bitfab 0.48.0 → 0.51.0

package/dist/index.d.cts

@@ -556,7 +556,7 @@ declare class HttpClient {
      * Start a replay session by fetching historical traces.
      * Blocking call - creates a test run 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, datasetId?: string, graderIds?: string[], dbBranchSettings?: DbBranchSettings, attempts?: number, includeOriginalMetadata?: boolean, onlyWithAssertions?: 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): Promise<StartReplayResponse>;
     /**
      * Fetch an external span by ID.
      * Blocking GET request.
@@ -1259,6 +1259,8 @@ declare class DatasetsClient {
 
 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 = ({
     traceId: string;
     originalTraceId?: never;
@@ -1285,11 +1287,72 @@ type LabelOutcome = {
     traceId: string;
     action: LabelAction;
 };
+type HumanLabelUpdate = {
+    traceId: string;
+    assertionId?: string;
+    label: boolean;
+    annotation: string;
+    confidence?: LabelConfidence;
+};
+type HumanLabelOutcome = {
+    traceId: string;
+    assertionId: string | null;
+    label: boolean;
+    action: "set";
+};
+type AssertionVerdict = {
+    assertionId: string;
+    assertion: string | null;
+    labelStatus: LabelStatus;
+    label: boolean | null;
+    annotation: string | null;
+    confidence: LabelConfidence | null;
+    labelSource: LabelSource;
+    approved: boolean;
+};
+type TraceLabels = {
+    traceId: string;
+    labelStatus: LabelStatus;
+    label: boolean | null;
+    annotation: string | null;
+    approved: boolean;
+    passed: number;
+    failed: number;
+    assertions: AssertionVerdict[];
+};
 declare class LabelsClient {
     private readonly httpClient;
     constructor(httpClient: HttpClient);
     save(update: LabelUpdate, testRunId?: string): Promise<LabelOutcome>;
     saveAll(updates: LabelUpdate[], testRunId?: string): Promise<LabelOutcome[]>;
+    saveHuman(update: HumanLabelUpdate): Promise<HumanLabelOutcome>;
+    saveHumanAll(updates: HumanLabelUpdate[]): Promise<HumanLabelOutcome[]>;
+    get(traceId: string): Promise<TraceLabels | null>;
+    getAll(traceIds: string[]): Promise<TraceLabels[]>;
+}
+
+type GraderLabelSource = "human" | "live_grader";
+type GraderLabel = {
+    traceId: string;
+    graderId: string;
+    graderName: string | null;
+    graderStatus: string;
+    label: boolean | null;
+    labelReason: string | null;
+    failureDiagnostic: string | null;
+    labelConfidence: LabelConfidence | null;
+    source: GraderLabelSource;
+    evaluatedAt: string | null;
+};
+type GetGraderLabelsParams = {
+    traceIds?: string[];
+    graderId?: string;
+    limit?: number;
+};
+declare class GradersClient {
+    private readonly httpClient;
+    constructor(httpClient: HttpClient);
+    getLabels(params: GetGraderLabelsParams): Promise<GraderLabel[]>;
 }
 
