If you want to target -12, go for it. Personally, I think there is no need to leave headroom in the expectation that sound pressure levels will quadruple.
I think there is some fallacy in the math but I haven't studied this enough. I've got piles of evidence that sets recorded at a peak of mostly -12 or lower can still have multiple points that approach or hit 0 db.
I think it all depends on what and where you record.
A loud amplified band from mid to back of a large room will be a fairly even wash and the peak level read will be pretty consistent. Unamplified music is far more diverse. Recording close adds another highly variable dynamic. The mics are also a factor...
All that said I usually try to record at levels as high as I can without hitting 0. I can live with an occasional transient redlining as that's not really perceptible, especially with a little Izotope clip repair if it actually clipped. I'd rather save myself the effort but have found I'm more willing to spend an extra few minutes of that editing than to record at what I consider excessively low levels (also as an editing tactic it makes sense and a better listen to duck those couple of major transients to achieve a higher overall output than to set final levels to preserve those couple of spikes). Often it's applause or a drum shot that drives the level. I don't like to set the levels to not clip the applause then have all the musical content 10dB lower than it needs to be (assuming the crowd doesn't loudly applaud during the songs).
Consistent brickwalling is of course a major problem to be studiously avoided.
On the M-10 I'm usually occasionally (but not regularly) blinking the green light but still often have transient peaks that may blink the red. OTOH I've recorded some rock shows even up close where I've had the green light essentially continuously lit but never a red. If the band plays loud nearly all the way it is very easy to narrow the margin.
I'm not a fan of hovering over the meters or continually adjusting during the performance. That does impact the accuracy of the document and the real dynamic range. I'm willing to smooth obvious things out in post but don't necessarily think smoothing it all out on the fly is the best approach. It is detrimental to my personal experience of the show and can be distracting to the crowd, crew and especially the band if you're near or on stage (so a big no-no there).
As a long-time audio professional I'm appalled at the number of people on this forum who advocate peak recorded levels of -12 dB or even lower. Peaks on digital recordings should go as high as possible without quite reaching 0 dB. If you're not sure you can judge the maximum sound levels that will occur, you might choose to aim initially for peaks at (say) -6 dB on the theory that louder moments might occur after that. But if so, the point is still to get peak levels of at least -6 dB and hopefully somewhat higher.
^ My goal.
It's an art, not a science. If we could reliably predict the outcome we wouldn't need to discuss the strategies.
The amount of safety margin will vary a lot based on factors above, and others. I think the best course is not to continually overestimate the required margin, but others may feel more conservative.
In practice I've found the noise floor that matters is the mics. You'll have more noise from the mics and ambient atmosphere at the show than you'll ever introduce from a decent recorder/pre-amp (as long as they're not driven beyond their capacity). The inherent noise in really quiet material is from the room and the mics. Recorded at -2 or -20 really quiet material will still be a little noisy when the editing is done. Loud music will never be noisy unless subpar equipment is involved.
As to setting levels during soundcheck that is never an accurate gauge. I learned quickly on the R-44 that whatever I thought a safe margin from a soundcheck needed to be reduced one full clickstop (and often two) for a performance with audience (so 6 to 12 db lower as it measures).