Like Paul Erdős, I resorted to experimentation. Very much unlike Paul Erdős the computer program I wrote contained a one-off error in its PRNG that coincidentally confirmed the wrong result. I then spent a day or so on Usenet insisting stubbornly on the wrong solution until some very kind person on sci.crypt made a complete truth table as a proof. That exhaustive proof immediately convinced me of my idiocy.
Lesson: If a bunch of smart and educated people tell you that you're wrong, then you're probably wrong.