Post-production almost universally puts the audio mix at the end, after picture is locked. The ordering is a consequence of how sound relates to timing rather than a matter of preference.
Every cut invalidates the work below it
A mix is a set of decisions tied to specific moments: a level change here, a transition there, a reverb tail that resolves at a particular point.
Move a cut by a few frames and every one of those decisions shifts out of alignment. Automation drifts, fades land in the wrong place and effects lose their sync.
Because the damage scales with how much mixing has been done, mixing early costs more the better the mix was.
Music has to be cut to a known duration
Scoring or tracking music requires a fixed length, since a piece is arranged so its structure resolves where the picture does.
A sequence that grows or shrinks after the music is placed forces the track to be re-edited, and edited music betrays itself quickly when the structure no longer resolves.
Composers therefore work to locked picture, and any change afterwards is a formal request with cost attached rather than an adjustment.
Dialogue decisions constrain picture decisions
Editors work with rough dialogue throughout, because whether a line is intelligible determines whether a shot can be used at all.
What is deferred is treatment rather than presence: noise reduction, matching between takes, and level balance all wait until the choices are final.
This is why a rough cut sounds uneven and the same cut sounds effortless later. The content did not change; the processing was applied once the target stopped moving.
The mix is judged against the whole programme
Loudness and dynamics are relative. A scene mixed in isolation sits at a level that only makes sense against the scenes on either side of it.
Delivery standards for loudness apply across the full programme, so a mix balanced piece by piece will not meet them without a final pass anyway.
Mixing whole is therefore not just cheaper but more accurate, because the reference the mixer needs is the finished sequence.
Where the rule bends
Music-led work inverts the order, because the track exists first and the picture is cut to it. Here the audio is the fixed element and the edit adapts.
Sound design that drives performance, such as an effect an actor reacts to, has to exist during shooting rather than after it.
Both are cases where sound is the structural element rather than a layer applied to structure, which is exactly the condition that reverses the sequence.