His thought - at least in my adaption - was: The performance of a lense depends on brightness of illumination of the subject, independent of post-processing. Say the target is illuminated to EV 6 or 10. I can compensate for the lower illumination by increasing the exposure time, achieving the same middle gray - but that will not help the lens to see lines. Or is it me who got it wrong?
The level of illumination
makes no difference to the lens, it's purpose is just a matter of refracting photons. Their quantity doesn't really matter, their wavelength does.
What his thoughts might have been, is that ideally one would do the deconvolution sharpening while in linear gamma space. When we do our Capture Sharpening however, we usually are already in an approx. gamma 1/2.2 space which stretches shadows, and compresses highlights. Therefore it does matter where on the gamma curve we measure contrast, and thus exposure
level (not illumination level) does matter.
I've taken two precautions to work around that. One is the grayscale stepwedge which, when properly exposed, spans a large range from reflected black to reflected white. It would be quite easy to already see if the exposure level was correct, otherwise we'd lose shadow or highlight detail. When properly exposed, like I assume one's regular images are, then the white patch should land around [235,235,235], the medium gray around [128,128,128], and black somewhere around [10,10,10], but these are not absolute goals because different Raw converters produce different tonalities. The point is that there is a good spread of brightness levels with visible details from dark to light. The second precaution I took was to evaluate that entire range (whatever it happens to be) as it also forms at the sub-pixel edge transitions, and fit a curve to it. I do not measure contrast, but a contrast transition curve shape
And as that curve shape shows, the transition is usually very well balanced and symmetrical along the brightness range, and almost perfectly follows a symmetrical differential Gaussian curve. The only deviation that can pop-up is some lens glare which raises the response of the shadows and lowers contrast. It is taken into account when fitting the curve, but it won't dominate the outcome.