> A far better option is to rely entirely on a transparent algorithmic model, in which automated tools like lists of trigger keywords and contextual analysis th

> A far better option is to rely entirely on a transparent algorithmic model, in which automated tools like lists of trigger keywords and contextual analysis that cause posts to be flagged for further review are clearly visible to the user audience. This is the worst idea I've read on HN and shows you've never actually dealt with users at scale. The moment your algorithm is visible, users will know how to beat it. Your list of 'trigger keywords' becomes a weapon both to harass normal users (by potentially tricking them into writing the trigger weapons) as well as by trolls because they know exactly how to modify the word to get around the filter. Social media is no different from a game, and when people figure out how a game works, they learn how to break it.