Actually I think I was mistaken. One of the tests was sort-of implicitly making the assumption that this works for subclasses in single table inheritance when of course it doesn’t.

I believe this can be closed.