This has been lurking -- we're not actually supposed to be calling getAverageThrustThrust when we're past the end of the thrustcurve (there's a test for whether the motor is active before calling it), but another bug in the parallel staging code exposed it.