It doesn't have to be the original. In most of his writing, the refs have been content-addressed (hash). It's fascinating to see him backpedal to location-based addressing (URL) here, given the extent to which he's railed against that concept in the past. I would have imagined at least a hashsum retained as a field in the xanadoc file.
I've read a fair amount of his stuff but hadn't seen him talk about hash-based addressing, do you know where you've seen it? In Literary Machines he wrote about a hierarchical addressing scheme called Tumblers:
I was under the impression that the point of transclusion is to ensure no actual copying occurs -- that every quotation or paraphrasing of a prior work is explicitly connected to to the prior work, ensuring the "truth" of the quotation and the connection to its own foundations.
If this doesn't point to a single version of something, if the provenance of the source isn't provable, I'm not sure I understand the point of the strong source linking. Unless I'm misunderstanding your comment.
Also, I think that HTTP got it right when encoding is returned with the document (rather than specified in the link), but if you've already made the mistake of dealing with byte offsets you're tied to a particular representation.
