Standalone

Run selected prompts on a schema, then score/rescore against their rubrics with a judge model. Ephemeral — nothing is browsed here later.