braintrust 3.30.0 → 3.31.0

package/dist/index.mjs

@@ -143,7 +143,7 @@ import {
   wrapMastraAgent,
   wrapTraced,
   zodToJsonSchema
-} from "./chunk-BU6SM54N.mjs";
+} from "./chunk-AJT4FR25.mjs";
 import {
   INSTRUMENTATION_NAMES,
   __export,
@@ -177,7 +177,7 @@ import {
   strandsAgentSDKChannels,
   voyageAIChannels,
   withSpanInstrumentationName
-} from "./chunk-N3JKQ3D3.mjs";
+} from "./chunk-CE3BLB2L.mjs";
 
 // src/exports.ts
 var exports_exports = {};
@@ -7170,8 +7170,8 @@ function normalizeScores(result) {
       (s) => s !== null && s !== void 0 && typeof s === "object" && isScore(s)
     );
   }
-  if (typeof result === "object" && result !== null && isScore(result)) {
-    return [result];
+  if (typeof result === "object" && result !== null && "score" in result) {
+    return [{ ...result, name: result.name ?? "score" }];
   }
   return [];
 }
@@ -9754,19 +9754,29 @@ function buildSpanScores(results) {
 function _internalPrepareEvaluatorScore(scoreValue, name) {
   if (scoreValue === null) return { results: null };
   if (Array.isArray(scoreValue)) {
+    const names = /* @__PURE__ */ new Set();
     for (const score of scoreValue) {
       if (!(typeof score === "object" && !isEmpty(score))) {
         throw new Error(
           `When returning an array of scores, each score must be a non-empty object. Got: ${JSON.stringify(score)}`
         );
       }
+      if (typeof score.name !== "string") {
+        throw new Error(
+          `When returning an array of scores, each score must have a name. Got: ${JSON.stringify(score)}`
+        );
+      }
+      if (names.has(score.name)) {
+        throw new Error(`Duplicate score name '${score.name}' in score array`);
+      }
+      names.add(score.name);
     }
   }
   let results;
   if (Array.isArray(scoreValue)) {
     results = scoreValue;
   } else if (typeof scoreValue === "object" && !isEmpty(scoreValue)) {
-    results = [scoreValue];
+    results = [{ ...scoreValue, name: scoreValue.name ?? name }];
   } else {
     results = [{ name, score: scoreValue }];
   }

