bitfab 0.58.5 → 0.59.0

package/dist/index.d.cts

@@ -1244,14 +1244,19 @@ interface GraderRerun {
 }
 interface RerunGradersOptions {
     graderIds?: string[];
-    wait?: boolean;
     timeoutMs?: number;
-    pollIntervalMs?: number;
 }
 interface RerunGradersResult {
     run: GraderRerun;
     joinedExisting: boolean;
 }
+declare class GraderRerunError extends Error {
+    readonly run: GraderRerun;
+    constructor(message: string, run: GraderRerun);
+}
+declare class GraderRerunTimeoutError extends GraderRerunError {
+    constructor(datasetId: string, run: GraderRerun, timeoutMs: number);
+}
 /**
  * Dataset operations for the authenticated organization, reached as
  * `client.datasets`. A dataset is a named bucket of traces scoped to one trace
@@ -1291,12 +1296,6 @@ declare class DatasetsClient {
     addGraders(datasetId: string, graderIds: string[]): Promise<AddDatasetGradersResult>;
     /** Unassign graders from the dataset. */
     removeGraders(datasetId: string, graderIds: string[]): Promise<RemoveDatasetGradersResult>;
-    /**
-     * Re-run graders over every trace in the dataset. Defaults to every assigned
-     * grader; an unassigned id is rejected. Waits for the run to finish (up to
-     * `timeoutMs`, default 90s) unless `wait` is `false`, and returns the last
-     * run state seen either way. A request matching an in-flight run joins it.
-     */
     rerunGraders(datasetId: string, options?: RerunGradersOptions): Promise<RerunGradersResult>;
     /**
      * The dataset's active grader re-run, or the run named by `runId`. Returns
@@ -1400,9 +1399,18 @@ type SaveAssertion = {
      */
     assigneeEmail?: string | null;
 };
+type AssertionGenerationStatus = "running" | "completed" | "failed";
+type AssertionGeneration = {
+    status: AssertionGenerationStatus;
+    error: string | null;
+    assertionIds: string[];
+    startedAt: string;
+    finishedAt: string | null;
+};
 type TraceAssertionsResult = {
     assertions: TraceAssertion[];
     inheritedFrom: string | null;
+    generation: AssertionGeneration | null;
 };
 type TraceAssertionsUpdate = {
     traceId: string;
@@ -1421,6 +1429,19 @@ type ArchiveAssertionsParams = {
     traceId: string;
     assertionIds: string[];
 };
+type GenerateAssertionsOptions = {
+    guidance?: string;
+    categoryIds?: string[];
+    regenerate?: boolean;
+    timeoutMs?: number;
+};
+declare class AssertionGenerationError extends Error {
+    readonly traceId: string;
+    constructor(message: string, traceId: string);
+}
+declare class AssertionGenerationTimeoutError extends AssertionGenerationError {
+    constructor(traceId: string, timeoutMs: number);
+}
 declare class TracesClient {
     private readonly httpClient;
     constructor(httpClient: HttpClient);
@@ -1428,6 +1449,7 @@ declare class TracesClient {
     getAssertions(traceId: string): Promise<TraceAssertionsResult>;
     saveAssertions(params: SaveAssertionsParams): Promise<TraceAssertion[]>;
     saveAssertionsAll(params: SaveAssertionsAllParams): Promise<TraceAssertion[]>;
+    generateAssertions(traceId: string, options?: GenerateAssertionsOptions): Promise<TraceAssertion[]>;
     archiveAssertions(params: ArchiveAssertionsParams): Promise<string[]>;
 }
 
@@ -3659,7 +3681,7 @@ interface CommitRef {
 /**
  * SDK version from package.json (injected at build time)
  */
-declare const __version__ = "0.58.5";
+declare const __version__ = "0.59.0";
 
