It might be a neat experiment to use ai to produce canonicalized paraphrasings of HN arguments so they could be compared directly and compress well.