Calibrating SKA comes down to removing the effects of the many bright sources in the field. This can only be done in a closed-loop system, using a model of the observed brightness distribution. This implies an extension of the well-known 'self-calibration' technique to solve for 'image-plane' effects, i.e. instrumental effects that depend on the position in the field. In other words, we have to start solving for the detailed shape of the voltage beams of the individual stations. A distinction must be made between the treatment of the main lobe, and of the sidelobes of these beams. This paper investigates some of the boundary conditions of this approach, and formulates requirements for the accuracy and stability of SKA hardware.