Scanning, it looks reasonable, but I know zip about svg. Do we have an svg expert we can call on for a review?

Do you want to add a test-only dependency on something like https://en.wikipedia.org/wiki/RDFLib to check that the metadata is correct? Otherwise it feels a bit hard to actually check that the implementation is correct, as I don't really want to read the RDF/Dublin Core specs either :)

PS: Looks like rdflib (which I just found from a quick goole search) may be one of the few python libs that is both older than matplotlib (it dates back to 2002) and still actively developed :) (I guess the winner there maybe PIL(low), which dates back to 1995, dunno if there's anything else?)

I tried this, but rdflib won't even read the RDF from Inkscape.

Let's perhaps consider another approach to testing to move this forward: if you write xpath queries (https://docs.python.org/3.8/library/xml.etree.elementtree.html#elementtree-xpath) testing that you can retrieve the correct metadata entries, I guess that I could be willing to believe you that the xpath queries themselves correspond to what the specs say.