The import of the filtered dumps on math26 (step 4 of T439313) fails for wikidatawiki and commonswiki. Their math is mostly in Wikibase content (items on Wikidata, MediaInfo on Commons), not in <math> tags in wikitext. importDump stops at the first page with such a content model, because Wikibase is not installed on the math server.
Proposal:
- Skip the pages with Wikibase content models of these two wikis at the import, so the wikitext pages are still imported.
- Get the formulae from the Wikibase content via SPARQL: the Wikidata Query Service for statements with the math datatype, and the Commons Query Service for MediaInfo.
- Add the formulae from SPARQL to the dataset of all formulae, with their source (item or file) instead of a page revision.