No confidence whatsoever in the following argument, but I think it's verrrrrrry interesting and some of the crucial assumptions seem rather plausible. Run it up the flagpole...
An argument that we ought to deny that by Φ-ing S does all the reasons require iff S rationally Φ’s. (I'm assuming that the conditions that determine rationality cannot go beyond those that determine what S's perspective is like.) It seems that reasons figure in reasoning and do so (often) because they are considerations that seem to favor a prospective course of action. These considerations are represented by the beliefs and are not considerations about the beliefs that provide premises for the purposes of practical deliberation. Suppose S is systematically deceived about the nature of his surroundings. It is possible to be fully rational in deciding to Φ under these circumstances and possible to be irrational in deciding to pursue some other course of action, say Ψ-ing. It is possible that on these occasions the subject does not respond correctly to the considerations the subject would identify as those that make Φ-ing favorable. But, on such an occasion we have already said that something explains why S’s decision to Φ would be rational and S’s decision to do Ψ irrational. That which makes Φ-ing rational and Ψ-ing irrational would be making demands of S, so they would be guiding reasons. Those would be beliefs or S’s or facts about those beliefs. But, S does not take his beliefs or facts about those beliefs to be considerations that favor Φ-ing so either S does not have a sense of what it would take for Φ-ing to be favorable or reasons do not figure in reasoning even when that reasoning culminates in the subject’s rationally deciding to Φ.
It is true that we sometimes reason from facts about beliefs instead of first-order beliefs about matters external to us. It seems clear that this reasoning is not the sort of reasoning we typically engage in because the rational requirements one must conform to in order for that reasoning to resolve rationally is different from the rational requirements that we must conform to in order for more familiar patterns of reasoning to resolve rationally. To see this, consider an example. If allowed to reason straight, as it were, from the premise that p, knowledge that q entails p is all I need to infer q and do so rationally (assuming of course that I arrive at this conclusion by means of competent deduction and lack reasons for denying q). If forced to reason from the fact that I believe p, I can still reason to the conclusion that q given ancillary premises that assure me it is unlikely that I would be mistaken about p such as the assumption that the fact that p is part of the best explanation as to why I would believe p. Depending on how strong this assurance is, I might rationally judge that q, but were I to deduce it from my starting premise that would be irrational.
The kinds of reasoning must be different because the rational requirements must be different. I would now like to argue that there is something oddly self-defeating about the view under consideration. Suppose we stick with the view that a reason’s demands cannot demand more than that we conclude deliberation in ways that are rational and accept the further view that reasons are not external considerations some might take to favor an action or attitude. The wages of such a view is scepticism. For the reasoning that reasons from facts about us, say, is reasoning governed by rational requirements that do not allow us to treat claims about the external surroundings as reasons. So, if p is a proposition about the external world, we are rationally compelled to reason from the fact that we believe p rather than the fact that p. Note, however, that knowledge of p’s truth is sufficient for p’s inclusion in practical and theoretical deliberation. So, if nothing suffices to warrant p’s inclusion in practical and theoretical deliberation, we do not know p. Knowing that we do not know p, we cannot rationally reason from the fact that we believe p because one cannot rationally persist in the belief that p whilst acknowledging that one does not know p. I’m not sure how much further we have to retreat, but I think we have already retreated too far.