I had not considered the hysteresis, that's a good point. I suppose the best material to use would be that used in electric motor rotors and transformer cores? I have not measured the readings from the new sensor, but the joystick output seems to give me repeatable centering. Assuming the rudder is deflected on average each way about 50% of the time, the centering should work over time. However this this does not hold true, especially when flying helicopters, so this new approach will drift over time. I think I will recover the sensor from the older setup and rebuild it with a new flux concentrator, this time in the shape of a ring running around the whole assembly, and still fixed to the magnets.