I think the article actually agrees with you, that their toxitiy is hostile to outsiders. "This entails, for example, that even if toxicity does not seem to ma

I think the article actually agrees with you, that their toxitiy is hostile to outsiders. "This entails, for example, that even if toxicity does not seem to make people leave conversations, it could still be a factor that discourages them from joining them." They also agree that the effects of the moderation impact the data set, and this is mentioned. That's part of why they tried to include many different sites which may have different policies and even so it's cited as a potential problem. What, specifically, from the conclusions do you think is wrong? I also think you are misunderstanding what the criteria is. It is not comments that cause someone to leave in that specific instance (something there's not data for anyways), but comments that are declared toxic according to a ML model where the idea is to find comments likely to repel others due to being rude or unreasoanable, regardless of the actual effect. This is kinda squishy and objectionable too, but not in the way that you're saying so I figured it was worth clarifying