Good to hear about your experience with this, Chuck. From similar things I've read, even setting up a good null can leave significant signal, though that may be 40 dB or more below the program level and would normally be masked and probably inaudible. I take the short and long of it to mean the value of this tool may be primarily subjective analysis of the quality of that difference along with determining the relative amount of difference between different tests.
I am somewhat suprised by your sine wave tests, as those should be identical, repeatable signals, while I'm not suprised by the pink noise tests, since that signal is essentially random energy, and therefore different for each individual test, even though it is filtered to a specific frequency profile to make it 'pink'.