Example: I create some localized spike in volume automation to emphasize a few notes. I hit space, I hear the temporary rise and fall in volume.
I route the resulting sound to an audio track and record it (so that I can File > Export > Audio Mixdown afterwards). When I play this audio data, the quick rise in volume is a little less pronounced, it feels different.
Does that make sense? Is there something related to the number of parameter updates per second which makes the recording a little different from what you hear when you press space?
This is a subtle but disturbing problem: all the small performance details you put in are then modified by the workflow…