@bitfab/sdk 0.50.0 → 0.52.0

package/dist/node.d.cts

@@ -1 +1 @@
-export { ActiveSpanContext, AdaptContext, AdaptInputsFn, AddDatasetGradersResult, AddDatasetTracesResult, AllowedEnvVars, ArchiveAssertionsParams, AssertionVerdict, BITFAB_PROGRESS_PREFIX, BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, BitfabConfig, BitfabError, BitfabFunction, BitfabLangChainCallbackHandler, BitfabLangChainCallbackHandler as BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabVercelAiHandler, CaptureSurface, CaptureWhen, CapturedSpan, CodeChangeFile, CurrentSpan, CurrentTrace, DEFAULT_SERVICE_URL, Dataset, DatasetGraderRef, DatasetTraceIds, DatasetsClient, DbBranchOptions, DbBranchReplayError, DbBranchTimings, DbSnapshotConfig, DbSnapshotProvider, DbSnapshotRef, DetachedTrace, GetGraderLabelsParams, GraderLabel, GraderLabelSource, GraderRerun, GraderRerunProgress, GraderRerunResult, GraderRerunStatus, GradersClient, HttpClient, HumanLabelOutcome, HumanLabelUpdate, LabelAction, LabelConfidence, LabelOutcome, LabelSource, LabelStatus, LabelUpdate, LabelsClient, LangGraphIntegrationOptions, ListDatasetsParams, MixedTracingError, MockOverride, MockOverrideCtx, MockOverrideInput, MockOverrideResolver, MockStrategy, MockValue, NO_MOCK_OVERRIDE, NodeMatcher, NodeMethodDecorator, NodeOptions, ProviderDefinition, RemoveDatasetGradersResult, RemoveDatasetTracesResult, ReplayBranch, ReplayError, ReplayItem, ReplayItemFinishProgress, ReplayItemStartProgress, ReplayOptions, ReplayOptionsFactory, ReplayProgress, ReplayProgressItem, ReplayRegistration, ReplayRegistry, ReplayRegistryContext, ReplayRegistryOptions, ReplayResult, RerunGradersOptions, RerunGradersResult, SUPPORTED_PROVIDERS, SaveAssertion, SaveAssertionsParams, SaveDatasetParams, SaveDatasetResult, SeedCase, SeedCaseOptions, SeedResult, SeedRunOptions, SpanLookup, SpanMethodDecorator, SpanMethodDecoratorContext, SpanNodeMeta, SpanOccurrence, SpanOptions, SpanType, TokenUsage, TraceAssertion, TraceAssertionSource, TraceAssertionsResult, TraceIngestionType, TraceLabels, TraceOutline, TraceOutlineSpan, TraceOutlineSpanError, TraceResponse, TraceTarget, TraceTargetOccurrence, TracesClient, TracingProcessor, VercelCallParams, VercelGenerateResult, VercelStreamResult, WrapBAMLOptions, WrappedBamlFn, __version__, defineReplayRegistry, finalizers, flushTraces, getCurrentReplayBranch, getCurrentSpan, getCurrentTrace, reportReplayProgress, seedFromRegistry, serializeReplayResult } from './index.cjs';
+export { ActiveSpanContext, AdaptContext, AdaptInputsFn, AddDatasetGradersResult, AddDatasetTracesResult, AllowedEnvVars, ArchiveAssertionsParams, AssertionVerdict, BITFAB_PROGRESS_PREFIX, BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, BitfabConfig, BitfabError, BitfabFunction, BitfabLangChainCallbackHandler, BitfabLangChainCallbackHandler as BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabVercelAiHandler, CaptureSurface, CaptureWhen, CapturedSpan, CodeChangeFile, CurrentSpan, CurrentTrace, DEFAULT_SERVICE_URL, Dataset, DatasetGraderRef, DatasetTraceIds, DatasetsClient, DbBranchOptions, DbBranchReplayError, DbBranchTimings, DbSnapshotConfig, DbSnapshotProvider, DbSnapshotRef, DetachedTrace, GetGraderLabelsParams, GraderLabel, GraderLabelSource, GraderRerun, GraderRerunProgress, GraderRerunResult, GraderRerunStatus, GradersClient, HttpClient, HumanLabelOutcome, HumanLabelUpdate, LabelAction, LabelConfidence, LabelOutcome, LabelSource, LabelStatus, LabelUpdate, LabelsClient, LangGraphIntegrationOptions, ListDatasetsParams, MixedTracingError, MockOverride, MockOverrideCtx, MockOverrideInput, MockOverrideResolver, MockStrategy, MockValue, NO_MOCK_OVERRIDE, NodeMatcher, NodeMethodDecorator, NodeOptions, ProviderDefinition, RemoveDatasetGradersResult, RemoveDatasetTracesResult, ReplayBranch, ReplayError, ReplayItem, ReplayItemFinishProgress, ReplayItemStartProgress, ReplayOptions, ReplayOptionsFactory, ReplayProgress, ReplayProgressItem, ReplayRegistration, ReplayRegistry, ReplayRegistryContext, ReplayRegistryOptions, ReplayResult, RerunGradersOptions, RerunGradersResult, SUPPORTED_PROVIDERS, SaveAssertion, SaveAssertionsAllParams, SaveAssertionsParams, SaveDatasetParams, SaveDatasetResult, SeedCase, SeedCaseOptions, SeedResult, SeedRunOptions, SpanLookup, SpanMethodDecorator, SpanMethodDecoratorContext, SpanNodeMeta, SpanOccurrence, SpanOptions, SpanType, TokenUsage, TraceAssertion, TraceAssertionSource, TraceAssertionsResult, TraceAssertionsUpdate, TraceIngestionType, TraceLabels, TraceOutline, TraceOutlineSpan, TraceOutlineSpanError, TraceResponse, TraceTarget, TraceTargetOccurrence, TracesClient, TracingProcessor, VercelCallParams, VercelGenerateResult, VercelStreamResult, WrapBAMLOptions, WrappedBamlFn, __version__, defineReplayRegistry, finalizers, flushTraces, getCurrentReplayBranch, getCurrentSpan, getCurrentTrace, reportReplayProgress, seedFromRegistry, serializeReplayResult } from './index.cjs';

