new GAN amplifier

As for subjective perceptions of audio, to paraphrase the local classic, "there are many, me including, who like tubes best".

Horses for courses. I don't care too much for tubes but I'm a fanatic for pristine treble, and that includes a ruler flat response out to >100kHz.   

Measurements of the purify look good but keep in mind those IMD measurements conducted at relatively low power and with a LPF in line. 

Btw, the new Halcro Eclipse is now the new "benchmark" for low distortion.   

>100kHz response?  Some species of bats indeed can detect 200kHz but they can't hear the human range sounds.
Then there is the question of transducers, I think I came across once a lab specimen that goes high ultrasonic. Then there is the problem of mics, top studio Telefunken U47 (9k $) goes only up to 20kHz.  So, other than Batman music recorded with lab mics and played by lab transducers it might be a tall horse to climb.

>100kHz response?  Some species of bats indeed can detect 200kHz but they can't hear the human range sounds.
Then there is the question of transducers, I think I came across once a lab specimen that goes high ultrasonic. Then there is the problem of mics, top studio Telefunken U47 (9k $) goes only up to 20kHz.  So, other than Batman music recorded with lab mics and played by lab transducers it might be a tall horse to climb.

Of course we can't hear 100kHz. The desirability of a flat response out to >100kHz has to do with minimizing the side effects from the cascading of low pass filters starting all the way from the recording process itself - as you correctly identify includes tranducers too.  

Always need to be mindful of the time domain. 

All digital audio is built around the sampling theorem, not the other way round.  There are places in the signal chain where high sampling rates make sense. The final reproduction stage is not one of them. Why, and how it can be harmful is well explained here:
https://people.xiph.org/~xiphmont/demo/neil-young.html

All digital audio is built around the sampling theorem, not the other way round

I wasn't singling out digital audio, but it's worth mentioning the use of oversampling to remove the necessity for analog brick wall filters in AD-DA conversions. Class D amplifiers don't require brick wall filters but they do still add multi-pole low pass filters to an already band filtered signal path for playback of a wide range of recorded music that's already passed through a diverse range of low pass filtering even before it leaves the mixing desk for digital recording and mastering.  

For the record (no pun intended), my highest fidelity recordings are all high res 24/192 digital. To my ears they're the closest thing to real live music.