Quantifying the plausibility gap in generative AI for surgical video generation with expert assessment

Generative artificial intelligence (AI) shows promise for physical simulation, yet its clinical application remains critically unexplored. We present SurgVeo, a pioneering expert-guided benchmark paired with the Surgical Plausibility Pyramid framework. Evaluating the Veo-3 and Wan2.2 models reveals a consistent plausibility gap: despite high visual fidelity, the majority of errors are surgical logic failures rather than visual quality deficits. By exposing this reliance on visual synthesis, this study provides a roadmap for developing surgical world models.