Right. There is some degree of transformation which is normal or necessary. For example, if one of {video, audio} is 24fps and the other is 23.976fps, the video will have its frame rate adjusted to match the audio. For basic video/audio mismatches you may need to trim audio, insert null audio, or add a mkv container delay.

These things are normally detailed in release notes. People write pretty detailed notes laying out exactly what they did.

The muxer can apply a timestamp offset to a track, so while the audio stream remains untouched it starts at some offset relative to the start of the video stream.

Agree once you're compensating for drift by scaling the audio track you've pushed yourself out of remux territory, but again these aren't hard and fast rules (evidenced by the disagreement over terminology).