 /**
@@ -1582,12 +1645,15 @@ interface DbBranchOptions {
 interface ReplayOptions {
     /**
      * Maximum number of traces to replay (1-5,000, default 5). Ignored when
-     * `traceIds` is passed (with a warning), or when `datasetId` is passed,
-     * because either source already determines how many traces replay.
+     * `traceIds` is passed (with a warning), or when `datasetId` / `datasetIds`
+     * is passed, because either source already determines how many traces replay.
      */
     limit?: number;
     attempts?: number;
-    /** Optional list of specific trace IDs to replay (max 100). Mutually exclusive with `datasetId`. */
+    /**
+     * Optional list of specific trace IDs to replay (max 100). Mutually exclusive
+     * with `datasetId` and `datasetIds`.
+     */
     traceIds?: string[];
     /** Optional display name for the resulting experiment/test run. */
     name?: string;
@@ -1644,13 +1710,14 @@ interface ReplayOptions {
     /** Group ID to associate this replay with an experiment group for live streaming in Studio. */
     experimentGroupId?: string;
     /**
-     * Dataset this replay runs against. Mutually exclusive with `traceIds`. When
-     * set, the resulting experiment is
+     * Dataset this replay runs against. Mutually exclusive with `traceIds` and
+     * with `datasetIds`. When set, the resulting experiment is
      * durably attributed to the dataset (stored on the test run), so it appears
      * under the dataset's experiments even if the trace lineage can't be
      * reconstructed. Validated server-side against the org.
      */
     datasetId?: string;
+    datasetIds?: string[];
     /**
      * Graders to attach directly to this experiment (test run), independent of any
      * graders already on the dataset. The resulting experiment is graded by the
@@ -1663,7 +1730,7 @@ interface ReplayOptions {
     /**
      * Replay only the selected traces that carry at least one assertion.
      *
-     * Narrows whatever `limit`, `traceIds` or `datasetId` selected, so a run that
+     * Narrows whatever `limit`, `traceIds` or `datasetIds` selected, so a run that
      * measures assertion outcomes does not pay to re-execute traces nothing can
      * grade. The server applies it as part of selection, which is what makes it
      * compose with `limit`: `{ limit: 10, onlyWithAssertions: true }` is the ten
@@ -2566,6 +2633,7 @@ declare class Bitfab {
     readonly datasets: DatasetsClient;
     readonly traces: TracesClient;
     readonly labels: LabelsClient;
+    readonly graders: GradersClient;
     private readonly bamlClient;
     private readonly dbSnapshot;
     private readonly autoTracePolicyRefreshes;
@@ -3211,7 +3279,7 @@ declare class BitfabFunction {
 /**
  * SDK version from package.json (injected at build time)
  */
-declare const __version__ = "0.48.0";
+declare const __version__ = "0.51.0";
 
 /**
  * Constants for the Bitfab SDK.
@@ -3341,4 +3409,4 @@ 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 ArchiveAssertionsParams, BITFAB_PROGRESS_PREFIX, type BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, type BitfabConfig, BitfabError, BitfabFunction, BitfabLangGraphCallbackHandler as BitfabLangChainCallbackHandler, BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, type BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabVercelAiHandler, type CaptureSurface, type CaptureWhen, type CapturedSpan, type CodeChangeFile, 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 GraderRerun, type GraderRerunProgress, type GraderRerunResult, type GraderRerunStatus, HttpClient, type LabelAction, type LabelConfidence, type LabelOutcome, 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 ProviderDefinition, type RemoveDatasetGradersResult, type RemoveDatasetTracesResult, ReplayBranch, ReplayError, type ReplayItem, type ReplayItemFinishProgress, type ReplayItemStartProgress, type ReplayOptions, type ReplayOptionsFactory, type ReplayProgress, type ReplayProgressItem, type ReplayRegistration, type ReplayRegistry, type ReplayRegistryContext, type ReplayRegistryOptions, type ReplayResult, type RerunGradersOptions, type RerunGradersResult, SUPPORTED_PROVIDERS, type SaveAssertion, type SaveAssertionsParams, type SaveDatasetParams, type SaveDatasetResult, type SeedCase, type SeedCaseOptions, type SeedResult, type SeedRunOptions, type SpanLookup, type SpanMethodDecorator, type SpanMethodDecoratorContext, type SpanNodeMeta, type SpanOccurrence, type SpanOptions, type SpanType, type TokenUsage, type TraceAssertion, type TraceAssertionSource, type TraceAssertionsResult, type TraceIngestionType, type TraceOutline, type TraceOutlineSpan, type TraceOutlineSpanError, type TraceResponse, 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, seedFromRegistry, serializeReplayResult };
+export { type ActiveSpanContext, type AdaptContext, type AdaptInputsFn, type AddDatasetGradersResult, type AddDatasetTracesResult, type AllowedEnvVars, type ArchiveAssertionsParams, type AssertionVerdict, BITFAB_PROGRESS_PREFIX, type BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, type BitfabConfig, BitfabError, BitfabFunction, BitfabLangGraphCallbackHandler as BitfabLangChainCallbackHandler, BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, type BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabVercelAiHandler, type CaptureSurface, type CaptureWhen, type CapturedSpan, type CodeChangeFile, 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 GetGraderLabelsParams, type GraderLabel, type GraderLabelSource, type GraderRerun, type GraderRerunProgress, type GraderRerunResult, type GraderRerunStatus, GradersClient, HttpClient, type HumanLabelOutcome, type HumanLabelUpdate, type LabelAction, type LabelConfidence, type LabelOutcome, type LabelSource, type LabelStatus, 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 ProviderDefinition, type RemoveDatasetGradersResult, type RemoveDatasetTracesResult, ReplayBranch, ReplayError, type ReplayItem, type ReplayItemFinishProgress, type ReplayItemStartProgress, type ReplayOptions, type ReplayOptionsFactory, type ReplayProgress, type ReplayProgressItem, type ReplayRegistration, type ReplayRegistry, type ReplayRegistryContext, type ReplayRegistryOptions, type ReplayResult, type RerunGradersOptions, type RerunGradersResult, SUPPORTED_PROVIDERS, type SaveAssertion, type SaveAssertionsParams, type SaveDatasetParams, type SaveDatasetResult, type SeedCase, type SeedCaseOptions, type SeedResult, type SeedRunOptions, type SpanLookup, type SpanMethodDecorator, type SpanMethodDecoratorContext, type SpanNodeMeta, type SpanOccurrence, type SpanOptions, type SpanType, type TokenUsage, type TraceAssertion, type TraceAssertionSource, type TraceAssertionsResult, type TraceIngestionType, type TraceLabels, type TraceOutline, type TraceOutlineSpan, type TraceOutlineSpanError, type TraceResponse, 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, seedFromRegistry, serializeReplayResult };

