I need some serious help

[Alibert]

When using a limited number of points for calibrating radius and rod length, I have found that there are multiple solutions that will fit those limited points. I have done this using a few combinatorial optimization techniques (gradient descent, downhill simplex and genetic algorithms). This explains why you can find 'crazy' combinations that should work, but do not in the spots you didn't include in the optimization. The printed size of the object directly relates to the diagonal rod length as well. If it is not the actual (real) rod length, the size of the printed object will scale to larger/smaller.

I have found that the best approach is to follow these steps in this specific order:

1) measure the diagonal rods carefully (I have a 12" digital caliper for this purpose) and use this value in the firmware
2) adjust the endstops so that all three towers just touch (as close as possible to the tower) at a given max z height
3) adjust radius value only to get a as flat as possible z-plane (using opposite points of the towers as well. You should be within 0.3mm over all of the build plate at this point. If not, redo steps 1-2.
4) print out the calibration object according to the link I gave above. Measure the correct angles between x,y and z tower and enter these in the firmware.
5) redo step 3, you only need very small adaptions
6) print out the calibration object again, but now only compensate the diagonal rod sizes for each tower using the calculation explained in the link. Leve the tower angles from step 4.

With Marlin firmware and following the above I am within 0.1 mm z-height error over nearly my whole build plate (280 mm diameter) and less than 0.05mm in X and Y direction. I am in the process of converting to Repertier 0.92 and will be able to tell you where to plug in the adjustments soon.