Tag
This paper presents a benchmark (DNBench) for evaluating LLMs in database normalization tasks and introduces a multi-agent reasoning framework (MARS) that improves accuracy by 82% over single-prompt methods.