The ONLY thing that is proven is that it is valid at MORE THAN (not exactly) twice the highest required upper frequency, BUT ONLY if the waveform is defined to plus and minus infinity (that is, doubly infinite). And the Cheung–Marks theorem (that is, proved) seems to indicate that being non doubly infinite and the presence of noise upset even this.
In other words, the Nyquist theorem sounds nice, and appears good to use for an argument for going for minimal specs, but it does NOT apply to ANY real-world sounds under ANY real-world conditions, so to continue to use it for ANY justification for one’s DAW sample rate choices is dubious.
I’m not saying I have answers here, but I refuse to accept INCORRECTLY applied theorems as EVIDENCE for ANYTHING.
Now, can we get down to real-world discussions so that we can actually understand what we are really dealing with in DAW work?