Everything is benchmaxxed. Whack-a-mole training is at least as representative of what is getting added to models as more general training advances.