Most AI Second Opinions Are Theater. I Built a System That Actually Fights Back.

The first time I ran two LLMs against the same pull request, 89% of their "debate" was fake. Not wrong. Not low-quality. Fake. The second model was replaying pre-generated text. Both models were excha