Now that I've mentioned the trust system a lot recently, there's been a fair amount of confusion and misinformation flying around, which I'm now going to try and quell:
users would list which people they really trust as being good people. (i.e., not assholes). you'd only do this if you're sure. if you just met somebody and they had a few neat posts or pretty pictures they posted, you shouldn't trust them. trust your closest friends ... or at least the polite ones. there are a number of people on my friends list I read and communicate with, but I wouldn't vouch for as being respectable netizens.
certifications (trust edges) would be private, as to not encourage false trust by pressure. (your pseudo-asshole friend bugging you until you mark them as trusted)
your public trust level is either "Trusted" or "Not yet trusted", not a numeric value. It's not a game. And the assumption is that everybody is a good person, and those that aren't soon will be, once they get some certs. It's important to choose the right wording. You can imagine some bad examples we won't use.
there will be no negative certifications.
How will this info be used? You'll have options like: "Who can comment?" Everybody, Registered users, Friends, or Trusted users? There will not be options like "trusted people of people I trust". That's too computationally intensive.
somebody mentioned they thought the advogato algorithm wasn't deterministic. i believe it is, but i'm not an expert at this stuff.