True. And in the early (internal) alpha days, false positives for wake word detection were quite high. So we felt the pain from eating our own dogwood; even though we spent huge amount of science / engineering resources on reducing those false positives, we knew it would never get to zero while maintaining acceptable true positive rates.