As I've been studying how the space-time scheduler behaves - in particular what it's phase-snapping after constraint-solving does - it seems that it's at its best when its choices are random with the probability being determined by headroom availability.
This is done.