This now produces graphs like the above (pdf), which is promising. For some reason it's not interleaving the solvers nicely which I'm slightly baffled by (note each individual tower priority differs from each other by ~10). I'll look into this more next week.
Attempt to reprioritise each group.