"most of the time dang unflags them" is that mechanism.

If I found a reliable way to automate the recognition of bad-faith user interactions I would sell it to google and retire before bringing it to HN.

edit: Also, apparently, users can "vouch" for posts which presumably is just the inverse of flagging a post.

You can vouch for a comment, but not a post. (Except by emailing dang.)

Well, for example, deducting karma (or whatever) from people who flag stuff that dang unflags.