Wow, fascinating - thank you!
I've done this myself in two ways:
- Commercial 24/96 file resampled and dithered by me in Audacity, then null-tested in Audacity; and
- Commercial 24/96 file and commercial CD rip of known-identical mastering (e.g. 2014/15 Led Zeppelin reissues), null-tested in Audacity.
My results typically have the difference RMS low, around -80dB, but with higher peaks, and with plenty of the difference in the 20Hz-20kHz region.
Based on what you've shown here, however, I would hypothesize that I could be trying this the wrong way, by directly comparing the 96k and 44.1k versions, where it is impossible to know exactly what the exact, proper alignment of the waveforms is in order to do the null test properly or fully. By converting a file from 96k to 44.1k and back again, one creates two 96k versions that can then be "perfectly" compared in a null test.
I'm going to try that now. Thanks!
EDIT: Holy moly - I did it with Audacity and got the same results as you - difference file shows information there, but inaudible because ultrasonic. Very cool. I don't buy much high-res material, but still, this is going to save me some money and storage space moving forward!