Regression modeling is a cornerstone of statistical analyses, when researchers aim to study the impact of some observed covariates on a given outcome. Often, researchers are interested in making some conjectures about a target parameter by means of a statistical test. In the first part of the talk, we will focus on (generalized) linear models; we will show how a recent procedure, known as Flipscores, can relax the heavy assumptions of the full parametric tests, with a huge improvement over the well-known sandwich-type approach. In the second part, we will see how this can be embedded in the cases of multiple responses when aiming to control the family-wise error rate. Finally, in the last part we will move towards quantile regression. In particular, we will focus on simultaneous inference over multiple quantiles. Despite the nonparametric nature of the problem, we will see how the default Wald-type test can be unreliable. Hence, working upon the Rankscores approach, we will show how a closed-testing procedure can be employed to outperform traditional Bonferroni-type corrections. Simulation studies and examples will be presented through the presentation.