package/dist/instrumentation/index.js

@@ -3963,6 +3963,7 @@ var ProjectSettings = import_v36.z.union([
     ]),
     disable_realtime_queries: import_v36.z.union([import_v36.z.boolean(), import_v36.z.null()]),
     monitor_charts_use_metrics_start: import_v36.z.union([import_v36.z.boolean(), import_v36.z.null()]),
+    blind_reviews: import_v36.z.union([import_v36.z.boolean(), import_v36.z.null()]),
     default_preprocessor: NullableSavedFunctionId
   }).partial(),
   import_v36.z.null()
@@ -4024,6 +4025,7 @@ var WindowedAutomationConfig = import_v36.z.object({
     auto_approve_tools: import_v36.z.array(import_v36.z.string().min(1)).optional().default([]),
     harness: import_v36.z.enum(["native", "codex", "claude-code"]).optional(),
     model: import_v36.z.string().min(1).optional(),
+    endpoint_name: import_v36.z.string().min(1).optional(),
     reasoning_effort: import_v36.z.enum(["none", "minimal", "low", "medium", "high", "xhigh", "max"]).optional()
   }).optional(),
   actions: import_v36.z.array(
@@ -5815,7 +5817,7 @@ var INSTRUMENTATION_NAMES = {
 var INTERNAL_SPAN_INSTRUMENTATION_NAME = /* @__PURE__ */ Symbol.for(
   "braintrust.spanInstrumentationName"
 );
-var SDK_VERSION = true ? "3.30.0" : "0.0.0";
+var SDK_VERSION = true ? "3.31.0" : "0.0.0";
 function withSpanInstrumentationName(args, instrumentationName) {
   return {
     ...args,
@@ -18220,6 +18222,12 @@ function prepareAISDKChildTracing(params, self, parentSpan, denyOutputPaths, aiS
       if (!activeEntry) {
         return result;
       }
+      const executionOptions = args[1];
+      const toolCallId = isObject(executionOptions) ? executionOptions.toolCallId : void 0;
+      const toolInput = {
+        input: serializeToolExecutionInput(args),
+        ...typeof toolCallId === "string" ? { metadata: { toolCallId } } : {}
+      };
       if (isAsyncGenerator(result)) {
         return (async function* () {
           const span = activeEntry.parentSpan.startSpan(
@@ -18233,7 +18241,7 @@ function prepareAISDKChildTracing(params, self, parentSpan, denyOutputPaths, aiS
               INSTRUMENTATION_NAMES.AI_SDK
             )
           );
-          span.log({ input: serializeToolExecutionInput(args) });
+          span.log(toolInput);
           try {
             let lastValue;
             for await (const value of result) {
@@ -18251,7 +18259,7 @@ function prepareAISDKChildTracing(params, self, parentSpan, denyOutputPaths, aiS
       }
       return activeEntry.parentSpan.traced(
         async (span) => {
-          span.log({ input: serializeToolExecutionInput(args) });
+          span.log(toolInput);
           const awaitedResult = await result;
           span.log({ output: awaitedResult });
           return awaitedResult;
@@ -32373,14 +32381,17 @@ function getMetricsFromResponse(response) {
     if (!isRecord(usageMetadata)) {
       continue;
     }
-    const inputTokenDetails = usageMetadata.input_token_details;
+    const inputTokenDetails = isRecord(usageMetadata.input_token_details) ? usageMetadata.input_token_details : {};
     const outputTokenDetails = usageMetadata.output_token_details;
     return normalizeTokenMetrics({
       total_tokens: usageMetadata.total_tokens,
       prompt_tokens: usageMetadata.input_tokens,
       completion_tokens: usageMetadata.output_tokens,
-      prompt_cache_creation_tokens: isRecord(inputTokenDetails) ? inputTokenDetails.cache_creation : void 0,
-      prompt_cached_tokens: isRecord(inputTokenDetails) ? inputTokenDetails.cache_read : void 0,
+      // Prefer TTL-specific cache writes over the aggregate when available.
+      prompt_cache_creation_tokens: inputTokenDetails.ephemeral_5m_input_tokens == null && inputTokenDetails.ephemeral_1h_input_tokens == null ? inputTokenDetails.cache_creation : void 0,
+      prompt_cache_creation_5m_tokens: inputTokenDetails.ephemeral_5m_input_tokens,
+      prompt_cache_creation_1h_tokens: inputTokenDetails.ephemeral_1h_input_tokens,
+      prompt_cached_tokens: inputTokenDetails.cache_read,
       completion_reasoning_tokens: isRecord(outputTokenDetails) ? outputTokenDetails.reasoning : void 0
     });
   }

