I find it strangely correlated with the way the camera is set up. If it uses the middle finger then the camera might not see correctly the cube's face. You can see it using it at the last resort.
But I don't see why this would matter in the simulation phase.