Adversarial collaboration has been championed as the gold standard for resolving scientific disputes but has gained relatively limited traction in neuroscience and allied fields. In this perspective, we argue that adversarial collaborative research has been stymied by an overly restrictive concern with the falsification of scientific theories. We advocate instead for a more expansive view that frames adversarial collaboration in terms of Bayesian belief updating, model comparison, and evidence accumulation. This framework broadens the scope of adversarial collaboration to accommodate a wide range of informative (but not necessarily definitive) studies while affording the requisite formal tools to guide experimental design and data analysis in the adversarial setting. We provide worked examples that demonstrate how these tools can be deployed to score theoretical models in terms of a common metric of evidence, thereby furnishing a means of tracking the amount of empirical support garnered by competing theories over time. Corcoran et al. present a Bayesian treatment of adversarial collaboration that operationalizes competing theoretical hypotheses as generative models. This approach enables the evaluation of alternative theories in terms of the evidential support their models garner across (potentially disparate) experimental settings.