bitfab 0.64.6 → 0.64.8

package/dist/index.d.cts

@@ -452,6 +452,18 @@ interface ReplayContext {
     dbSnapshotAccessed?: boolean;
 }
 
+type FlagScalar = string | number | boolean | null;
+type FlagValue = FlagScalar | FlagScalar[];
+type ReplayEnvironment = "local" | "ci" | "cloud_replay";
+interface ReplayInvocation {
+    flags: Record<string, FlagValue>;
+    sdk: {
+        language: "typescript";
+        version: string;
+    };
+    environment: ReplayEnvironment;
+}
+
 interface SimulationPlanNode {
     traceFunctionKey: string;
     name: string;
@@ -697,7 +709,7 @@ declare class HttpClient {
      * Start a replay session by fetching historical traces.
      * Blocking call - creates an experiment and returns lightweight item references.
      */
-    startReplay(traceFunctionKey: string, limit: number | undefined, traceIds?: string[], name?: string, codeChangeDescription?: string | null, codeChangeFiles?: CodeChangeFile[] | null, includeDbBranchLease?: boolean, experimentGroupId?: string, datasetIds?: string[], graderIds?: string[], dbBranchSettings?: DbBranchSettings, attempts?: number, includeOriginalMetadata?: boolean, onlyWithAssertions?: boolean, skipAssertionJudging?: boolean, experimentalSelectiveReplay?: SelectiveReplayOptions, notes?: string, metadata?: Record<string, string>, dryRun?: boolean): Promise<StartReplayResponse>;
+    startReplay(traceFunctionKey: string, limit: number | undefined, traceIds?: string[], name?: string, codeChangeDescription?: string | null, codeChangeFiles?: CodeChangeFile[] | null, includeDbBranchLease?: boolean, experimentGroupId?: string, datasetIds?: string[], graderIds?: string[], dbBranchSettings?: DbBranchSettings, attempts?: number, includeOriginalMetadata?: boolean, onlyWithAssertions?: boolean, skipAssertionJudging?: boolean, experimentalSelectiveReplay?: SelectiveReplayOptions, notes?: string, metadata?: Record<string, string>, dryRun?: boolean, invocation?: ReplayInvocation): Promise<StartReplayResponse>;
     resumeReplay(experimentId: string, traceFunctionKey: string, codeChangeFiles?: CodeChangeFile[] | null, force?: boolean): Promise<ResumeReplayResponse>;
     /**
      * Fetch an external span by ID.
@@ -1660,7 +1672,7 @@ type LabelConfidence = "VeryLow" | "Low" | "Medium" | "High" | "VeryHigh";
 type LabelAction = "set" | "archived" | "no-active-label" | "skipped";
 type LabelStatus = "labeled" | "skipped" | "unlabeled";
 type LabelSource = "human" | "agent";
-type LabelTarget = ({
+type LabelAddress = {
     traceId: string;
     originalTraceId?: never;
     attempt?: never;
@@ -1668,7 +1680,11 @@ type LabelTarget = ({
     originalTraceId: string;
     attempt?: number;
     traceId?: never;
-}) & {
+};
+type LabelTarget = LabelAddress & {
+    assertionId: string;
+};
+type ArchiveLabelTarget = LabelAddress & {
     assertionId?: string;
 };
 type LabelVerdict = {
@@ -1679,10 +1695,10 @@ type LabelVerdict = {
 } | {
     skip: true;
     annotation?: string;
-} | {
-    archive: true;
 };
-type LabelUpdate = LabelTarget & LabelVerdict;
+type LabelUpdate = (LabelTarget & LabelVerdict) | (ArchiveLabelTarget & {
+    archive: true;
+});
 type LabelOutcome = {
     key: string;
     traceId: string;
@@ -1690,7 +1706,7 @@ type LabelOutcome = {
 };
 type HumanLabelUpdate = {
     traceId: string;
-    assertionId?: string;
+    assertionId: string;
     label: boolean;
     annotation: string;
     confidence?: LabelConfidence;
@@ -1729,12 +1745,10 @@ declare class LabelsClient {
     constructor(httpClient: HttpClient);
     save(update: LabelUpdate, experimentId?: string): Promise<LabelOutcome>;
     saveAll(updates: LabelUpdate[], experimentId?: string): Promise<LabelOutcome[]>;
-    /** Withhold a verdict for a trace or replay attempt whose check did not run. */
     skip(target: LabelTarget & {
         annotation?: string;
     }, experimentId?: string): Promise<LabelOutcome>;
-    /** Clear a previous verdict for a trace, replay attempt, or assertion. */
-    archive(target: LabelTarget, experimentId?: string): Promise<LabelOutcome>;
+    archive(target: ArchiveLabelTarget, experimentId?: string): Promise<LabelOutcome>;
     saveHuman(update: HumanLabelUpdate): Promise<HumanLabelOutcome>;
     saveHumanAll(updates: HumanLabelUpdate[]): Promise<HumanLabelOutcome[]>;
     get(traceId: string): Promise<TraceLabels | null>;
@@ -3980,7 +3994,7 @@ interface CommitRef {
 /**
  * SDK version from package.json (injected at build time)
  */
-declare const __version__ = "0.64.6";
+declare const __version__ = "0.64.8";
 
 /**
  * Constants for the Bitfab SDK.
@@ -4119,4 +4133,4 @@ declare function seedFromRegistry(registry: ReplayRegistry, pipeline: string, ca
     run?: boolean;
 }): Promise<SeedResult>;
 
