It seems to me that there is an epistemological asymmetry when relying on the good or bad actions of others. Karla has a sterling character, and is an expert on the relevant topic, and he says p. I conclude that p. Patrick has a horrible character, and is an expert on the relevant topic, and it would clearly be in his interest to lie about this to me, and he says q. I conclude that not-q. The two conclusions may have extremely high probability based on the evidence about Karla's and Patrick's past behavior, but it is plausible to me that only the inference in Karla's case gives knowledge. Why? Because we have a right to count on someone's good character but have no right to count on someone's bad character, even if the probabilities of acting out of character are equally small.
Similarly, given past behavior, I think sometimes we are justified in believing that a particular person will do the right thing in some circumstances, not just that she will very likely do the right thing. But we are never justified in believing of a person that she will do the wrong thing in some circumstances, though we may be justified in believing that she will very likely do the wrong thing. It would be uncharitable to do so--she might reasonably complain that we shouldn't have formed such a belief, given her free will.
One might think the second case, and maybe the first, confuses moral justification with epistemic justification. But morality governs every aspect of our lives (this is clearest from a Jewish or Christian standpoint--we are to love God with every aspect of our lives), and in particular governs our believings.