Separate fixing with model logic from post-processing the model's responses for reporting #5
Description
Either provide a way to re-read previous runs and validate the results to check fix status or create a pipeline from different steps where we can run only post-processing.
Use case: we discover a new way that the model responds to a query and we want to account for this structure when processing the response. Right now, we'd have to re-run everything (including querying), which doesn't make sense.