-export { type ActiveSpanContext, type AdaptContext, type AdaptInputsFn, type AddDatasetGradersResult, type AddDatasetTracesResult, type AllowedEnvVars, type ApprovalFields, type ApprovalState, type Approver, type ArchiveAssertionsParams, AssertionCategoriesClient, type AssertionCategory, type AssertionCategorySummary, type AssertionEvidenceParameter, type AssertionGeneration, AssertionGenerationError, type AssertionGenerationStatus, AssertionGenerationTimeoutError, type AssertionLabelEvidence, type AssertionVerdict, type Assignee, BITFAB_PROGRESS_PREFIX, type BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, type BitfabConfig, BitfabError, BitfabFunction, BitfabLangGraphCallbackHandler as BitfabLangChainCallbackHandler, BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, type BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabRivetIntegration, BitfabVercelAiHandler, type CaptureSurface, type CaptureWhen, type CapturedSpan, type CodeChangeFile, type CommitRef, type CurrentSpan, type CurrentTrace, DEFAULT_SERVICE_URL, type Dataset, type DatasetGraderRef, type DatasetTraceIds, DatasetsClient, type DbBranchOptions, DbBranchReplayError, type DbBranchTimings, type DbSnapshotConfig, type DbSnapshotProvider, type DbSnapshotRef, type DetachedTrace, type Experiment, type ExperimentCategoryTally, type ExperimentCodeChangeTotals, type ExperimentJitter, type ExperimentPage, type ExperimentRollup, type ExperimentRollupResult, type ExperimentRunBy, type ExperimentStatus, type ExperimentTally, type ExperimentTotals, ExperimentsClient, type GenerateAssertionsOptions, type GetGraderLabelsParams, type GraderLabel, type GraderLabelOutcome, type GraderLabelSource, type GraderLabelUpdate, type GraderRerun, GraderRerunError, type GraderRerunProgress, type GraderRerunResult, type GraderRerunStatus, GraderRerunTimeoutError, GradersClient, HttpClient, type HumanLabelOutcome, type HumanLabelUpdate, type Justification, type JustificationSpan, type LabelAction, type LabelConfidence, type LabelOutcome, type LabelSource, type LabelStatus, type LabelTarget, type LabelUpdate, LabelsClient, type LangGraphIntegrationOptions, type ListDatasetsParams, type ListExperimentsParams, MixedTracingError, type MockOverride, type MockOverrideCtx, type MockOverrideInput, type MockOverrideResolver, type MockStrategy, type MockValue, NO_MOCK_OVERRIDE, type NodeMatcher, type NodeMethodDecorator, type NodeOptions, type OrganizationMember, OrganizationMembersClient, type ProviderDefinition, type RemoveDatasetGradersResult, type RemoveDatasetTracesResult, ReplayBranch, ReplayError, type ReplayExperiment, type ReplayInterrupt, type ReplayInterruptSignal, type ReplayItem, type ReplayItemFinishProgress, type ReplayItemStartProgress, type ReplayNodeIdentity, type ReplayOptions, type ReplayOptionsFactory, type ReplayProgress, type ReplayProgressItem, type ReplayRegistration, type ReplayRegistry, type ReplayRegistryContext, type ReplayRegistryOptions, type ReplayResult, type RerunGradersOptions, type RerunGradersResult, type ReseedOptions, type ReseedRegistryResult, type ReseedResult, type RivetContextLike, type RivetEffectType, type RivetIntegrationOptions, SUPPORTED_PROVIDERS, type SaveAssertion, type SaveAssertionCategoryParams, type SaveAssertionsAllParams, type SaveAssertionsParams, type SaveDatasetParams, type SaveDatasetResult, type SeedCase, type SeedCaseOptions, type SeedResult, type SeedRunOptions, type SelectiveReplayDecision, type SelectiveReplayOptions, type SelectiveReplayReport, type SpanLookup, type SpanMethodDecorator, type SpanMethodDecoratorContext, type SpanNodeMeta, type SpanOccurrence, type SpanOptions, type SpanType, type TokenUsage, type TraceAssertion, type TraceAssertionSource, type TraceAssertionsResult, type TraceAssertionsUpdate, type TraceIngestionType, type TraceLabels, type TraceResponse, type TraceSearchFilters, type TraceSearchParams, type TraceSearchResult, type TraceTarget, type TraceTargetOccurrence, TracesClient, type TracingProcessor, type VercelCallParams, type VercelGenerateResult, type VercelStreamResult, type WrapBAMLOptions, type WrappedBamlFn, __version__, defineReplayRegistry, finalizers, flushTraces, getCurrentReplayBranch, getCurrentSpan, getCurrentTrace, reportReplayProgress, reseedFromRegistry, seedFromRegistry, serializeReplayResult };
+export { type ActiveSpanContext, type AdaptContext, type AdaptInputsFn, type AddDatasetGradersResult, type AddDatasetTracesResult, type AllowedEnvVars, type ApprovalFields, type ApprovalState, type Approver, type ArchiveAssertionsParams, type ArchiveLabelTarget, AssertionCategoriesClient, type AssertionCategory, type AssertionCategorySummary, type AssertionEvidenceParameter, type AssertionGeneration, AssertionGenerationError, type AssertionGenerationStatus, AssertionGenerationTimeoutError, type AssertionLabelEvidence, type AssertionVerdict, type Assignee, BITFAB_PROGRESS_PREFIX, type BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, type BitfabConfig, BitfabError, BitfabFunction, BitfabLangGraphCallbackHandler as BitfabLangChainCallbackHandler, BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, type BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabRivetIntegration, BitfabVercelAiHandler, type CaptureSurface, type CaptureWhen, type CapturedSpan, type CodeChangeFile, type CommitRef, type CurrentSpan, type CurrentTrace, DEFAULT_SERVICE_URL, type Dataset, type DatasetGraderRef, type DatasetTraceIds, DatasetsClient, type DbBranchOptions, DbBranchReplayError, type DbBranchTimings, type DbSnapshotConfig, type DbSnapshotProvider, type DbSnapshotRef, type DetachedTrace, type Experiment, type ExperimentCategoryTally, type ExperimentCodeChangeTotals, type ExperimentJitter, type ExperimentPage, type ExperimentRollup, type ExperimentRollupResult, type ExperimentRunBy, type ExperimentStatus, type ExperimentTally, type ExperimentTotals, ExperimentsClient, type GenerateAssertionsOptions, type GetGraderLabelsParams, type GraderLabel, type GraderLabelOutcome, type GraderLabelSource, type GraderLabelUpdate, type GraderRerun, GraderRerunError, type GraderRerunProgress, type GraderRerunResult, type GraderRerunStatus, GraderRerunTimeoutError, GradersClient, HttpClient, type HumanLabelOutcome, type HumanLabelUpdate, type Justification, type JustificationSpan, type LabelAction, type LabelConfidence, type LabelOutcome, type LabelSource, type LabelStatus, type LabelTarget, type LabelUpdate, LabelsClient, type LangGraphIntegrationOptions, type ListDatasetsParams, type ListExperimentsParams, MixedTracingError, type MockOverride, type MockOverrideCtx, type MockOverrideInput, type MockOverrideResolver, type MockStrategy, type MockValue, NO_MOCK_OVERRIDE, type NodeMatcher, type NodeMethodDecorator, type NodeOptions, type OrganizationMember, OrganizationMembersClient, type ProviderDefinition, type RemoveDatasetGradersResult, type RemoveDatasetTracesResult, ReplayBranch, ReplayError, type ReplayExperiment, type ReplayInterrupt, type ReplayInterruptSignal, type ReplayItem, type ReplayItemFinishProgress, type ReplayItemStartProgress, type ReplayNodeIdentity, type ReplayOptions, type ReplayOptionsFactory, type ReplayProgress, type ReplayProgressItem, type ReplayRegistration, type ReplayRegistry, type ReplayRegistryContext, type ReplayRegistryOptions, type ReplayResult, type RerunGradersOptions, type RerunGradersResult, type ReseedOptions, type ReseedRegistryResult, type ReseedResult, type RivetContextLike, type RivetEffectType, type RivetIntegrationOptions, SUPPORTED_PROVIDERS, type SaveAssertion, type SaveAssertionCategoryParams, type SaveAssertionsAllParams, type SaveAssertionsParams, type SaveDatasetParams, type SaveDatasetResult, type SeedCase, type SeedCaseOptions, type SeedResult, type SeedRunOptions, type SelectiveReplayDecision, type SelectiveReplayOptions, type SelectiveReplayReport, type SpanLookup, type SpanMethodDecorator, type SpanMethodDecoratorContext, type SpanNodeMeta, type SpanOccurrence, type SpanOptions, type SpanType, type TokenUsage, type TraceAssertion, type TraceAssertionSource, type TraceAssertionsResult, type TraceAssertionsUpdate, type TraceIngestionType, type TraceLabels, type TraceResponse, type TraceSearchFilters, type TraceSearchParams, type TraceSearchResult, type TraceTarget, type TraceTargetOccurrence, TracesClient, type TracingProcessor, type VercelCallParams, type VercelGenerateResult, type VercelStreamResult, type WrapBAMLOptions, type WrappedBamlFn, __version__, defineReplayRegistry, finalizers, flushTraces, getCurrentReplayBranch, getCurrentSpan, getCurrentTrace, reportReplayProgress, reseedFromRegistry, seedFromRegistry, serializeReplayResult };

