I develop a broadly reliabilist theory of non-ideal epistemic rationality and argue that if it is correct we should reject the recently popular idea that the standards of non-ideal epistemic rationality are mere social conventions.