Hi Connor,
I'm afraid it's not moving in right direction. It's partially caused by returning success and NULL window in some of the tests. While it may be interesting to test we end testing corner behavior only. Some paths in native implementation accepts NULL HWND while other not - it may lead to wrong conclusions.
What do you think about something like in attached diff (generated on top of your patch)? Please note that I didn't test it well. It would be good to understand what SID_AccessibilityUnk is for.
Thanks, Piotr