For scandium through zinc, the #4s# orbitals fill AFTER the #3d# orbitals, AND the #4s# electrons are lost before the #3d# electrons (last in, first out). See here for an explanation that does not depend on "half-filled subshells" for stability.
See how the #3d# orbitals are lower in energy than the #4s# for the first-row transition metals here (Appendix B.9):
All the Aufbau Principle predicts is that electron orbitals are filled from lower energy to higher energy... whatever order that may entail.
The #4s# orbitals are higher in energy for these transition metals, so naturally they tend to fill LAST (especially for late transition metals, where #V_(3d) "<<" V_(4s)#), not first.
Hence, it makes sense that the #4s# electron is predicted to be ionized first (barring #3d-4s# interactions), at least for the first ionization.