This is pretty inchoate, but I was just recently struck by the possibility of an interesting puzzle here.
One the one hand, we clearly need to include as evidence for clearly justified occurrent beliefs propositions stored in memory.
On the other, we clearly sometimes are justified in believing p on the basis of occurrent evidence r *right up until* we remember that, say, we know s and that s entails ~p (or maybe we just remember ~p).
The latter requires, it seems, that not just any proposition stored in memory is included in the evidence for just any instance of occurent belief. I’ve hardly thought about a solution, but it does seem like a potentially interesting puzzle.