OK - although my cge-pro is not available I was able to do some handcontroller tests with an Edge8 on evolution.
This is an alt-az mount but I told it to be in eq mode for the test.
I aimed it at ra 0, dec 0 and read the ra/dec values after 3 slews. They were consistently off by -1.4 or -1.5s in RA and +5" in dec. The sign of the error was also constant. This corresponds to a fixed offset of the landing from the target - according to its own encoders. Multipying the ra error in seconds by 15 to get arc seconds, this is a miss of 23" - just based on the encoder landing error alone. Nothing about the goto accuracy or the model.
I then proceeded to follow the routine of the pseudocode I provided here long ago - to compensate for the previous error and keep trying.
I therefor aimed first for +1.4s, -5" instead of 0, 0 - and immediately landed at -0.6s, 0. This means the dec error was nulled completely and the ra error was more than cut in half - so the total miss is now 9" instead of being stuck at 23".
Repeated application of the routine kept the miss at 0 in dec, while the RA error varied a bit from a minimum of 1.5" to a maximum of 16.5" as it re-aimed at the target. So although it did not converge perfectly and exactly - it immediately removed the main bias - and after that was hitting on either side of the target with most hits within 10" of the target and as low as only 1.5".
This does not test the total sgp system including plate solve - but it does test the ability to offset the target goto location to remove most, and nearly all, of the error that causes the centering routine to get stuck at some location off the target. And this is with a fairly basic alt-az mount - as opposed to the much more accurate cge-pro.
Anyone can do this with a mount indoors as long as you can do goto's to a particular ra/dec value and read the result.
Frank