Layer audio files instead of joining them. Each track gets a level, a balance, and a start time, and the export is what they sum to.
Levels add up
Two tracks of the same material at the same level sum 6 dB louder than either one alone. Two unrelated tracks land somewhere between 3 and 6 dB above the louder of the two, depending on how much their peaks coincide.
Nothing is scaled back automatically to avoid it. The mix is written as summed, and the peak of the result is reported afterwards along with the number of samples that went over full scale. Solving for a gain that fits would mean measuring every source before writing anything, which doubles the decoding for a number you can read off the result and dial in once.
If the mix is hot, drop the master level by the amount the peak reading names, or send the file to the limiter.
Balance, not pan
Centre passes a track through at exactly the level it went in. Moving it to one side takes it out of the other channel and leaves the near channel alone.
That is a balance control rather than a constant power pan, and the difference matters here: a pan law would drop a centred track by 3 dB, so layering one file under another would change the level of the file you did not touch.
A mono source is fed to both output channels first, so balance moves it across the field. A stereo source keeps its own two channels and balance acts on them as a left/right trim.
Start times
Start at delays a track from the beginning of the mix. It is one direction only: a track cannot start before zero, because the mix has no earlier point to put it at.
To move a track earlier relative to another, raise the other track’s start time instead. To line two takes up sample for sample, measure the offset with the time aligner first.
What the output inherits
Sample rate and channel count are the highest of the set. Resampling a 22 kHz file up to 48 kHz adds nothing and loses nothing; folding a 48 kHz file down to 22 kHz would throw away a track that was fine, so the mix never does it.
The length is the track that ends last, start offset included. Silence before the first track and after the last is written out as part of the file.
Track limit
8 tracks. Each one holds a decoder and a block of decoded audio while the mix runs, so memory stays flat in the length of the files and only grows with the count. Past 8 the useful next step is usually to export a submix and bring it back as a single track.