Convene an adversarial multi-model panel (Gemini/OpenAI/Opus, overridable) on an in-progress squint review — each model a distinct lens, cross-examining each other's findings, survivors fold in. Use for high-stakes PRs that warrant a deep second and third opinion.
