We study frequentist properties of Bayesian and L0 model selection, with a focus on (potentially non-linear) high-dimensional regression. We propose a construction to study how posterior probabilities and normalized L0 ...