Epistemic defeat has traditionally been classified into opposing and undercutting types, and recent formal work has sought to capture this distinction within a Bayesian framework. However, we argue that purely probabilistic characterizations face a principled limitation: there exist cases of defeat that are intuitively distinct from both opposing and undercutting, yet cannot be adequately distinguished from them on probabilistic grounds alone. These cases elude existing probabilistic taxonomies not due to technical deficiencies, but rather to the fact that probabilistic relations abstract away from the causal structures that determine how evidence bears on hypotheses. To address this limitation, we shift the analysis from probabilistic relations to underlying causal mechanisms. Using structural causal models, we identify a distinct type of defeater—screening-off—and show that it cannot be adequately characterized in probabilistic terms alone. We provide necessary and sufficient causal conditions for opposing, undercutting, and screening-off defeaters in simple cases, thereby yielding a principled and exhaustive classification at the causal level. Finally, we suggest that even complex cases of epistemic defeat can be systematically understood as combinations of these basic causal patterns, offering a unified framework that both extends and clarifies existing Bayesian accounts.
Shigaku Nogami (Mon,) studied this question.