Frege believed that the unrestricted comprehension axiom is true, and it is sad, since the axiom leads to paradox. If you are inclined toward coherentism, the rationality of Frege’s belief causes a problem, since it is logically inconsistent.
I’ve been working on the problem for coherentism of justified inconsistent beliefs, and this is one version of the problem. I’m tempted, though, to think it isn’t an epistemological problem, but rather a philosophy of language issue. Here’s why.
The problem concerns the problem of cognitive significance, which standard propositional theories don’t solve. They don’t, that is, give us a solution to Frege’s puzzle of how a=a can be trivial whereas a=b is substantive. The same reasons that lead Frege to appeal to sense in addition to reference should cause us to appeal to modes of presentation in addition to propositional content. This is essentially the line taken by triadic theories of intentional attitudes of the sort articulated by Nathan Salmon.
What would be helpful here is if some of the work on 2D semantics could help make more sense of the mysterious notion of a mode of presentation, but here I’m skeptical. Even the most epistemologically sensitive versions of 2D semantics focuses too much on what is knowable a priori, and such a focus won’t help with the problem of cognitive significance. To see this, notice that the information involved in apriority is information that need not be present in the noetic structure of the person in question. It may be detectable by reflection alone, but it can have that feature without being present in the noetic structure itself. So it will be more akin to the external defeaters constitutive of Gettier situations than to internal defeaters of the sort that undermine justification itself. To the extent that this diagnosis is correct, the 2D work won’t help solve the problem of Frege’s sad belief.