set in model lattice gradient = 1/calibration * K use this model to fit errors and see if smaller K are needed.