package/dist/index.d.ts

@@ -452,6 +452,18 @@ interface ReplayContext {
     dbSnapshotAccessed?: boolean;
 }
 
+type FlagScalar = string | number | boolean | null;
+type FlagValue = FlagScalar | FlagScalar[];
+type ReplayEnvironment = "local" | "ci" | "cloud_replay";
+interface ReplayInvocation {
+    flags: Record<string, FlagValue>;
+    sdk: {
+        language: "typescript";
+        version: string;
+    };
+    environment: ReplayEnvironment;
+}
+
 interface SimulationPlanNode {
     traceFunctionKey: string;
     name: string;
@@ -697,7 +709,7 @@ declare class HttpClient {
      * Start a replay session by fetching historical traces.
      * Blocking call - creates an experiment and returns lightweight item references.
      */
-    startReplay(traceFunctionKey: string, limit: number | undefined, traceIds?: string[], name?: string, codeChangeDescription?: string | null, codeChangeFiles?: CodeChangeFile[] | null, includeDbBranchLease?: boolean, experimentGroupId?: string, datasetIds?: string[], graderIds?: string[], dbBranchSettings?: DbBranchSettings, attempts?: number, includeOriginalMetadata?: boolean, onlyWithAssertions?: boolean, skipAssertionJudging?: boolean, experimentalSelectiveReplay?: SelectiveReplayOptions, notes?: string, metadata?: Record<string, string>, dryRun?: boolean): Promise<StartReplayResponse>;
+    startReplay(traceFunctionKey: string, limit: number | undefined, traceIds?: string[], name?: string, codeChangeDescription?: string | null, codeChangeFiles?: CodeChangeFile[] | null, includeDbBranchLease?: boolean, experimentGroupId?: string, datasetIds?: string[], graderIds?: string[], dbBranchSettings?: DbBranchSettings, attempts?: number, includeOriginalMetadata?: boolean, onlyWithAssertions?: boolean, skipAssertionJudging?: boolean, experimentalSelectiveReplay?: SelectiveReplayOptions, notes?: string, metadata?: Record<string, string>, dryRun?: boolean, invocation?: ReplayInvocation): Promise<StartReplayResponse>;
     resumeReplay(experimentId: string, traceFunctionKey: string, codeChangeFiles?: CodeChangeFile[] | null, force?: boolean): Promise<ResumeReplayResponse>;
     /**
      * Fetch an external span by ID.
@@ -1660,7 +1672,7 @@ type LabelConfidence = "VeryLow" | "Low" | "Medium" | "High" | "VeryHigh";
 type LabelAction = "set" | "archived" | "no-active-label" | "skipped";
 type LabelStatus = "labeled" | "skipped" | "unlabeled";
 type LabelSource = "human" | "agent";
-type LabelTarget = ({
+type LabelAddress = {
     traceId: string;
     originalTraceId?: never;
     attempt?: never;
@@ -1668,7 +1680,11 @@ type LabelTarget = ({
     originalTraceId: string;
     attempt?: number;
     traceId?: never;
-}) & {
+};
+type LabelTarget = LabelAddress & {
+    assertionId: string;
+};
+type ArchiveLabelTarget = LabelAddress & {
     assertionId?: string;
 };
 type LabelVerdict = {
@@ -1679,10 +1695,10 @@ type LabelVerdict = {
 } | {
     skip: true;
     annotation?: string;
-} | {
-    archive: true;
 };
-type LabelUpdate = LabelTarget & LabelVerdict;
+type LabelUpdate = (LabelTarget & LabelVerdict) | (ArchiveLabelTarget & {
+    archive: true;
+});
 type LabelOutcome = {
     key: string;
     traceId: string;
@@ -1690,7 +1706,7 @@ type LabelOutcome = {
 };
 type HumanLabelUpdate = {
     traceId: string;
-    assertionId?: string;
+    assertionId: string;
     label: boolean;
     annotation: string;
     confidence?: LabelConfidence;
@@ -1729,12 +1745,10 @@ declare class LabelsClient {
     constructor(httpClient: HttpClient);
     save(update: LabelUpdate, experimentId?: string): Promise<LabelOutcome>;
     saveAll(updates: LabelUpdate[], experimentId?: string): Promise<LabelOutcome[]>;
-    /** Withhold a verdict for a trace or replay attempt whose check did not run. */
     skip(target: LabelTarget & {
         annotation?: string;
     }, experimentId?: string): Promise<LabelOutcome>;
-    /** Clear a previous verdict for a trace, replay attempt, or assertion. */
-    archive(target: LabelTarget, experimentId?: string): Promise<LabelOutcome>;
+    archive(target: ArchiveLabelTarget, experimentId?: string): Promise<LabelOutcome>;
     saveHuman(update: HumanLabelUpdate): Promise<HumanLabelOutcome>;
     saveHumanAll(updates: HumanLabelUpdate[]): Promise<HumanLabelOutcome[]>;
     get(traceId: string): Promise<TraceLabels | null>;
@@ -3980,7 +3994,7 @@ interface CommitRef {
 /**
  * SDK version from package.json (injected at build time)
  */
-declare const __version__ = "0.64.6";
+declare const __version__ = "0.64.8";
 
 /**
  * Constants for the Bitfab SDK.
@@ -4119,4 +4133,4 @@ declare function seedFromRegistry(registry: ReplayRegistry, pipeline: string, ca
     run?: boolean;
 }): Promise<SeedResult>;
 
