Hi Lars,
Stars 2 and 5 are most likely outliers. Mask them.
I would assume you may have performed a meridian flip at star 26 onwards.
Whatever tool chain you are using is not conveying to TPoint that the observation
is from the flipped position.
For example, take star 26 itself. The scope RA/Dec position is 22:31:50.8 -57:02:32.1.
The angular separation on this star is 6641 arc seconds.
Would I be right in assuming this is where you first performed a flip?
If so, the scope RA/Dec position should be "flipped" to 10:31:50 -122:57:28.
Likewise the scope position for star 27 should be "flipped" to 08:44:53 -140:35:05.
I only converted 26 and 27 by hand to verify this but when you mask 28 to 45,
the results are excellent, with an RMS of 25 arc seconds using only 7 terms,
namely IH, ID, CH, ME, MA, HDCH1, HHCH3.
HTML Code:
* fit
coeff change value sigma
1 IH -5.899 -6439.18 14.304
2 ID -64.536 -21.79 10.861
3 CH +2.153 +3066.14 13.963
4 ME +3.271 -22.63 21.493
5 MA +14.153 -48.96 12.070
6 HDCH1 +53.635 +91.58 23.079
7 HHCH3 -2.473 -27.28 10.134
Sky RMS = 25.26
Popn SD = 29.77
The mount has virtually no discernible NP, so I did not include it. You can see in
the fit above that the CH value of 3066 arc seconds is the most significant, so
it may simply be that the OTA is not "square" to the mount.
I would not be surprised that if the remaining samples, that is 28 to 45, had
their scope positions "flipped", that the results may continue to stay good.
So in a nutshell, you should first do the following -
Take care to avoid outliers.
Review how your tool chain conveys "flipped" mount positions to TPoint and
correct it. That appears to be the major source of your problems.
As you can appreciate, GEM's have the ability to view the same place in the
sky from two distinct mechanical orientations. The mount co-ordinates sent
to TPoint need to convey when the OTA is observing from the flipped position.
One could post convert the positions by hand, which is tedious, or by way
of a script, or better still, determine a tool chain flow that might allow the
flipped co-ordinates to be embedded in the data.