This strikes me as a similar technique to trying to jailbreak an LLM to answer your questions without all the safety neutering -- you're trying to convince the judges to eval your new rules of debate, and then interpret your utterances in that new context.