-export { type ActiveSpanContext, type AdaptContext, type AdaptInputsFn, type AddDatasetGradersResult, type AddDatasetTracesResult, type AllowedEnvVars, type ApprovalFields, type ApprovalState, type Approver, type ArchiveAssertionsParams, AssertionCategoriesClient, type AssertionCategory, type AssertionCategorySummary, type AssertionEvidenceParameter, type AssertionGeneration, AssertionGenerationError, type AssertionGenerationStatus, AssertionGenerationTimeoutError, type AssertionLabelEvidence, type AssertionVerdict, type Assignee, BITFAB_PROGRESS_PREFIX, type BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, type BitfabConfig, BitfabError, BitfabFunction, BitfabLangGraphCallbackHandler as BitfabLangChainCallbackHandler, BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, type BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabRivetIntegration, BitfabVercelAiHandler, type CaptureSurface, type CaptureWhen, type CapturedSpan, type CodeChangeFile, type CommitRef, type CurrentSpan, type CurrentTrace, DEFAULT_SERVICE_URL, type Dataset, type DatasetGraderRef, type DatasetTraceIds, DatasetsClient, type DbBranchOptions, DbBranchReplayError, type DbBranchTimings, type DbSnapshotConfig, type DbSnapshotProvider, type DbSnapshotRef, type DetachedTrace, type Experiment, type ExperimentCategoryTally, type ExperimentCodeChangeTotals, type ExperimentJitter, type ExperimentPage, type ExperimentRollup, type ExperimentRollupResult, type ExperimentRunBy, type ExperimentStatus, type ExperimentTally, type ExperimentTotals, ExperimentsClient, type GenerateAssertionsOptions, type GetGraderLabelsParams, type GraderLabel, type GraderLabelOutcome, type GraderLabelSource, type GraderLabelUpdate, type GraderRerun, GraderRerunError, type GraderRerunProgress, type GraderRerunResult, type GraderRerunStatus, GraderRerunTimeoutError, GradersClient, HttpClient, type HumanLabelOutcome, type HumanLabelUpdate, type Justification, type JustificationSpan, type LabelAction, type LabelConfidence, type LabelOutcome, type LabelSource, type LabelStatus, type LabelTarget, type LabelUpdate, LabelsClient, type LangGraphIntegrationOptions, type ListDatasetsParams, type ListExperimentsParams, MixedTracingError, type MockOverride, type MockOverrideCtx, type MockOverrideInput, type MockOverrideResolver, type MockStrategy, type MockValue, NO_MOCK_OVERRIDE, type NodeMatcher, type NodeMethodDecorator, type NodeOptions, type OrganizationMember, OrganizationMembersClient, type ProviderDefinition, type RemoveDatasetGradersResult, type RemoveDatasetTracesResult, ReplayBranch, ReplayError, type ReplayExperiment, type ReplayInterrupt, type ReplayInterruptSignal, type ReplayItem, type ReplayItemFinishProgress, type ReplayItemStartProgress, type ReplayNodeIdentity, type ReplayOptions, type ReplayOptionsFactory, type ReplayProgress, type ReplayProgressItem, type ReplayRegistration, type ReplayRegistry, type ReplayRegistryContext, type ReplayRegistryOptions, type ReplayResult, type RerunGradersOptions, type RerunGradersResult, type ReseedOptions, type ReseedRegistryResult, type ReseedResult, type RivetContextLike, type RivetEffectType, type RivetIntegrationOptions, SUPPORTED_PROVIDERS, type SaveAssertion, type SaveAssertionCategoryParams, type SaveAssertionsAllParams, type SaveAssertionsParams, type SaveDatasetParams, type SaveDatasetResult, type SeedCase, type SeedCaseOptions, type SeedResult, type SeedRunOptions, type SelectiveReplayDecision, type SelectiveReplayOptions, type SelectiveReplayReport, type SpanLookup, type SpanMethodDecorator, type SpanMethodDecoratorContext, type SpanNodeMeta, type SpanOccurrence, type SpanOptions, type SpanType, type TokenUsage, type TraceAssertion, type TraceAssertionSource, type TraceAssertionsResult, type TraceAssertionsUpdate, type TraceIngestionType, type TraceLabels, type TraceResponse, type TraceSearchFilters, type TraceSearchParams, type TraceSearchResult, type TraceTarget, type TraceTargetOccurrence, TracesClient, type TracingProcessor, type VercelCallParams, type VercelGenerateResult, type VercelStreamResult, type WrapBAMLOptions, type WrappedBamlFn, __version__, defineReplayRegistry, finalizers, flushTraces, getCurrentReplayBranch, getCurrentSpan, getCurrentTrace, reportReplayProgress, reseedFromRegistry, seedFromRegistry, serializeReplayResult };
