You are probably right, eg Losslesscut in Linux does exactly that: no transcoding, no overlays nor any editing other than cutting and moving around sequences from the same video. Editing and exporting such videos is blazingly fast.

Adding a bit technical context (for readers) to your comment on inserting ads at "random places": many codecs use the concept of an initial frame [1] which would place technical restrictions on where the ad could be inserted. If Youtube believes that ad timing is too important, they might not accept technical restrictions on the timing.

[1] https://en.wikipedia.org/wiki/Video_compression_picture_type...