Running a preliminary proof of concept on a dataset of about 500,000 entities, reveals that <~0.02 % are actually being excluded by this condition (471524 out of 471591 are not instances of scholarly articles or astronomical objects). Should we consider checking for instances of transitive subclasses as well (direct subclasses of Q13442814 or Q6999 as well as subclasses of those subclasses). The following SPARQL Query reveals that there are 1577 of those: https://w.wiki/F2fy.

I ran some presto queries on the full dump in the analytics cluster, here are the actual entity counts as of 2025-07-28:

Notebooks are at: https://gitlab.wikimedia.org/repos/wmde/wikibase-reuse/wikidata-dump-subsetting/-/tree/main?ref_type=heads