Bandit Multiclass PAC Learning: Corrected Lower Bounds, Exact Families, and a Confidence Direct-Sum Phenomenon
The open question in the negative under its uniform-constant reading is answered, and the failure is intrinsic: for an explicit affine multiplexer class, a confidence direct-sum regime where a multiplicative $\log(1/\delta)$ cost is information-theoretically necessary, followed by a rank-saturation phase transition.