+export { type ActiveSpanContext, type AdaptContext, type AdaptInputsFn, type AddDatasetGradersResult, type AddDatasetTracesResult, type AllowedEnvVars, type ApprovalFields, type ApprovalState, type Approver, type ArchiveAssertionsParams, type ArchiveLabelTarget, AssertionCategoriesClient, type AssertionCategory, type AssertionCategorySummary, type AssertionEvidenceParameter, type AssertionGeneration, AssertionGenerationError, type AssertionGenerationStatus, AssertionGenerationTimeoutError, type AssertionLabelEvidence, type AssertionVerdict, type Assignee, BITFAB_PROGRESS_PREFIX, type BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, type BitfabConfig, BitfabError, BitfabFunction, BitfabLangGraphCallbackHandler as BitfabLangChainCallbackHandler, BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, type BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabRivetIntegration, BitfabVercelAiHandler, type CaptureSurface, type CaptureWhen, type CapturedSpan, type CodeChangeFile, type CommitRef, type CurrentSpan, type CurrentTrace, DEFAULT_SERVICE_URL, type Dataset, type DatasetGraderRef, type DatasetTraceIds, DatasetsClient, type DbBranchOptions, DbBranchReplayError, type DbBranchTimings, type DbSnapshotConfig, type DbSnapshotProvider, type DbSnapshotRef, type DetachedTrace, type Experiment, type ExperimentCategoryTally, type ExperimentCodeChangeTotals, type ExperimentJitter, type ExperimentPage, type ExperimentRollup, type ExperimentRollupResult, type ExperimentRunBy, type ExperimentStatus, type ExperimentTally, type ExperimentTotals, ExperimentsClient, type GenerateAssertionsOptions, type GetGraderLabelsParams, type GraderLabel, type GraderLabelOutcome, type GraderLabelSource, type GraderLabelUpdate, type GraderRerun, GraderRerunError, type GraderRerunProgress, type GraderRerunResult, type GraderRerunStatus, GraderRerunTimeoutError, GradersClient, HttpClient, type HumanLabelOutcome, type HumanLabelUpdate, type Justification, type JustificationSpan, type LabelAction, type LabelConfidence, type LabelOutcome, type LabelSource, type LabelStatus, type LabelTarget, type LabelUpdate, LabelsClient, type LangGraphIntegrationOptions, type ListDatasetsParams, type ListExperimentsParams, MixedTracingError, type MockOverride, type MockOverrideCtx, type MockOverrideInput, type MockOverrideResolver, type MockStrategy, type MockValue, NO_MOCK_OVERRIDE, type NodeMatcher, type NodeMethodDecorator, type NodeOptions, type OrganizationMember, OrganizationMembersClient, type ProviderDefinition, type RemoveDatasetGradersResult, type RemoveDatasetTracesResult, ReplayBranch, ReplayError, type ReplayExperiment, type ReplayInterrupt, type ReplayInterruptSignal, type ReplayItem, type ReplayItemFinishProgress, type ReplayItemStartProgress, type ReplayNodeIdentity, type ReplayOptions, type ReplayOptionsFactory, type ReplayProgress, type ReplayProgressItem, type ReplayRegistration, type ReplayRegistry, type ReplayRegistryContext, type ReplayRegistryOptions, type ReplayResult, type RerunGradersOptions, type RerunGradersResult, type ReseedOptions, type ReseedRegistryResult, type ReseedResult, type RivetContextLike, type RivetEffectType, type RivetIntegrationOptions, SUPPORTED_PROVIDERS, type SaveAssertion, type SaveAssertionCategoryParams, type SaveAssertionsAllParams, type SaveAssertionsParams, type SaveDatasetParams, type SaveDatasetResult, type SeedCase, type SeedCaseOptions, type SeedResult, type SeedRunOptions, type SelectiveReplayDecision, type SelectiveReplayOptions, type SelectiveReplayReport, type SpanLookup, type SpanMethodDecorator, type SpanMethodDecoratorContext, type SpanNodeMeta, type SpanOccurrence, type SpanOptions, type SpanType, type TokenUsage, type TraceAssertion, type TraceAssertionSource, type TraceAssertionsResult, type TraceAssertionsUpdate, type TraceIngestionType, type TraceLabels, type TraceResponse, type TraceSearchFilters, type TraceSearchParams, type TraceSearchResult, type TraceTarget, type TraceTargetOccurrence, TracesClient, type TracingProcessor, type VercelCallParams, type VercelGenerateResult, type VercelStreamResult, type WrapBAMLOptions, type WrappedBamlFn, __version__, defineReplayRegistry, finalizers, flushTraces, getCurrentReplayBranch, getCurrentSpan, getCurrentTrace, reportReplayProgress, reseedFromRegistry, seedFromRegistry, serializeReplayResult };

