Good point. I shall have to think about this. Alternative web-of-trust, that doesn't have recursion issues. Originally intended for an up/downvote system, but can be tweaked for hubski. Like yours, this is per-user. For each user, you calculate a correlation coefficient for all other users. Simplest being Pearson's product-moment correlation, with upvote being 1, downvote being -1, and no vote being 0, but there are others. Then you multiply their vote by their correlation. Yes, this means that if someone tends to vote oppositely of you, their upvote may be counted as a downvote for you. In actuality, you don't want to have to keep n^2 correlation coefficients calculated. There are probably a bunch of other approximations that would hep reduce the load to a manageable level. I disagree with this. Putting moderation into the hands of the users is what makes hubski so great.