What we can do rationally is what we can do by means of a reasoning procedure, and our reasoning procedures can only operate on our actual representations. Only when a potential action is (at least minimally) rationalized by their beliefs can agents perform it rationally. If I do not represent the state of affairs at which an action is aimed as desirable, or as furthering the satisfaction of my goals or ends, then I cannot settle on that action by means of a reasoning procedure; if I perform that action, I will do so only as a consequence of a failure of my reasoning.
But this neglects the obvious possibility that the agent could revise their unjustified beliefs and desires. Ex hypothesi, Potter's false normative beliefs are unreasonable. So there is available evidence that could - and rationally should - move Potter to form more reasonable normative beliefs instead. These in turn should rationally move Potter to act less selfishly. Where in this picture is the alleged "failure of... reasoning"? Potter is now reasoning (and responding to evidence) better than ever before!
Levy seems to be making the mistake of confusing internalism with subjectivism. It's true that rational assessment must be 'internal' in a sense: it depends on what evidence and information the agent possesses, rather than just on what's objective true. But, as my linked post explains, that's a far cry from saying that an agent is rational whenever they believe themselves to be so (or that such self-satisfaction entails that they rationally shouldn't revise their attitudes).
See also: 'Rational Akrasia' and On Rosen's Moral Responsibility Skepticism.