Removing the saturation assumption in a posteriori error analysis