Langsmith experiment trace as empty Lambda

LangSmith Product Help Observability & Evals
1

Hi, I’m creating experiments on datasets using the langgraph studio sdk and it creates experiments but these are the traces from it. what’s going wrong that this is happening?
I have all of these env variables set:

LANGCHAIN_TRACING_V2: ‘true’,

LANGCHAIN_API_KEY

LANGCHAIN_ENDPOINT: eu.api….

LANGCHAIN_PROJECT: `experiment-${launchConfig.experimentId}`,

LANGSMITH_TRACING: ‘true’,

LANGSMITH_API_KEY,

LANGSMITH_ENDPOINT

image
image1462×426 25.7 KB

2

Hi @hesamzkr , so its quite hard to determine what is wrong without code.

You’re likely seeing two different things:

  • is just the run name because your experiment target is an anonymous function (which is expected).
  • The trace looks “empty” when nested calls are not traced/attached (often from not forwarding config in RunnableLambda/custom wrappers).

Here are a quick fixes:

  1. Give the target a named function (or explicit run name) instead of anonymous lambda
  2. In wrapper functions, accept config and pass it to every inner .invoke/.ainvoke call
  3. Use 1. LANGSMITH_* env vars consistently (endpoint/key/project/tracing), and avoid mixing old LANGCHAIN_*

If you share your SDK versions and some code snippet, I could point you to the exact line causing it.