Hello Peter,
I missed the bit about the flattener. Ok then, aligning without the Gemini is probably more difficult than it's worth since I presume you only have adapters and spacers to fit the Gemini+camera to the flattener. This also means my side note bit about field curvature is not of any use, but it does bring up the question, are the star shapes and sizes in the corners of the sensor for your best aligned config so far (the 270deg one?) satisfactory? What is making the stars bigger off-axis, coma/astig or still round just bigger?
A flat frame would be misleading IMO for collimation and tilt adjustments. For example the peak of my flats are off-center by a few millimeters, I think because of a non-perfectly orthogonal fixed primary mirror, but the system still collimates perfectly and the star images to the corners of the 16803 are very similar. The causes of vignetting and optical correction are just different.
I understand your trepidation of being out all night playing with little fiddly bits, that's why a reliable measurement system that you can trust is golden here.
Since you are comfortable using CCDI, and before you try another software, validate CCDI for your situation. That means: introduce known amounts of tilt with shims (forget the rotation bit for now just use one position angle) and see what it reports. Is the output consistent with your known input?
Use two values of tilt and look at the difference, that way you can ignore the unknown started value. Ideally you would be increasing the tilt to go against the starting value to not make it super extreme. This could take a little bit of guess work at first to go in the right direction.
If CCDIs numbers do not match up with your input then proceed with great caution and maybe you will have to use something like Platesolve2, which I can help you with but that's another learning curve.
I wouldn't make more adapters yet, unless you are certain that what you have mechanically just won't do, but it might not be the limiting factor here
Best,
EB