Model-Value Inconsistency as a Signal for Epistemic Uncertainty