It has been recognised for 10 years that the observed evolution of galaxy star formation rates at a given stellar mass does not agree with predictions from models and simulations. I will examine the various physical processes which play a role in these predictions, and argue that current observations imply that the efficiency of feedback must evolve with cosmic time. I will demonstrate different options for how this can be achieved empirically, and conclude by examining the behaviour of the Eagle simulations to see if there is a scenario which is preferred by current hydrodynamical simulations.