package/dist/node.d.ts

@@ -1 +1 @@
-export { ActiveSpanContext, AdaptContext, AdaptInputsFn, AddDatasetGradersResult, AddDatasetTracesResult, AllowedEnvVars, ArchiveAssertionsParams, AssertionVerdict, BITFAB_PROGRESS_PREFIX, BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, BitfabConfig, BitfabError, BitfabFunction, BitfabLangChainCallbackHandler, BitfabLangChainCallbackHandler as BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabVercelAiHandler, CaptureSurface, CaptureWhen, CapturedSpan, CodeChangeFile, CurrentSpan, CurrentTrace, DEFAULT_SERVICE_URL, Dataset, DatasetGraderRef, DatasetTraceIds, DatasetsClient, DbBranchOptions, DbBranchReplayError, DbBranchTimings, DbSnapshotConfig, DbSnapshotProvider, DbSnapshotRef, DetachedTrace, GetGraderLabelsParams, GraderLabel, GraderLabelSource, GraderRerun, GraderRerunProgress, GraderRerunResult, GraderRerunStatus, GradersClient, HttpClient, HumanLabelOutcome, HumanLabelUpdate, LabelAction, LabelConfidence, LabelOutcome, LabelSource, LabelStatus, LabelUpdate, LabelsClient, LangGraphIntegrationOptions, ListDatasetsParams, MixedTracingError, MockOverride, MockOverrideCtx, MockOverrideInput, MockOverrideResolver, MockStrategy, MockValue, NO_MOCK_OVERRIDE, NodeMatcher, NodeMethodDecorator, NodeOptions, ProviderDefinition, RemoveDatasetGradersResult, RemoveDatasetTracesResult, ReplayBranch, ReplayError, ReplayItem, ReplayItemFinishProgress, ReplayItemStartProgress, ReplayOptions, ReplayOptionsFactory, ReplayProgress, ReplayProgressItem, ReplayRegistration, ReplayRegistry, ReplayRegistryContext, ReplayRegistryOptions, ReplayResult, RerunGradersOptions, RerunGradersResult, SUPPORTED_PROVIDERS, SaveAssertion, SaveAssertionsParams, SaveDatasetParams, SaveDatasetResult, SeedCase, SeedCaseOptions, SeedResult, SeedRunOptions, SpanLookup, SpanMethodDecorator, SpanMethodDecoratorContext, SpanNodeMeta, SpanOccurrence, SpanOptions, SpanType, TokenUsage, TraceAssertion, TraceAssertionSource, TraceAssertionsResult, TraceIngestionType, TraceLabels, TraceOutline, TraceOutlineSpan, TraceOutlineSpanError, TraceResponse, TraceTarget, TraceTargetOccurrence, TracesClient, TracingProcessor, VercelCallParams, VercelGenerateResult, VercelStreamResult, WrapBAMLOptions, WrappedBamlFn, __version__, defineReplayRegistry, finalizers, flushTraces, getCurrentReplayBranch, getCurrentSpan, getCurrentTrace, reportReplayProgress, seedFromRegistry, serializeReplayResult } from './index.js';
+export { ActiveSpanContext, AdaptContext, AdaptInputsFn, AddDatasetGradersResult, AddDatasetTracesResult, AllowedEnvVars, ArchiveAssertionsParams, AssertionVerdict, BITFAB_PROGRESS_PREFIX, BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, BitfabConfig, BitfabError, BitfabFunction, BitfabLangChainCallbackHandler, BitfabLangChainCallbackHandler as BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabVercelAiHandler, CaptureSurface, CaptureWhen, CapturedSpan, CodeChangeFile, CurrentSpan, CurrentTrace, DEFAULT_SERVICE_URL, Dataset, DatasetGraderRef, DatasetTraceIds, DatasetsClient, DbBranchOptions, DbBranchReplayError, DbBranchTimings, DbSnapshotConfig, DbSnapshotProvider, DbSnapshotRef, DetachedTrace, GetGraderLabelsParams, GraderLabel, GraderLabelSource, GraderRerun, GraderRerunProgress, GraderRerunResult, GraderRerunStatus, GradersClient, HttpClient, HumanLabelOutcome, HumanLabelUpdate, LabelAction, LabelConfidence, LabelOutcome, LabelSource, LabelStatus, LabelUpdate, LabelsClient, LangGraphIntegrationOptions, ListDatasetsParams, MixedTracingError, MockOverride, MockOverrideCtx, MockOverrideInput, MockOverrideResolver, MockStrategy, MockValue, NO_MOCK_OVERRIDE, NodeMatcher, NodeMethodDecorator, NodeOptions, ProviderDefinition, RemoveDatasetGradersResult, RemoveDatasetTracesResult, ReplayBranch, ReplayError, ReplayItem, ReplayItemFinishProgress, ReplayItemStartProgress, ReplayOptions, ReplayOptionsFactory, ReplayProgress, ReplayProgressItem, ReplayRegistration, ReplayRegistry, ReplayRegistryContext, ReplayRegistryOptions, ReplayResult, RerunGradersOptions, RerunGradersResult, SUPPORTED_PROVIDERS, SaveAssertion, SaveAssertionsAllParams, SaveAssertionsParams, SaveDatasetParams, SaveDatasetResult, SeedCase, SeedCaseOptions, SeedResult, SeedRunOptions, SpanLookup, SpanMethodDecorator, SpanMethodDecoratorContext, SpanNodeMeta, SpanOccurrence, SpanOptions, SpanType, TokenUsage, TraceAssertion, TraceAssertionSource, TraceAssertionsResult, TraceAssertionsUpdate, TraceIngestionType, TraceLabels, TraceOutline, TraceOutlineSpan, TraceOutlineSpanError, TraceResponse, TraceTarget, TraceTargetOccurrence, TracesClient, TracingProcessor, VercelCallParams, VercelGenerateResult, VercelStreamResult, WrapBAMLOptions, WrappedBamlFn, __version__, defineReplayRegistry, finalizers, flushTraces, getCurrentReplayBranch, getCurrentSpan, getCurrentTrace, reportReplayProgress, seedFromRegistry, serializeReplayResult } from './index.js';

