The thing about political discussion—the identifying feature of it, which is the same thing that makes it such a problem—is that it sucks people into it. There can be four productive subthreads of a discussion all peacefully chatting, and then a political subthread gets started and from then on, all the comments end up on the political tangent subthread and discussion of the original non-tangential topic is forgotten.
This is a structural pattern to how posts happen—the kind of thing you can identify with linear regression on time series of post metadata. Politics is almost literally a "black hole": it has greater "mass" than other subjects, sucking commenters away from their original intent toward political subthreads. If you model subthreads as having that sort of "gravitational force", you can predict the ones with the highest "pull" will be political in nature. (Well, that, or a subthread indulging HN's fun habit of primary sources showing up to make comments on stories about them. But those subthreads likely have very low "controversy" metrics [in the Reddit ranking algorithm sense], which can be used to filter those out from political consideration.)