we're manually clearing scope in the middleware but if there's an exception in the job this might not run. We should just use push_scope instead to ensure cleanup.

Try this out with a proper test case and fix it.