package/dist/instrumentation/index.mjs

@@ -3917,6 +3917,7 @@ var ProjectSettings = z6.union([
     ]),
     disable_realtime_queries: z6.union([z6.boolean(), z6.null()]),
     monitor_charts_use_metrics_start: z6.union([z6.boolean(), z6.null()]),
+    blind_reviews: z6.union([z6.boolean(), z6.null()]),
     default_preprocessor: NullableSavedFunctionId
   }).partial(),
   z6.null()
@@ -3978,6 +3979,7 @@ var WindowedAutomationConfig = z6.object({
     auto_approve_tools: z6.array(z6.string().min(1)).optional().default([]),
     harness: z6.enum(["native", "codex", "claude-code"]).optional(),
     model: z6.string().min(1).optional(),
+    endpoint_name: z6.string().min(1).optional(),
     reasoning_effort: z6.enum(["none", "minimal", "low", "medium", "high", "xhigh", "max"]).optional()
   }).optional(),
   actions: z6.array(
@@ -5771,7 +5773,7 @@ var INSTRUMENTATION_NAMES = {
 var INTERNAL_SPAN_INSTRUMENTATION_NAME = /* @__PURE__ */ Symbol.for(
   "braintrust.spanInstrumentationName"
 );
-var SDK_VERSION = true ? "3.30.0" : "0.0.0";
+var SDK_VERSION = true ? "3.31.0" : "0.0.0";
 function withSpanInstrumentationName(args, instrumentationName) {
   return {
     ...args,
@@ -18176,6 +18178,12 @@ function prepareAISDKChildTracing(params, self, parentSpan, denyOutputPaths, aiS
       if (!activeEntry) {
         return result;
       }
+      const executionOptions = args[1];
+      const toolCallId = isObject(executionOptions) ? executionOptions.toolCallId : void 0;
+      const toolInput = {
+        input: serializeToolExecutionInput(args),
+        ...typeof toolCallId === "string" ? { metadata: { toolCallId } } : {}
+      };
       if (isAsyncGenerator(result)) {
         return (async function* () {
           const span = activeEntry.parentSpan.startSpan(
@@ -18189,7 +18197,7 @@ function prepareAISDKChildTracing(params, self, parentSpan, denyOutputPaths, aiS
               INSTRUMENTATION_NAMES.AI_SDK
             )
           );
-          span.log({ input: serializeToolExecutionInput(args) });
+          span.log(toolInput);
           try {
             let lastValue;
             for await (const value of result) {
@@ -18207,7 +18215,7 @@ function prepareAISDKChildTracing(params, self, parentSpan, denyOutputPaths, aiS
       }
       return activeEntry.parentSpan.traced(
         async (span) => {
-          span.log({ input: serializeToolExecutionInput(args) });
+          span.log(toolInput);
           const awaitedResult = await result;
           span.log({ output: awaitedResult });
           return awaitedResult;
@@ -32329,14 +32337,17 @@ function getMetricsFromResponse(response) {
     if (!isRecord(usageMetadata)) {
       continue;
     }
-    const inputTokenDetails = usageMetadata.input_token_details;
+    const inputTokenDetails = isRecord(usageMetadata.input_token_details) ? usageMetadata.input_token_details : {};
     const outputTokenDetails = usageMetadata.output_token_details;
     return normalizeTokenMetrics({
       total_tokens: usageMetadata.total_tokens,
       prompt_tokens: usageMetadata.input_tokens,
       completion_tokens: usageMetadata.output_tokens,
-      prompt_cache_creation_tokens: isRecord(inputTokenDetails) ? inputTokenDetails.cache_creation : void 0,
-      prompt_cached_tokens: isRecord(inputTokenDetails) ? inputTokenDetails.cache_read : void 0,
+      // Prefer TTL-specific cache writes over the aggregate when available.
+      prompt_cache_creation_tokens: inputTokenDetails.ephemeral_5m_input_tokens == null && inputTokenDetails.ephemeral_1h_input_tokens == null ? inputTokenDetails.cache_creation : void 0,
+      prompt_cache_creation_5m_tokens: inputTokenDetails.ephemeral_5m_input_tokens,
+      prompt_cache_creation_1h_tokens: inputTokenDetails.ephemeral_1h_input_tokens,
+      prompt_cached_tokens: inputTokenDetails.cache_read,
       completion_reasoning_tokens: isRecord(outputTokenDetails) ? outputTokenDetails.reasoning : void 0
     });
   }

package/dist/vitest-evals-reporter.js

@@ -4,11 +4,11 @@
 
 
 
-var _chunkTWCDSYJNjs = require('./chunk-TWCDSYJN.js');
-require('./chunk-2XDW3UCG.js');
+var _chunkV32LW47Zjs = require('./chunk-V32LW47Z.js');
+require('./chunk-4AQGZS6X.js');
 
