[S]ome imagined reactions are a better guide to real reactions than others. Imagined shape reactions are a good guide, you say, and you are probably right. But it is hard to see how the knowledge that they are a good guide could be a priori. If the mind’s eye sees one sort of property roughly as real eyes do, while its take on another sort of property tends to be off the mark, that is an empirical fact known on the basis of empirical evidence. I know not to trust my imagined reactions to arrangements of furniture, because they have often been wrong; now that I see the wardrobe in the room, I realize it is far too big. It is only because they have generally been right that I am entitled to trust my imagined judgments of shape. (458)
I think, however, that there is a conflation here between knowledge and knowledge of knowledge. It is a controversial thesis that, in order to know via the deliverances of an evidential source, one must know that source to be reliable. But even if that controversial thesis is true, it doesn't entail that in order for one to know a priori via the deliverance of some source, one must know a priori that that source is reliable; that would be a much stronger claim, and not obviously a plausible one.