A classical detector and several language models scan the same code for duplicate methods. A judge model weighs their answers. You see where they agree, disagree, and where the models catch clones the baseline misses.