 // src/wrappers/vitest-evals/reporter.ts
-_chunkTWCDSYJNjs.configureNode.call(void 0, );
+_chunkV32LW47Zjs.configureNode.call(void 0, );
 var RESERVED_NORMALIZED_SPAN_ATTRIBUTE_KEYS = /* @__PURE__ */ new Set([
   "name",
   "type",
@@ -43,7 +43,7 @@ var BraintrustVitestEvalsReporter = class {
     for (const { test, meta } of evalTests) {
       logEvalTest(experiment, test, meta);
     }
-    await _chunkTWCDSYJNjs.summarizeAndFlush.call(void 0, experiment, {
+    await _chunkV32LW47Zjs.summarizeAndFlush.call(void 0, experiment, {
       displaySummary: this.options.displaySummary
     });
     this.experiment = void 0;
@@ -59,7 +59,7 @@ var BraintrustVitestEvalsReporter = class {
       );
     }
     const experimentName = _nullishCoalesce(this.options.experimentName, () => ( `vitest-evals-${(/* @__PURE__ */ new Date()).toISOString()}`));
-    this.experiment = _chunkTWCDSYJNjs.initExperiment.call(void 0, {
+    this.experiment = _chunkV32LW47Zjs.initExperiment.call(void 0, {
       ...projectId ? { projectId } : { project: projectName },
       experiment: experimentName,
       metadata: this.options.metadata,
@@ -274,7 +274,7 @@ function logToolCallSpans(rootSpan, calls) {
   }
 }
 function readEvalTaskMeta(input) {
-  if (!_chunkTWCDSYJNjs.isObject.call(void 0, input)) return void 0;
+  if (!_chunkV32LW47Zjs.isObject.call(void 0, input)) return void 0;
   const evalMeta = readEvalMeta(input.eval);
   const harnessMeta = readHarnessMeta(input.harness);
   if (!evalMeta && !harnessMeta) return void 0;
@@ -284,7 +284,7 @@ function readEvalTaskMeta(input) {
   };
 }
 function readEvalMeta(input) {
-  if (!_chunkTWCDSYJNjs.isObject.call(void 0, input)) return void 0;
+  if (!_chunkV32LW47Zjs.isObject.call(void 0, input)) return void 0;
   const avgScore = readFiniteOrNull(input.avgScore);
   const scores = Array.isArray(input.scores) ? input.scores.map(readEvalScore).filter(isDefined) : void 0;
   const toolCalls = Array.isArray(input.toolCalls) ? input.toolCalls.map(readToolCall).filter(isDefined) : void 0;
@@ -298,23 +298,23 @@ function readEvalMeta(input) {
   };
 }
 function readEvalScore(input) {
-  if (!_chunkTWCDSYJNjs.isObject.call(void 0, input)) return void 0;
+  if (!_chunkV32LW47Zjs.isObject.call(void 0, input)) return void 0;
   const score = readFiniteOrNull(input.score);
   return {
     ...typeof input.name === "string" ? { name: input.name } : {},
     ...score !== void 0 ? { score } : {},
-    ..._chunkTWCDSYJNjs.isObject.call(void 0, input.metadata) ? { metadata: input.metadata } : {}
+    ..._chunkV32LW47Zjs.isObject.call(void 0, input.metadata) ? { metadata: input.metadata } : {}
   };
 }
 function readHarnessMeta(input) {
-  if (!_chunkTWCDSYJNjs.isObject.call(void 0, input)) return void 0;
+  if (!_chunkV32LW47Zjs.isObject.call(void 0, input)) return void 0;
   return {
     ...typeof input.name === "string" ? { name: input.name } : {},
-    ..._chunkTWCDSYJNjs.isObject.call(void 0, input.run) ? { run: input.run } : {}
+    ..._chunkV32LW47Zjs.isObject.call(void 0, input.run) ? { run: input.run } : {}
   };
 }
 function readToolCall(input) {
-  if (!_chunkTWCDSYJNjs.isObject.call(void 0, input)) return void 0;
+  if (!_chunkV32LW47Zjs.isObject.call(void 0, input)) return void 0;
   return input;
 }
 function filteredNormalizedSpanAttributes(attributes) {
@@ -374,10 +374,10 @@ function epochSeconds(value) {
 }
 function logReporterError(span, error) {
   if (error instanceof Error) {
-    _chunkTWCDSYJNjs.logError.call(void 0, span, error);
+    _chunkV32LW47Zjs.logError.call(void 0, span, error);
     return;
   }
-  if (_chunkTWCDSYJNjs.isObject.call(void 0, error)) {
+  if (_chunkV32LW47Zjs.isObject.call(void 0, error)) {
     const message = typeof error.message === "string" ? error.message : void 0;
     const stack = typeof error.stack === "string" ? error.stack : void 0;
     if (message !== void 0 || stack !== void 0) {
@@ -389,10 +389,10 @@ ${stack}` : message
       return;
     }
   }
-  _chunkTWCDSYJNjs.logError.call(void 0, span, error);
+  _chunkV32LW47Zjs.logError.call(void 0, span, error);
 }
 function formatErrorMessage(error) {
-  if (_chunkTWCDSYJNjs.isObject.call(void 0, error)) {
+  if (_chunkV32LW47Zjs.isObject.call(void 0, error)) {
     if (typeof error.message === "string") return error.message;
     if (typeof error.stack === "string") return error.stack;
   }

package/dist/vitest-evals-reporter.mjs

@@ -4,8 +4,8 @@ import {
   isObject,
   logError,
   summarizeAndFlush
-} from "./chunk-BU6SM54N.mjs";
-import "./chunk-N3JKQ3D3.mjs";
+} from "./chunk-AJT4FR25.mjs";
+import "./chunk-CE3BLB2L.mjs";
 