 /**
  * Constants for the Bitfab SDK.
@@ -3798,4 +3820,4 @@ 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 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 GetGraderLabelsParams, type GraderLabel, type GraderLabelSource, type GraderRerun, type GraderRerunProgress, type GraderRerunResult, type GraderRerunStatus, 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 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, 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 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 TraceOutline, type TraceOutlineSpan, type TraceOutlineSpanError, 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 GenerateAssertionsOptions, type GetGraderLabelsParams, type GraderLabel, type GraderLabelSource, 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 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, 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 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 TraceOutline, type TraceOutlineSpan, type TraceOutlineSpanError, 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

@@ -1244,14 +1244,19 @@ interface GraderRerun {
 }
 interface RerunGradersOptions {
     graderIds?: string[];
-    wait?: boolean;
     timeoutMs?: number;
-    pollIntervalMs?: number;
 }
 interface RerunGradersResult {
     run: GraderRerun;
     joinedExisting: boolean;
 }
+declare class GraderRerunError extends Error {
+    readonly run: GraderRerun;
+    constructor(message: string, run: GraderRerun);
+}
+declare class GraderRerunTimeoutError extends GraderRerunError {
+    constructor(datasetId: string, run: GraderRerun, timeoutMs: number);
+}
 /**
  * Dataset operations for the authenticated organization, reached as
  * `client.datasets`. A dataset is a named bucket of traces scoped to one trace
@@ -1291,12 +1296,6 @@ declare class DatasetsClient {
     addGraders(datasetId: string, graderIds: string[]): Promise<AddDatasetGradersResult>;
     /** Unassign graders from the dataset. */
     removeGraders(datasetId: string, graderIds: string[]): Promise<RemoveDatasetGradersResult>;
-    /**
-     * Re-run graders over every trace in the dataset. Defaults to every assigned
-     * grader; an unassigned id is rejected. Waits for the run to finish (up to
-     * `timeoutMs`, default 90s) unless `wait` is `false`, and returns the last
-     * run state seen either way. A request matching an in-flight run joins it.
-     */
     rerunGraders(datasetId: string, options?: RerunGradersOptions): Promise<RerunGradersResult>;
     /**
      * The dataset's active grader re-run, or the run named by `runId`. Returns
@@ -1400,9 +1399,18 @@ type SaveAssertion = {
      */
     assigneeEmail?: string | null;
 };
+type AssertionGenerationStatus = "running" | "completed" | "failed";
+type AssertionGeneration = {
+    status: AssertionGenerationStatus;
+    error: string | null;
+    assertionIds: string[];
+    startedAt: string;
+    finishedAt: string | null;
+};
 type TraceAssertionsResult = {
     assertions: TraceAssertion[];
     inheritedFrom: string | null;
+    generation: AssertionGeneration | null;
 };
 type TraceAssertionsUpdate = {
     traceId: string;
@@ -1421,6 +1429,19 @@ type ArchiveAssertionsParams = {
     traceId: string;
     assertionIds: string[];
 };
+type GenerateAssertionsOptions = {
+    guidance?: string;
+    categoryIds?: string[];
+    regenerate?: boolean;
+    timeoutMs?: number;
+};
+declare class AssertionGenerationError extends Error {
+    readonly traceId: string;
+    constructor(message: string, traceId: string);
+}
+declare class AssertionGenerationTimeoutError extends AssertionGenerationError {
+    constructor(traceId: string, timeoutMs: number);
+}
 declare class TracesClient {
     private readonly httpClient;
     constructor(httpClient: HttpClient);
@@ -1428,6 +1449,7 @@ declare class TracesClient {
     getAssertions(traceId: string): Promise<TraceAssertionsResult>;
     saveAssertions(params: SaveAssertionsParams): Promise<TraceAssertion[]>;
     saveAssertionsAll(params: SaveAssertionsAllParams): Promise<TraceAssertion[]>;
+    generateAssertions(traceId: string, options?: GenerateAssertionsOptions): Promise<TraceAssertion[]>;
     archiveAssertions(params: ArchiveAssertionsParams): Promise<string[]>;
 }
 
