This should solve #179 (closed)
For some integrators, like the Velocity Verlet, it remains unclear how a completely correct version (with the expected integration order) should be realized. The general problem is that that the transformations from World Frame to Body Frame and vice versa depend on the rotation, which however is to be updated during integration. This dependency opens up many possibilities regarding the "correct" order.
Fixing this issue completely will probably require some substantially more work and investigation. Also literature seems to be scarce on that problem and if mentioned at all, a rather complicated leap-frog / predictor-corrector like scheme is proposed. See e.g. the notes in Wachs, 2019, doi: 10.1007/s00707-019-02389-9, around Eq. 105.