package/dist/node.js

@@ -18,7 +18,7 @@ import {
   getCurrentSpan,
   getCurrentTrace,
   seedFromRegistry
-} from "./chunk-BT4RNCQL.js";
+} from "./chunk-VEDTNQQE.js";
 import "./chunk-ZUD7OFYB.js";
 import {
   BITFAB_PROGRESS_PREFIX,
@@ -27,7 +27,7 @@ import {
   ReplayError,
   reportReplayProgress,
   serializeReplayResult
-} from "./chunk-6LG3OFIX.js";
+} from "./chunk-SJ65MW4K.js";
 import {
   BitfabError,
   DEFAULT_SERVICE_URL,
@@ -35,7 +35,7 @@ import {
   MixedTracingError,
   __version__,
   flushTraces
-} from "./chunk-XZDYNFI5.js";
+} from "./chunk-4EOEBN3M.js";
 import {
   assertAsyncStorageRegistered,
   registerAsyncLocalStorageClass

package/dist/{replay-AFX7ENZI.js → replay-SY74BXDK.js}

@@ -7,8 +7,8 @@ import {
   serializeReplayResult,
   sleepForReplayPersistence,
   waitForReplayPersistence
-} from "./chunk-6LG3OFIX.js";
-import "./chunk-XZDYNFI5.js";
+} from "./chunk-SJ65MW4K.js";
+import "./chunk-4EOEBN3M.js";
 import "./chunk-H6LZRFMN.js";
 export {
   BITFAB_PROGRESS_PREFIX,
@@ -20,4 +20,4 @@ export {
   sleepForReplayPersistence,
   waitForReplayPersistence
 };
-//# sourceMappingURL=replay-AFX7ENZI.js.map
+//# sourceMappingURL=replay-SY74BXDK.js.map

package/dist/replayCli.js

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   BitfabError
-} from "./chunk-W5P5JAH7.js";
+} from "./chunk-KXBXRR4I.js";
 
