Evaluating the Reliability and Fidelity of Automated Judgment Systems of Large Language Models