@@ -3659,7 +3681,7 @@ interface CommitRef {
 /**
  * SDK version from package.json (injected at build time)
  */
-declare const __version__ = "0.58.5";
+declare const __version__ = "0.59.0";
 
 /**
  * Constants for the Bitfab SDK.
@@ -3798,4 +3820,4 @@ 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 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 GetGraderLabelsParams, type GraderLabel, type GraderLabelSource, type GraderRerun, type GraderRerunProgress, type GraderRerunResult, type GraderRerunStatus, 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 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, 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 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 TraceOutline, type TraceOutlineSpan, type TraceOutlineSpanError, 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 GenerateAssertionsOptions, type GetGraderLabelsParams, type GraderLabel, type GraderLabelSource, 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 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, 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 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 TraceOutline, type TraceOutlineSpan, type TraceOutlineSpanError, 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

@@ -9,6 +9,8 @@ if (typeof globalThis.__bitfab_deprecation_warned === "undefined") {
 }
 import {
   AssertionCategoriesClient,
+  AssertionGenerationError,
+  AssertionGenerationTimeoutError,
   Bitfab,
   BitfabClaudeAgentHandler,
   BitfabFunction,
@@ -19,6 +21,8 @@ import {
   BitfabRivetIntegration,
   BitfabVercelAiHandler,
   DatasetsClient,
+  GraderRerunError,
+  GraderRerunTimeoutError,
   GradersClient,
   LabelsClient,
   OrganizationMembersClient,
@@ -31,7 +35,7 @@ import {
   getCurrentTrace,
   reseedFromRegistry,
   seedFromRegistry
-} from "./chunk-UYAQRSCB.js";
+} from "./chunk-RYSCYPHQ.js";
 import "./chunk-ETZUSSF6.js";
 import {
   BITFAB_PROGRESS_PREFIX,
@@ -40,7 +44,7 @@ import {
   ReplayError,
   reportReplayProgress,
   serializeReplayResult
-} from "./chunk-TAEWAXPA.js";
+} from "./chunk-FHEEAPG2.js";
 import {
   BitfabError,
   DEFAULT_SERVICE_URL,
@@ -48,10 +52,12 @@ import {
   MixedTracingError,
   __version__,
   flushTraces
-} from "./chunk-7SPNHC33.js";
+} from "./chunk-KFZ5OIN5.js";
 import "./chunk-BAOIRE5T.js";
 export {
   AssertionCategoriesClient,
+  AssertionGenerationError,
+  AssertionGenerationTimeoutError,
   BITFAB_PROGRESS_PREFIX,
   Bitfab,
   BitfabClaudeAgentHandler,
@@ -67,6 +73,8 @@ export {
   DEFAULT_SERVICE_URL,
   DatasetsClient,
   DbBranchReplayError,
+  GraderRerunError,
+  GraderRerunTimeoutError,
   GradersClient,
   HttpClient,
   LabelsClient,

package/dist/node.cjs

@@ -97,7 +97,7 @@ var __version__, __packageName__;
 var init_version_generated = __esm({
   "src/version.generated.ts"() {
     "use strict";
-    __version__ = "0.58.5";
+    __version__ = "0.59.0";
     __packageName__ = "bitfab";
   }
 });
@@ -4406,6 +4406,8 @@ var init_replay = __esm({
 var node_exports = {};
 __export(node_exports, {
   AssertionCategoriesClient: () => AssertionCategoriesClient,
+  AssertionGenerationError: () => AssertionGenerationError,
+  AssertionGenerationTimeoutError: () => AssertionGenerationTimeoutError,
   BITFAB_PROGRESS_PREFIX: () => BITFAB_PROGRESS_PREFIX,
   Bitfab: () => Bitfab,
   BitfabClaudeAgentHandler: () => BitfabClaudeAgentHandler,
@@ -4421,6 +4423,8 @@ __export(node_exports, {
   DEFAULT_SERVICE_URL: () => DEFAULT_SERVICE_URL,
   DatasetsClient: () => DatasetsClient,
   DbBranchReplayError: () => DbBranchReplayError,
+  GraderRerunError: () => GraderRerunError,
+  GraderRerunTimeoutError: () => GraderRerunTimeoutError,
   GradersClient: () => GradersClient,
   HttpClient: () => HttpClient,
   LabelsClient: () => LabelsClient,
@@ -5855,11 +5859,39 @@ init_constants();
 
