QuestionQ136

VerifiedSolution Design

You are testing an agent that condenses lengthy legal documents into key bullet points. Because multiple summaries may be valid and there is no single correct answer, which evaluation method should be used?

Explanation

Model-graded evaluation can judge open-ended summaries against semantic quality criteria, including coverage of important points, coherence, and faithfulness to the source. Exact string metrics or comparison with a single reference summary would incorrectly penalize valid alternative phrasings and structures.

Community Discussion

No comments yet. Be the first to start the discussion!