We do this currently so we can test internal implementations. However, this is really brittle - it breaks when someone adds a new import aliased a certain way. We also recently had a case where a unit test was added modeled after this one incorrectly, which ended up hiding the fact that its implementation wasn't quite working right.

We should figure out a way to either expose the relevant bits that are internal or make sure we can test the relevant bits through the public API.