batsbrew
so, if they're not identical,
it's lossy, eh?
I think it might be 1 bit lossy. The difference goes either all the way up, zero, or all the way down after normalization. To me this means that something is either 1, 0 or -1. Those are the variance. So, using Flac 24 bit seems to lose exactly 1 bit of resolution. For a lot of digital audio, the most significant bit is omitted (assumed to be 1). This is the reason in floating point you have -0 and +0 values because neither can be exactly zero if there is always a presumed 1 at the most significant bit. Normally this gets you one more bit of resolution.
I don't know if the 24 bit waves count as floats. I think they don't. I'm not positive though.
In any case, the Flac seems to lose exactly 1 bit of resolution as compared to the waves.
In this 1 bit, I am able to pick out the song well enough to recognize it. (It sounds awful but you can hear it.)