 // src/wrappers/vitest-evals/reporter.ts
 configureNode();

package/dist/workerd.js

@@ -4031,6 +4031,7 @@ var ProjectSettings = import_v36.z.union([
     ]),
     disable_realtime_queries: import_v36.z.union([import_v36.z.boolean(), import_v36.z.null()]),
     monitor_charts_use_metrics_start: import_v36.z.union([import_v36.z.boolean(), import_v36.z.null()]),
+    blind_reviews: import_v36.z.union([import_v36.z.boolean(), import_v36.z.null()]),
     default_preprocessor: NullableSavedFunctionId
   }).partial(),
   import_v36.z.null()
@@ -4092,6 +4093,7 @@ var WindowedAutomationConfig = import_v36.z.object({
     auto_approve_tools: import_v36.z.array(import_v36.z.string().min(1)).optional().default([]),
     harness: import_v36.z.enum(["native", "codex", "claude-code"]).optional(),
     model: import_v36.z.string().min(1).optional(),
+    endpoint_name: import_v36.z.string().min(1).optional(),
     reasoning_effort: import_v36.z.enum(["none", "minimal", "low", "medium", "high", "xhigh", "max"]).optional()
   }).optional(),
   actions: import_v36.z.array(
@@ -5883,7 +5885,7 @@ var INSTRUMENTATION_NAMES = {
 var INTERNAL_SPAN_INSTRUMENTATION_NAME = /* @__PURE__ */ Symbol.for(
   "braintrust.spanInstrumentationName"
 );
-var SDK_VERSION = true ? "3.30.0" : "0.0.0";
+var SDK_VERSION = true ? "3.31.0" : "0.0.0";
 function withSpanInstrumentationName(args, instrumentationName) {
   return {
     ...args,
@@ -20034,6 +20036,12 @@ function prepareAISDKChildTracing(params, self, parentSpan, denyOutputPaths, aiS
       if (!activeEntry) {
         return result;
       }
+      const executionOptions = args[1];
+      const toolCallId = isObject(executionOptions) ? executionOptions.toolCallId : void 0;
+      const toolInput = {
+        input: serializeToolExecutionInput(args),
+        ...typeof toolCallId === "string" ? { metadata: { toolCallId } } : {}
+      };
       if (isAsyncGenerator(result)) {
         return (async function* () {
           const span = activeEntry.parentSpan.startSpan(
@@ -20047,7 +20055,7 @@ function prepareAISDKChildTracing(params, self, parentSpan, denyOutputPaths, aiS
               INSTRUMENTATION_NAMES.AI_SDK
             )
           );
-          span.log({ input: serializeToolExecutionInput(args) });
+          span.log(toolInput);
           try {
             let lastValue;
             for await (const value of result) {
@@ -20065,7 +20073,7 @@ function prepareAISDKChildTracing(params, self, parentSpan, denyOutputPaths, aiS
       }
       return activeEntry.parentSpan.traced(
         async (span) => {
-          span.log({ input: serializeToolExecutionInput(args) });
+          span.log(toolInput);
           const awaitedResult = await result;
           span.log({ output: awaitedResult });
           return awaitedResult;
@@ -34187,14 +34195,17 @@ function getMetricsFromResponse(response) {
     if (!isRecord(usageMetadata)) {
