How are teams handling evals when agent pipelines span multiple LangSmith projects?

What I have found really useful is the following strategy:

For tracing projects, we only use:

  1. Production
  2. Staging

However, to divide the individual traces with an isolated unit for Team or Project, we are using metadata tags, e.g., “project_name:xyz”.
We do this as this allows for easy filtering and allows full visibility across different team/projects (but be mindful I am part of a small startup, so in a big organisation this might be different).

For my other Mental Model about evaluation, I have already given a comprehensive answer in this thread