[7777777phil]

The classifiers I used are definitely conflating technical criticism with genuine negativity, and that's a real limitation. When I say "technical critique reads differently than personal attacks," I probably should have been clearer that the models aren't making that distinction well.