Thanks, it's interesting to learn what is a practical problem when growing a discussion website.
Is it possible to deter troll behavior by using a reputation system and investigating bad behavior? Reputation that takes a long time of good behavior to build but a single act of malice to destroy.
It also has the statistical beauty of adding a prior probability estimate to the likelihood estimate that sorts Reddit comments.