArtifacts(options);
@@ -48312,6 +48266,7 @@ export {
   orchestrateBenchmarkRuns,
   pairedDeltaConfidenceInterval,
   parseBenchmarkArtifact,
+  parseCanaryFloor,
   parseCodexJsonlUsage,
   parseCustomBenchmark,
   parseLocalLabManifest,
@@ -48348,8 +48303,10 @@ export {
   resolveBenchmarkResultReference,
   resolveBenchmarkRunCount,
   resolveBenchmarkRunId,
+  resolveCanaryFloorFromEnv,
   resolveCodexCreditBudgetConfig,
   resolveCommittedH6FixtureDirectory,
+  resolveEffectiveCanaryFloor,
   resolveLocalLabJudgeProviderConfig,
   resolveLocalLabProfile,
   resolveLocalLabRole,

package/dist/{repeated-failure-suite-runner-5BL2FBIP.js → repeated-failure-suite-runner-ONGURPYC.js}

@@ -11,7 +11,7 @@ import {
   runRepeatedFailureCliCommand,
   runRepeatedFailureSuite,
   verifyPreregistrationBinding
-} from "./chunk-VJ4YENME.js";
+} from "./chunk-PWFYSAUK.js";
 export {
   baseEvidence,
   buildDesign,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@remnic/bench",
-  "version": "9.69.35",
+  "version": "9.69.36",
   "description": "Retrieval latency ladder benchmarks + CI regression gates for @remnic/core",
   "type": "module",
   "main": "./dist/index.js",
@@ -41,8 +41,8 @@
     "hyparquet": "^1.25.7",
     "yaml": "^2.4.2",
     "zod": "^3.24.0",
-    "@remnic/coding-graph": "^9.69.35",
-    "@remnic/core": "^9.69.35"
+    "@remnic/coding-graph": "^9.69.36",
+    "@remnic/core": "^9.69.36"
   },
   "devDependencies": {
     "tsup": "^8.5.1",