I missed that list in the docs - thanks for pointing it out. That, along with the others' suggestions to improve the reporting on failed tests, gives me someplace meaningful to look for a cause even if I'm not able to interact directly with the problem machine.