 // src/replayCli.ts
 import { tsImport } from "tsx/esm/api";
@@ -79,6 +79,7 @@ var VALUE_FLAGS = /* @__PURE__ */ new Set([
   "--code-change",
   "--experiment-group-id",
   "--dataset-id",
+  "--dataset-ids",
   "--grader-ids",
   "--mock",
   "--param",
@@ -104,7 +105,7 @@ Options:
   --concurrency N, --max-concurrency N
   --code-change PATH, --no-code-change
   --experiment-group-id UUID
-  --dataset-id UUID
+  --dataset-ids uuid1,uuid2 (--dataset-id is the same flag)
   --grader-ids id1,id2
   --only-with-assertions (replay only traces carrying an assertion)
   --mock none|all|marked
@@ -172,7 +173,7 @@ async function seedFromRegistry(registry, pipeline, cases, options = {}) {
       sessionId: seedCase.sessionId
     })
   );
-  const { flushTraces: flushTraces2 } = await import("./http-W2YPLY7H.js");
+  const { flushTraces: flushTraces2 } = await import("./http-4A2JEPBD.js");
   await flushTraces2(3e4);
   return { pipeline, traceFunctionKey, traceIds };
 }
@@ -300,7 +301,8 @@ ${usage(registry)}`
         parsed.experimentGroupId = raw;
         break;
       case "--dataset-id":
-        parsed.datasetId = raw;
+      case "--dataset-ids":
+        parsed.datasetIds = commaSeparated(flag, raw);
         break;
       case "--grader-ids":
         parsed.graderIds = commaSeparated(flag, raw);
@@ -321,9 +323,9 @@ ${usage(registry)}`
         break;
     }
   }
-  if (parsed.traceIds !== void 0 && parsed.datasetId !== void 0) {
+  if (parsed.traceIds !== void 0 && parsed.datasetIds !== void 0) {
     throw new BitfabError(
-      "--trace-ids and --dataset-id select different replay sources and cannot be used together."
+      "--trace-ids and --dataset-ids select different replay sources and cannot be used together."
     );
   }
   if (parsed.codeChangePath !== void 0 && parsed.noCodeChange === true) {
@@ -333,6 +335,27 @@ ${usage(registry)}`
   }
   return parsed;
 }
+function resolveRegistrationDatasetIds(options) {
+  if (options.datasetId !== void 0 && options.datasetIds !== void 0) {
+    throw new BitfabError(
+      "Replay registry options datasetId and datasetIds are the same selector: set one of them."
+    );
+  }
+  if (options.datasetId !== void 0) {
+    return [options.datasetId];
+  }
+  return options.datasetIds;
+}
+function describeSelection(options) {
+  if (options.traceIds !== void 0) {
+    return `${options.traceIds.length} traces`;
+  }
+  if (options.limit !== void 0) {
+    return `${options.limit} traces`;
+  }
+  const datasetCount = options.datasetIds?.length ?? 1;
+  return datasetCount > 1 ? `${datasetCount} datasets' traces` : "dataset traces";
+}
 function resolveTraceFunctionKey(registration) {
   const wrappedKey = registration.fn._bitfabTraceFunctionKey;
   const key = registration.traceFunctionKey ?? wrappedKey;
@@ -486,11 +509,13 @@ async function runReplayCli(registry, argv = typeof process === "undefined" ? []
     ...registration.options,