package/dist/index.js

@@ -36,7 +36,7 @@ import {
   getCurrentTrace,
   reseedFromRegistry,
   seedFromRegistry
-} from "./chunk-6NUSTEQ6.js";
+} from "./chunk-2UDEQITO.js";
 import "./chunk-YWXQM6J3.js";
 import {
   BITFAB_PROGRESS_PREFIX,
@@ -45,7 +45,7 @@ import {
   ReplayError,
   reportReplayProgress,
   serializeReplayResult
-} from "./chunk-B363XZTL.js";
+} from "./chunk-2U4Y5FHT.js";
 import {
   BitfabError,
   DEFAULT_SERVICE_URL,
@@ -53,7 +53,7 @@ import {
   MixedTracingError,
   __version__,
   flushTraces
-} from "./chunk-ROZEHSGO.js";
+} from "./chunk-YGDXDDTX.js";
 import "./chunk-MT36GJMW.js";
 export {
   AssertionCategoriesClient,

package/dist/node.cjs

@@ -97,7 +97,7 @@ var __version__, __packageName__;
 var init_version_generated = __esm({
   "src/version.generated.ts"() {
     "use strict";
-    __version__ = "0.64.6";
+    __version__ = "0.64.8";
     __packageName__ = "bitfab";
   }
 });
@@ -2911,7 +2911,7 @@ var init_http = __esm({
        * Start a replay session by fetching historical traces.
        * Blocking call - creates an experiment and returns lightweight item references.
        */
-      async startReplay(traceFunctionKey, limit, traceIds, name, codeChangeDescription, codeChangeFiles, includeDbBranchLease, experimentGroupId, datasetIds, graderIds, dbBranchSettings, attempts, includeOriginalMetadata, onlyWithAssertions, skipAssertionJudging, experimentalSelectiveReplay, notes, metadata, dryRun) {
+      async startReplay(traceFunctionKey, limit, traceIds, name, codeChangeDescription, codeChangeFiles, includeDbBranchLease, experimentGroupId, datasetIds, graderIds, dbBranchSettings, attempts, includeOriginalMetadata, onlyWithAssertions, skipAssertionJudging, experimentalSelectiveReplay, notes, metadata, dryRun, invocation) {
         const payload = { traceFunctionKey };
         if (experimentalSelectiveReplay !== void 0) {
           payload.experimentalSelectiveReplay = experimentalSelectiveReplay;
@@ -2972,6 +2972,9 @@ var init_http = __esm({
         if (dryRun) {
           payload.dryRun = true;
         }
+        if (invocation !== void 0) {
+          payload.invocation = invocation;
+        }
         const git = await resolvedGitState();
         if (git && !isEmptyGitState(git)) {
           payload.git = git;
@@ -3767,6 +3770,113 @@ var init_replayInterrupt = __esm({
   }
 });
 
