bitfab 0.62.3 → 0.62.5

package/dist/index.d.cts

@@ -697,7 +697,7 @@ 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, judgeAssertions?: boolean, experimentalSelectiveReplay?: SelectiveReplayOptions, notes?: string): 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, judgeAssertions?: boolean, experimentalSelectiveReplay?: SelectiveReplayOptions, notes?: string, metadata?: Record<string, string>): Promise<StartReplayResponse>;
     resumeReplay(experimentId: string, traceFunctionKey: string, codeChangeFiles?: CodeChangeFile[] | null, force?: boolean): Promise<ResumeReplayResponse>;
     /**
      * Fetch an external span by ID.
@@ -1397,6 +1397,114 @@ declare class DatasetsClient {
     getGraderRerun(datasetId: string, runId?: string): Promise<GraderRerun | null>;
 }
 
+type ExperimentStatus = "pending" | "completed" | "failed" | "interrupted";
+type ExperimentRunBy = {
+    id: string;
+    fullName: string | null;
+    email: string | null;
+    imageUrl: string | null;
+};
+type ExperimentCodeChangeTotals = {
+    files: number;
+    added: number;
+    removed: number;
+};
+type Experiment = {
+    id: string;
+    name: string | null;
+    notes: string | null;
+    status: ExperimentStatus;
+    traceFunctionKey: string | null;
+    createdAt: string;
+    completedAt: string | null;
+    codeChangeDescription: string | null;
+    codeChangeTotals: ExperimentCodeChangeTotals | null;
+    experimentGroupId: string | null;
+    experimentGroupName: string | null;
+    datasetId: string | null;
+    datasetName: string | null;
+    datasetIds: string[];
+    datasetNames: string[];
+    attempts: number;
+    plannedTraceCount: number | null;
+    runBy: ExperimentRunBy | null;
+    githubEmail: string | null;
+    gitBranch: string | null;
+    commitSha: string | null;
+    baseSha: string | null;
+    experimentSha: string | null;
+    metadata: Record<string, string>;
+};
+type ExperimentPage = {
+    experiments: Experiment[];
+    nextCursor: string | null;
+    hasMore: boolean;
+};
+type ListExperimentsParams = {
+    datasetId?: string;
+    traceFunctionKey?: string;
+    gitBranch?: string;
+    experimentGroupId?: string;
+    status?: ExperimentStatus;
+    metadata?: Record<string, string>;
+    createdAfter?: Date | string;
+    createdBefore?: Date | string;
+    cursor?: string;
+    limit?: number;
+};
+type ExperimentTally = {
+    fixed: number;
+    stillPassing: number;
+    regressed: number;
+    stillFailing: number;
+    originalUnlabeled: number;
+    unpaired: number;
+    classified: number;
+    passing: number;
+    failing: number;
+};
+type ExperimentTotals = {
+    total: number;
+    succeeded: number;
+    failed: number;
+    pending: number;
+    awaitingLabels: number;
+    errored: number;
+    withErrors: number;
+    ungradable: number;
+    skipped: number;
+};
+type ExperimentCategoryTally = {
+    category: AssertionCategorySummary;
+    labels: ExperimentTally;
+};
+type ExperimentJitter = {
+    disagreeing: number;
+    attempts: number;
+    traces: number;
+};
+type ExperimentRollup = {
+    traces: ExperimentTally;
+    skippedTraces: number;
+    labels: ExperimentTally;
+    checks: ExperimentTally;
+    categories: ExperimentCategoryTally[];
+    jitter: ExperimentJitter;
+};
+type ExperimentRollupResult = {
+    experimentId: string;
+    totals: ExperimentTotals;
+    rollup: ExperimentRollup;
+};
+declare class ExperimentsClient {
+    private readonly httpClient;
+    constructor(httpClient: HttpClient);
+    list(params?: ListExperimentsParams): Promise<ExperimentPage>;
+    get(id: string): Promise<Experiment>;
+    getRollup(id: string): Promise<ExperimentRollupResult>;
+    getRollupAll(ids: string[]): Promise<ExperimentRollupResult[]>;
+}
+
 type TraceTargetOccurrence = "first" | "last" | number;
 type TraceTarget = {
     kind: "output";
@@ -2028,6 +2136,7 @@ interface ReplayOptions {
      * override or a forced feature flag. Kept on the experiment next to its name.
      */
     notes?: string;
+    metadata?: Record<string, string>;
     /** Positive integer limiting parallel items. Set to 1 for sequential. Default 10. */
     maxConcurrency?: number;
     /**
@@ -3163,6 +3272,7 @@ declare class Bitfab {
     readonly traces: TracesClient;
     readonly labels: LabelsClient;
     readonly graders: GradersClient;
+    readonly experiments: ExperimentsClient;
     private readonly bamlClient;
     private readonly dbSnapshot;
     private readonly simulationPlan;
@@ -3867,7 +3977,7 @@ interface CommitRef {
 /**
  * SDK version from package.json (injected at build time)
  */
-declare const __version__ = "0.62.3";
+declare const __version__ = "0.62.5";
 