 // src/datasets.ts
 var DEFAULT_RERUN_TIMEOUT_MS = 9e4;
-var DEFAULT_RERUN_POLL_INTERVAL_MS = 1e3;
+var RERUN_POLL_INTERVAL_MS = 1e3;
 var TERMINAL_RERUN_STATUSES = /* @__PURE__ */ new Set([
   "completed",
   "errored"
 ]);
+var GraderRerunError = class extends Error {
+  constructor(message, run) {
+    super(message);
+    this.run = run;
+    this.name = "GraderRerunError";
+  }
+};
+var GraderRerunTimeoutError = class extends GraderRerunError {
+  constructor(datasetId, run, timeoutMs) {
+    super(
+      `Grader re-run ${run.id} on dataset ${datasetId} was still ${run.status} after ${timeoutMs}ms. It keeps running on Bitfab; read it later with getGraderRerun.`,
+      run
+    );
+    this.name = "GraderRerunTimeoutError";
+  }
+};
+function finishedRerun(datasetId, run, joinedExisting) {
+  if (!TERMINAL_RERUN_STATUSES.has(run.status)) {
+    return null;
+  }
+  if (run.status === "errored") {
+    throw new GraderRerunError(
+      `Grader re-run ${run.id} on dataset ${datasetId} errored: ${run.error ?? "no reason recorded"}`,
+      run
+    );
+  }
+  return { run, joinedExisting };
+}
 function sleep(ms) {
   return new Promise((resolve) => setTimeout(resolve, ms));
 }
@@ -5959,28 +5991,25 @@ var DatasetsClient = class {
       { graderIds }
     );
   }
-  /**
-   * Re-run graders over every trace in the dataset. Defaults to every assigned
-   * grader; an unassigned id is rejected. Waits for the run to finish (up to
-   * `timeoutMs`, default 90s) unless `wait` is `false`, and returns the last
-   * run state seen either way. A request matching an in-flight run joins it.
-   */
   async rerunGraders(datasetId, options = {}) {
     const started = await this.httpClient.request(
       datasetPath(datasetId, "/rerunGraders"),
       options.graderIds === void 0 ? {} : { graderIds: options.graderIds }
     );
-    if (options.wait === false) {
-      return started;
-    }
-    const deadline = Date.now() + (options.timeoutMs ?? DEFAULT_RERUN_TIMEOUT_MS);
-    const interval = options.pollIntervalMs ?? DEFAULT_RERUN_POLL_INTERVAL_MS;
+    const timeoutMs = options.timeoutMs ?? DEFAULT_RERUN_TIMEOUT_MS;
+    const deadline = Date.now() + timeoutMs;
     let run = started.run;
-    while (!TERMINAL_RERUN_STATUSES.has(run.status) && Date.now() < deadline) {
-      await sleep(interval);
+    while (true) {
+      const finished = finishedRerun(datasetId, run, started.joinedExisting);
+      if (finished !== null) {
+        return finished;
+      }
+      if (Date.now() >= deadline) {
+        throw new GraderRerunTimeoutError(datasetId, run, timeoutMs);
+      }
+      await sleep(RERUN_POLL_INTERVAL_MS);
       run = await this.getGraderRerun(datasetId, run.id) ?? run;
     }
-    return { run, joinedExisting: started.joinedExisting };
   }
   /**
    * The dataset's active grader re-run, or the run named by `runId`. Returns
@@ -7836,6 +7865,46 @@ init_traceMetadata();
 // src/traces.ts
 var ASSERTIONS_PATH = "/api/sdk/traces/assertions";
 var SEARCH_PATH = "/api/sdk/traces/search";
+var DEFAULT_GENERATION_TIMEOUT_MS = 6e5;
+var GENERATION_POLL_INTERVAL_MS = 2e3;
+var AssertionGenerationError = class extends Error {
+  constructor(message, traceId) {
+    super(message);
