How about creating rules? A talks to B B blocks A A registers as C *not too much time passes* C talks to B At this point C is considered to be a har

How about creating rules? A talks to B B blocks A A registers as C *not too much time passes* C talks to B At this point C is considered to be a harasser and therefore it will be hidden for B. B can of course lift the ban if it's just a false positive. To further specify who's who in this graph, the logic can find the smallest time between ban and the creation of a new user with the same behavior (writes to B) as the reported one. I think this is way more easier than handling those who abuse the actual service you're selling.