Inaudibility is fake news. There are many undergoing researches about ultrasounds and their impact on humans. Your claims only rely on audiometry using air conduction. Science is not settled about an audible limit. Science is not settled about anything in fact not even what is a women or a men. It uses the 20-20khz conventionally because this is what is commonly observed in practice.
We have means to perceive ultrasounds. Some do exist, like fairly solid bone conduction(the paper I read had an apparatus stuck on the jaw in the mouth). Or blasting ultrasounds right on the skin(I seem to remember a need for ultrasound above 100dB SPL, not sure what the exact magnitude was?). We can also by any mean necessary, make our eyes shake in a way we become aware of. And probably other parts of the body can react to high enough dozes of ultrasounds in some ways we might notice while it happens, but those are TBD or at least I didn't not read about them since the embassy guys getting sick and people suspecting ultrasonic attacks(I'd watch a spy movie on this).
And of course, some young kids and a very rare amount of adults, simply can hear higher than 20kHz tones(until they stop being able to).
So it's reasonable to just agree that noticing ultrasounds can happen. It doesn't really matter to audiophiles how exactly we perceive them. It's interesting but not that important for our hobby. What matters is that we do, and to determine under which conditions and if those conditions apply to typical music listening.
So I propose this as the question to answer. The hypothesis being that ultrasounds in music do have a consciously noticed impact on our experience of the music. I say conscious because of all the audiophiles who keep claiming to clearly hear the difference. Unconscious impact, long term impact, those may or may not be significant, I have no idea how to test it and would not care for random "I know what I heard" empty anecdotes on the topic. So I suggest to leave it out of the picture unless someone actually has some fact based information on the topic for a change.
Sighted impressions aren't even up to grade school experiments loosely supervised by a teacher. So please let's not discuss that while pretending that it's anything remotely conclusive. It's not, a sighted impression could be right or wrong. As there is no way to tell without performing an actual listening test, it's worthless. Worst than worthless, it wastes everybody's time in this sub-forum.
Do you agree that if I think I notice something different when my hires file has been stripped of ultrasounds, I can test that proposition and demonstrate it to be true(or not)?
No matter how my body makes me aware of it, if I'm aware of it, then I can put that perception to the test with a blind test. Do we agree on that? If so, we have an hypothesis. We can test it, hermaggerd we're almost starting to do science!
We can consider many tests, answering slightly different questions. For example we could just do a blind test with hires file and 16/44 file of the album. But if we pass that, the actual cause could be many things. It could be that the playback lagged a little on the heavier file. It could be that the source resampled really poorly one of the files. It could be that they come from different masters. It could be that the DAC has a fairly "colored" filter for 44.1kHz files(trebles roll off, or lot of aliasing). And some other reasons(format decoding issues, bit depth difference, buffer issues with the bigger file...). So this wouldn't really test our hypothesis. Ironically that's what almost everybody doing sighted "tests" are using, as if a sighted impression wasn't enough of a problem already.
A better test would be a blind test between an original hires file in .wav(to cross all the t's) and that file converted to 44 or 48kHz, then converted back to the original sample rate. That way everything is identical(file size, format, source and DAC's way of handling it) except for one file that has lost its ultrasonic content.
If we can tell those apart consistently in a blind test then we can confidently say that the conversion caused an audible difference. It doesn't 100% prove that what we're noticing is the ultrasonic content specifically, as it could still be IMD from the ultrasounds coming back in the audible range, or maybe the converter and settings caused something other than just the removal of ultrasonic range(maybe a really crappy band limiting, maybe just a change in gain). But we can try to investigate with other tests or measurements and probably rule out those options. So it's not a dead end. At the very least, passing such a test would confirm that we're hearing a difference when the music has ultrasonic content. That would already be a bog step up in term of demonstrated fact. That in turn could make a good argument about the impact of using hires files when listening to music, at least on our specific playback system with our ears. Which isn't much for the world, but is usually everything that really counts to us Headfiers.
Have you considered trying such tests instead of jumping head first into the marvelously self gratifying adventures of only seeking confirmation on the web and almost always striking gold no matter how bad the idea actually is? I've done perhaps 30 blind tests(pretty conservative guesstimate), and about half of them were attempts at perceiving changes with hires files, because of course I'd like to hear it if there is something worth hearing that improves realism, soundstage and whatever random stuff we read audiophiles claiming to have heard improve with hires. Each test being usually made of 20 trials where I'd try to identify the files for as long as I deemed necessary.
I have yet to pass even one such test. And I haven't met anybody IRL who would accept to try and could pass. Doesn't mean nobody can, but the group claiming it can strangely also happens to be the same group that usually refuses to properly test anything. Saying I'm skeptical is putting it nicely.
On the other hand, I've passed several blind tests when using a redbook file against an hires file, usually I could determine rapidly with some measurements that the cause was one of the reasons I mentioned above. Trying to eliminate that cause also eliminated my ability to pass the test. Back to square one.
And of course when casually listening to music and switching between hires file and the one at 44.1kHz while knowing at all time what it is I'm listening to because I'm seeing the files, I, like most people, have felt all sorts of things. But when I feel like I can tell the difference between 64kbps mp3 and 320 in casual listening, I can then pass a blind test to confirm it. Same with certain levels of background noise, same when trying to tell Celine Dion apart from Barry white by ear. And of course I can! Because I'm actually perceiving differences even without looking at them. With high sample rate files, when I stop looking I also stop being able to tell the difference. But a test between 44.1kHz and a 8kHz conversion and back, is something I can pass. What do you think is the reason for that? Maybe because this time, the frequency range removed is something I could actually notice without looking at the filename? I think that's a pretty fair assumption.
Most people I know who honestly tried .wav against the downsampled then upsampled back version, couldn't pass. It's been done for decades now and failing this is IMO a very strong argument against our initial hypothesis. So until demonstrated otherwise, that's my position on the subject.