I'm more familiar with a IIIF canvas than a TEI surface, but the TEI guidelines suggest that they match. But let's explore a use case I understand from IIIF and ask how TEI treats it to make sure, though:
A particular leaf--pretend it's a letter written by Abraham Lincoln--has been separated into fragments by an autograph dealer, and that those fragments have been imaged separately from each other. In IIIF, we'd create a single canvas to represent one page of the letter, then use one image annotation for the text fragment and another image annotation for the signature fragment. Both images would annotate the canvas, which is itself a somewhat hypothetical representation of the original page. Transcripts would be textual annotations on the canvas as well, rather than using either image as a target. (The verso page could have its own canvas with image annotations as well.)
Does surface work the same way?