Runs a LangSmith offline experiment over datasetName. For each example (one chart-review
question) the target runs that question against the EVAL BRANCH clone (useEvalNamespaces + the
branch supabase client) and the evaluator scores the run's status against the example's expected
status.
The target reuses the real runChartReviewSingleTemplate pipeline (it writes an audit_run /
audit_response onto the disposable branch), so what's scored is exactly what production runs. The
evaluated "application version" is settings + metadata — stamp the git commit / prompt tag so
experiments stay comparable across changes.
Runs a LangSmith offline experiment over
datasetName. For each example (one chart-review question) the target runs that question against the EVAL BRANCH clone (useEvalNamespaces+ the branch supabase client) and the evaluator scores the run's status against the example's expected status.The target reuses the real
runChartReviewSingleTemplatepipeline (it writes an audit_run / audit_response onto the disposable branch), so what's scored is exactly what production runs. The evaluated "application version" issettings+metadata— stamp the git commit / prompt tag so experiments stay comparable across changes.