ORIGINAL: Marah Mag
As I understand it, it's not that you're not supposed to not go above -12, it's that if you are recording at 24 bits, then the noise floor is so low that you can still get acceptable quality even if you're only peaking at -12.
But that doesn't mean that you should avoid using those 12 dbs.
What this means in practice is that you can be less concerned about setting a precise level and worrying that an extra hard guitar or drum hit is going to go over 0db by some small amount because you can just shift your entire recording range down and still get a good sound.
In other words, there's more margin for error in anticipating dynamic range. This makes it easier to just get on with it, and less need to play "chicken" with 0db. But there's no reason not to come as close as you can, as per CJ.
At least that's how I understand it.
That's pretty close to my view of the world... we may both be wrong<G>... but there is one little bit in there I'd like to highlight.
Just because you are recording with 24 bits of resolution does not mean you have oodles of dynamic range to play with. First, almost every vendor of low to mid-range converters uses the converter chip S/N ratio specification, not the device specification. Now to be fair, that's partly to make their product look good, partly because everyone else does it, and partly because there is no really well defined specification for S/N ratio as it crosses over from A into D (or vica-versa!)
The other thing to consider is that nothing out there has a 144 dB S/N ratio, so even if your converter were capable of perfect operation (and it isn't), your preamplifiers, monitor amplifier, etc aren't anywhere near that good.
As an experiment I built a line level buffer, no gain, really a pretty pointless device, just to see how quiet I could make it. I was pretty surprised - the test set I was using at the time, an AP Portable One Plus, was struggling to measure the noise floor with the input terminated. So I don't really know what I ended up with, but it was scary quiet. To get there I took great care with the circuit layout, I used all metal resistors (both of them??), and no capacitors. I also overbuilt the power supply such that noise and ripple were unmeasurable.
It didn't really prove anything except that it was possible to build a circuit that was really quiet. So I added 70 dB of gain, no adjustment, just enough gain to bring a ribbon microphone up to line level, and I added a passive ladder attenuator on the output. All of the sudden I had no problem measuring S/N ratio<G>!
What's all this have to do with numbers of bits? Just that you can't assume that all your bits are pristine, there is a least significant bit, and it's probably much higher than you'd like. So you do want to keep your levels reasonable.
At the same time you don't want to run too hot because there are all sorts of ways that a signal can distort, and as a rule most listeners are more aware of distortion than they are of noise, at least according to studies I've read. The really odd part is that they are less bothered by certain types of distortion than others, or noise. If the distortion product is harmonically related to the signal it often sounds pleasing. Go figure.
There is no agreement on a reference level for digital audio recording or processing. Some folks work as low as -18dBFS, others as high as -6dBFS, and some fall in the middle.
Me? I love tranisients, and I figure they are probably going to get mucked up somewhere along the way anyway, but I like to give them a fighting chance, so I use -18dBFS or maybe -16dBFS.
YMMV. and really, if you are busy counting bits you probably aren't paying enough attention to your mix<G>!