 /**
  * Constants for the Bitfab SDK.
@@ -4006,4 +4116,4 @@ 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 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, 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, 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

@@ -697,7 +697,7 @@ 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, judgeAssertions?: boolean, experimentalSelectiveReplay?: SelectiveReplayOptions, notes?: string): 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, judgeAssertions?: boolean, experimentalSelectiveReplay?: SelectiveReplayOptions, notes?: string, metadata?: Record<string, string>): Promise<StartReplayResponse>;
     resumeReplay(experimentId: string, traceFunctionKey: string, codeChangeFiles?: CodeChangeFile[] | null, force?: boolean): Promise<ResumeReplayResponse>;
     /**
      * Fetch an external span by ID.
@@ -1397,6 +1397,114 @@ declare class DatasetsClient {
     getGraderRerun(datasetId: string, runId?: string): Promise<GraderRerun | null>;
 }
 
+type ExperimentStatus = "pending" | "completed" | "failed" | "interrupted";
+type ExperimentRunBy = {
+    id: string;
+    fullName: string | null;
+    email: string | null;
+    imageUrl: string | null;
+};
+type ExperimentCodeChangeTotals = {
+    files: number;
+    added: number;
+    removed: number;
+};
+type Experiment = {
+    id: string;
+    name: string | null;
+    notes: string | null;
+    status: ExperimentStatus;
+    traceFunctionKey: string | null;
+    createdAt: string;
+    completedAt: string | null;
+    codeChangeDescription: string | null;
+    codeChangeTotals: ExperimentCodeChangeTotals | null;
+    experimentGroupId: string | null;
+    experimentGroupName: string | null;
+    datasetId: string | null;
+    datasetName: string | null;
+    datasetIds: string[];
+    datasetNames: string[];
+    attempts: number;
+    plannedTraceCount: number | null;
+    runBy: ExperimentRunBy | null;
+    githubEmail: string | null;
+    gitBranch: string | null;
+    commitSha: string | null;
+    baseSha: string | null;
+    experimentSha: string | null;
+    metadata: Record<string, string>;
+};
+type ExperimentPage = {
+    experiments: Experiment[];
+    nextCursor: string | null;
+    hasMore: boolean;
+};
+type ListExperimentsParams = {
+    datasetId?: string;
+    traceFunctionKey?: string;
+    gitBranch?: string;
+    experimentGroupId?: string;
+    status?: ExperimentStatus;
+    metadata?: Record<string, string>;
+    createdAfter?: Date | string;
+    createdBefore?: Date | string;
+    cursor?: string;
+    limit?: number;
+};
+type ExperimentTally = {
+    fixed: number;
+    stillPassing: number;
+    regressed: number;
+    stillFailing: number;
+    originalUnlabeled: number;
+    unpaired: number;
+    classified: number;
+    passing: number;
+    failing: number;
+};
+type ExperimentTotals = {
+    total: number;
+    succeeded: number;
+    failed: number;
+    pending: number;
+    awaitingLabels: number;
+    errored: number;
+    withErrors: number;
+    ungradable: number;
+    skipped: number;
+};
+type ExperimentCategoryTally = {
+    category: AssertionCategorySummary;
+    labels: ExperimentTally;
+};
+type ExperimentJitter = {
+    disagreeing: number;
+    attempts: number;
+    traces: number;
+};
+type ExperimentRollup = {
+    traces: ExperimentTally;
+    skippedTraces: number;
+    labels: ExperimentTally;
+    checks: ExperimentTally;
+    categories: ExperimentCategoryTally[];
+    jitter: ExperimentJitter;
+};
+type ExperimentRollupResult = {
+    experimentId: string;
+    totals: ExperimentTotals;
+    rollup: ExperimentRollup;
+};
+declare class ExperimentsClient {
+    private readonly httpClient;
+    constructor(httpClient: HttpClient);
+    list(params?: ListExperimentsParams): Promise<ExperimentPage>;
+    get(id: string): Promise<Experiment>;
+    getRollup(id: string): Promise<ExperimentRollupResult>;
+    getRollupAll(ids: string[]): Promise<ExperimentRollupResult[]>;
+}
+
 type TraceTargetOccurrence = "first" | "last" | number;
 type TraceTarget = {
     kind: "output";
@@ -2028,6 +2136,7 @@ interface ReplayOptions {
      * override or a forced feature flag. Kept on the experiment next to its name.
      */
     notes?: string;
+    metadata?: Record<string, string>;
     /** Positive integer limiting parallel items. Set to 1 for sequential. Default 10. */
     maxConcurrency?: number;
     /**
@@ -3163,6 +3272,7 @@ declare class Bitfab {
     readonly traces: TracesClient;
     readonly labels: LabelsClient;
     readonly graders: GradersClient;
+    readonly experiments: ExperimentsClient;
     private readonly bamlClient;
     private readonly dbSnapshot;
     private readonly simulationPlan;
@@ -3867,7 +3977,7 @@ interface CommitRef {
 /**
  * SDK version from package.json (injected at build time)
  */
-declare const __version__ = "0.62.3";
+declare const __version__ = "0.62.5";
 
