LLM As Judge

How does a chain-of-thought judge differ from a direct-score judge, and when is each better?

What is the primary advantage of chain-of-thought (CoT) judge prompts over direct-score prompts?

Explain the structural reason BLEU fails for open-ended LLM evaluation

How can an agent confidently produce a wrong answer without raising any runtime exception?

Why is scoring only the final answer a weak way to evaluate agents?

Which of these are valid concerns when using LLM-as-judge for evaluation?