A separate AI system that evaluates another system's decisions and outputs a confidence or correctness score.