package/dist/index.d.ts

@@ -556,7 +556,7 @@ declare class HttpClient {
      * Start a replay session by fetching historical traces.
      * Blocking call - creates a test run 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, datasetId?: string, graderIds?: string[], dbBranchSettings?: DbBranchSettings, attempts?: number, includeOriginalMetadata?: boolean, onlyWithAssertions?: 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): Promise<StartReplayResponse>;
     /**
      * Fetch an external span by ID.
      * Blocking GET request.
@@ -1259,6 +1259,8 @@ declare class DatasetsClient {
 
 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 = ({
     traceId: string;
     originalTraceId?: never;
@@ -1285,11 +1287,72 @@ type LabelOutcome = {
     traceId: string;
     action: LabelAction;
 };
+type HumanLabelUpdate = {
+    traceId: string;
+    assertionId?: string;
+    label: boolean;
+    annotation: string;
+    confidence?: LabelConfidence;
+};
+type HumanLabelOutcome = {
+    traceId: string;
+    assertionId: string | null;
+    label: boolean;
+    action: "set";
+};
+type AssertionVerdict = {
+    assertionId: string;
+    assertion: string | null;
+    labelStatus: LabelStatus;
+    label: boolean | null;
+    annotation: string | null;
+    confidence: LabelConfidence | null;
+    labelSource: LabelSource;
+    approved: boolean;
+};
+type TraceLabels = {
+    traceId: string;
+    labelStatus: LabelStatus;
+    label: boolean | null;
+    annotation: string | null;
+    approved: boolean;
+    passed: number;
+    failed: number;
+    assertions: AssertionVerdict[];
+};
 declare class LabelsClient {
     private readonly httpClient;
     constructor(httpClient: HttpClient);
     save(update: LabelUpdate, testRunId?: string): Promise<LabelOutcome>;
     saveAll(updates: LabelUpdate[], testRunId?: string): Promise<LabelOutcome[]>;
+    saveHuman(update: HumanLabelUpdate): Promise<HumanLabelOutcome>;
+    saveHumanAll(updates: HumanLabelUpdate[]): Promise<HumanLabelOutcome[]>;
+    get(traceId: string): Promise<TraceLabels | null>;
+    getAll(traceIds: string[]): Promise<TraceLabels[]>;
+}
+
+type GraderLabelSource = "human" | "live_grader";
+type GraderLabel = {
+    traceId: string;
+    graderId: string;
+    graderName: string | null;
+    graderStatus: string;
+    label: boolean | null;
+    labelReason: string | null;
+    failureDiagnostic: string | null;
+    labelConfidence: LabelConfidence | null;
+    source: GraderLabelSource;
+    evaluatedAt: string | null;
+};
+type GetGraderLabelsParams = {
+    traceIds?: string[];
+    graderId?: string;
+    limit?: number;
+};
+declare class GradersClient {
+    private readonly httpClient;
+    constructor(httpClient: HttpClient);
+    getLabels(params: GetGraderLabelsParams): Promise<GraderLabel[]>;
 }
 