 /**
  * Constants for the Bitfab SDK.
@@ -4006,4 +4116,4 @@ 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 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, 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, 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

@@ -21,6 +21,7 @@ import {
   BitfabRivetIntegration,
   BitfabVercelAiHandler,
   DatasetsClient,
+  ExperimentsClient,
   GraderRerunError,
   GraderRerunTimeoutError,
   GradersClient,
@@ -35,7 +36,7 @@ import {
   getCurrentTrace,
   reseedFromRegistry,
   seedFromRegistry
-} from "./chunk-BPOAXWVS.js";
+} from "./chunk-SDEDJXKM.js";
 import "./chunk-K4XCWQBC.js";
 import {
   BITFAB_PROGRESS_PREFIX,
@@ -44,7 +45,7 @@ import {
   ReplayError,
   reportReplayProgress,
   serializeReplayResult
-} from "./chunk-T6O3MFM4.js";
+} from "./chunk-XADZNIOO.js";
 import {
   BitfabError,
   DEFAULT_SERVICE_URL,
@@ -52,7 +53,7 @@ import {
   MixedTracingError,
   __version__,
   flushTraces
-} from "./chunk-DRRJRCVG.js";
+} from "./chunk-UUB5LHWU.js";
 import "./chunk-BAOIRE5T.js";
 export {
   AssertionCategoriesClient,
@@ -73,6 +74,7 @@ export {
   DEFAULT_SERVICE_URL,
   DatasetsClient,
   DbBranchReplayError,
+  ExperimentsClient,
   GraderRerunError,
   GraderRerunTimeoutError,
   GradersClient,

package/dist/node.cjs

@@ -97,7 +97,7 @@ var __version__, __packageName__;
 var init_version_generated = __esm({
   "src/version.generated.ts"() {
     "use strict";
-    __version__ = "0.62.3";
+    __version__ = "0.62.5";
     __packageName__ = "bitfab";
   }
 });
@@ -2901,7 +2901,7 @@ 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, judgeAssertions, experimentalSelectiveReplay, notes) {
+      async startReplay(traceFunctionKey, limit, traceIds, name, codeChangeDescription, codeChangeFiles, includeDbBranchLease, experimentGroupId, datasetIds, graderIds, dbBranchSettings, attempts, includeOriginalMetadata, onlyWithAssertions, judgeAssertions, experimentalSelectiveReplay, notes, metadata) {
         const payload = { traceFunctionKey };
         if (experimentalSelectiveReplay !== void 0) {
           payload.experimentalSelectiveReplay = experimentalSelectiveReplay;
@@ -2918,6 +2918,9 @@ var init_http = __esm({
         if (notes !== void 0) {
           payload.notes = notes;
         }
+        if (metadata !== void 0) {
+          payload.metadata = metadata;
+        }
         if (codeChangeDescription !== void 0) {
           payload.codeChangeDescription = codeChangeDescription;
         }
@@ -4691,7 +4694,8 @@ async function replay(httpClient, serviceUrl, traceFunctionKey, fn, options, reg
       options?.onlyWithAssertions,
       options?.judgeAssertions,
       options?.experimentalSelectiveReplay,
-      options?.notes
+      options?.notes,
+      options?.metadata
     );
     if (started.items.length === 0) {
       try {
@@ -5138,6 +5142,7 @@ __export(node_exports, {
   DEFAULT_SERVICE_URL: () => DEFAULT_SERVICE_URL,
   DatasetsClient: () => DatasetsClient,
   DbBranchReplayError: () => DbBranchReplayError,
+  ExperimentsClient: () => ExperimentsClient,
   GraderRerunError: () => GraderRerunError,
   GraderRerunTimeoutError: () => GraderRerunTimeoutError,
   GradersClient: () => GradersClient,
@@ -6759,6 +6764,76 @@ function buildSnapshotRef(config, sdkWallClockBeforeFn) {
 // src/client.ts
 init_errors();
 
