When tracking from the analog world, all of your gain staging decisions should be about getting as little noise as possible without unwanted distortion. It doesn't matter what the meters read as long as you don't clip your converters. If it's a little "too loud" or "too quiet", you have plenty of perfectly clean, flat, noiseless ways to adjust that once it's recorded.
In mixing the only meter that really matters is the one on whatever track you plan to render your mix from. Most times that'll just be the Master. As long as that doesn't hit 0, you can feel safe that your final fixed point render will be clean. You really kind of do want to "use all the bits" here though, especially if it's going to be mastered afterwards. This advice is contrary to common wisdom today, but I think if you're going to mess with the dynamic range later, you should provide as much as possible to work with. Push it up to almost 0. If your mastering engineer needs to turn it down, he will. Course, if you're rendering a mix to be mastered in a separate session, you could render to floating point, and it won't matter at all.
So... It's still nice to stay close to but not over 0 at your mix bus. It is true that you can usually adjust the mix bus fader to "fix" the level once the mix is good, but it turns out that if you set your track levels to average around -18, in most mixes, you probably won't have to adjust that bus fader as much. But it's still not really about the meters.
Here's what I do in a basic rock mix:
Rough in a mix of the drums and then push that up until the mix bus is hitting around 0 on really loud peaks. If that puts any of the track faders too far off from unity, I adjust the level at some point before the fader and reset it.
Then I bring the other instruments up until the mix sounds about right. If any of those faders end up way off from 0, I again go adjust something upstream. That's really just personal preference. On most mixers both analog and digital, you have more fine control right around unity. There's no sonic benefit to this in the box at all.
By now it might be hitting above 0 in places where the whole band hits hard together. That usually tells me that I need to put some work into controlling the dynamic range of individual tracks (especially drums, probably bass, definitely vocals), so I go back and do that. EQ, compression, saturation, automation, whatever it takes. We are sort of trying to get it to stop clipping without fucking up the mix too much, but at this point we're still not super worried about that meter. No matter what we do, there still maybe like one or two samples along the line where everything accidentally pushes the same direction at exactly the same time and it goes over.
Those things I consider to be part of mastering. If I'm doing an album, I'll render to 32bit FP at this point. If it's just a one-off, I just slap my mastering chain on, adjust as necessary, and render the final fixed point file.
So what happens in mastering? By this point, you really should be pretty close to what you want for dynamic range, but again you maybe have one or two big spikes, and I for one like just a little bit of "glue", a bit of "squshiness". So I add a slow RMS compressor doing almost nothing. It very subtly evens out the overall dynamic so that if the kick hits one beat alone and then the next with the bass, it's just that little bit less louder than it "should" be. When the guitar kicks into the loud chorus, the whole band is just a little closer to that quiet verse.
Then you start looking at your meters. Most of the time, we want our loudest peak to be about as loud as it can be - as close to 0 as we feel comfortable. I prefer -0.6 just out of habit. This is not so much to "use all the bits" as to "meet standards". Like, everybody else is that loud, and it's what your listeners tend to expect. But we also want our average levels to be in the ballpark of other similar tracks. Where that sits completely does depend on the individual mix, the genre, and expectations of the artist. So look at your meters and for the moment don't think about the actual value of either peak or RMS, but look at the difference between them.
If that "dynamic range" is right for your genre (and it sounds good!), then you just adjust the mix bus fader (or output of one of the plugs on the mix bus) until the loudest leak is wherever you decided you want it to be and go. If it's too big, but it sounds good, it's probably because of that one or two little spikes that we let slide before. You'll probably even see on playback that your leaks will sit at a certain level until one little place where it suddenly jumps way up. Here's where we try to knock those down. I personally won't use anything that calls itself a limiter for this. This is almost always just a sample of two, and anything with a time constant - attack, release, RMS, even oversampling - will be incapable of promising a consistent output level. I use a "curvy" saturation that works on a sample to sample basis and just won't let anything get louder than the ceiling I set. Adjust that ceiling to the loudest peak you want, and then adjust the level going into it until both your RMS and peaks are where you want them. If you can't do that without distorting the mix too much, then you need to go back a few steps and control the dynamics of individual tracks more tightly and/or adjust the mix bus compressor to squish things together a little better.
OTOH - if your dynamic range is too small, maybe the RMS compressor is doing to much, or else you'll have to go back to your individual tracks.
But you wanted specific numbers. Understand the DR numbers depend on everything in the mix, and RMS is measured over time, so how much time you're sampling makes a real difference. The last full rock album I mastered had integrated (full song) RMS ranging from -15 up to -9. The shortish song that was just balls to the walls came out overall loudest, but the quietest one is has long stretches of minimal, softer material and sudden bursts of things that are as loud or louder. Even if you're mixing square waves, the one that plays more often is going to have a higher integrated RMS.
So that probably raises about more questions, but you know, this topic has been beaten to death and back a number of times all over the Internet.