But is it reasonable to believe the trade-off is the smearing and ugliness at a part of the right side of the image?
I don't know... I can't explain the situation really. Would it be reasonable to believe a grinding machine at Sigma making an aspherical piece of glass is in need of some adjustment? But surely Sigma should have noticed? Then what?
In any case the lens is a no-go for some of my needs, no matter how nice it is.
Here is a close-up of patch No. 6:
Don't mean to say that you've made a mistake, just want to say that this is _reeeallly_ hard without an almost lab-grade setup.
Jim Kasson has some great resources on how to easier test lenses with a single large target that you can frame within the center and sides at appropriate distance here. With a single target for a 35mm f/2 lens, 42mpix you'd need to be about 40m away, but that eliminates any and all alignment errors by allowing the camera to rotate edge-to-edge while keeping target within DoF.
I did some quick napkin math to check which tolerances you need to align to, (disclaimer: I'm prone to make math mistakes ):
If you take half the sensor pitch of a 24mpix full-frame camera as acceptable circle of confusion for depth of field calculations (~3 microns), you're left with a depth of field of 6mm at 0.85m (and oops, that's at f/2.8, not f/2).
I calculated the maximum angle deviation at 0.85m to be 0.81 degrees to stay within the DoF of the lens. That means your measurements of patches 1 to hotshoe and 7 to hotshoe need to be accurate to under 7mm over a distance of ~1m. That's less than five 1 cent coins (don't know about about krona, but 4 quarters for the rest of us ) stacked, or about 1/3 the width of the A7 hotshoe.
And edit: I assume you shot wide-open, sorry if I missed that the lens was stopped down for the tests - that would make it easier to calibrate.
):