Is this something a limited or partially mocked integration test could catch instead?

I don't mind whether Quibble uses it or not (either is fine I suppose), but it tends to scale better and benefit other run-times' test coverage if we "test" it directly.

Indeed, sounds better to have some test written for DatabaseDomain :)

A unit tests for DatabaseDomain might be too low-level, a slightly higher-level of integration (whilst still not requiring an actual database) might be better.

But, it depends on whether the original issue would've been prevented. Perhaps a test case for DatabaseDomain would suffice, in which case that should suffice for this task as well.