You need 8 bits to do 48dB, so if you want your music to have 96dB of dynamic range and you want your editor to be able to slide things around 48dB without perceptible loss of quality, you need to record at 24 bits of depth.

If you record at 16 bits of depth you are pretty much automatically forcing your engineer to choose between quality and range when they are normalizing the recording.

Listening to music with 96 dB of dynamic range is extreme though, right?

You seem quite determined to leave inadequate margin for error. :(

The most important rule of audio recording is "always be ready to roll tape". What the engineer does is of little consequence compared to the talent. Don't ever fuck up a keeper take.

Recording with some extra bits means that even if your gain staging was bad, you just get some extra uncorrelated noise, not correlated quantization distortion.

As for whether this stuff is perceptible, there are two issues at play. First, differences which are not perceptible during production can result in a perceptible difference in the final product. This is easiest to understand with a visual analogy: if you have inadequate resolution of an image during processing, the end result may be slightly defocused and smeared.

Second, as I described elsethread, quantization distortion is a particularly pernicious form of degradation, poisonous in small amounts. https://www.youtube.com/watch?v=8KkS4NSv-0M It is hard to listen for it during production, so we defend against it by giving ourselves lots and lots of headroom.

I get it. I’m asking what dynamic range you need for the completed production, and how much additional headroom you need when recording. I don’t think you normally would need 96 dB for the completed production, because that would be very impractical to listen to.

16 bit certainly suffices as a delivery format. Many listening environments provide far less dynamic range (e.g. cars) and environments where the lowest level details aren't buried in the ambient noise floor are going to be few and far between. Even though it is possible to encode information below the noise floor in practice information that far down isn't going to be perceptible. (A practical definition of "perceptible" might be "able to be distinguished in an ABX test by a trained listener".)

As for how much safety margin is "enough" during initial capture... We're looking at a stark choice between either 2 or 3 bytes per sample. 2 bytes seems to be borderline, while 3 bytes seems to be plenty. It would be hard to quantify a safety margin more precisely than that.

Say that I need a 300x400 image for a webpage. Does it make sense to set my camera to take 300x400 pictures? What happens when we crop and postprocess?

If we want the best quality end product, free from low-res artifacting, we need to preserve an excess of resolution all the way through production until the final export. Exactly how headroom much is enough? Enough that artifacts arising from insufficient resolution are not perceivable in the end product.

Yes, I understand the reasons to have enough dynamic range and enough headroom. My question is about what specifically is enough.

I think 16-bit quantization noise level is usually lower than the noise of all the analog hardware before the ADC, but 24-bit quantization noise seems to be lower than even the internal noise of audio ADCs so there seems to be no reason to go higher.