The SDAM spec defines two kinds of tests:

• unit tests that ensure the SDAM state machine implementation is correct
• integration tests that ensure server responses are properly propagated through the state machine, and the effects on the connection pool are observable

The server currently only runs the former, which leaves the RSM susceptible to bugs when dealing with actual responses from other servers. As an example, the RSM makes no distinction between errors that occur before or after the handshake as required by the server-montioring specification, yet it passes all the unittests because this isn't a problem with the state machine.

We should also run the SDAM integration tests to ensure bugs like this aren't missed. This may involve writing some driver-like wrapper of the server's networking internals in order to implement the test runner.