Nice, I think it makes sense to me. So you'd provide it for example with a csv with prompt and expected completion, and then a test file, and then you can run

Nice, I think it makes sense to me. So you'd provide it for example with a csv with prompt and expected completion, and then a test file, and then you can run that test file and see how it does on those sample prompts.