       continue;
     }
-    const inputTokenDetails = usageMetadata.input_token_details;
+    const inputTokenDetails = isRecord(usageMetadata.input_token_details) ? usageMetadata.input_token_details : {};
     const outputTokenDetails = usageMetadata.output_token_details;
     return normalizeTokenMetrics({
       total_tokens: usageMetadata.total_tokens,
       prompt_tokens: usageMetadata.input_tokens,
       completion_tokens: usageMetadata.output_tokens,
-      prompt_cache_creation_tokens: isRecord(inputTokenDetails) ? inputTokenDetails.cache_creation : void 0,
-      prompt_cached_tokens: isRecord(inputTokenDetails) ? inputTokenDetails.cache_read : void 0,
+      // Prefer TTL-specific cache writes over the aggregate when available.
+      prompt_cache_creation_tokens: inputTokenDetails.ephemeral_5m_input_tokens == null && inputTokenDetails.ephemeral_1h_input_tokens == null ? inputTokenDetails.cache_creation : void 0,
+      prompt_cache_creation_5m_tokens: inputTokenDetails.ephemeral_5m_input_tokens,
+      prompt_cache_creation_1h_tokens: inputTokenDetails.ephemeral_1h_input_tokens,
+      prompt_cached_tokens: inputTokenDetails.cache_read,
       completion_reasoning_tokens: isRecord(outputTokenDetails) ? outputTokenDetails.reasoning : void 0
     });
   }
@@ -45206,8 +45217,8 @@ function normalizeScores(result) {
       (s) => s !== null && s !== void 0 && typeof s === "object" && isScore(s)
     );
   }
-  if (typeof result === "object" && result !== null && isScore(result)) {
-    return [result];
+  if (typeof result === "object" && result !== null && "score" in result) {
+    return [{ ...result, name: result.name ?? "score" }];
   }
   return [];
 }
@@ -47790,19 +47801,29 @@ function buildSpanScores(results) {
 function _internalPrepareEvaluatorScore(scoreValue, name) {
   if (scoreValue === null) return { results: null };
   if (Array.isArray(scoreValue)) {
+    const names = /* @__PURE__ */ new Set();
     for (const score of scoreValue) {
       if (!(typeof score === "object" && !isEmpty2(score))) {
         throw new Error(
           `When returning an array of scores, each score must be a non-empty object. Got: ${JSON.stringify(score)}`
         );
       }
+      if (typeof score.name !== "string") {
+        throw new Error(
+          `When returning an array of scores, each score must have a name. Got: ${JSON.stringify(score)}`
+        );
+      }
+      if (names.has(score.name)) {
+        throw new Error(`Duplicate score name '${score.name}' in score array`);
+      }
+      names.add(score.name);
     }
   }
   let results;
   if (Array.isArray(scoreValue)) {
     results = scoreValue;
   } else if (typeof scoreValue === "object" && !isEmpty2(scoreValue)) {
-    results = [scoreValue];
+    results = [{ ...scoreValue, name: scoreValue.name ?? name }];
   } else {
     results = [{ name, score: scoreValue }];
   }

