That said, they might have 2000 fails and 2 working tests for all we know. But again, I don't think Google would be deceiving us that way. Not really in their engineering blood.
It would be interesting to know how many real world tests they have done. One could safely assume that there would have been many thousands of tests in laboratory conditions, but we won't know how well this really works in real world conditions until we see some robust statistics about its performance in that uncontrolled environment.