Poisson local statistics of eigenvalues is widely accepted as a necessary signature of Anderson localization, but so far has been rigorously established only for random systems. We will argue that this paradigm is wrong, and the reality is a lot more complex and interesting, by presenting both rigorous results for the Harper and Maryland models and numerics for other quasiperiodic and similar models with localization. We will also discuss a conjecture on what the distribution is in the general ergodic situation.