I show that accepting Moss’s (Philos Rev 122:1–43, 2013) claim that features of a rational agent’s credence function can constitute knowledge, together with the claim (put forward by several philosophers) that a rational agent should only act on the basis of reasons that he knows, predicts and explains evidential decision theory’s failure to recommend the right choice for the Newcomb problem. The Newcomb problem can be seen, in light of Moss’s suggestion, as a manifestation of a Gettier case in the domain of choice. This serves as strong evidence for both Moss’s claim and the knowledge-based action approach.