package/dist/workerd.mjs

@@ -3818,6 +3818,7 @@ var ProjectSettings = z6.union([
     ]),
     disable_realtime_queries: z6.union([z6.boolean(), z6.null()]),
     monitor_charts_use_metrics_start: z6.union([z6.boolean(), z6.null()]),
+    blind_reviews: z6.union([z6.boolean(), z6.null()]),
     default_preprocessor: NullableSavedFunctionId
   }).partial(),
   z6.null()
@@ -3879,6 +3880,7 @@ var WindowedAutomationConfig = z6.object({
     auto_approve_tools: z6.array(z6.string().min(1)).optional().default([]),
     harness: z6.enum(["native", "codex", "claude-code"]).optional(),
     model: z6.string().min(1).optional(),
+    endpoint_name: z6.string().min(1).optional(),
     reasoning_effort: z6.enum(["none", "minimal", "low", "medium", "high", "xhigh", "max"]).optional()
   }).optional(),
   actions: z6.array(
@@ -5672,7 +5674,7 @@ var INSTRUMENTATION_NAMES = {
 var INTERNAL_SPAN_INSTRUMENTATION_NAME = /* @__PURE__ */ Symbol.for(
   "braintrust.spanInstrumentationName"
 );
-var SDK_VERSION = true ? "3.30.0" : "0.0.0";
+var SDK_VERSION = true ? "3.31.0" : "0.0.0";
 function withSpanInstrumentationName(args, instrumentationName) {
   return {
     ...args,
@@ -19823,6 +19825,12 @@ function prepareAISDKChildTracing(params, self, parentSpan, denyOutputPaths, aiS
       if (!activeEntry) {
         return result;
       }
+      const executionOptions = args[1];
+      const toolCallId = isObject(executionOptions) ? executionOptions.toolCallId : void 0;
+      const toolInput = {
+        input: serializeToolExecutionInput(args),
+        ...typeof toolCallId === "string" ? { metadata: { toolCallId } } : {}
+      };
       if (isAsyncGenerator(result)) {
         return (async function* () {
           const span = activeEntry.parentSpan.startSpan(
@@ -19836,7 +19844,7 @@ function prepareAISDKChildTracing(params, self, parentSpan, denyOutputPaths, aiS
               INSTRUMENTATION_NAMES.AI_SDK
             )
           );
-          span.log({ input: serializeToolExecutionInput(args) });
+          span.log(toolInput);
           try {
             let lastValue;
             for await (const value of result) {
@@ -19854,7 +19862,7 @@ function prepareAISDKChildTracing(params, self, parentSpan, denyOutputPaths, aiS
       }
       return activeEntry.parentSpan.traced(
         async (span) => {
-          span.log({ input: serializeToolExecutionInput(args) });
+          span.log(toolInput);
           const awaitedResult = await result;
           span.log({ output: awaitedResult });
           return awaitedResult;
@@ -33976,14 +33984,17 @@ function getMetricsFromResponse(response) {
     if (!isRecord(usageMetadata)) {
       continue;
     }
-    const inputTokenDetails = usageMetadata.input_token_details;
+    const inputTokenDetails = isRecord(usageMetadata.input_token_details) ? usageMetadata.input_token_details : {};
     const outputTokenDetails = usageMetadata.output_token_details;
     return normalizeTokenMetrics({
       total_tokens: usageMetadata.total_tokens,
       prompt_tokens: usageMetadata.input_tokens,
       completion_tokens: usageMetadata.output_tokens,
-      prompt_cache_creation_tokens: isRecord(inputTokenDetails) ? inputTokenDetails.cache_creation : void 0,
-      prompt_cached_tokens: isRecord(inputTokenDetails) ? inputTokenDetails.cache_read : void 0,
+      // Prefer TTL-specific cache writes over the aggregate when available.
+      prompt_cache_creation_tokens: inputTokenDetails.ephemeral_5m_input_tokens == null && inputTokenDetails.ephemeral_1h_input_tokens == null ? inputTokenDetails.cache_creation : void 0,
+      prompt_cache_creation_5m_tokens: inputTokenDetails.ephemeral_5m_input_tokens,
+      prompt_cache_creation_1h_tokens: inputTokenDetails.ephemeral_1h_input_tokens,
+      prompt_cached_tokens: inputTokenDetails.cache_read,
       completion_reasoning_tokens: isRecord(outputTokenDetails) ? outputTokenDetails.reasoning : void 0
     });
   }
@@ -44995,8 +45006,8 @@ function normalizeScores(result) {
       (s) => s !== null && s !== void 0 && typeof s === "object" && isScore(s)
     );
   }
-  if (typeof result === "object" && result !== null && isScore(result)) {
-    return [result];
+  if (typeof result === "object" && result !== null && "score" in result) {
+    return [{ ...result, name: result.name ?? "score" }];
   }
   return [];
 }
@@ -47579,19 +47590,29 @@ function buildSpanScores(results) {
 function _internalPrepareEvaluatorScore(scoreValue, name) {
   if (scoreValue === null) return { results: null };
   if (Array.isArray(scoreValue)) {
+    const names = /* @__PURE__ */ new Set();
     for (const score of scoreValue) {
       if (!(typeof score === "object" && !isEmpty2(score))) {
         throw new Error(
           `When returning an array of scores, each score must be a non-empty object. Got: ${JSON.stringify(score)}`
         );
       }
+      if (typeof score.name !== "string") {
+        throw new Error(
+          `When returning an array of scores, each score must have a name. Got: ${JSON.stringify(score)}`
+        );
+      }
+      if (names.has(score.name)) {
+        throw new Error(`Duplicate score name '${score.name}' in score array`);
+      }
+      names.add(score.name);
     }
   }
   let results;
   if (Array.isArray(scoreValue)) {
     results = scoreValue;
   } else if (typeof scoreValue === "object" && !isEmpty2(scoreValue)) {
-    results = [scoreValue];
+    results = [{ ...scoreValue, name: scoreValue.name ?? name }];
   } else {
     results = [{ name, score: scoreValue }];
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "braintrust",
-  "version": "3.30.0",
+  "version": "3.31.0",
   "description": "SDK for integrating Braintrust",
   "repository": {
     "type": "git",