Justification and Explanatory Normalcy

In his very interesting 'Justification, Normalcy and Evidential Probability', Martin Smith argues that justification requires one's belief to be normally true, rather than just very likely true, given one's evidence.  The relevant sense of 'normalcy' is explanatory rather than statistical: for the belief to turn out false would call out for explanation.  Beliefs based on perception, for example, are generally "normically supported" in this way -- for the belief to be false, something weird must have happened (perhaps the agent hallucinated, or was tricked somehow -- there must be some further explanation).  By way of contrast, the belief that your lottery ticket is going to lose might be as likely as can be (stretch the odds of the lottery as much as you like), still it would require no special explanation were this ticket to turn out to be the winning one -- some or other ticket must win, after all, it's just a matter of random chance.

It's an interesting proposal, and seems to capture common usage pretty nicely, but I wonder about the normative significance of normic support.  It seems to me that we are better off, epistemically speaking, with beliefs that are very likely true than we are with beliefs that are normally true (given our evidence).  On p.18, Smith offers the following:

If one believes that a proposition P is true, based upon evidence that normically supports it then, while one’s belief is not assured to be true, this much is assured: If one’s belief turns out to be false, then the error has to be explicable in terms of disobliging environmental conditions, deceit, cognitive or perceptual malfunction or some other interfering factor. In short, the error must be attributable to mitigating circumstances and thus excusable, after a fashion. Errors that do not fall into this category are naturally regarded as errors for which one must bear full responsibility – errors for which there is no excuse. And if error could not be excused, then belief cannot be permitted.

But if I'm making a high-stakes decision, I would (I hope!) prefer any mistake on my part to be unlikely rather than excusable.  We should want to get things right, and not to merely offload responsibility onto "disobliging environmental conditions".  And the best, most reliable way to get things right is to follow the probabilities, rather than rely on cooperative environmental conditions by taking perceptual evidence (and the like) at face value.

A background point: I'm a little unsure about the significance of so-called "all-out belief", as opposed to credence (or degrees of belief).  So, rather than claiming that Smith is mistaken about what justifies all-out beliefs, I might instead say that we shouldn't be interested in them at all.  In an uncertain world, rational decisions should be informed by our credences, not our beliefs.  That would be another way to express my main point.  But however we say it, the crucial point is just that likelihood, rather than normalcy, seems to be what really matters, epistemically speaking.

The practical importance of this comes out in Smith's example of privileging (notoriously unreliable) eye-witness testimony over statistical evidence.  His account captures standard practices very well, but this seems like an issue that calls out for a more revisionist approach.  If we care about getting accurate verdicts, we should want to reform the legal system (and others like it) to rely less on eye-witness accounts, and more on statistical evidence that has a higher probability of seeing us right.

What do you think?