This brings up a newb question for me actually. Say you were recording 4-channel for a known 50-50 matrix in post. Should you record both stereo pairs at their optimal levels (close to clipping without going over), then when mixing turn them down so when added together they don't clip? Or is it better to leave headroom during aquisition on both stereo pairs so when you mix in post they add together and don't clip. My guess is for noise floor reasons, you'd always want to do the former, but honestly I'm not sure, curious though.
If you are recording in 24-bit, whether you are matrixing or not, I would highly recommend leaving more headroom, there simply is no need to crank the gain past -12 dB, the extra dynamic range you get is theoretically not as much as you think it is, somebody put some numbers on this board explaining why, but I forget them and where they are. If you are recording in 16-bit, it is up to you and your preamp whether the extra noise introduced is worth it to get your peaks up to -3 / -5 dB.
When you know that you will be mixing two different stereo signals in the post, I would say that there is no need to push the gain (and risk clipping and extra noise) when you know that you would have to attenuate in the post just so you can mix the two signals later. However, I will try to get optimal levels (based on my comment in my first paragraph) on the aud feed if it is onstage (which it usually is for me), just so I can capture the wonderful onstage dynamics as purely as possible, and I'll leave the sbd feed relatively safe and low. If it's a PA, I still try to get somewhat optimal levels but I'm much more reluctant to push them as high as an onstage recording.
These are just my experiences/preferences, I hope someone who has a lot more experience and knowledge will offer their point of view.
Isn't there a balance here though? I mean, yeah, if you are maxing out the gain on your preamp, most likely you are introducing a bunch of noise, so of course you wouldn't want to do that. But on the other hand, if you apply way too little gain, the recorded music's volume will be down closer to the noise floor, so when you normalize in post, you'll be bringing up both the music and the noise floor together. So, it seems to me, there is a sweet spot.
My take on this is that you should apply gain as far up until the point of clipping (but obviously NOT clipping) as you can as long as you aren't maxing out your preamp to the point it starts getting noisy. For concerts, this is usually pretty easy to do without going too far on the gain. This results in the recorded music being as far above the noise floor as possible in your recording. Then, if you are mixing two sources, you'll have to reduce the levels such that the sum of the two sources don't add up to a clip. But even though you have to reduce the volume during the mixdown, at least you are preserving the spread between the recorded music and the noise floor. Again, only if you don't have to crank the gain past your pre's ability to add lots of noise.
Doesn't that make sense? Or am I missing something here?