Tag
BAER introduces a backbone-adaptive evidence routing method for robust pairwise LLM judging, achieving higher accuracy across benchmarks by dynamically selecting evidence mechanisms per condition.