+    this.traceId = traceId;
+    this.name = "AssertionGenerationError";
+  }
+};
+var AssertionGenerationTimeoutError = class extends AssertionGenerationError {
+  constructor(traceId, timeoutMs) {
+    super(
+      `Assertion generation for trace ${traceId} did not finish within ${timeoutMs}ms. Generation keeps running on Bitfab. Call getAssertions and read its generation field to check its status.`,
+      traceId
+    );
+    this.name = "AssertionGenerationTimeoutError";
+  }
+};
+function sleep2(ms) {
+  return new Promise((resolve) => setTimeout(resolve, ms));
+}
+function isGenerationFinished(traceId, generation) {
+  if (generation === null || generation.status === "running") {
+    return false;
+  }
+  if (generation.status === "failed") {
+    throw new AssertionGenerationError(
+      `Assertion generation for trace ${traceId} failed: ${generation.error ?? "no reason recorded"}. Pass regenerate: true to try again.`,
+      traceId
+    );
+  }
+  return true;
+}
+function draftsInGenerationOrder(assertions, assertionIds) {
+  const byId = new Map(assertions.map((assertion) => [assertion.id, assertion]));
+  return assertionIds.flatMap((id) => {
+    const assertion = byId.get(id);
+    return assertion === void 0 ? [] : [assertion];
+  });
+}
 function traceAssertionsPath(traceId, suffix = "") {
   return `/api/sdk/traces/${encodeURIComponent(traceId)}/assertions${suffix}`;
 }
@@ -7867,6 +7936,34 @@ var TracesClient = class {
     });
     return response.assertions;
   }
+  async generateAssertions(traceId, options = {}) {
+    const started = await this.httpClient.request(
+      traceAssertionsPath(traceId, "/generate"),
+      {
+        ...options.guidance === void 0 ? {} : { guidance: options.guidance },
+        ...options.categoryIds === void 0 ? {} : { categoryIds: options.categoryIds },
+        ...options.regenerate === void 0 ? {} : { regenerate: options.regenerate }
+      }
+    );
+    if (isGenerationFinished(traceId, started.generation)) {
+      return started.assertions;
+    }
+    const timeoutMs = options.timeoutMs ?? DEFAULT_GENERATION_TIMEOUT_MS;
+    const deadline = Date.now() + timeoutMs;
+    while (true) {
+      if (Date.now() >= deadline) {
+        throw new AssertionGenerationTimeoutError(traceId, timeoutMs);
+      }
+      await sleep2(GENERATION_POLL_INTERVAL_MS);
+      const read2 = await this.getAssertions(traceId);
+      if (isGenerationFinished(traceId, read2.generation)) {
+        return draftsInGenerationOrder(
+          read2.assertions,
+          read2.generation.assertionIds
+        );
+      }
+    }
+  }
   async archiveAssertions(params) {
     const response = await this.httpClient.request(
       traceAssertionsPath(params.traceId, "/archive"),
@@ -11089,6 +11186,8 @@ assertAsyncStorageRegistered();
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   AssertionCategoriesClient,
+  AssertionGenerationError,
+  AssertionGenerationTimeoutError,
   BITFAB_PROGRESS_PREFIX,
   Bitfab,
   BitfabClaudeAgentHandler,
@@ -11104,6 +11203,8 @@ assertAsyncStorageRegistered();
   DEFAULT_SERVICE_URL,
   DatasetsClient,
   DbBranchReplayError,
+  GraderRerunError,
+  GraderRerunTimeoutError,
   GradersClient,
   HttpClient,
   LabelsClient,