Fix input/output getting misrouted by mixed usage-schema keys

The previous cached-tokens commit added cache_read_input_tokens (a
flat-schema key) alongside prompt_tokens/completion_tokens (OpenAI's
native schema) in the same usageDetails object. Langfuse's ingestion
explicitly does not support mixing those two schemas, and doing so
broke generation.update()'s output field - confirmed live: prompt/
response content was landing in the Usage panel instead of Input/
Output.

Fixed by staying entirely within OpenAI's native schema: cached
tokens now nest under prompt_tokens_details.cached_tokens instead of
using the flat cache_read_input_tokens key. Cast past the installed
@langfuse/tracing SDK's usageDetails type (which only exposes a
3-field camelCase shape with no _details nesting, unlike the wire
schema in @langfuse/core) since the nesting is valid on the wire even
though the SDK's public type doesn't model it.

Verified: full suite (82/82) passing, tsc/eslint clean.

src/core/langfuse/LangfuseTracer.ts

@@ -11,27 +11,35 @@ import type {
   TraceHandle,
 } from '../llm/Tracer';
 
+interface OpenAiUsageDetails {
+  prompt_tokens?: number;
+  completion_tokens?: number;
+  total_tokens?: number;
+  prompt_tokens_details?: { cached_tokens: number };
+}
+
 /**
- * Langfuse only recognizes usage in a flat schema (input/output/...) or
- * OpenAI's native snake_case schema (prompt_tokens/completion_tokens/...);
- * our own ChatResult.usage uses camelCase for our internal contract, so it
- * needs translating at this boundary or Langfuse silently shows zero usage.
+ * Langfuse recognizes usage in either a flat generic schema (input/output/
+ * cache_read_input_tokens/...) or OpenAI's native nested schema
+ * (prompt_tokens/completion_tokens/prompt_tokens_details.cached_tokens/...)
+ * - but NOT a mix of the two ("do not mix formats" per Langfuse's docs).
+ * Mixing them previously caused the whole generation.update() payload
+ * (including output) to land wrong, so this stays entirely within the
+ * OpenAI-native schema.
  *
- * Cached tokens use the flat schema's cache_read_input_tokens key rather
- * than OpenAI's nested prompt_tokens_details.cached_tokens: the installed
- * @langfuse/tracing SDK's usageDetails type is `Record<string, number> |
- * OpenAiUsage` (a 3-field camelCase shape with no _details nesting) even
- * though the wire schema (@langfuse/core's OpenAiCompletionUsageSchema)
- * supports the nested form — keeping everything as flat numbers avoids
- * fighting that gap in the SDK's public types.
+ * The installed @langfuse/tracing SDK's usageDetails type is
+ * `Record<string, number> | OpenAiUsage` (a 3-field camelCase shape with
+ * no _details nesting) even though the wire schema (@langfuse/core's
+ * OpenAiCompletionUsageSchema) supports this nested form - a gap in the
+ * SDK's public types, not a real runtime restriction, hence the cast.
  */
-function toUsageDetails(usage: GenerationOutput['usage']): Record<string, number> | undefined {
+function toUsageDetails(usage: GenerationOutput['usage']): OpenAiUsageDetails | undefined {
   if (!usage) return undefined;
-  const details: Record<string, number> = {};
+  const details: OpenAiUsageDetails = {};
   if (usage.promptTokens !== undefined) details.prompt_tokens = usage.promptTokens;
   if (usage.completionTokens !== undefined) details.completion_tokens = usage.completionTokens;
   if (usage.totalTokens !== undefined) details.total_tokens = usage.totalTokens;
-  if (usage.cachedTokens !== undefined) details.cache_read_input_tokens = usage.cachedTokens;
+  if (usage.cachedTokens !== undefined) details.prompt_tokens_details = { cached_tokens: usage.cachedTokens };
   return Object.keys(details).length > 0 ? details : undefined;
 }
 
@@ -64,7 +72,7 @@ export class LangfuseTracer extends Tracer {
           end(output: GenerationOutput) {
             generation.update({
               output: output.output,
-              usageDetails: toUsageDetails(output.usage),
+              usageDetails: toUsageDetails(output.usage) as Record<string, number> | undefined,
               metadata: output.metadata,
             });
             generation.end();

test/core/langfuse/LangfuseTracer.test.ts

@@ -78,7 +78,7 @@ describe('LangfuseTracer', () => {
     expect(generationEnd).toHaveBeenCalledTimes(1);
   });
 
-  it('maps cachedTokens to the flat cache_read_input_tokens key', () => {
+  it('maps cachedTokens to the OpenAI-native nested prompt_tokens_details.cached_tokens, not a mixed flat key', () => {
     const tracer = new LangfuseTracer({ publicKey: 'pk', secretKey: 'sk', baseUrl: 'http://lf' });
     const handle = tracer.startTrace('my-pipeline');
     const gen = handle.startGeneration({ name: 'step-1', model: 'gpt-4o-mini', input: ['hi'] });
@@ -93,7 +93,7 @@ describe('LangfuseTracer', () => {
           prompt_tokens: 100,
           completion_tokens: 20,
           total_tokens: 120,
-          cache_read_input_tokens: 80,
+          prompt_tokens_details: { cached_tokens: 80 },
         },
       })
     );