The use of the baryonic acoustic oscillations (BAO) datasets offers a unique opportunity to connect the early universe and the late one. In this proceeding, we discuss recent results that used a marginalised likelihood to remove the $H_0-r_d$ degeneracy and then tested it on different dark energy (DE) models. It was found that this approach which does not rely on calibration on $r_d$ or $H_0$, allows us to obtain results, comparable to the ones calculated with standard likelihoods. Here we emphasize on the major differences that we observed for the two different BAO datasets that we employed -- a transversal one, containing only angular BAO measurements, and a mixed one, containing both angular and radial BAO measurements. We see that the two datasets have different statistical preferences for DE models and also different preference for the curvature of the universe.