 /**
@@ -1582,12 +1645,15 @@ interface DbBranchOptions {
 interface ReplayOptions {
     /**
      * Maximum number of traces to replay (1-5,000, default 5). Ignored when
-     * `traceIds` is passed (with a warning), or when `datasetId` is passed,
-     * because either source already determines how many traces replay.
+     * `traceIds` is passed (with a warning), or when `datasetId` / `datasetIds`
+     * is passed, because either source already determines how many traces replay.
      */
     limit?: number;
     attempts?: number;
-    /** Optional list of specific trace IDs to replay (max 100). Mutually exclusive with `datasetId`. */
+    /**
+     * Optional list of specific trace IDs to replay (max 100). Mutually exclusive
+     * with `datasetId` and `datasetIds`.
+     */
     traceIds?: string[];
     /** Optional display name for the resulting experiment/test run. */
     name?: string;
@@ -1644,13 +1710,14 @@ interface ReplayOptions {
     /** Group ID to associate this replay with an experiment group for live streaming in Studio. */
     experimentGroupId?: string;
     /**
-     * Dataset this replay runs against. Mutually exclusive with `traceIds`. When
-     * set, the resulting experiment is
+     * Dataset this replay runs against. Mutually exclusive with `traceIds` and
+     * with `datasetIds`. When set, the resulting experiment is
      * durably attributed to the dataset (stored on the test run), so it appears
      * under the dataset's experiments even if the trace lineage can't be
      * reconstructed. Validated server-side against the org.
      */
     datasetId?: string;
+    datasetIds?: string[];
     /**
      * Graders to attach directly to this experiment (test run), independent of any
      * graders already on the dataset. The resulting experiment is graded by the
@@ -1663,7 +1730,7 @@ interface ReplayOptions {
     /**
      * Replay only the selected traces that carry at least one assertion.
      *
-     * Narrows whatever `limit`, `traceIds` or `datasetId` selected, so a run that
+     * Narrows whatever `limit`, `traceIds` or `datasetIds` selected, so a run that
      * measures assertion outcomes does not pay to re-execute traces nothing can
      * grade. The server applies it as part of selection, which is what makes it
      * compose with `limit`: `{ limit: 10, onlyWithAssertions: true }` is the ten
@@ -2566,6 +2633,7 @@ declare class Bitfab {
     readonly datasets: DatasetsClient;
     readonly traces: TracesClient;
     readonly labels: LabelsClient;
+    readonly graders: GradersClient;
     private readonly bamlClient;
     private readonly dbSnapshot;
     private readonly autoTracePolicyRefreshes;
@@ -3211,7 +3279,7 @@ declare class BitfabFunction {
 /**
  * SDK version from package.json (injected at build time)
  */
-declare const __version__ = "0.48.0";
+declare const __version__ = "0.51.0";
 
