I had to ctrl+f to navigate enough to find it but they mention that their method for identifying toxic comments is under the methods and materials section. Its

I had to ctrl+f to navigate enough to find it but they mention that their method for identifying toxic comments is under the methods and materials section. Its just some kind of trained ML model. In that section they link to a page (Id share it directly but it clearly has some annoying "heres your temporary access key" query parameters that would probably cause the link to break super fast) that gives examples of how it ranks different kinds of comments, and I think that list of examples is enough to trust that it does at least well enough at classifying to draw at least light conclusions from.