AdversarialDebate

MIT-licensed open-source tool that runs two language models over the same artifact in isolation, requires each to commit a verdict with cited evidence, then opens a bounded debate between them and reports unresolved disagreement.