How to pick the right regression model without fooling yourself?

When fitting regression models, testing many candidate variables inflates false discovery risk—a problem classical statistics solves with multiplicity correction. This work builds that logic into Bayesian model priors directly, proposing a prior distribution that naturally penalizes adding too many variables. The approach outperforms several recommended "objective" priors at avoiding spurious model inclusion, suggesting practitioners need stronger sparsity constraints than typically recommended.