Evaluation playground

Run a complete evaluation against a sample dataset, inspect case-level evidence, then try the same workflow with your own output.