So, there are two basic approaches you can make to this:

1. Delegate moderation powers only to a select group of individuals who know and will uphold the site's standards. Effectively: and editorial board.

2. Allow all users to moderate. But score each on how well the result of their moderation adheres to a specified goal -- that is, for a given submission, was it 1) appropriate to the site and 2) did it create high-level engagement? Users might correlate positively or negatively, strongly or weakly. That is: some people will vote up content that's not desirable, and downvote content that is. Others simply can't tell ass from teakettle. In the first case, you simply reverse the sign, in the second, you set a low correlation value. And of course, those who are good and accurate predictors get a higher correlation value.

With the 2nd approach, everyone's "vote" counts, though it may not be as they expected. You've also got to re-normalize moderation against a target goal.

It's more computationally intensive, but I think it might actually be a better way of doing things.

  > ... hn's built in control of making users
  > earn a bunch of karma before letting them
  > downvote.

Since all of this is talking about the classification of submissions, this is irrelevant, because you can't downvote submissions, only comments.

At least, I don't yet have enough karma to downvote submissions.

Submissions can be flagged.

I'd argue they should be downvotable as well, though you're right, they're not.

Incidentally, comments can also be flagged (on the comment link view only, not in the forum view).