We propose that the flexibility offered by modern event-generator tuning tools allows for more than just obtaining “best fits” to a collection of data. In particular, we argue that the universality of the underlying physics model can be tested by performing several, mutually independent, optimizations of the generator parameters in different physical regions. For regions in which these optimizations return similar and self-consistent parameter values, the model can be considered universal. Deviations from this behavior can be associated with a breakdown of the modeling, with the nature of the deviations giving clues as to the nature of the breakdown. We apply this procedure to study the energy scaling of a class of minimum-bias models based on multiple parton interactions (MPI) and p⊥-ordered showers, implemented in the PYTHIA 6.4 generator. We find that a parameter controlling the strength of color reconnections in the final state is the most important source of non-universality in this model.