All good points, worth some discussion. Unfortunately, any meaningful conversation in this environment will get crapped on by those 'newcomers':
2) "In the frequency domain the triangle of music is supposed to include at least potentially stuff over 48kHz" - I am not arguing that the '[music] triangle" does not go above 48kHz... But why would we care, with the hearing limit like this (the black, triangled curve) [from Meridien paper]:
View attachment 133170
3) "...and even here there is the possibility of encoding through what Jim lesurf has called the leaky decimation filter" - there is always a possibility for encoding. The issue there - and it too has been pointed out a lot - your baseband (especially if it's a 16-bit one) simply runs out information capacity to store any this-particular-tune related information - first the (<22kHz) baseband itself, then the 'first', and then this 'second' unfold. No matter what codec and how 'leaky' your filters, you only carry information (a) in your filter shape (a constant) convoluted with (b) some information ultimately taken from the initially-baseband samples. So, yes, the information budget of the 16bit/22kHz or 24bit/22kHz baseband needs to be looked at (and that's what the 'triangle' does, but unfortunately not quantitatively, plus see #2.)
.