Imagine you've tracked guitar, bass, vocals, and so on, and have multiple takes, all at slightly (or maybe radically) different recorded levels. Perhaps the vocalist positioned themselves further away from the mic on one take, for example, or the bass player just happened to play a lighter, quieter take. How and when would you advise dealing with this?
Via SOS web site
SOS contributor Mike Senior replies: This is a very common issue, but on commercial productions it's often dealt with by gofers behind the scenes, so there tends to be very little said about it by higher-profile engineers.
As with most things recording-related, it's preferable to try to minimise these kinds of variations between takes at source, but on real-world sessions, that kind of thing will be the least of your concerns. Capturing a great performance must be the most important thing, and that usually involves faffing about as little as possible with technical matters.
To my mind, the best time to deal with inconsistencies between the takes is during the comping process. If nothing else, it's easier to judge which bits of each take are the keepers if all the potential candidates are heard at comparable levels. Compression is usually too blunt a tool for this job, so I like to apply manual gain changes to different takes — or, indeed, individual snippets of takes — while putting together the final comp. My preferred gain-adjustment method is to use offline changes applied to the audio regions themselves, something that's easy to do in the DAW systems I'm most familiar with (Cubase, Logic and Reaper), and certainly possible in almost all others. That way each gain change sticks with its respective take while I'm still deciding which one I like most. 
Offline or region-specific gain and EQ adjustments are very handy for reducing inconsistencies between takes during the comping process.
Bear in mind, too, that level isn't the only thing that can be a bit different between takes, because the frequency balance can also shift quite dramatically — for example, if the vocalist moves much closer to a directional microphone — triggering heavier proximity-effect bass boost. Again, I prefer to deal with this using region-specific offline EQ processing: a few decibels of high/low shelving one way or the other is usually all it needs.