TurnTrout comments on Alignment allows “nonrobust” decision-influences and doesn’t require robust grading

[TurnTrout]

See the value-child speculative story for detail there. I have specific example structures in mind but don’t yet know how to compactly communicate them in an intro.