 /**
  * Constants for the Bitfab SDK.
@@ -3341,4 +3409,4 @@ 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 ArchiveAssertionsParams, BITFAB_PROGRESS_PREFIX, type BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, type BitfabConfig, BitfabError, BitfabFunction, BitfabLangGraphCallbackHandler as BitfabLangChainCallbackHandler, BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, type BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabVercelAiHandler, type CaptureSurface, type CaptureWhen, type CapturedSpan, type CodeChangeFile, 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 GraderRerun, type GraderRerunProgress, type GraderRerunResult, type GraderRerunStatus, HttpClient, type LabelAction, type LabelConfidence, type LabelOutcome, 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 ProviderDefinition, type RemoveDatasetGradersResult, type RemoveDatasetTracesResult, ReplayBranch, ReplayError, type ReplayItem, type ReplayItemFinishProgress, type ReplayItemStartProgress, type ReplayOptions, type ReplayOptionsFactory, type ReplayProgress, type ReplayProgressItem, type ReplayRegistration, type ReplayRegistry, type ReplayRegistryContext, type ReplayRegistryOptions, type ReplayResult, type RerunGradersOptions, type RerunGradersResult, SUPPORTED_PROVIDERS, type SaveAssertion, type SaveAssertionsParams, type SaveDatasetParams, type SaveDatasetResult, type SeedCase, type SeedCaseOptions, type SeedResult, type SeedRunOptions, type SpanLookup, type SpanMethodDecorator, type SpanMethodDecoratorContext, type SpanNodeMeta, type SpanOccurrence, type SpanOptions, type SpanType, type TokenUsage, type TraceAssertion, type TraceAssertionSource, type TraceAssertionsResult, type TraceIngestionType, type TraceOutline, type TraceOutlineSpan, type TraceOutlineSpanError, type TraceResponse, 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, seedFromRegistry, serializeReplayResult };
+export { type ActiveSpanContext, type AdaptContext, type AdaptInputsFn, type AddDatasetGradersResult, type AddDatasetTracesResult, type AllowedEnvVars, type ArchiveAssertionsParams, type AssertionVerdict, BITFAB_PROGRESS_PREFIX, type BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, type BitfabConfig, BitfabError, BitfabFunction, BitfabLangGraphCallbackHandler as BitfabLangChainCallbackHandler, BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, type BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabVercelAiHandler, type CaptureSurface, type CaptureWhen, type CapturedSpan, type CodeChangeFile, 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 GetGraderLabelsParams, type GraderLabel, type GraderLabelSource, type GraderRerun, type GraderRerunProgress, type GraderRerunResult, type GraderRerunStatus, GradersClient, HttpClient, type HumanLabelOutcome, type HumanLabelUpdate, type LabelAction, type LabelConfidence, type LabelOutcome, type LabelSource, type LabelStatus, 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 ProviderDefinition, type RemoveDatasetGradersResult, type RemoveDatasetTracesResult, ReplayBranch, ReplayError, type ReplayItem, type ReplayItemFinishProgress, type ReplayItemStartProgress, type ReplayOptions, type ReplayOptionsFactory, type ReplayProgress, type ReplayProgressItem, type ReplayRegistration, type ReplayRegistry, type ReplayRegistryContext, type ReplayRegistryOptions, type ReplayResult, type RerunGradersOptions, type RerunGradersResult, SUPPORTED_PROVIDERS, type SaveAssertion, type SaveAssertionsParams, type SaveDatasetParams, type SaveDatasetResult, type SeedCase, type SeedCaseOptions, type SeedResult, type SeedRunOptions, type SpanLookup, type SpanMethodDecorator, type SpanMethodDecoratorContext, type SpanNodeMeta, type SpanOccurrence, type SpanOptions, type SpanType, type TokenUsage, type TraceAssertion, type TraceAssertionSource, type TraceAssertionsResult, type TraceIngestionType, type TraceLabels, type TraceOutline, type TraceOutlineSpan, type TraceOutlineSpanError, type TraceResponse, 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, seedFromRegistry, serializeReplayResult };

