Epistemic uncertainty estimation methods are fundamentally incomplete
We argue that current epistemic uncertainty methods are fundamentally incomplete: unaccounted bias contaminates aleatoric estimates, and existing methods capture only partial variance contributions.