How do you avoid judging actual human discussions subjectively? I remember being a forum moderator and struggling with exactly the same issues. No matter what guidelines we'd set, there'd be essentially legitimate discussions that were way over the line superficially, and on the other you'd have neo-nazis acting in ways that weren't technically bad, but were clearly leading there.

Facebook moderators have an even harder job than that because the inherent scale of the platform prevents the kinds of personal insights and contextual understanding I had.