Comparison
Exact match and rule-based scoringvsLLM-as-judge
Exact match and rule-based scoring
the answer is either the right invoice number or it is not, so you assert on it instead of paying a judge model to have an opinion.
Scoring by string equality, regex, numeric tolerance or schema checks. Cheap, deterministic and ideal for extraction, classification and code that either runs or does not. Use it wherever the task permits before reaching for a judge.
Full entry →LLM-as-judge
you have a second model score the first one's output against a rubric, because there is no exact answer to diff against.
Using a strong model to score outputs against a rubric when correctness is fuzzy: answer quality, faithfulness to sources, tone. Powerful and biased; calibrate the judge against a sample of human labels before trusting its numbers, and keep the judge model pinned.
Full entry →