package/dist/index.js

@@ -17,6 +17,7 @@ import {
   BitfabOpenAITracingProcessor,
   BitfabVercelAiHandler,
   DatasetsClient,
+  GradersClient,
   LabelsClient,
   SUPPORTED_PROVIDERS,
   TracesClient,
@@ -26,7 +27,7 @@ import {
   getCurrentSpan,
   getCurrentTrace,
   seedFromRegistry
-} from "./chunk-JXDMOPST.js";
+} from "./chunk-2RAX75MI.js";
 import "./chunk-ZUD7OFYB.js";
 import {
   BITFAB_PROGRESS_PREFIX,
@@ -35,7 +36,7 @@ import {
   ReplayError,
   reportReplayProgress,
   serializeReplayResult
-} from "./chunk-I6SVO5KX.js";
+} from "./chunk-TCOMLVHK.js";
 import {
   BitfabError,
   DEFAULT_SERVICE_URL,
@@ -43,7 +44,7 @@ import {
   MixedTracingError,
   __version__,
   flushTraces
-} from "./chunk-3VHHLXEJ.js";
+} from "./chunk-4Z5FCOMO.js";
 import "./chunk-H6LZRFMN.js";
 export {
   BITFAB_PROGRESS_PREFIX,
@@ -60,6 +61,7 @@ export {
   DEFAULT_SERVICE_URL,
   DatasetsClient,
   DbBranchReplayError,
+  GradersClient,
   HttpClient,
   LabelsClient,
   MixedTracingError,

package/dist/node.cjs

@@ -97,7 +97,7 @@ var __version__, __packageName__;
 var init_version_generated = __esm({
   "src/version.generated.ts"() {
     "use strict";
-    __version__ = "0.48.0";
+    __version__ = "0.51.0";
     __packageName__ = "bitfab";
   }
 });
@@ -1863,7 +1863,7 @@ var init_http = __esm({
        * Start a replay session by fetching historical traces.
        * Blocking call - creates a test run and returns lightweight item references.
        */
-      async startReplay(traceFunctionKey, limit, traceIds, name, codeChangeDescription, codeChangeFiles, includeDbBranchLease, experimentGroupId, datasetId, graderIds, dbBranchSettings, attempts, includeOriginalMetadata, onlyWithAssertions) {
+      async startReplay(traceFunctionKey, limit, traceIds, name, codeChangeDescription, codeChangeFiles, includeDbBranchLease, experimentGroupId, datasetIds, graderIds, dbBranchSettings, attempts, includeOriginalMetadata, onlyWithAssertions) {
         const payload = { traceFunctionKey };
         if (limit !== void 0) {
           payload.limit = limit;
@@ -1887,8 +1887,12 @@ var init_http = __esm({
         if (experimentGroupId !== void 0) {
           payload.experimentGroupId = experimentGroupId;
         }
-        if (datasetId !== void 0) {
-          payload.datasetId = datasetId;
+        if (datasetIds !== void 0) {
+          if (datasetIds.length === 1) {
+            payload.datasetId = datasetIds[0];
+          } else {
+            payload.datasetIds = datasetIds;
+          }
         }
         if (graderIds !== void 0) {
           payload.graderIds = graderIds;
@@ -2490,6 +2494,23 @@ function resolveAttempts(attempts) {
   }
   return attempts;
 }
+function resolveDatasetIds(options) {
+  if (options?.datasetId !== void 0 && options?.datasetIds !== void 0) {
+    throw new BitfabError(
+      "datasetId and datasetIds are the same selector: pass one dataset through datasetId, or several through datasetIds."
+    );
+  }
+  if (options?.datasetId !== void 0) {
+    return [options.datasetId];
+  }
+  if (options?.datasetIds === void 0) {
+    return void 0;
+  }
+  if (options.datasetIds.length === 0) {
+    throw new BitfabError("datasetIds must contain at least one dataset ID.");
+  }
+  return [...new Set(options.datasetIds)];
+}
 function dbBranchEnabled(dbBranch) {
   return dbBranch !== void 0 && dbBranch !== false;
 }
@@ -2922,9 +2943,10 @@ async function replay(httpClient, serviceUrl, traceFunctionKey, fn, options, reg
       );
     }
   }
-  if (options?.traceIds !== void 0 && options?.datasetId !== void 0) {
+  const datasetIds = resolveDatasetIds(options);
+  if (options?.traceIds !== void 0 && datasetIds !== void 0) {
     throw new BitfabError(
-      "traceIds and datasetId select different replay sources and cannot be used together."
+      "traceIds and datasetIds select different replay sources and cannot be used together."
     );
   }
   if (options?.limit !== void 0 && options?.traceIds !== void 0) {
@@ -2966,7 +2988,7 @@ async function replay(httpClient, serviceUrl, traceFunctionKey, fn, options, reg
     // before it could report its resolved inputs.
     dbBranchEnabled(options?.dbBranch) && options?.dryRun !== true,
     options?.experimentGroupId,
-    options?.datasetId,
+    datasetIds,
     options?.graderIds,
     resolveDbBranchSettings(options?.dbBranch),
     attempts,
@@ -3288,6 +3310,7 @@ __export(node_exports, {
   DEFAULT_SERVICE_URL: () => DEFAULT_SERVICE_URL,
   DatasetsClient: () => DatasetsClient,
   DbBranchReplayError: () => DbBranchReplayError,
+  GradersClient: () => GradersClient,
   HttpClient: () => HttpClient,
   LabelsClient: () => LabelsClient,
   MixedTracingError: () => MixedTracingError,
@@ -4590,10 +4613,45 @@ function buildSnapshotRef(config, sdkWallClockBeforeFn) {
 
