What should a statistical mechanics satisfy to reflect nature?

There is no compelling reason imposing that the methods of statistical mechanics should be restricted to the dynamical systems which follow the usual Boltzmann–Gibbs prescriptions. More specifically, ubiquitous natural and artificial systems exhibit generic stationary states which are not ergodic nor close to it, in any geometrically simple subset of the a priori allowed phase space, in any (even extended) trivial sense. A vast class of such systems appears, nevertheless, to be tractable within thermostatistical methods completely analogous to the usual ones. The question posed in the title arises then naturally. Some answer to this complex question is advanced in the present review of nonextensive statistical mechanics and its recent connections.