     ...dynamicOptions
   };
-  if (registrationOptions.traceIds !== void 0 && registrationOptions.datasetId !== void 0) {
+  const registrationDatasetIds = resolveRegistrationDatasetIds(registrationOptions);
+  if (registrationOptions.traceIds !== void 0 && registrationDatasetIds !== void 0) {
     throw new BitfabError(
-      "Replay registry options traceIds and datasetId select different sources and cannot be used together."
+      "Replay registry options traceIds and datasetIds select different sources and cannot be used together."
     );
   }
+  const selectedDatasetIds = args.datasetIds ?? registrationDatasetIds;
   const codeChange = await loadCodeChange(args.codeChangePath, readFile);
   const options = {
     ...registrationOptions,
@@ -498,7 +523,8 @@ async function runReplayCli(registry, argv = typeof process === "undefined" ? []
     ...args.attempts === void 0 ? {} : { attempts: args.attempts },
     ...args.maxConcurrency === void 0 ? {} : { maxConcurrency: args.maxConcurrency },
     ...args.experimentGroupId === void 0 ? {} : { experimentGroupId: args.experimentGroupId },
-    ...args.datasetId === void 0 ? {} : { datasetId: args.datasetId },
+    datasetId: void 0,
+    datasetIds: selectedDatasetIds,
     ...args.graderIds === void 0 ? {} : { graderIds: args.graderIds },
     ...args.onlyWithAssertions === void 0 ? {} : { onlyWithAssertions: args.onlyWithAssertions },
     ...args.mock === void 0 ? {} : { mock: args.mock },
@@ -516,23 +542,23 @@ async function runReplayCli(registry, argv = typeof process === "undefined" ? []
   };
   if (args.traceIds !== void 0) {
     options.traceIds = args.traceIds;
-    options.datasetId = void 0;
+    options.datasetIds = void 0;
     options.limit = void 0;
-  } else if (args.datasetId !== void 0) {
+  } else if (args.datasetIds !== void 0) {
     options.traceIds = void 0;
     options.limit = void 0;
   } else if (args.limit !== void 0) {
     options.traceIds = void 0;
-    options.datasetId = void 0;
+    options.datasetIds = void 0;
     options.limit = args.limit;
-  } else if (registrationOptions.traceIds !== void 0 || registrationOptions.datasetId !== void 0) {
+  } else if (registrationOptions.traceIds !== void 0 || registrationDatasetIds !== void 0) {
     options.limit = void 0;
   } else {
     options.limit = registrationOptions.limit ?? 10;
   }
   const attemptsSuffix = options.attempts !== void 0 && options.attempts > 1 ? ` (${options.attempts} attempts each)` : "";
   stderr(
-    `[replay] ${args.dryRun === true ? "Resolving inputs for" : "Replaying"} ${options.traceIds?.length ?? options.limit ?? "dataset"} traces from "${traceFunctionKey}"${attemptsSuffix}...`
+    `[replay] ${args.dryRun === true ? "Resolving inputs for" : "Replaying"} ${describeSelection(options)} from "${traceFunctionKey}"${attemptsSuffix}...`
   );
   try {
     const result = await registration.client.replay(