i think these models can spot contradictions in some of your philosophical articles as well that a human reader would agree look like real contradictions. not necessarily decisive ones, but ones worth addressing. wouldn’t that be step 3 working in the ideas domain too?