So, another possible error source has been eliminated. That's progress. A few more tests changing some of the variables might be a good next step. See if you can make predictions about the things actual performance behavior. Ultimately, I hope they'll stick it to a small satelite and see if they can measurably influence its orbit. If that works, all bets are off. We still might not get how the thing actually works, but we'd know that it works... Which would still surprise me, but I think the concept deserves the opportunity after a few more vacuum tests.