Service Tests are our way of testing the interfaces of a service by poking at it as a client of that service. In an ideal world, different implementations of the service would be able to seamlessly reuse that service's service tests, i.e., they could serve as a form of conformance test. However, currently the decision of which service implementation is under test is baked into the service test itself, as it's the service test that spins up the service.

It would be good to determine an architecture that gracefully separates these concerns so that one could change the service implementation under test by e.g. setting a GN variable. This would also have the extremely nice side benefit that it would *enforce* that service tests don't reach into a service implementation's internals (since they would have no access to those internals).