> Anything above 44100hz is just headroom for mixing.

No it's not, you're confusing it with how many bits are used.

Once you are above the Nyquist frequency you can perfectly represent the sample. Any more frequency does absolutely nothing.

But the number of bits of resolution that's what actually determines how perfectly the digital wave form matches the original.

ah, thank you! yes. every bit adds about 6db of headroom right?

Yes, that's correct.

Wikipedia claims that more than 21 bits is pointless for actual playback since there are no circuits that can be that accurate (quiet).

But more bit depth can help with mixing headroom - but you don't need a lot. "24 bits is enough for anyone." © :)

Yes, for playback. For some people there's a lot of confusion about bit depth since it can represent a range of integers (like how the data is stored in a wav file or what dacs process) or a range of floats.

Internally, audio apps typically represent signals as sequences of 32bit floats between -1 and +1.

Headroom has nothing to do with sampling frequency (bit depth do). But it has been explained here already.

While you're absolutely right that listening to stuff at 96khz is pointless, it's worth noting that these high sampling frequencies are useful for other audio applications that are not on the consumer side (like recording, mixing, instruments sampling, and more).