+// src/replayInvocation.ts
+function replayEnvironment() {
+  if (readEnv(REPLAY_EXECUTION_TARGET_ENV) === HOSTED_EXECUTION_TARGET) {
+    return "cloud_replay";
+  }
+  if (CI_TRUE_VALUES.has((readEnv(CI_ENV) ?? "").toLowerCase())) {
+    return "ci";
+  }
+  if (CI_PRESENCE_ENV_VARS.some((name) => (readEnv(name) ?? "") !== "")) {
+    return "ci";
+  }
+  return "local";
+}
+function flagScalar(value) {
+  if (typeof value === "string") {
+    return value.slice(0, MAX_FLAG_STRING_LENGTH);
+  }
+  if (typeof value === "number") {
+    return Number.isFinite(value) ? value : void 0;
+  }
+  if (typeof value === "boolean" || value === null) {
+    return value;
+  }
+  return void 0;
+}
+function flagValue(value) {
+  if (Array.isArray(value)) {
+    const items = value.slice(0, MAX_FLAG_LIST_LENGTH).map(flagScalar);
+    return items.every((item) => item !== void 0) ? items : true;
+  }
+  const scalar = flagScalar(value);
+  return scalar === void 0 ? true : scalar;
+}
+function optionFlagName(option) {
+  return OPTION_FLAG_NAMES[option] ?? option.replace(/[A-Z]/g, (letter) => `-${letter.toLowerCase()}`);
+}
+function metadataKeys(value) {
+  const keys = Array.isArray(value) ? value.map((pair) => String(pair).split("=")[0]) : typeof value === "object" && value !== null ? Object.keys(value) : void 0;
+  return keys === void 0 ? true : [...new Set(keys)].sort();
+}
+function invocationFlags(entries) {
+  const flags = {};
+  const seen = /* @__PURE__ */ new Set();
+  for (const [name, value] of entries) {
+    if (seen.size >= MAX_FLAGS) {
+      break;
+    }
+    if (value === void 0 || seen.has(name) || name.length === 0 || name.length > MAX_FLAG_NAME_LENGTH || SECRET_FLAG_NAME.test(name)) {
+      continue;
+    }
+    flags[name] = flagValue(
+      name === METADATA_FLAG ? metadataKeys(value) : value
+    );
+    seen.add(name);
+  }
+  return flags;
+}
+function buildReplayInvocation(entries = []) {
+  return {
+    flags: invocationFlags(entries),
+    sdk: { language: "typescript", version: __version__ },
+    environment: replayEnvironment()
+  };
+}
+function safeReplayInvocation(build) {
+  try {
+    return build();
+  } catch {
+    return void 0;
+  }
+}
+function cliFlags(options) {
+  return options?.[CLI_FLAGS];
+}
+var REPLAY_EXECUTION_TARGET_ENV, HOSTED_EXECUTION_TARGET, CI_ENV, CI_TRUE_VALUES, CI_PRESENCE_ENV_VARS, MAX_FLAGS, MAX_FLAG_NAME_LENGTH, MAX_FLAG_STRING_LENGTH, MAX_FLAG_LIST_LENGTH, SECRET_FLAG_NAME, CLI_FLAGS, METADATA_FLAG, OPTION_FLAG_NAMES;
+var init_replayInvocation = __esm({
+  "src/replayInvocation.ts"() {
+    "use strict";
+    init_constants();
+    init_readEnv();
+    REPLAY_EXECUTION_TARGET_ENV = "BITFAB_REPLAY_EXECUTION_TARGET";
+    HOSTED_EXECUTION_TARGET = "hosted";
+    CI_ENV = "CI";
+    CI_TRUE_VALUES = /* @__PURE__ */ new Set(["true", "1"]);
+    CI_PRESENCE_ENV_VARS = [
+      "GITHUB_ACTIONS",
+      "GITLAB_CI",
+      "CIRCLECI",
+      "BUILDKITE",
+      "JENKINS_URL",
+      "TF_BUILD"
+    ];
+    MAX_FLAGS = 100;
+    MAX_FLAG_NAME_LENGTH = 100;
+    MAX_FLAG_STRING_LENGTH = 1e3;
+    MAX_FLAG_LIST_LENGTH = 100;
+    SECRET_FLAG_NAME = /key|token|secret|password/i;
+    CLI_FLAGS = /* @__PURE__ */ Symbol("bitfab.replayCliFlags");
+    METADATA_FLAG = "metadata";
+    OPTION_FLAG_NAMES = {
+      maxConcurrency: "concurrency",
+      datasetId: "dataset-ids",
+      datasetIds: "dataset-ids"
+    };
+  }
+});
+
 // src/selectiveReplay.ts
 function replayNodeKey(node) {
   return JSON.stringify([node.traceFunctionKey, node.spanName]);
@@ -4196,6 +4306,16 @@ function resolveDbBranchSettings(dbBranch) {
   };
   return Object.keys(settings).length === 0 ? void 0 : settings;
 }
+function explicitReplayOptions(options) {
+  var _a;
+  const explicit = {};
+  for (const name of REPLAY_OPTION_NAMES) {
+    if (options?.[name] !== void 0) {
+      explicit[_a = optionFlagName(name)] ?? (explicit[_a] = options[name]);
+    }
+  }
+  return explicit;
+}
 function reportReplayProgress(progress) {
   const stderr = typeof process !== "undefined" ? process.stderr : void 0;
   if (!stderr) {
@@ -4707,7 +4827,12 @@ async function replay(httpClient, serviceUrl, traceFunctionKey, fn, options, reg
       options?.experimentalSelectiveReplay,
       options?.notes,
       options?.metadata,
-      options?.dryRun === true
+      options?.dryRun === true,
+      safeReplayInvocation(
+        () => buildReplayInvocation(
+          Object.entries(cliFlags(options) ?? explicitReplayOptions(options))
+        )
+      )
     );
     if (started.items.length === 0) {
       try {
@@ -5099,7 +5224,7 @@ async function writeReplayResultFile(result) {
     }
   }
 }
-var REPLAY_PERSISTENCE_TIMEOUT_MS, MAX_REPLAY_ATTEMPTS, RESUME_CONFLICTING_OPTIONS, CARRIED_OVER_ERROR, BITFAB_PROGRESS_PREFIX, ReplayError, DbBranchReplayError, JUDGE_ASSERTIONS_DEPRECATION_WARNING;
+var REPLAY_PERSISTENCE_TIMEOUT_MS, MAX_REPLAY_ATTEMPTS, RESUME_CONFLICTING_OPTIONS, CARRIED_OVER_ERROR, REPLAY_OPTION_NAMES, BITFAB_PROGRESS_PREFIX, ReplayError, DbBranchReplayError, JUDGE_ASSERTIONS_DEPRECATION_WARNING;
 var init_replay = __esm({
   "src/replay.ts"() {
     "use strict";
@@ -5112,6 +5237,7 @@ var init_replay = __esm({
     init_replayContext();
     init_replayHeartbeat();
     init_replayInterrupt();
+    init_replayInvocation();
     init_selectiveReplayRuntime();
     init_serialize();
     init_warnOnce();
@@ -5129,6 +5255,37 @@ var init_replay = __esm({
       "experimentalSelectiveReplay"
     ];
     CARRIED_OVER_ERROR = "This item errored in an earlier run of this experiment, before it was resumed.";
+    REPLAY_OPTION_NAMES = [
+      "experimentalSelectiveReplay",
+      "limit",
+      "attempts",
+      "traceIds",
+      "name",
+      "notes",
+      "metadata",
+      "maxConcurrency",
+      "codeChangeDescription",
+      "codeChangeFiles",
+      "mock",
+      "mockOverride",
+      "dbBranch",
+      "experimentGroupId",
+      "datasetId",
+      "datasetIds",
+      "graderIds",
+      "onlyWithAssertions",
+      "skipAssertionJudging",
+      "judgeAssertions",
+      "adaptInputs",
+      "dryRun",
+      "onItemFinish",
+      "onProgress",
+      "onItemStart",
+      "onExperimentStart",
+      "onInterrupt",
+      "resume",
+      "force"
+    ];
     BITFAB_PROGRESS_PREFIX = "@@bitfab:progress ";
     ReplayError = class extends BitfabError {
       constructor(message, items, experimentId, experimentUrl, cause) {
@@ -6912,9 +7069,34 @@ var GradersClient = class {
 init_http();
 
