Про сенсоры сони, интересное. Частично объясняет чому никон из них больше выжимает, а сапог и на старых выезжает:

Should have been superb
This lens design has the potential to be among the extremely few designs shorter than 85mm capable of the superb optical performance, i.e. corner to corner excellent detail on a plane at infinity at f/2.8, which photography of starry skies often demands, especially for astro-landscape, as opposed to pure astro work where no land is in the image and longer lenses are employed, on average, and everything is done on an equatorial mount/clock drive. For that kind of work, f/4 is more commonly used and works out pretty well with some of the best lenses, though sometimes 5.6 is used as well, at least thats my sense. For astro landscape, even larger apertures like 2.0 and bigger are often used, though not without further sacrifice. Equatorial mounts are of great usefulness for astro-landscape as well, if you want to get anywhere near to the best possible results. Apart from lens design, the assembly of the lens plays a huge part in the resulting quality, but so does the cameras sensor to lens mount alignment. Unfortunately the Sony Alpha sensors are relatively poorly aligned, on average, if my sample of four A7r IIs is any indication, not to mention the inherently inferior method of Sonys sensor alignment, using shims which come in fixed 50 micron increments. An error in a teardown article suggested it was 5 microns - it isnt. It looks as though Sony cannot, given this method, individually adjust cameras the way Canon can, by using their spring-tensioned screw design, on the assembly line. The four A7r IIs that I and a colleague very carefully measured were out of alignment by between 30 and almost 50 microns. The one EOS, a 5D II, which I measured, was off by about 6 microns. Under 10 to 12 microns is pretty tolerable for this insanely demanding optical problem, but 6 is highly desirable. After months of hassle I was able to adjust my own A7r II to within ~10 microns by shimming the lens mount ring. Long story. But on that account, and on account of the performance of several other lenses which I own, and my own system for measuring camera sensor to mount alignment, I was able to confidently conclude that the focus tilt I saw in three copies of the Sigma 40 ART was a defect of the lens in by far the greatest part, or rather of the calibration of the lens which is likely done with shims under its rear mounting ring, and not a defect of the Sonys sensor alignment, except to a small degree. The best of three samples, which were otherwise superb, focussed at about 80 feet in the top right, while focussing at about 130 feet at the lower middle edge, which is a 25 micron error and more than can be overcome by best possible compromise focussing at f/2.8, if stars are to appear to be in focus. F/4 might work with that copy. Given tight calibration of this issue, these lenses would work at f/2.0 and even wide open, though certainly not as well and with too much falloff wide open. The worst of the three had perhaps twice that tilt error, roughly. I checked to see whether the sheer weight of this lens on the Sonys lens mount ring was changing the focus and it was not. I could see zero difference in focus by pressing upward on the front end of the lens. Also if it had been, it would have been causing the opposite tilt problem. Some people allege that copies of Sigma lenses purchased in Japan are great, and that they suspect that cherry picking is being done in that way. My limited sample of having tested only four copies of two types, total, is far too small to tell, but what Ive seen at least doesnt contradict that possibility, as the sample variation problem in each of those four lenses has been larger than the fundamental optics problems which Sigma has successfully overcome, and to an increasing degree over time. My sense of the one Sigma 50 ART which I examined was that had it been a good copy, it would have been usable for stars at f/4. Pretty good, but not good enough, still. For a million other uses, no problem, but for stars, most of the time, forget it. 2.8 or bust. When the Sigma 40s were refocussed, each part of each copy produced a superb image, including with point light sources (bright city lights at over 10 miles or bright stars), not having more than a very little bit of the usual batwings, spaceships, angels and other shapes caused by the various aberrations which lenses exhibit to varying degree around their periphery and which are routinely referred to as coma in reviews. This is extraordinary and a sign of a superb design. I could see none of the focus shift when stopping down from 1.4 to 2.0 or 2.8 which others have alleged, so it was at least very small, and the curvature of field was fairly small though certainly visible, and seemingly smaller than the tilted focus problem. Why bother to make the lens capable of superb detail at 1.4 or 2.0 or 2.8 yet not capable of being in focus on a plane at 2.8? If all you want is in-focus eyes in a portrait, thats easy for lens designers, though not for AF system designers. Perhaps its just too hard. Given that Sigmas entire push into very fast lenses was in part motivated by inspiration from within the company to make lenses for this very purpose, its puzzling to me why such a relatively easy-to-fix kind of flaw would be allowed to spoil an otherwise stunning performance. It could be that looking at more copies would lead to finding one which happens to be right, but for now Im going to put on hold my search for a 35 or 40mm lens to use for stitching starry skies, as no clear candidates seem to exist. This after my colleague and I have examined, either in person or by studying reviews and samples, essentially every such lens on the market. The vaunted EOS 35 f/1.4L II required nine copies for a colleague and I to find one marginally usable copy, mostly on account of strong field curvature. The three copies that I examined would require stopping down to f/6.8 to become usable, making them a total waste of time, despite this lens being widely regarded as greatly superior to the 35 ART, for example. And so on. I hope Sigma ups its QC game further because I really like what theyre trying to do. The same goes for Sony and the Alpha full-frame bodies. They really need to switch to Canons system, even if it means making the body a few mm thicker, where each body is quickly and easily calibrated on the assembly line to within something like 6 microns using three spring-tensioned screws and a scanning laser micrometer setup to give instant, real-time feedback on the sensor position. 100 microns is the thickness of a typical sheet of laser paper! The requirements are ridiculously tight for this kind of work, much more so than what is implied by the night sky work that we see online, usually at very low resolutions, almost all of which is a mess when you get to see it at 100%, from both lens and camera optical limitations and noise issues as well. If you want to see what can be achieved, look at what the pure astrophotography photographers manage. Its pretty often seemingly flawless, though it usually involves a huge and complex effort to make a great portrait of a given nebula, for example.

https://www.bhphotovideo.com/c/product/1436292-REG/sigma_40mm_f_1_4_dg_hsm.html