First, I wanted to improve collimation. In order to remove as many sources of errors as possible, I started just with eyepieces.
First using the 40mm eyepiece and then the 4mm, I collimated to scope as good as possible out of focus. Seeing wasn't great - when I tried to use in-focus collimation, I could hardly see any Airy disk.
2. Camera directly on scope
I removed the focal reducer, spacers, filter wheel, OAG. I couldn't get the scope to focus anymore, but that was OK. I used CCDInspector's out-of-focus collimation. I used ζ Cygni - it was very high up and has no other bright stars around it. Used 30 sec exposures to minimize atmospheric and other distortions. And always waited at least 3 images with similar errors.
I started with a 16" collimation error! I turned the camera around to make sure that the error that I'm seeing is purely the scope and not the camera. Moved it slowly down to 4". Then it changed direction. And then it got funny: for the other 2 collimation screws, the star moved away from the screw when I tightened it. But for the third, it moved towards the screw when I tightened it. That certainly explains the difficulties that I had to improve collimation earlier. After I figured this out, I could bring collimation below 1" (then it changed direction from picture to picture).
Also, at this point the image had no tilt at all!
Now I could have slowly added equipment to it.
... but instead, I'd rather focus on Hyperstar imaging. I know now that the camera is completely orthogonal, so I can use it for collimation of the Hyperstar lens. So, reconfigure the entire scope (cables, secondary mirror, hyperstar lens).
Collimating the Hyperstar lens turns out to be a major pain in the ...
I have to use multi-star collimation - because of the wide field of the Hyperstar lens, I couldn't find a brigh enough start that didn't have other stars in its vicinity. Furthermore, to defocus a start enough for out-of focus collimation with CCDInspector is outside of the range of my focuser!
Because of stray light and the moon light, I have to use a shield on the scope to get correct collimation. But that means that I have to take the damn shield off whenever I need to adjust the Hyperstar lens, then put it back on.
This is the initial state:
Tilt and quite a large collimation error.
This is the final result:
Almost no collimation: good!
Almost no image tile: good!
But look at that curvature!!! I have absolutely no idea what to do about that...