What Rubber Duck establishes institutionally is that GitHub now treats model-family diversity as a prerequisite for code review integrity, not an optimization. The cross-family review agent now supporting more models formalizes a principle that individual model providers have been reluctant to state publicly: a GPT-family model reviewing GPT-family output inherits the same blind spots. Routing review through an adversarial family is the closest available substitute for human judgment at scale. The catch is that it also routes liability through a second autonomous decision-maker — and code ownership in AI-generated work…