+// src/experiments.ts
+var EXPERIMENTS_PATH = "/api/sdk/experiments";
+function isoTimestamp(value) {
+  return value instanceof Date ? value.toISOString() : value;
+}
+function buildListQuery(params) {
+  const query = new URLSearchParams();
+  if (params.datasetId !== void 0) {
+    query.set("datasetId", params.datasetId);
+  }
+  if (params.traceFunctionKey !== void 0) {
+    query.set("traceFunctionKey", params.traceFunctionKey);
+  }
+  if (params.gitBranch !== void 0) {
+    query.set("gitBranch", params.gitBranch);
+  }
+  if (params.experimentGroupId !== void 0) {
+    query.set("experimentGroupId", params.experimentGroupId);
+  }
+  if (params.status !== void 0) {
+    query.set("status", params.status);
+  }
+  if (params.metadata !== void 0) {
+    query.set("metadata", JSON.stringify(params.metadata));
+  }
+  if (params.createdAfter !== void 0) {
+    query.set("createdAfter", isoTimestamp(params.createdAfter));
+  }
+  if (params.createdBefore !== void 0) {
+    query.set("createdBefore", isoTimestamp(params.createdBefore));
+  }
+  if (params.cursor !== void 0) {
+    query.set("cursor", params.cursor);
+  }
+  if (params.limit !== void 0) {
+    query.set("limit", String(params.limit));
+  }
+  const encoded = query.toString();
+  return encoded.length === 0 ? "" : `?${encoded}`;
+}
+var ExperimentsClient = class {
+  constructor(httpClient) {
+    this.httpClient = httpClient;
+  }
+  async list(params = {}) {
+    return this.httpClient.get(
+      `${EXPERIMENTS_PATH}${buildListQuery(params)}`
+    );
+  }
+  async get(id) {
+    const response = await this.httpClient.get(
+      `${EXPERIMENTS_PATH}/${encodeURIComponent(id)}`
+    );
+    return response.experiment;
+  }
+  async getRollup(id) {
+    return this.httpClient.get(
+      `${EXPERIMENTS_PATH}/${encodeURIComponent(id)}/rollup`
+    );
+  }
+  async getRollupAll(ids) {
+    if (ids.length === 0) {
+      return [];
+    }
+    const query = new URLSearchParams({ experimentIds: ids.join(",") });
+    const response = await this.httpClient.get(`${EXPERIMENTS_PATH}/rollups?${query.toString()}`);
+    return response.rollups;
+  }
+};
+
 // src/graders.ts
 init_errors();
 var GRADER_LABELS_PATH = "/api/sdk/graderLabels";
@@ -9673,6 +9748,7 @@ var Bitfab = class {
     this.traces = new TracesClient(this.httpClient);
     this.labels = new LabelsClient(this.httpClient);
     this.graders = new GradersClient(this.httpClient);
+    this.experiments = new ExperimentsClient(this.httpClient);
   }
   /**
    * Decorate a class method as an automatically expanded trace root.
@@ -11957,6 +12033,7 @@ assertAsyncStorageRegistered();
   DEFAULT_SERVICE_URL,
   DatasetsClient,
   DbBranchReplayError,
+  ExperimentsClient,
   GraderRerunError,
   GraderRerunTimeoutError,
   GradersClient,