Why I Added a Validator to a Small JSONL Candidate Ranker

A text similarity score can look perfectly reasonable even when it's wrong. Suppose a script ranks a...