It seems like you think I'm trying to attack you, which is not my intention. I'm also not sure how I was "changing what you said" per se.
As to the "highly questionable assertion", perhaps you're meaning the comment about being able to build a machine which can pass a static THD or IMD test easily but not be able to reproduce music well. We already see cases of existing audio chains that have "resonance peaks" which resist changing frequencies once driven, so suppose you built a machine which produced sets (say some limited number) of pure tones via pure tone generators, and only locked on to the tones over a period of time (enough to easily be perceptible to humans). It would likely pass a static THD or IMD test just fine. I would also agree it's pretty ridiculous, but in my meager defense of the example it's not too unlike some early audio hardware experiments that existed.
A reasonable argument is that my example is specifically tuned to defeat the test... but the point in the example is to find out if traditional THD/IMD measurement numbers can track correctness in full audio reproduction.
Perhaps your argument to me (as I am indeed the one less experienced in audio science) is that there is some proof that I am unaware of that IMD and THD measurements somehow cover all those cases and the machine I'm postulating wouldn't pass the test due to some failure of imagination on my part, to which I am happy to learn about, as I've not done any kind of thorough search of the literature.
It will probably take me a few days/a week-ish of evenings, but arguably it's on me to try to build at least an electronic simulation of my fake proposed machine, and I might do it for fun.