Let's say you see some spam. You could have the software tell you which part of the web has made the spam trusted, and then you could manually mark that part of the web as untrusted. If there are only a few breaks in the chain like that, it'd be a workable solution.