We have recently tried 3.5.0 beta3 and got some problems with delta calibration/probing. After a downgrade to 3.4.5 the problems were resolved without changing any parameters.
We calibrate our delta machine with a detachable probe connected to a CAN board.
M558 P8 C"^121.io0.in" H5 F300 T18000
When I start delta calibration, the head homes and subsequently crashes the probe/head into the bed. If I compensate the homed height to make the machine 15mm "shorter", the head stops before crashing into the bed but the initial probe of the sequence will have an upward direction. If I stop the calibration and try again, behaviour remains the same. However if i let the sequence finish (even though it does not touch the bed) and subsequently do a new calibration, the initial probe point direction is set to downwards again, however the machine does not reach the bed because of the decreased homed height. If I then adjust the homed height again to make the machine “longer” I can eventually get it to calibrate without errors. Problem is that after a power cycle all behaviour seems reset and the head/probe crash into the bed again.
Hope my explanation makes sense but can't see how to describe this any better. As this is a production machine it will be complicated for me to do further investigation, but if needed I'll see what i can do...