 // src/client.ts
 init_errors();
+
+// src/graders.ts
+init_errors();
+var GRADER_LABELS_PATH = "/api/sdk/graderLabels";
+var GradersClient = class {
+  constructor(httpClient) {
+    this.httpClient = httpClient;
+  }
+  async getLabels(params) {
+    const hasTraceIds = params.traceIds !== void 0 && params.traceIds.length > 0;
+    const hasGraderId = params.graderId !== void 0;
+    if (!hasTraceIds && !hasGraderId) {
+      throw new BitfabError(
+        "Pass traceIds, graderId, or both: there is nothing to read otherwise."
+      );
+    }
+    const query = new URLSearchParams();
+    if (params.traceIds !== void 0 && params.traceIds.length > 0) {
+      query.set("traceIds", params.traceIds.join(","));
+    }
+    if (params.graderId !== void 0) {
+      query.set("graderId", params.graderId);
+    }
+    if (params.limit !== void 0) {
+      query.set("limit", String(params.limit));
+    }
+    const response = await this.httpClient.get(
+      `${GRADER_LABELS_PATH}?${query.toString()}`
+    );
+    return response.labels;
+  }
+};
+
+// src/client.ts
 init_http();
 
 // src/labels.ts
 var LABELS_PATH = "/api/sdk/traces/labels";
+var HUMAN_LABELS_PATH = "/api/sdk/traces/labels/human";
 var LabelsClient = class {
   constructor(httpClient) {
     this.httpClient = httpClient;
@@ -4612,6 +4670,28 @@ var LabelsClient = class {
     );
     return response.labels;
   }
+  async saveHuman(update) {
+    const [outcome] = await this.saveHumanAll([update]);
+    return outcome;
+  }
+  async saveHumanAll(updates) {
+    const response = await this.httpClient.request(HUMAN_LABELS_PATH, { labels: updates });
+    return response.labels;
+  }
+  async get(traceId) {
+    const [labels] = await this.getAll([traceId]);
+    return labels ?? null;
+  }
+  async getAll(traceIds) {
+    if (traceIds.length === 0) {
+      return [];
+    }
+    const query = new URLSearchParams({ traceIds: traceIds.join(",") });
+    const response = await this.httpClient.get(
+      `${LABELS_PATH}?${query.toString()}`
+    );
+    return response.labels;
+  }
 };
 
 // src/langgraph.ts
@@ -6389,6 +6469,7 @@ var Bitfab = class {
     this.datasets = new DatasetsClient(this.httpClient);
     this.traces = new TracesClient(this.httpClient);
     this.labels = new LabelsClient(this.httpClient);
+    this.graders = new GradersClient(this.httpClient);
   }
   /**
    * Decorate a class method as an automatically expanded trace root.
@@ -8293,6 +8374,7 @@ assertAsyncStorageRegistered();
   DEFAULT_SERVICE_URL,
   DatasetsClient,
   DbBranchReplayError,
+  GradersClient,
   HttpClient,
   LabelsClient,
   MixedTracingError,