Superadmin-only: launch a LangSmith offline experiment over the eval dataset — each question is
run against the eval-branch clone and scored against its graded ground truth. Gated behind the
global enableEvalRuns flag (environment-targeted, not per-entity).
Superadmin-only: launch a LangSmith offline experiment over the eval dataset — each question is run against the eval-branch clone and scored against its graded ground truth. Gated behind the global
enableEvalRunsflag (environment-targeted, not per-entity).