 // src/labels.ts
+init_errors();
 init_http();
 var LABELS_PATH = "/api/sdk/traces/labels";
 var HUMAN_LABELS_PATH = "/api/sdk/traces/labels/human";
+function assertEveryVerdictNamesAnAssertion(updates) {
+  const unscoped = updates.filter(
+    (update) => !("archive" in update && update.archive === true) && (typeof update.assertionId !== "string" || update.assertionId === "")
+  );
+  if (unscoped.length > 0) {
+    const targets = unscoped.map(
+      (update) => update.traceId ?? update.originalTraceId
+    );
+    throw new BitfabError(
+      `Agent verdicts must name an assertion: ${targets.join(", ")}. Pass the assertionId of the assertion each verdict scores to labels.save, saveAll, and skip. Only archive may omit it, to clear an old whole-trace verdict.`
+    );
+  }
+}
+function assertEveryHumanVerdictNamesAnAssertion(updates) {
+  const unscoped = updates.filter(
+    (update) => typeof update.assertionId !== "string" || update.assertionId === ""
+  );
+  if (unscoped.length > 0) {
+    const targets = unscoped.map((update) => update.traceId);
+    throw new BitfabError(
+      `Human verdicts must name an assertion: ${targets.join(", ")}. Pass the assertionId of the assertion each verdict scores to labels.saveHuman and saveHumanAll.`
+    );
+  }
+}
 function labelEvidencePath(traceId, assertionId) {
   return `/api/sdk/traces/${encodeURIComponent(traceId)}/assertions/${encodeURIComponent(assertionId)}/evidence`;
 }
@@ -6927,6 +7109,7 @@ var LabelsClient = class {
     return outcome;
   }
   async saveAll(updates, experimentId) {
+    assertEveryVerdictNamesAnAssertion(updates);
     const response = await this.httpClient.request(
       LABELS_PATH,
       {
@@ -6936,11 +7119,9 @@ var LabelsClient = class {
     );
     return response.labels;
   }
-  /** Withhold a verdict for a trace or replay attempt whose check did not run. */
   async skip(target, experimentId) {
     return this.save({ ...target, skip: true }, experimentId);
   }
-  /** Clear a previous verdict for a trace, replay attempt, or assertion. */
   async archive(target, experimentId) {
     return this.save({ ...target, archive: true }, experimentId);
   }
@@ -6949,6 +7130,7 @@ var LabelsClient = class {
     return outcome;
   }
   async saveHumanAll(updates) {
+    assertEveryHumanVerdictNamesAnAssertion(updates);
     const response = await this.httpClient.request(HUMAN_LABELS_PATH, { labels: updates });
     return response.labels;
   }
@@ -11986,6 +12168,7 @@ init_replay();
 init_errors();
 init_replay();
 init_replayInterrupt();
+init_replayInvocation();
 init_warnOnce();
 function defineReplayRegistry(registry) {
   validateReplayRegistry(registry);