I've checked the latest all.ttl. The only axiom causing problems is the one from PCO we identified above. Is it possible to drop PCO from the load? More drastically we could just drop fbcv/dpo. This would be fine for now.
BTW - One possible strategy for making file sizes manageable: The major cause of inconsistent individuals is likely to be type statements on anatomical individuals. It would be easy to specify a sparql query to select only these individuals + type axioms, use limit to split them into chunks and then test each chunk for consistency.
I suggest we do this as a parallel pipeline.
From slack: