This is a nice hack! I once thought of a more systematic and cleaner solution would be to produce the scattering matrix whose modes are identical aside of spin degeneracy. This wouldn't rely on using the slower and less stable Green's functions. Forcing extra requirements on the modes in the lead is really quite easy and can be preferable since you can probably save computational costs also for the modes. For example in  we've done it for time-reversal symmetry, and in  Michael used it to calculate Andreev conductance in a neater way than what is suggested by the Kwant tutorial.
I started implementing such leads once, but didn't have enough time to come to a reasonable point.
: http://arxiv.org/src/1408.1563v2/anc/trijunction.py (see class TRIInfiniteSystem) : http://wwwhome.lorentz.leidenuniv.nl/~wimmer/tcscxviii/phlead.py (was used for example in http://nbviewer.ipython.org/url/wwwhome.lorentz.leidenuniv.nl/~wimmer/tcscxv...)
On Fri, Jul 17, 2015 at 7:44 PM, Joseph Weston email@example.com wrote:
This time *with* the recipe attached