LLM-as-judge

Your code graders from the last lesson confirm the enquiries are well-formed, grounded and complete against the planted issues. They can't tell you whether an enquiry is necessary. "Please confirm whether there have been any disputes or complaints with neighbours" is a perfectly grounded, schema-valid enquiry, and it's useless if the seller has already answered "No" on the TA6 and nothing else in the pack suggests otherwise. Deciding that requires reading two documents and judging whether one answers the other. That is the job of an LLM judge: a model, a prompt and a rubric that turn a judgement call into a score you can track.