It's clear that any normative facts must supervene on the descriptive facts. And I've previously argued that this connection must be conceptual rather than synthetic, a priori rather than a posteriori. But then we seem to end up with mere "descriptivism", according to which normative claims are just shorthand for certain descriptive claims. Their evaluative aspect seems to have been eliminated in the reduction process.
I think Kit Fine states the problem especially well in his excellent article (which I keep returning to!), 'The Varieties of Necessity' (Modality and Tense, p.251):
If there is a correct analysis of good, say, as what promotes pleasure over pain, then something's being good must consist in nothing more than its promoting pleasure over pain. But we have a strong intuition that it does consist in something more. Here we are not relying on the purported epistemic status of a correct analysis, as Moore does, but on its metaphysical consequences.
This argument, moreover, can be strengthened. For suppose one merely takes it to be a conceptual necessity that something is good if it promotes pleasure over pain. Now, if this is true, then presumably it must also be true that something is good in virtue of promoting pleasure over pain. Indeed, it is only because something is good in virtue of promoting pleasure over pain that there is the conceptual connection between the one and the other. But now what is this in-virtue-of relationship that accounts for the conceptual connection? The only possible answer, it seems, is that it is the relationship of one thing consisting in no more than some other; for this would appear to be the only in-virtue-of relationship capable of sustaining a conceptual connection. But if this is right, then the argument can also be taken to apply to statements of conceptual implication, and not merely to analyses.
Tricky! I'd replace the hedonism with a welfarist account of the good. But even then, aren't we saying two different things when we call something either "good" or "conducive to general welfare"? The former seems to come with an evaluative force that the latter, merely "descriptive" phrase lacks.
If there is a conceptual connection between ethics and ideal rationality -- as seems plausible -- then perhaps this could provide the requisite normative force. Supervenience would then arise because rationality requires treating like cases alike (and identical cases identically). Nevertheless, it is a substantive ('synthetic'?) fact just which values would survive ideal rational reflection. It is still a priori in the sense that an ideally rational agent wouldn't need to know which world he's in in order to come to the right conclusions. But this doesn't seem to be a merely analytic fact about the meaning of the word "rational". So perhaps there can be synthetic a priori truths along these lines. (If so, I should retract my earlier opposition to synthetic ethical naturalism. All my arguments show is that it's not a posteriori. If a priori synthetic statements are possible, then that's fine.)
If we accept this kind of "constructivist non-cognitivism", then it seems we are able to find room for supervening ethics within a naturalistic framework. An act's "rightness" holds in virtue of its promoting welfare, but it does not merely consist in this descriptive fact. Rather, it consists in its being the ideally rational action.